| 1 | //! Account-owned remote control for the active TUI session. |
| 2 | //! |
| 3 | //! This is deliberately a typed relay, not a remote shell. The control plane |
| 4 | //! may send prompts, approval decisions, and run-control requests for the exact |
| 5 | //! enrolled target. Provider credentials, paths, environment variables, and |
| 6 | //! arbitrary command strings never cross this boundary. |
| 7 | |
| 8 | use std::{ |
| 9 | collections::{BTreeMap, HashMap, HashSet}, |
| 10 | fs::File, |
| 11 | path::{Path, PathBuf}, |
| 12 | time::{Duration, Instant, SystemTime, UNIX_EPOCH}, |
| 13 | }; |
| 14 | |
| 15 | use reqwest::Url; |
| 16 | use reqwest::{Client, Method, StatusCode}; |
| 17 | use serde::{Deserialize, Serialize}; |
| 18 | use serde_json::{Value, json}; |
| 19 | use sha2::{Digest, Sha256}; |
| 20 | use tokio::sync::mpsc; |
| 21 | |
| 22 | use codewhale_models::{ContentBlock, Message}; |
| 23 | |
| 24 | use crate::{ |
| 25 | core::events::{Event as EngineEvent, TurnOutcomeStatus}, |
| 26 | runtime_chat_relay::{ |
| 27 | RuntimeChatControlScope, RuntimeChatProjection, RuntimeChatPrompt, RuntimeChatRelayHost, |
| 28 | }, |
| 29 | }; |
| 30 | |
| 31 | const PRODUCTION_CONTROL_PLANE: &str = "https://api.codewhale.net/"; |
| 32 | const ENROLLMENT_SECRET_SLOT: &str = "cwc-remote-control-enrollment-v1"; |
| 33 | /// Machine-stable device identity. It outlives individual enrollments so the |
| 34 | /// control plane can fold every folder enrolled from this terminal into one |
| 35 | /// computer instead of one row per `/rc`. |
| 36 | const DEVICE_IDENTITY_SECRET_SLOT: &str = "cwc-remote-control-device-v1"; |
| 37 | /// The only web origin whose session links the terminal will surface or open. |
| 38 | const APP_ORIGIN_HOST: &str = "app.codewhale.net"; |
| 39 | const HEARTBEAT_INTERVAL: Duration = Duration::from_secs(25); |
| 40 | const SYNC_INTERVAL: Duration = Duration::from_millis(1_200); |
| 41 | const MAX_RESPONSE_BYTES: usize = 1024 * 1024; |
| 42 | const MAX_RUNS: usize = 64; |
| 43 | const MAX_COMMANDS: usize = 128; |
| 44 | const JS_MAX_SAFE_INTEGER: u64 = 9_007_199_254_740_991; |
| 45 | const MAX_RUNTIME_ENVELOPE_BYTES: usize = 128 * 1024; |
| 46 | const SNAPSHOT_ENVELOPE_BYTE_BUDGET: usize = 120 * 1024; |
| 47 | const MAX_SNAPSHOT_MESSAGES: usize = 64; |
| 48 | const MAX_SNAPSHOT_MESSAGE_CHARS: usize = 128 * 1024; |
| 49 | const MIN_TRUNCATED_MESSAGE_CHARS: usize = 32; |
| 50 | const MAX_REMOTE_ERROR_MESSAGE_BYTES: usize = 4 * 1024; |
| 51 | const RUNTIME_UPLOAD_RETRY_INTERVAL: Duration = Duration::from_millis(250); |
| 52 | const RUNTIME_UPLOAD_MAX_BACKOFF: Duration = Duration::from_secs(5); |
| 53 | const RUNTIME_CHAT_RELAY_PROTOCOL: &str = "codewhale.runtime-chat-relay.v1"; |
| 54 | const RUNTIME_CHAT_CATALOG_TIMESTAMP: &str = "1970-01-01T00:00:00Z"; |
| 55 | const CAPABILITIES: &[&str] = &["evidence-ledger", "fim", "git", "shell"]; |
| 56 | /// How long an aborted or failed relay keeps local input locked. Matches the |
| 57 | /// server-side runner lease expiry with margin; local input never returns |
| 58 | /// while the server could still consider a remote owner live. |
| 59 | const OWNERSHIP_LOCK_AFTER_FAILURE: Duration = Duration::from_secs(95); |
| 60 | /// Ceiling for draining unacknowledged runtime events during `/rc stop`. |
| 61 | /// Deliberately below `OWNERSHIP_LOCK_AFTER_FAILURE` so a failed drain still |
| 62 | /// resolves into the ownership-locked path before the lease question is moot. |
| 63 | const STOP_DRAIN_DEADLINE: Duration = Duration::from_secs(45); |
| 64 | const JOURNAL_SCHEMA_VERSION: u64 = 2; |
| 65 | const CLASSIC_SESSION_STATE_SCHEMA_VERSION: u64 = 3; |
| 66 | /// Hard bounds for the crash-recoverable unacknowledged-envelope journal. |
| 67 | const MAX_JOURNAL_EVENTS: usize = 256; |
| 68 | const MAX_JOURNAL_ENCODED_BYTES: usize = 4 * 1024 * 1024; |
| 69 | /// Capacity held back exclusively for integrity-critical envelopes (terminal |
| 70 | /// turn state, approvals, failures, resynchronization snapshots). Ordinary |
| 71 | /// deltas may never consume this headroom. |
| 72 | const JOURNAL_RESERVED_INTEGRITY_EVENTS: usize = 64; |
| 73 | const JOURNAL_RESERVED_INTEGRITY_BYTES: usize = 1024 * 1024; |
| 74 | /// A deferred (not yet handed to transport) delta envelope may grow to this |
| 75 | /// encoded size through coalescing before it is forced onto the wire. |
| 76 | const DELTA_COALESCE_BYTE_CAP: usize = 32 * 1024; |
| 77 | const JOURNAL_SETUP_ERROR: &str = "Remote control could not prepare its private delivery journal."; |
| 78 | const JOURNAL_UNTRUSTED_ERROR: &str = "The saved remote-control delivery journal could not be trusted; it was set aside. The account run may show an incomplete turn."; |
| 79 | const JOURNAL_LEGACY_SCOPE_ERROR: &str = "A saved legacy remote-control delivery journal is not bound to this workspace and cannot be replayed safely; it was preserved for explicit recovery."; |
| 80 | const CLASSIC_LEASE_SCOPE_ERROR: &str = "This saved session still owns an unfinished account turn in its original workspace; reconnect that workspace or create a new local session."; |
| 81 | |
| 82 | #[cfg(test)] |
| 83 | static TEST_JOURNAL_PERSIST_FAILURES: std::sync::Mutex<Vec<(PathBuf, usize)>> = |
| 84 | std::sync::Mutex::new(Vec::new()); |
| 85 | |
| 86 | #[cfg(test)] |
| 87 | fn inject_journal_persist_failures(path: &Path, count: usize) { |
| 88 | assert!(count > 0); |
| 89 | TEST_JOURNAL_PERSIST_FAILURES |
| 90 | .lock() |
| 91 | .unwrap_or_else(std::sync::PoisonError::into_inner) |
| 92 | .push((path.to_path_buf(), count)); |
| 93 | } |
| 94 | |
| 95 | #[cfg(test)] |
| 96 | fn take_journal_persist_failure(path: &Path) -> bool { |
| 97 | let mut failures = TEST_JOURNAL_PERSIST_FAILURES |
| 98 | .lock() |
| 99 | .unwrap_or_else(std::sync::PoisonError::into_inner); |
| 100 | let Some(index) = failures.iter().position(|(target, _)| target == path) else { |
| 101 | return false; |
| 102 | }; |
| 103 | if failures[index].1 > 1 { |
| 104 | failures[index].1 -= 1; |
| 105 | } else { |
| 106 | failures.remove(index); |
| 107 | } |
| 108 | true |
| 109 | } |
| 110 | |
| 111 | /// Envelopes whose loss would strand account-side truth: terminal turn state, |
| 112 | /// approval requests, failure records, and resynchronization snapshots. They |
| 113 | /// draw on reserved journal capacity, are never dropped silently, and gate |
| 114 | /// `/rc stop` until the server cursor covers them. |
| 115 | fn integrity_critical_event(event: &str) -> bool { |
| 116 | matches!( |
| 117 | event, |
| 118 | "turn.completed" |
| 119 | | "approval.required" |
| 120 | | "approval.resolved" |
| 121 | | "item.failed" |
| 122 | | "session.snapshot" |
| 123 | | "runtime.catalog" |
| 124 | ) |
| 125 | } |
| 126 | |
| 127 | fn runtime_envelope_event(envelope: &Value) -> Option<&str> { |
| 128 | envelope.get("event").and_then(Value::as_str) |
| 129 | } |
| 130 | |
| 131 | #[derive(Debug, Clone, PartialEq, Eq)] |
| 132 | pub enum RemoteControlAction { |
| 133 | Start, |
| 134 | Stop, |
| 135 | } |
| 136 | |
| 137 | #[derive(Clone)] |
| 138 | pub struct RemoteStart { |
| 139 | pub workspace_label: String, |
| 140 | pub target_ref: String, |
| 141 | pub session_id: String, |
| 142 | pub runtime_version: String, |
| 143 | pub runtime_commit: String, |
| 144 | /// Directory that holds the crash-recoverable delivery journal. `None` |
| 145 | /// runs memory-only and is reserved for tests; production callers must |
| 146 | /// always provide a private directory under the Codewhale home. |
| 147 | pub journal_dir: Option<PathBuf>, |
| 148 | /// Observed `owner/name` from `git remote get-url origin`, when the folder |
| 149 | /// is a Git checkout. This is a display receipt, never a path or GitHub App |
| 150 | /// grant. |
| 151 | pub git_remote: Option<String>, |
| 152 | } |
| 153 | |
| 154 | #[derive(Debug, Clone)] |
| 155 | pub enum RemoteEvent { |
| 156 | Notice(String), |
| 157 | Connected { |
| 158 | account_ref: String, |
| 159 | runner_id: String, |
| 160 | target_ref: String, |
| 161 | attachment: RemoteAttachment, |
| 162 | links: RemoteLinks, |
| 163 | }, |
| 164 | Attachment { |
| 165 | account_ref: String, |
| 166 | target_ref: String, |
| 167 | attachment: RemoteAttachment, |
| 168 | links: RemoteLinks, |
| 169 | }, |
| 170 | RuntimeCursor { |
| 171 | run_id: String, |
| 172 | cursor: u64, |
| 173 | }, |
| 174 | Command { |
| 175 | run_id: String, |
| 176 | seq: u64, |
| 177 | command: RemoteCommand, |
| 178 | }, |
| 179 | RuntimeChatProjection(RuntimeChatProjection), |
| 180 | /// Internal handoff receipt: the connected worker has dropped its clone |
| 181 | /// of the old immutable Runtime Chat host, so the controller can safely |
| 182 | /// reopen the same durable scope with a freshly supplied provider config. |
| 183 | RuntimeChatHostReleased, |
| 184 | Failed(String), |
| 185 | /// The relay died before any server-confirmed lease existed — during |
| 186 | /// enrollment, device authorization, or the first connect. No lease can |
| 187 | /// still be live, so nothing is locked and `/rc` can retry immediately. |
| 188 | FailedPreLease(String), |
| 189 | Stopped, |
| 190 | OwnershipRestored { |
| 191 | approvals: Vec<PendingRemoteApproval>, |
| 192 | }, |
| 193 | } |
| 194 | |
| 195 | #[derive(Debug, Clone, PartialEq, Eq)] |
| 196 | pub struct RemoteAttachment { |
| 197 | pub run_id: String, |
| 198 | pub workspace_id: String, |
| 199 | pub runtime_cursor: u64, |
| 200 | pub snapshot_present: bool, |
| 201 | pub runtime_chat_relay_protocol: String, |
| 202 | pub runtime_chat_relay_challenge: String, |
| 203 | } |
| 204 | |
| 205 | /// Web links the control plane advertises for the attached session. Both are |
| 206 | /// optional: an older control plane omits them and the terminal then simply |
| 207 | /// shows no link. Links are validated against the Codewhale app origin before |
| 208 | /// they are ever displayed or opened; the terminal never invents one. |
| 209 | #[derive(Debug, Clone, Default, PartialEq, Eq)] |
| 210 | pub struct RemoteLinks { |
| 211 | /// `https://app.codewhale.net/session?run=<runId>` for the live run. |
| 212 | pub run_url: Option<String>, |
| 213 | /// `https://app.codewhale.net/settings?section=workspaces` for this computer. |
| 214 | pub computer_url: Option<String>, |
| 215 | } |
| 216 | |
| 217 | #[derive(Debug, Clone, PartialEq, Eq)] |
| 218 | struct RunnerConnection { |
| 219 | runner_id: String, |
| 220 | attachment: RemoteAttachment, |
| 221 | links: RemoteLinks, |
| 222 | } |
| 223 | |
| 224 | #[derive(Debug, Clone, PartialEq, Eq)] |
| 225 | pub enum RemoteCommand { |
| 226 | Prompt { |
| 227 | turn_id: String, |
| 228 | prompt: String, |
| 229 | }, |
| 230 | RuntimeChatPrompt(Box<RuntimeChatPrompt>), |
| 231 | Approval { |
| 232 | gate: String, |
| 233 | approved: bool, |
| 234 | }, |
| 235 | Control { |
| 236 | action: RemoteControlRequest, |
| 237 | turn_id: Option<String>, |
| 238 | runtime_chat: Option<RuntimeChatControlScope>, |
| 239 | }, |
| 240 | } |
| 241 | |
| 242 | #[derive(Debug, Clone, Copy, PartialEq, Eq)] |
| 243 | pub enum RemoteControlRequest { |
| 244 | Interrupt, |
| 245 | Cancel, |
| 246 | } |
| 247 | |
| 248 | enum RelayPhase { |
| 249 | /// Everything before the first server-confirmed lease: control-plane base |
| 250 | /// resolution, enrollment, device authorization, and the first connect. |
| 251 | Enrolling, |
| 252 | /// The server confirmed a lease (Connected was emitted). Any failure now |
| 253 | /// is a lost-after-lease disconnect and stays fail-closed. |
| 254 | Leased, |
| 255 | } |
| 256 | |
| 257 | impl RelayPhase { |
| 258 | fn lease_confirmed(&self) -> bool { |
| 259 | matches!(self, Self::Leased) |
| 260 | } |
| 261 | } |
| 262 | |
| 263 | #[derive(Clone)] |
| 264 | enum WorkerCommand { |
| 265 | Upload { |
| 266 | run_id: String, |
| 267 | acknowledgements: Vec<CommandAcknowledgement>, |
| 268 | envelopes: Vec<Value>, |
| 269 | }, |
| 270 | ReleaseRuntimeChatHost, |
| 271 | InstallRuntimeChatHost(RuntimeChatRelayHost), |
| 272 | Stop, |
| 273 | } |
| 274 | |
| 275 | struct PendingRuntimeChatConfiguration { |
| 276 | config: crate::config::Config, |
| 277 | plugin_registry: std::sync::Arc<crate::plugins::PluginRegistry>, |
| 278 | private_root: PathBuf, |
| 279 | target_ref: String, |
| 280 | session_id: String, |
| 281 | } |
| 282 | |
| 283 | #[derive(Debug, Default)] |
| 284 | struct RuntimeTransportOutbox { |
| 285 | events: BTreeMap<(String, u64), Value>, |
| 286 | } |
| 287 | |
| 288 | /// One unacknowledged runtime envelope owned by the controller. |
| 289 | /// |
| 290 | /// `handed_off` records whether the envelope may already have reached the |
| 291 | /// server through the transport worker. Once true the envelope is immutable: |
| 292 | /// ambiguous retries must resend byte-identical JSON. |
| 293 | #[derive(Debug, Clone, PartialEq)] |
| 294 | struct PendingRuntimeEnvelope { |
| 295 | envelope: Value, |
| 296 | encoded_len: usize, |
| 297 | integrity: bool, |
| 298 | handed_off: bool, |
| 299 | } |
| 300 | |
| 301 | /// Crash-recoverable journal of unacknowledged runtime envelopes. |
| 302 | /// |
| 303 | /// The file name is hash-derived so nothing about the workspace or session |
| 304 | /// leaks through the path; the directory is private and the file owner-only. |
| 305 | /// Neither the path nor the contents are ever reported to the control plane |
| 306 | /// or written to logs. Acknowledged prefixes are compacted on every persist, |
| 307 | /// and a journal that cannot be verified fails closed at load time. |
| 308 | struct RuntimeEventJournal { |
| 309 | path: PathBuf, |
| 310 | scope_tag: String, |
| 311 | legacy_path: PathBuf, |
| 312 | legacy_unscoped_path: PathBuf, |
| 313 | active_index_path: PathBuf, |
| 314 | classic_lease: parking_lot::Mutex<Option<ClassicRunLease>>, |
| 315 | _classic_session_lock: ClassicSessionOwnerLock, |
| 316 | } |
| 317 | |
| 318 | #[derive(Debug)] |
| 319 | struct ClassicSessionOwnerLock { |
| 320 | _file: File, |
| 321 | } |
| 322 | |
| 323 | impl ClassicSessionOwnerLock { |
| 324 | fn acquire(path: &Path) -> Result<Self, String> { |
| 325 | let mut options = std::fs::OpenOptions::new(); |
| 326 | options.create(true).read(true).write(true); |
| 327 | #[cfg(unix)] |
| 328 | { |
| 329 | use std::os::unix::fs::OpenOptionsExt as _; |
| 330 | options.mode(0o600).custom_flags(libc::O_NOFOLLOW); |
| 331 | } |
| 332 | let file = options |
| 333 | .open(path) |
| 334 | .map_err(|_| JOURNAL_SETUP_ERROR.to_string())?; |
| 335 | #[cfg(unix)] |
| 336 | { |
| 337 | use std::os::fd::AsRawFd as _; |
| 338 | // SAFETY: `file` is open and live. |
| 339 | if unsafe { libc::flock(file.as_raw_fd(), libc::LOCK_EX | libc::LOCK_NB) } != 0 { |
| 340 | return Err(CLASSIC_LEASE_SCOPE_ERROR.to_string()); |
| 341 | } |
| 342 | } |
| 343 | #[cfg(windows)] |
| 344 | { |
| 345 | use std::os::windows::io::AsRawHandle as _; |
| 346 | use windows_sys::Win32::Storage::FileSystem::LockFile; |
| 347 | // SAFETY: `file` is open and live. |
| 348 | if unsafe { LockFile(file.as_raw_handle() as _, 0, 0, u32::MAX, u32::MAX) } == 0 { |
| 349 | return Err(CLASSIC_LEASE_SCOPE_ERROR.to_string()); |
| 350 | } |
| 351 | } |
| 352 | Ok(Self { _file: file }) |
| 353 | } |
| 354 | } |
| 355 | |
| 356 | fn runtime_journal_scope_tag(target_ref: &str, session_id: &str) -> String { |
| 357 | let mut hasher = Sha256::new(); |
| 358 | hasher.update(b"cwc-remote-control-journal.v2\0"); |
| 359 | hasher.update(target_ref.as_bytes()); |
| 360 | hasher.update(b"\0"); |
| 361 | hasher.update(session_id.as_bytes()); |
| 362 | bytes_to_hex(&hasher.finalize())[..32].to_string() |
| 363 | } |
| 364 | |
| 365 | fn runtime_journal_session_tag(session_id: &str) -> String { |
| 366 | let mut hasher = Sha256::new(); |
| 367 | hasher.update(b"cwc-remote-control-journal\0"); |
| 368 | hasher.update(session_id.as_bytes()); |
| 369 | bytes_to_hex(&hasher.finalize())[..32].to_string() |
| 370 | } |
| 371 | |
| 372 | fn classic_recovery_turn_id(run_id: &str, lease_id: &str) -> String { |
| 373 | let mut hasher = Sha256::new(); |
| 374 | hasher.update(b"codewhale.classic-recovery-turn.v1\0"); |
| 375 | hasher.update(run_id.as_bytes()); |
| 376 | hasher.update(b"\0"); |
| 377 | hasher.update(lease_id.as_bytes()); |
| 378 | format!("turn_recovered_{}", &bytes_to_hex(&hasher.finalize())[..24]) |
| 379 | } |
| 380 | |
| 381 | impl RuntimeEventJournal { |
| 382 | fn open(dir: &Path, target_ref: &str, session_id: &str) -> Result<Self, String> { |
| 383 | let scope_tag = runtime_journal_scope_tag(target_ref, session_id); |
| 384 | let legacy_session_tag = runtime_journal_session_tag(session_id); |
| 385 | std::fs::create_dir_all(dir).map_err(|_| JOURNAL_SETUP_ERROR.to_string())?; |
| 386 | #[cfg(unix)] |
| 387 | { |
| 388 | use std::os::unix::fs::PermissionsExt; |
| 389 | std::fs::set_permissions(dir, std::fs::Permissions::from_mode(0o700)) |
| 390 | .map_err(|_| JOURNAL_SETUP_ERROR.to_string())?; |
| 391 | } |
| 392 | let legacy_path = dir.join(format!("journal_{legacy_session_tag}.json")); |
| 393 | let active_index_path = dir.join(format!("active_classic_{legacy_session_tag}.json")); |
| 394 | let classic_session_lock = |
| 395 | ClassicSessionOwnerLock::acquire(&active_index_path.with_extension("lock"))?; |
| 396 | if !active_index_path.exists() |
| 397 | && (legacy_path.exists() || legacy_path.with_extension("unscoped").exists()) |
| 398 | { |
| 399 | return Err(JOURNAL_LEGACY_SCOPE_ERROR.to_string()); |
| 400 | } |
| 401 | let journal = Self { |
| 402 | path: dir.join(format!("journal_{scope_tag}.json")), |
| 403 | scope_tag, |
| 404 | legacy_unscoped_path: legacy_path.with_extension("unscoped"), |
| 405 | legacy_path, |
| 406 | active_index_path, |
| 407 | classic_lease: parking_lot::Mutex::new(None), |
| 408 | _classic_session_lock: classic_session_lock, |
| 409 | }; |
| 410 | if journal.active_index_path.exists() { |
| 411 | let lease = journal.read_active_index()?; |
| 412 | *journal.classic_lease.lock() = lease; |
| 413 | } else { |
| 414 | journal.write_active_index(None)?; |
| 415 | } |
| 416 | Ok(journal) |
| 417 | } |
| 418 | |
| 419 | /// Loads every journaled envelope, or fails closed when the journal |
| 420 | /// cannot be trusted (corrupt, oversized, or written for another |
| 421 | /// session). A missing file is an ordinary empty journal. |
| 422 | fn load(&self) -> Result<HashMap<String, BTreeMap<u64, Value>>, String> { |
| 423 | let bytes = match std::fs::read(&self.path) { |
| 424 | Ok(bytes) => bytes, |
| 425 | Err(error) if error.kind() == std::io::ErrorKind::NotFound => { |
| 426 | if self.legacy_path.exists() || self.legacy_unscoped_path.exists() { |
| 427 | // Schema 1 authenticated only a session id. The same |
| 428 | // saved session can be opened from another target, so its |
| 429 | // envelopes cannot be relabelled into this v2 |
| 430 | // target+session scope without server-side ownership |
| 431 | // proof. Preserve it and fail closed. |
| 432 | return Err(JOURNAL_LEGACY_SCOPE_ERROR.to_string()); |
| 433 | } |
| 434 | return Ok(HashMap::new()); |
| 435 | } |
| 436 | Err(_) => return Err(JOURNAL_UNTRUSTED_ERROR.to_string()), |
| 437 | }; |
| 438 | if bytes.len() > MAX_JOURNAL_ENCODED_BYTES.saturating_mul(2) { |
| 439 | return Err(JOURNAL_UNTRUSTED_ERROR.to_string()); |
| 440 | } |
| 441 | let value: Value = |
| 442 | serde_json::from_slice(&bytes).map_err(|_| JOURNAL_UNTRUSTED_ERROR.to_string())?; |
| 443 | let identity_valid = value.get("schemaVersion").and_then(Value::as_u64) |
| 444 | == Some(JOURNAL_SCHEMA_VERSION) |
| 445 | && value.get("scope").and_then(Value::as_str) == Some(self.scope_tag.as_str()); |
| 446 | if !identity_valid { |
| 447 | return Err(JOURNAL_UNTRUSTED_ERROR.to_string()); |
| 448 | } |
| 449 | let runs = value |
| 450 | .get("runs") |
| 451 | .and_then(Value::as_object) |
| 452 | .ok_or_else(|| JOURNAL_UNTRUSTED_ERROR.to_string())?; |
| 453 | let mut restored: HashMap<String, BTreeMap<u64, Value>> = HashMap::new(); |
| 454 | let mut total_events = 0usize; |
| 455 | let mut total_bytes = 0usize; |
| 456 | for (run_id, envelopes) in runs { |
| 457 | if !valid_opaque_ref(run_id) { |
| 458 | return Err(JOURNAL_UNTRUSTED_ERROR.to_string()); |
| 459 | } |
| 460 | let envelopes = envelopes |
| 461 | .as_array() |
| 462 | .ok_or_else(|| JOURNAL_UNTRUSTED_ERROR.to_string())?; |
| 463 | let mut events = BTreeMap::new(); |
| 464 | for envelope in envelopes { |
| 465 | let seq = runtime_envelope_seq(envelope) |
| 466 | .ok_or_else(|| JOURNAL_UNTRUSTED_ERROR.to_string())?; |
| 467 | let encoded_len = serde_json::to_vec(envelope) |
| 468 | .map(|body| body.len()) |
| 469 | .unwrap_or(usize::MAX); |
| 470 | if encoded_len > MAX_RUNTIME_ENVELOPE_BYTES { |
| 471 | return Err(JOURNAL_UNTRUSTED_ERROR.to_string()); |
| 472 | } |
| 473 | total_events += 1; |
| 474 | total_bytes = total_bytes.saturating_add(encoded_len); |
| 475 | if total_events > MAX_JOURNAL_EVENTS || total_bytes > MAX_JOURNAL_ENCODED_BYTES { |
| 476 | return Err(JOURNAL_UNTRUSTED_ERROR.to_string()); |
| 477 | } |
| 478 | if events.insert(seq, envelope.clone()).is_some() { |
| 479 | return Err(JOURNAL_UNTRUSTED_ERROR.to_string()); |
| 480 | } |
| 481 | } |
| 482 | if !events.is_empty() { |
| 483 | restored.insert(run_id.clone(), events); |
| 484 | } |
| 485 | } |
| 486 | Ok(restored) |
| 487 | } |
| 488 | |
| 489 | /// Atomically replaces the journal with the current unacknowledged set. |
| 490 | /// An empty set removes the file entirely (prompt compaction). |
| 491 | fn persist( |
| 492 | &self, |
| 493 | pending: &HashMap<String, BTreeMap<u64, PendingRuntimeEnvelope>>, |
| 494 | ) -> Result<(), String> { |
| 495 | #[cfg(test)] |
| 496 | if take_journal_persist_failure(&self.path) { |
| 497 | return Err(JOURNAL_SETUP_ERROR.to_string()); |
| 498 | } |
| 499 | let classic_lease = self.classic_lease.lock().clone(); |
| 500 | if pending.values().all(BTreeMap::is_empty) && classic_lease.is_none() { |
| 501 | self.remove(); |
| 502 | return Ok(()); |
| 503 | } |
| 504 | let mut runs = serde_json::Map::new(); |
| 505 | for (run_id, events) in pending { |
| 506 | if events.is_empty() { |
| 507 | continue; |
| 508 | } |
| 509 | runs.insert( |
| 510 | run_id.clone(), |
| 511 | Value::Array( |
| 512 | events |
| 513 | .values() |
| 514 | .map(|entry| entry.envelope.clone()) |
| 515 | .collect(), |
| 516 | ), |
| 517 | ); |
| 518 | } |
| 519 | let body = serde_json::to_vec(&json!({ |
| 520 | "schemaVersion": JOURNAL_SCHEMA_VERSION, |
| 521 | "scope": self.scope_tag, |
| 522 | "runs": runs, |
| 523 | })) |
| 524 | .map_err(|_| JOURNAL_SETUP_ERROR.to_string())?; |
| 525 | crate::utils::write_atomic(&self.path, &body).map_err(|_| JOURNAL_SETUP_ERROR.to_string()) |
| 526 | } |
| 527 | |
| 528 | fn remove(&self) { |
| 529 | if self.classic_lease.lock().is_some() { |
| 530 | return; |
| 531 | } |
| 532 | let _ = std::fs::remove_file(&self.path); |
| 533 | let _ = std::fs::remove_file(self.path.with_extension("tmp")); |
| 534 | } |
| 535 | |
| 536 | /// Moves an untrusted journal aside so the failure is explicit and a |
| 537 | /// deliberate later `/rc` start can proceed from a clean slate. |
| 538 | fn quarantine_current(&self) { |
| 539 | let _ = std::fs::rename(&self.path, self.path.with_extension("corrupt")); |
| 540 | } |
| 541 | |
| 542 | fn quarantine_legacy(&self) { |
| 543 | if self.legacy_path.exists() { |
| 544 | let _ = std::fs::rename(&self.legacy_path, &self.legacy_unscoped_path); |
| 545 | } |
| 546 | } |
| 547 | |
| 548 | fn classic_lease(&self) -> Option<ClassicRunLease> { |
| 549 | self.classic_lease.lock().clone() |
| 550 | } |
| 551 | |
| 552 | fn read_active_index(&self) -> Result<Option<ClassicRunLease>, String> { |
| 553 | let bytes = std::fs::read(&self.active_index_path) |
| 554 | .map_err(|_| JOURNAL_UNTRUSTED_ERROR.to_string())?; |
| 555 | if bytes.len() > 2048 { |
| 556 | return Err(JOURNAL_UNTRUSTED_ERROR.to_string()); |
| 557 | } |
| 558 | let value: Value = |
| 559 | serde_json::from_slice(&bytes).map_err(|_| JOURNAL_UNTRUSTED_ERROR.to_string())?; |
| 560 | if value.get("schemaVersion").and_then(Value::as_u64) |
| 561 | != Some(CLASSIC_SESSION_STATE_SCHEMA_VERSION) |
| 562 | || value.get("scope").and_then(Value::as_str) != Some(self.scope_tag.as_str()) |
| 563 | { |
| 564 | return Err(CLASSIC_LEASE_SCOPE_ERROR.to_string()); |
| 565 | } |
| 566 | match value.get("lease") { |
| 567 | None | Some(Value::Null) => Ok(None), |
| 568 | Some(value) => { |
| 569 | let lease: ClassicRunLease = serde_json::from_value(value.clone()) |
| 570 | .map_err(|_| JOURNAL_UNTRUSTED_ERROR.to_string())?; |
| 571 | if !lease.valid() { |
| 572 | return Err(JOURNAL_UNTRUSTED_ERROR.to_string()); |
| 573 | } |
| 574 | Ok(Some(lease)) |
| 575 | } |
| 576 | } |
| 577 | } |
| 578 | |
| 579 | fn write_active_index(&self, lease: Option<&ClassicRunLease>) -> Result<(), String> { |
| 580 | #[cfg(test)] |
| 581 | if take_journal_persist_failure(&self.active_index_path) { |
| 582 | return Err(JOURNAL_SETUP_ERROR.to_string()); |
| 583 | } |
| 584 | let index = serde_json::to_vec(&json!({ |
| 585 | "schemaVersion": CLASSIC_SESSION_STATE_SCHEMA_VERSION, |
| 586 | "scope": self.scope_tag, |
| 587 | "lease": lease, |
| 588 | })) |
| 589 | .map_err(|_| JOURNAL_SETUP_ERROR.to_string())?; |
| 590 | crate::utils::write_atomic(&self.active_index_path, &index) |
| 591 | .map_err(|_| JOURNAL_SETUP_ERROR.to_string()) |
| 592 | } |
| 593 | |
| 594 | fn set_classic_lease( |
| 595 | &self, |
| 596 | lease: Option<ClassicRunLease>, |
| 597 | pending: &HashMap<String, BTreeMap<u64, PendingRuntimeEnvelope>>, |
| 598 | ) -> Result<(), String> { |
| 599 | let previous = self.classic_lease.lock().clone(); |
| 600 | let indexed = self.read_active_index()?; |
| 601 | if indexed != previous { |
| 602 | return Err(CLASSIC_LEASE_SCOPE_ERROR.to_string()); |
| 603 | } |
| 604 | if let Some(lease) = &lease { |
| 605 | self.write_active_index(Some(lease))?; |
| 606 | } |
| 607 | *self.classic_lease.lock() = lease.clone(); |
| 608 | if let Err(error) = self.persist(pending) { |
| 609 | *self.classic_lease.lock() = previous.clone(); |
| 610 | let _ = self.write_active_index(previous.as_ref()); |
| 611 | return Err(error); |
| 612 | } |
| 613 | if lease.is_none() |
| 614 | && let Err(error) = self.write_active_index(None) |
| 615 | { |
| 616 | *self.classic_lease.lock() = previous.clone(); |
| 617 | let _ = self.write_active_index(previous.as_ref()); |
| 618 | return Err(error); |
| 619 | } |
| 620 | Ok(()) |
| 621 | } |
| 622 | |
| 623 | /// Advances the canonical sequence floor for the active classic Work run. |
| 624 | /// |
| 625 | /// The target journal can compact an acknowledged prefix, so the |
| 626 | /// session-wide lease must retain the highest server/local sequence while |
| 627 | /// a provider turn is still active. Recovery uses this floor before it |
| 628 | /// synthesizes a terminal event and can therefore never reuse a sequence |
| 629 | /// the server already acknowledged. |
| 630 | fn advance_classic_seq_floor(&self, run_id: &str, seq: u64) -> Result<(), String> { |
| 631 | let Some(mut lease) = self.classic_lease.lock().clone() else { |
| 632 | return Ok(()); |
| 633 | }; |
| 634 | if lease.run_id != run_id || seq <= lease.seq_floor { |
| 635 | return Ok(()); |
| 636 | } |
| 637 | let indexed = self.read_active_index()?; |
| 638 | if indexed.as_ref() != self.classic_lease.lock().as_ref() { |
| 639 | return Err(CLASSIC_LEASE_SCOPE_ERROR.to_string()); |
| 640 | } |
| 641 | lease.seq_floor = seq; |
| 642 | self.write_active_index(Some(&lease))?; |
| 643 | *self.classic_lease.lock() = Some(lease); |
| 644 | Ok(()) |
| 645 | } |
| 646 | } |
| 647 | |
| 648 | #[derive(Debug, Clone, Copy, PartialEq, Eq)] |
| 649 | enum RuntimePostOutcome { |
| 650 | Accepted(u64), |
| 651 | Retryable, |
| 652 | AccessTokenExpired, |
| 653 | } |
| 654 | |
| 655 | #[derive(Debug, Clone, PartialEq, Eq)] |
| 656 | enum RuntimeFlushOutcome { |
| 657 | Idle, |
| 658 | Accepted { run_id: String, cursor: u64 }, |
| 659 | Retryable, |
| 660 | AccessTokenExpired, |
| 661 | } |
| 662 | |
| 663 | #[derive(Debug, Clone, Serialize)] |
| 664 | #[serde(rename_all = "camelCase")] |
| 665 | struct CommandAcknowledgement { |
| 666 | command_seq: u64, |
| 667 | command_type: String, |
| 668 | status: String, |
| 669 | #[serde(skip_serializing_if = "Option::is_none")] |
| 670 | turn_id: Option<String>, |
| 671 | #[serde(skip_serializing_if = "Option::is_none")] |
| 672 | error: Option<String>, |
| 673 | } |
| 674 | |
| 675 | #[derive(Clone, Deserialize, Serialize)] |
| 676 | #[serde(rename_all = "camelCase", deny_unknown_fields)] |
| 677 | struct PersistedEnrollment { |
| 678 | schema_version: u64, |
| 679 | control_plane_base: String, |
| 680 | runner_enrollment_id: String, |
| 681 | account_ref: String, |
| 682 | device_id: String, |
| 683 | target_ref: String, |
| 684 | target_grant_ref: String, |
| 685 | runtime_version: String, |
| 686 | runtime_commit: String, |
| 687 | bootstrap_secret: String, |
| 688 | } |
| 689 | |
| 690 | /// The machine-stable device identity persisted independently of any |
| 691 | /// enrollment. Enrollments are deleted and re-created when the target or |
| 692 | /// runtime changes; this record is only ever created once per keychain. |
| 693 | #[derive(Debug, Clone, Deserialize, Serialize, PartialEq, Eq)] |
| 694 | #[serde(rename_all = "camelCase", deny_unknown_fields)] |
| 695 | struct PersistedDeviceIdentity { |
| 696 | schema_version: u64, |
| 697 | device_id: String, |
| 698 | } |
| 699 | |
| 700 | impl PersistedDeviceIdentity { |
| 701 | fn valid(&self) -> bool { |
| 702 | self.schema_version == 1 && valid_opaque_ref(&self.device_id) |
| 703 | } |
| 704 | } |
| 705 | |
| 706 | /// Pick the device id this terminal presents to the control plane. A valid |
| 707 | /// saved identity always wins; otherwise the device id of an existing |
| 708 | /// enrollment is adopted (so upgrading terminals keep their computer row); |
| 709 | /// otherwise a fresh id is minted. Returns the id and whether it must be |
| 710 | /// saved. |
| 711 | fn resolve_device_identity( |
| 712 | saved: Option<PersistedDeviceIdentity>, |
| 713 | enrollment_device_id: Option<&str>, |
| 714 | ) -> (String, bool) { |
| 715 | if let Some(saved) = saved.filter(PersistedDeviceIdentity::valid) { |
| 716 | return (saved.device_id, false); |
| 717 | } |
| 718 | if let Some(existing) = enrollment_device_id.filter(|value| valid_opaque_ref(value)) { |
| 719 | return (existing.to_string(), true); |
| 720 | } |
| 721 | (format!("device_{}", uuid::Uuid::new_v4().simple()), true) |
| 722 | } |
| 723 | |
| 724 | #[derive(Clone)] |
| 725 | struct LiveEnrollment { |
| 726 | persisted: PersistedEnrollment, |
| 727 | access_token: String, |
| 728 | } |
| 729 | |
| 730 | #[derive(Debug, Clone)] |
| 731 | struct ActiveRelayRun { |
| 732 | run_id: String, |
| 733 | turn_id: String, |
| 734 | } |
| 735 | |
| 736 | #[derive(Debug, Clone, Deserialize, Serialize, PartialEq, Eq)] |
| 737 | #[serde(rename_all = "camelCase", deny_unknown_fields)] |
| 738 | struct ClassicRunLease { |
| 739 | run_id: String, |
| 740 | turn_id: Option<String>, |
| 741 | lease_id: String, |
| 742 | seq_floor: u64, |
| 743 | } |
| 744 | |
| 745 | impl ClassicRunLease { |
| 746 | fn valid(&self) -> bool { |
| 747 | valid_opaque_ref(&self.run_id) |
| 748 | && self.turn_id.as_deref().is_none_or(valid_opaque_ref) |
| 749 | && valid_opaque_ref(&self.lease_id) |
| 750 | && self.seq_floor <= JS_MAX_SAFE_INTEGER |
| 751 | } |
| 752 | } |
| 753 | |
| 754 | #[derive(Debug, Clone)] |
| 755 | pub struct PendingRemoteApproval { |
| 756 | pub tool_id: String, |
| 757 | } |
| 758 | |
| 759 | #[derive(Debug, Default, Clone, Copy, PartialEq, Eq)] |
| 760 | enum Status { |
| 761 | #[default] |
| 762 | Off, |
| 763 | Connecting, |
| 764 | Connected, |
| 765 | Stopping, |
| 766 | Failed, |
| 767 | } |
| 768 | |
| 769 | /// UI-thread owner for remote-control state and typed transport channels. |
| 770 | pub struct RemoteControlController { |
| 771 | status: Status, |
| 772 | status_detail: String, |
| 773 | account_ref: Option<String>, |
| 774 | target_ref: Option<String>, |
| 775 | links: RemoteLinks, |
| 776 | /// Latest server-confirmed run attachment. Kept separately from |
| 777 | /// `active_run`: an attachment can exist while the session is idle, and a |
| 778 | /// mid-turn `/rc` can bind the already-running local turn to it without |
| 779 | /// inventing a second prompt. |
| 780 | attached_run_id: Option<String>, |
| 781 | attached_workspace_id: Option<String>, |
| 782 | active_run: Option<ActiveRelayRun>, |
| 783 | /// A local dispatch that was already in flight when the web attachment |
| 784 | /// became ready, but whose typed `TurnStarted` event has not landed yet. |
| 785 | /// The first such event promotes this exact run into `active_run`. |
| 786 | pending_local_turn_run: Option<String>, |
| 787 | event_seq: HashMap<String, u64>, |
| 788 | uploaded_snapshots: HashSet<String>, |
| 789 | pending_runtime_events: HashMap<String, BTreeMap<u64, PendingRuntimeEnvelope>>, |
| 790 | pending_approvals: HashMap<String, PendingRemoteApproval>, |
| 791 | command_fingerprints: HashMap<(String, u64), String>, |
| 792 | worker_tx: Option<mpsc::UnboundedSender<WorkerCommand>>, |
| 793 | event_rx: Option<mpsc::UnboundedReceiver<RemoteEvent>>, |
| 794 | worker: Option<tokio::task::JoinHandle<()>>, |
| 795 | applying_remote_command: bool, |
| 796 | ownership_blocked_until: Option<Instant>, |
| 797 | journal: Option<RuntimeEventJournal>, |
| 798 | /// At most one deferred (unsent, still coalescible) delta seq per run. |
| 799 | deferred_delta: HashMap<String, u64>, |
| 800 | /// Runs whose deltas were shed under pressure; truth is restored with a |
| 801 | /// bounded snapshot at the next terminal boundary. |
| 802 | resync_required: HashSet<String>, |
| 803 | /// Runs that crossed their terminal boundary with `resync_required` set; |
| 804 | /// the UI drains these via `take_pending_resync`. |
| 805 | resync_ready: Vec<String>, |
| 806 | pending_event_count: usize, |
| 807 | pending_encoded_bytes: usize, |
| 808 | /// Separate native Runtime host for account Chat. It is configured by the |
| 809 | /// TUI composition root and never points at the current interactive turn. |
| 810 | runtime_chat: Option<RuntimeChatRelayHost>, |
| 811 | /// A same-scope reconnect may need the old immutable host long enough to |
| 812 | /// drain terminal/catalog rows behind the server cursor. New prompts stay |
| 813 | /// fail-closed while this fresh configuration waits for an acknowledged |
| 814 | /// boundary and a worker/controller host handoff. |
| 815 | pending_runtime_chat_configuration: Option<PendingRuntimeChatConfiguration>, |
| 816 | runtime_chat_refresh_release_requested: bool, |
| 817 | runtime_chat_attachment: Option<RemoteAttachment>, |
| 818 | uploaded_runtime_catalog_receipt: HashMap<String, String>, |
| 819 | } |
| 820 | |
| 821 | impl Default for RemoteControlController { |
| 822 | fn default() -> Self { |
| 823 | Self { |
| 824 | status: Status::Off, |
| 825 | status_detail: "off".to_string(), |
| 826 | account_ref: None, |
| 827 | target_ref: None, |
| 828 | links: RemoteLinks::default(), |
| 829 | attached_run_id: None, |
| 830 | attached_workspace_id: None, |
| 831 | active_run: None, |
| 832 | pending_local_turn_run: None, |
| 833 | event_seq: HashMap::new(), |
| 834 | uploaded_snapshots: HashSet::new(), |
| 835 | pending_runtime_events: HashMap::new(), |
| 836 | pending_approvals: HashMap::new(), |
| 837 | command_fingerprints: HashMap::new(), |
| 838 | worker_tx: None, |
| 839 | event_rx: None, |
| 840 | worker: None, |
| 841 | applying_remote_command: false, |
| 842 | ownership_blocked_until: None, |
| 843 | journal: None, |
| 844 | deferred_delta: HashMap::new(), |
| 845 | resync_required: HashSet::new(), |
| 846 | resync_ready: Vec::new(), |
| 847 | pending_event_count: 0, |
| 848 | pending_encoded_bytes: 0, |
| 849 | runtime_chat: None, |
| 850 | pending_runtime_chat_configuration: None, |
| 851 | runtime_chat_refresh_release_requested: false, |
| 852 | runtime_chat_attachment: None, |
| 853 | uploaded_runtime_catalog_receipt: HashMap::new(), |
| 854 | } |
| 855 | } |
| 856 | } |
| 857 | |
| 858 | impl RemoteControlController { |
| 859 | /// Acquires and validates the saved-session authority before any Runtime |
| 860 | /// Chat host/provider configuration is opened. |
| 861 | /// |
| 862 | /// The session-wide lifetime lock is intentionally acquired first. This |
| 863 | /// fixes the lock order across processes: a competing workspace cannot |
| 864 | /// make us release an existing Runtime store owner and only later discover |
| 865 | /// that the saved session is permanently bound elsewhere. |
| 866 | pub(crate) fn prepare_remote_control_session_journal( |
| 867 | &mut self, |
| 868 | dir: &Path, |
| 869 | target_ref: &str, |
| 870 | session_id: &str, |
| 871 | ) -> Result<(), String> { |
| 872 | let requested_scope = runtime_journal_scope_tag(target_ref, session_id); |
| 873 | let requested_session_path = dir.join(format!( |
| 874 | "active_classic_{}.json", |
| 875 | runtime_journal_session_tag(session_id) |
| 876 | )); |
| 877 | if let Some(current) = self.journal.as_ref() { |
| 878 | if current.active_index_path == requested_session_path { |
| 879 | if current.scope_tag != requested_scope { |
| 880 | return Err(CLASSIC_LEASE_SCOPE_ERROR.to_string()); |
| 881 | } |
| 882 | return Ok(()); |
| 883 | } |
| 884 | if self.has_active_run() || self.has_unacknowledged_integrity_events() { |
| 885 | return Err( |
| 886 | "The previous saved session still owns account delivery; reconnect it before opening another local session." |
| 887 | .to_string(), |
| 888 | ); |
| 889 | } |
| 890 | } |
| 891 | drop(self.journal.take()); |
| 892 | let journal = RuntimeEventJournal::open(dir, target_ref, session_id)?; |
| 893 | match journal.load() { |
| 894 | Ok(restored) => { |
| 895 | self.reset_pending_from(restored); |
| 896 | self.journal = Some(journal); |
| 897 | Ok(()) |
| 898 | } |
| 899 | Err(error) => { |
| 900 | if error == JOURNAL_LEGACY_SCOPE_ERROR { |
| 901 | journal.quarantine_legacy(); |
| 902 | } else { |
| 903 | journal.quarantine_current(); |
| 904 | } |
| 905 | Err(error) |
| 906 | } |
| 907 | } |
| 908 | } |
| 909 | |
| 910 | pub(crate) fn configure_runtime_chat( |
| 911 | &mut self, |
| 912 | config: crate::config::Config, |
| 913 | plugin_registry: std::sync::Arc<crate::plugins::PluginRegistry>, |
| 914 | private_root: PathBuf, |
| 915 | target_ref: String, |
| 916 | session_id: String, |
| 917 | ) -> Result<(), String> { |
| 918 | if matches!( |
| 919 | self.status, |
| 920 | Status::Connecting | Status::Connected | Status::Stopping |
| 921 | ) { |
| 922 | return Err("Remote control is already active.".to_string()); |
| 923 | } |
| 924 | let requested = PendingRuntimeChatConfiguration { |
| 925 | config, |
| 926 | plugin_registry, |
| 927 | private_root, |
| 928 | target_ref, |
| 929 | session_id, |
| 930 | }; |
| 931 | if let Some(host) = self.runtime_chat.as_ref() { |
| 932 | if host.scope_matches(&requested.target_ref, &requested.session_id) |
| 933 | && (host.has_any_unsettled_turns() || self.has_unacknowledged_integrity_events()) |
| 934 | { |
| 935 | // Reuse the existing lifetime owner and durable manager only |
| 936 | // while recovery still needs it. The fresh config is retained |
| 937 | // separately and every new prompt is rejected until the |
| 938 | // terminal/catalog backlog is server-acked and the connected |
| 939 | // worker has explicitly dropped its clone. |
| 940 | self.pending_runtime_chat_configuration = Some(requested); |
| 941 | self.runtime_chat_refresh_release_requested = false; |
| 942 | return Ok(()); |
| 943 | } |
| 944 | if host.has_any_unsettled_turns() { |
| 945 | return Err( |
| 946 | "An isolated Runtime Chat turn is still active; reconnect its original remote-control session before opening another." |
| 947 | .to_string(), |
| 948 | ); |
| 949 | } |
| 950 | } |
| 951 | if self.has_unacknowledged_integrity_events() { |
| 952 | return Err( |
| 953 | "The current remote-control session still has unacknowledged terminal events; reconnect it before opening another session." |
| 954 | .to_string(), |
| 955 | ); |
| 956 | } |
| 957 | // A completed `/rc` session deliberately keeps its durable host until |
| 958 | // the next start. Release that lifetime owner lock before reopening the |
| 959 | // same account/session namespace with a fresh provider configuration. |
| 960 | self.install_runtime_chat_configuration(requested) |
| 961 | } |
| 962 | |
| 963 | fn install_runtime_chat_configuration( |
| 964 | &mut self, |
| 965 | requested: PendingRuntimeChatConfiguration, |
| 966 | ) -> Result<(), String> { |
| 967 | drop(self.runtime_chat.take()); |
| 968 | let host = RuntimeChatRelayHost::open( |
| 969 | requested.config, |
| 970 | requested.plugin_registry, |
| 971 | requested.private_root, |
| 972 | requested.target_ref, |
| 973 | requested.session_id, |
| 974 | )?; |
| 975 | self.runtime_chat = Some(host); |
| 976 | self.pending_runtime_chat_configuration = None; |
| 977 | self.runtime_chat_refresh_release_requested = false; |
| 978 | self.uploaded_runtime_catalog_receipt.clear(); |
| 979 | Ok(()) |
| 980 | } |
| 981 | |
| 982 | fn schedule_runtime_chat_refresh_if_ready(&mut self) -> Result<(), String> { |
| 983 | if self.pending_runtime_chat_configuration.is_none() |
| 984 | || self.runtime_chat_refresh_release_requested |
| 985 | || self.has_unacknowledged_integrity_events() |
| 986 | || self |
| 987 | .runtime_chat |
| 988 | .as_ref() |
| 989 | .is_some_and(RuntimeChatRelayHost::has_any_unsettled_turns) |
| 990 | { |
| 991 | return Ok(()); |
| 992 | } |
| 993 | if self.status != Status::Connected { |
| 994 | return Ok(()); |
| 995 | } |
| 996 | let tx = self.worker_tx.as_ref().ok_or_else(|| { |
| 997 | "Runtime Chat could not refresh its provider configuration safely.".to_string() |
| 998 | })?; |
| 999 | tx.send(WorkerCommand::ReleaseRuntimeChatHost) |
| 1000 | .map_err(|_| { |
| 1001 | "Runtime Chat could not refresh its provider configuration safely.".to_string() |
| 1002 | })?; |
| 1003 | self.runtime_chat_refresh_release_requested = true; |
| 1004 | self.status_detail = "refreshing the local Runtime model configuration".to_string(); |
| 1005 | Ok(()) |
| 1006 | } |
| 1007 | |
| 1008 | fn complete_runtime_chat_refresh(&mut self) -> Result<(), String> { |
| 1009 | if !self.runtime_chat_refresh_release_requested |
| 1010 | || self.has_unacknowledged_integrity_events() |
| 1011 | || self |
| 1012 | .runtime_chat |
| 1013 | .as_ref() |
| 1014 | .is_some_and(RuntimeChatRelayHost::has_any_unsettled_turns) |
| 1015 | { |
| 1016 | return Err("Runtime Chat received an invalid provider-refresh handoff.".to_string()); |
| 1017 | } |
| 1018 | let requested = self |
| 1019 | .pending_runtime_chat_configuration |
| 1020 | .take() |
| 1021 | .ok_or_else(|| { |
| 1022 | "Runtime Chat received an invalid provider-refresh handoff.".to_string() |
| 1023 | })?; |
| 1024 | // The worker has already removed its clone. Dropping the controller's |
| 1025 | // final old host now releases both the relay scope and native store |
| 1026 | // owner locks before the fresh immutable configuration is opened. |
| 1027 | drop(self.runtime_chat.take()); |
| 1028 | let host = RuntimeChatRelayHost::open( |
| 1029 | requested.config, |
| 1030 | requested.plugin_registry, |
| 1031 | requested.private_root, |
| 1032 | requested.target_ref, |
| 1033 | requested.session_id, |
| 1034 | )?; |
| 1035 | if let (Some(account_ref), Some(target_ref)) = |
| 1036 | (self.account_ref.as_deref(), self.target_ref.as_deref()) |
| 1037 | { |
| 1038 | host.bind_account(account_ref, target_ref)?; |
| 1039 | } |
| 1040 | if let Some(run_id) = self.attached_run_id.as_deref() { |
| 1041 | host.authorize_run(run_id)?; |
| 1042 | } |
| 1043 | self.runtime_chat = Some(host.clone()); |
| 1044 | self.runtime_chat_refresh_release_requested = false; |
| 1045 | self.uploaded_runtime_catalog_receipt.clear(); |
| 1046 | self.worker_tx |
| 1047 | .as_ref() |
| 1048 | .ok_or_else(|| { |
| 1049 | "Runtime Chat could not install its refreshed provider configuration.".to_string() |
| 1050 | })? |
| 1051 | .send(WorkerCommand::InstallRuntimeChatHost(host)) |
| 1052 | .map_err(|_| { |
| 1053 | "Runtime Chat could not install its refreshed provider configuration.".to_string() |
| 1054 | })?; |
| 1055 | if let Some(attachment) = self.runtime_chat_attachment.clone() { |
| 1056 | self.upload_runtime_chat_catalog(&attachment)?; |
| 1057 | } |
| 1058 | self.status_detail = "web mirror connected".to_string(); |
| 1059 | Ok(()) |
| 1060 | } |
| 1061 | |
| 1062 | pub fn start(&mut self, start: RemoteStart) -> Result<(), String> { |
| 1063 | if matches!( |
| 1064 | self.status, |
| 1065 | Status::Connecting | Status::Connected | Status::Stopping |
| 1066 | ) { |
| 1067 | return Err("Remote control is already active.".to_string()); |
| 1068 | } |
| 1069 | if self.status == Status::Failed |
| 1070 | && self |
| 1071 | .ownership_blocked_until |
| 1072 | .is_some_and(|deadline| Instant::now() < deadline) |
| 1073 | { |
| 1074 | return Err( |
| 1075 | "The previous remote lease may still be active; wait for ownership to return before reconnecting." |
| 1076 | .to_string(), |
| 1077 | ); |
| 1078 | } |
| 1079 | if self.active_run.is_some() || self.pending_local_turn_run.is_some() { |
| 1080 | return Err( |
| 1081 | "The local Work turn is still running; wait for its terminal receipt before reconnecting remote control." |
| 1082 | .to_string(), |
| 1083 | ); |
| 1084 | } |
| 1085 | if self.has_unacknowledged_integrity_events() |
| 1086 | && self |
| 1087 | .runtime_chat |
| 1088 | .as_ref() |
| 1089 | .is_none_or(|host| !host.scope_matches(&start.target_ref, &start.session_id)) |
| 1090 | { |
| 1091 | return Err( |
| 1092 | "The previous session still has unacknowledged terminal events; reconnect that exact session first." |
| 1093 | .to_string(), |
| 1094 | ); |
| 1095 | } |
| 1096 | if !valid_runtime_version(&start.runtime_version) |
| 1097 | || !valid_runtime_commit(&start.runtime_commit) |
| 1098 | || !valid_opaque_ref(&start.target_ref) |
| 1099 | || !valid_session_ref(&start.session_id) |
| 1100 | { |
| 1101 | return Err("This build or session does not have an enrollable identity.".to_string()); |
| 1102 | } |
| 1103 | if !self.pending_runtime_events.is_empty() |
| 1104 | && let Some(journal) = &self.journal |
| 1105 | { |
| 1106 | // A prior send-first fault is impossible now, but a disk failure |
| 1107 | // can leave a classic Work terminal retained only in live memory. |
| 1108 | // Repair that exact pending set before loading/replacing the same |
| 1109 | // session journal; never let reconnect clear the sole copy. |
| 1110 | journal.persist(&self.pending_runtime_events)?; |
| 1111 | } |
| 1112 | match &start.journal_dir { |
| 1113 | Some(dir) => self.prepare_remote_control_session_journal( |
| 1114 | dir, |
| 1115 | &start.target_ref, |
| 1116 | &start.session_id, |
| 1117 | )?, |
| 1118 | None => self.journal = None, |
| 1119 | } |
| 1120 | self.recover_classic_lease_before_worker()?; |
| 1121 | self.recover_runtime_chat_ownership_before_worker()?; |
| 1122 | let (worker_tx, worker_rx) = mpsc::unbounded_channel(); |
| 1123 | let (event_tx, event_rx) = mpsc::unbounded_channel(); |
| 1124 | self.stop_worker(); |
| 1125 | self.status = Status::Connecting; |
| 1126 | self.status_detail = "waiting for account authorization".to_string(); |
| 1127 | self.target_ref = Some(start.target_ref.clone()); |
| 1128 | self.attached_run_id = None; |
| 1129 | self.attached_workspace_id = None; |
| 1130 | self.runtime_chat_attachment = None; |
| 1131 | self.uploaded_runtime_catalog_receipt.clear(); |
| 1132 | self.active_run = None; |
| 1133 | self.pending_local_turn_run = None; |
| 1134 | self.worker_tx = Some(worker_tx); |
| 1135 | self.event_rx = Some(event_rx); |
| 1136 | let runtime_chat = self.runtime_chat.clone(); |
| 1137 | self.worker = Some(tokio::spawn(async move { |
| 1138 | let mut phase = RelayPhase::Enrolling; |
| 1139 | if let Err(error) = |
| 1140 | relay_worker(start, runtime_chat, worker_rx, event_tx.clone(), &mut phase).await |
| 1141 | { |
| 1142 | let _ = if phase.lease_confirmed() { |
| 1143 | event_tx.send(RemoteEvent::Failed(error)) |
| 1144 | } else { |
| 1145 | event_tx.send(RemoteEvent::FailedPreLease(error)) |
| 1146 | }; |
| 1147 | } |
| 1148 | })); |
| 1149 | Ok(()) |
| 1150 | } |
| 1151 | |
| 1152 | fn recover_runtime_chat_ownership_before_worker(&self) -> Result<(), String> { |
| 1153 | let Some(host) = &self.runtime_chat else { |
| 1154 | return Ok(()); |
| 1155 | }; |
| 1156 | if host.has_any_unsettled_turns() || self.has_unacknowledged_integrity_events() { |
| 1157 | // A prior process may have stopped either while native inference |
| 1158 | // was still running or after terminal projection was durable but |
| 1159 | // before CWC acknowledged its account-delivery envelope. Reclaim |
| 1160 | // the exclusive attached-run writer synchronously before worker |
| 1161 | // enrollment (and therefore before any autonomous provider work) |
| 1162 | // can begin. |
| 1163 | host.recover_inference_ownership_for_pending_delivery()?; |
| 1164 | } |
| 1165 | Ok(()) |
| 1166 | } |
| 1167 | |
| 1168 | fn recover_classic_lease_before_worker(&mut self) -> Result<(), String> { |
| 1169 | let Some(lease) = self |
| 1170 | .journal |
| 1171 | .as_ref() |
| 1172 | .and_then(RuntimeEventJournal::classic_lease) |
| 1173 | else { |
| 1174 | return Ok(()); |
| 1175 | }; |
| 1176 | self.event_seq |
| 1177 | .entry(lease.run_id.clone()) |
| 1178 | .and_modify(|known| *known = (*known).max(lease.seq_floor)) |
| 1179 | .or_insert(lease.seq_floor); |
| 1180 | let recovery_turn_id = lease |
| 1181 | .turn_id |
| 1182 | .clone() |
| 1183 | .unwrap_or_else(|| classic_recovery_turn_id(&lease.run_id, &lease.lease_id)); |
| 1184 | let already_terminal = |
| 1185 | self.pending_runtime_events |
| 1186 | .get(&lease.run_id) |
| 1187 | .is_some_and(|events| { |
| 1188 | events.iter().any(|(seq, entry)| { |
| 1189 | *seq > lease.seq_floor |
| 1190 | && runtime_envelope_event(&entry.envelope) == Some("turn.completed") |
| 1191 | && entry.envelope.get("turn_id").and_then(Value::as_str) |
| 1192 | == Some(recovery_turn_id.as_str()) |
| 1193 | }) |
| 1194 | }); |
| 1195 | if !already_terminal { |
| 1196 | let seq = self.next_runtime_seq(&lease.run_id); |
| 1197 | let envelope = runtime_envelope( |
| 1198 | seq, |
| 1199 | "turn.completed", |
| 1200 | Some(&recovery_turn_id), |
| 1201 | chrono::Utc::now().to_rfc3339(), |
| 1202 | json!({ |
| 1203 | "turn": { |
| 1204 | "status": "failed", |
| 1205 | "usage": {}, |
| 1206 | "error": "The local Runtime restarted before terminal delivery." |
| 1207 | } |
| 1208 | }), |
| 1209 | ); |
| 1210 | if !self.queue_runtime_envelope(&lease.run_id, envelope) { |
| 1211 | return Err( |
| 1212 | "Remote control could not recover its interrupted account turn.".to_string(), |
| 1213 | ); |
| 1214 | } |
| 1215 | } |
| 1216 | self.journal |
| 1217 | .as_ref() |
| 1218 | .ok_or_else(|| "Remote control lost its delivery journal during recovery.".to_string())? |
| 1219 | .set_classic_lease(None, &self.pending_runtime_events)?; |
| 1220 | Ok(()) |
| 1221 | } |
| 1222 | |
| 1223 | /// Why `/rc stop` must currently be refused, if any reason exists. |
| 1224 | /// |
| 1225 | /// Stopping is only safe once no remote turn is active and every |
| 1226 | /// integrity-critical envelope (terminal turn state, approvals, failures, |
| 1227 | /// resynchronization snapshots) is behind the server-confirmed cursor. |
| 1228 | pub fn stop_refusal(&self) -> Option<String> { |
| 1229 | if self.has_active_run() { |
| 1230 | return Some( |
| 1231 | "Finish or interrupt the active remote turn before stopping remote control." |
| 1232 | .to_string(), |
| 1233 | ); |
| 1234 | } |
| 1235 | if self |
| 1236 | .runtime_chat |
| 1237 | .as_ref() |
| 1238 | .is_some_and(RuntimeChatRelayHost::has_any_unsettled_turns) |
| 1239 | { |
| 1240 | return Some( |
| 1241 | "Finish or interrupt the active Runtime Chat turn before stopping remote control." |
| 1242 | .to_string(), |
| 1243 | ); |
| 1244 | } |
| 1245 | if self.has_unacknowledged_integrity_events() { |
| 1246 | return Some( |
| 1247 | "The server has not yet acknowledged this session's terminal or approval events; try /rc stop again in a moment." |
| 1248 | .to_string(), |
| 1249 | ); |
| 1250 | } |
| 1251 | None |
| 1252 | } |
| 1253 | |
| 1254 | fn has_unacknowledged_integrity_events(&self) -> bool { |
| 1255 | self.pending_runtime_events |
| 1256 | .values() |
| 1257 | .flat_map(BTreeMap::values) |
| 1258 | .any(|entry| entry.integrity) |
| 1259 | } |
| 1260 | |
| 1261 | pub(crate) fn runtime_chat_blocks_local_dispatch(&self) -> bool { |
| 1262 | self.has_durable_classic_lease() |
| 1263 | || self.pending_runtime_chat_configuration.is_some() |
| 1264 | || self.runtime_chat.as_ref().is_some_and(|host| { |
| 1265 | host.has_any_unsettled_turns() || self.has_unacknowledged_integrity_events() |
| 1266 | }) |
| 1267 | } |
| 1268 | |
| 1269 | #[cfg(test)] |
| 1270 | pub(crate) fn block_runtime_chat_dispatch_for_tests(&mut self) { |
| 1271 | self.pending_runtime_chat_configuration = Some(PendingRuntimeChatConfiguration { |
| 1272 | config: crate::config::Config::default(), |
| 1273 | plugin_registry: std::sync::Arc::new(crate::plugins::PluginRegistry::empty(Path::new( |
| 1274 | ".", |
| 1275 | ))), |
| 1276 | private_root: PathBuf::from("runtime-chat-test-block"), |
| 1277 | target_ref: "target_test_block".to_string(), |
| 1278 | session_id: "session_test_block".to_string(), |
| 1279 | }); |
| 1280 | } |
| 1281 | |
| 1282 | fn attachment_switch_refusal(&self, next_run_id: &str) -> Option<String> { |
| 1283 | let current_run_id = self.attached_run_id.as_deref()?; |
| 1284 | if current_run_id == next_run_id { |
| 1285 | return None; |
| 1286 | } |
| 1287 | if self.has_active_run() { |
| 1288 | return Some( |
| 1289 | "Finish the active Work turn before attaching another account run.".to_string(), |
| 1290 | ); |
| 1291 | } |
| 1292 | let has_unacknowledged_terminal = self |
| 1293 | .pending_runtime_events |
| 1294 | .get(current_run_id) |
| 1295 | .is_some_and(|events| events.values().any(|entry| entry.integrity)); |
| 1296 | has_unacknowledged_terminal.then(|| { |
| 1297 | "The current Runtime Chat turn must be acknowledged before attaching another run." |
| 1298 | .to_string() |
| 1299 | }) |
| 1300 | } |
| 1301 | |
| 1302 | pub fn stop(&mut self) { |
| 1303 | if let Some(refusal) = self.stop_refusal() { |
| 1304 | self.status_detail = refusal; |
| 1305 | return; |
| 1306 | } |
| 1307 | if self.status == Status::Connecting { |
| 1308 | // The worker may have completed its server-side connect just before |
| 1309 | // the UI consumed RemoteEvent::Connected. Aborting it cannot prove |
| 1310 | // that no lease exists, so retain the ownership lock through the |
| 1311 | // server expiry instead of returning local input immediately. |
| 1312 | self.stop_worker(); |
| 1313 | self.status = Status::Failed; |
| 1314 | self.status_detail = |
| 1315 | "authorization cancelled; waiting for any server lease to expire safely" |
| 1316 | .to_string(); |
| 1317 | self.ownership_blocked_until = Some(Instant::now() + OWNERSHIP_LOCK_AFTER_FAILURE); |
| 1318 | } else if self.status == Status::Connected { |
| 1319 | // Hand every deferred delta to the transport first so the worker's |
| 1320 | // pre-heartbeat drain covers the complete unacknowledged set. |
| 1321 | self.hand_off_all_deferred(); |
| 1322 | let queued = self |
| 1323 | .worker_tx |
| 1324 | .as_ref() |
| 1325 | .is_some_and(|tx| tx.send(WorkerCommand::Stop).is_ok()); |
| 1326 | self.worker_tx = None; |
| 1327 | if queued { |
| 1328 | self.status = Status::Stopping; |
| 1329 | self.status_detail = "confirming the runner is offline".to_string(); |
| 1330 | } else { |
| 1331 | self.status = Status::Failed; |
| 1332 | self.status_detail = |
| 1333 | "waiting for the last server lease to expire safely".to_string(); |
| 1334 | self.ownership_blocked_until = Some(Instant::now() + OWNERSHIP_LOCK_AFTER_FAILURE); |
| 1335 | } |
| 1336 | } |
| 1337 | if self.status == Status::Failed && self.ownership_blocked_until.is_none() { |
| 1338 | // A pre-lease failure holds no lease; stopping is an ordinary |
| 1339 | // reset, not a drain confirmation. |
| 1340 | self.status = Status::Off; |
| 1341 | } |
| 1342 | if self.status == Status::Off { |
| 1343 | self.account_ref = None; |
| 1344 | self.links = RemoteLinks::default(); |
| 1345 | self.attached_run_id = None; |
| 1346 | self.attached_workspace_id = None; |
| 1347 | self.runtime_chat_attachment = None; |
| 1348 | self.active_run = None; |
| 1349 | self.pending_local_turn_run = None; |
| 1350 | self.pending_approvals.clear(); |
| 1351 | self.command_fingerprints.clear(); |
| 1352 | self.ownership_blocked_until = None; |
| 1353 | } |
| 1354 | } |
| 1355 | |
| 1356 | fn stop_worker(&mut self) { |
| 1357 | if let Some(worker) = self.worker.take() { |
| 1358 | worker.abort(); |
| 1359 | } |
| 1360 | if let Some(host) = &self.runtime_chat { |
| 1361 | // Projection claims are only an in-process handoff lease between |
| 1362 | // the worker and this controller. Aborting the worker or dropping |
| 1363 | // its receiver must make every unjournaled native event eligible |
| 1364 | // for replay on the same durable host. |
| 1365 | host.release_all_projection_claims(); |
| 1366 | } |
| 1367 | self.worker_tx = None; |
| 1368 | self.event_rx = None; |
| 1369 | self.runtime_chat_refresh_release_requested = false; |
| 1370 | } |
| 1371 | |
| 1372 | fn quarantine_rejected_authority(&mut self, error: String) -> RemoteEvent { |
| 1373 | // Once an attachment fails its account/run/catalog authority checks, |
| 1374 | // no later event from that worker can be trusted. In particular, the |
| 1375 | // classic Prompt and Approval command paths predate Runtime Chat and |
| 1376 | // must not keep consuming commands from a rejected re-enrollment. |
| 1377 | self.stop_worker(); |
| 1378 | self.status = Status::Failed; |
| 1379 | self.status_detail = error.clone(); |
| 1380 | self.ownership_blocked_until = Some(Instant::now() + OWNERSHIP_LOCK_AFTER_FAILURE); |
| 1381 | RemoteEvent::Failed(error) |
| 1382 | } |
| 1383 | |
| 1384 | pub fn try_next_event(&mut self) -> Option<RemoteEvent> { |
| 1385 | // Coalescing ends at the next UI poll: hand any deferred delta to the |
| 1386 | // transport so live viewers never wait more than one tick. |
| 1387 | self.hand_off_all_deferred(); |
| 1388 | if self.status == Status::Failed |
| 1389 | && self |
| 1390 | .ownership_blocked_until |
| 1391 | .is_some_and(|deadline| Instant::now() >= deadline) |
| 1392 | { |
| 1393 | if self.has_active_run() |
| 1394 | || self.has_unacknowledged_integrity_events() |
| 1395 | || self |
| 1396 | .runtime_chat |
| 1397 | .as_ref() |
| 1398 | .is_some_and(RuntimeChatRelayHost::has_any_unsettled_turns) |
| 1399 | { |
| 1400 | // The server lease may have expired, but provider-backed work |
| 1401 | // still owns this private Runtime store. Keep the host/scope |
| 1402 | // lock and allow only a same-session reconnect to resume |
| 1403 | // projection/drain; never transition Off and reopen a second |
| 1404 | // manager over the active store. |
| 1405 | self.stop_worker(); |
| 1406 | self.status_detail = |
| 1407 | "this session still has unsettled or unacknowledged Runtime Chat work; reconnect it to drain" |
| 1408 | .to_string(); |
| 1409 | self.ownership_blocked_until = None; |
| 1410 | return None; |
| 1411 | } |
| 1412 | let approvals = self |
| 1413 | .pending_approvals |
| 1414 | .drain() |
| 1415 | .map(|(_, value)| value) |
| 1416 | .collect(); |
| 1417 | self.stop_worker(); |
| 1418 | self.status = Status::Off; |
| 1419 | self.status_detail = "off".to_string(); |
| 1420 | self.ownership_blocked_until = None; |
| 1421 | return Some(RemoteEvent::OwnershipRestored { approvals }); |
| 1422 | } |
| 1423 | let event = self.event_rx.as_mut()?.try_recv().ok()?; |
| 1424 | match &event { |
| 1425 | RemoteEvent::Connected { |
| 1426 | account_ref, |
| 1427 | target_ref, |
| 1428 | attachment, |
| 1429 | links, |
| 1430 | .. |
| 1431 | } => { |
| 1432 | if let Some(error) = self.attachment_switch_refusal(&attachment.run_id) { |
| 1433 | return Some(self.quarantine_rejected_authority(error)); |
| 1434 | } |
| 1435 | if let Some(host) = &self.runtime_chat |
| 1436 | && let Err(error) = host |
| 1437 | .bind_account(account_ref, target_ref) |
| 1438 | .and_then(|()| host.authorize_run(&attachment.run_id)) |
| 1439 | { |
| 1440 | return Some(self.quarantine_rejected_authority(error)); |
| 1441 | } |
| 1442 | self.apply_attachment(attachment); |
| 1443 | if self.pending_runtime_chat_configuration.is_none() |
| 1444 | && self.runtime_chat.is_some() |
| 1445 | && let Err(error) = self.upload_runtime_chat_catalog(attachment) |
| 1446 | { |
| 1447 | return Some(self.quarantine_rejected_authority(error)); |
| 1448 | } |
| 1449 | self.links = links.clone(); |
| 1450 | // Journal recovery may hold unacknowledged envelopes for runs |
| 1451 | // beyond this attachment; resend every pending run now. |
| 1452 | self.flush_all_pending(); |
| 1453 | self.status = Status::Connected; |
| 1454 | self.status_detail = "web mirror connected".to_string(); |
| 1455 | self.ownership_blocked_until = None; |
| 1456 | self.account_ref = Some(account_ref.clone()); |
| 1457 | self.target_ref = Some(target_ref.clone()); |
| 1458 | if let Err(error) = self.schedule_runtime_chat_refresh_if_ready() { |
| 1459 | return Some(self.quarantine_rejected_authority(error)); |
| 1460 | } |
| 1461 | } |
| 1462 | RemoteEvent::Attachment { |
| 1463 | account_ref, |
| 1464 | target_ref, |
| 1465 | attachment, |
| 1466 | links, |
| 1467 | } => { |
| 1468 | if let Some(error) = self.attachment_switch_refusal(&attachment.run_id) { |
| 1469 | return Some(self.quarantine_rejected_authority(error)); |
| 1470 | } |
| 1471 | if let Some(host) = &self.runtime_chat |
| 1472 | && let Err(error) = host |
| 1473 | .bind_account(account_ref, target_ref) |
| 1474 | .and_then(|()| host.authorize_run(&attachment.run_id)) |
| 1475 | { |
| 1476 | return Some(self.quarantine_rejected_authority(error)); |
| 1477 | } |
| 1478 | self.apply_attachment(attachment); |
| 1479 | if self.pending_runtime_chat_configuration.is_none() |
| 1480 | && self.runtime_chat.is_some() |
| 1481 | && let Err(error) = self.upload_runtime_chat_catalog(attachment) |
| 1482 | { |
| 1483 | return Some(self.quarantine_rejected_authority(error)); |
| 1484 | } |
| 1485 | self.links = links.clone(); |
| 1486 | self.account_ref = Some(account_ref.clone()); |
| 1487 | self.target_ref = Some(target_ref.clone()); |
| 1488 | if let Err(error) = self.schedule_runtime_chat_refresh_if_ready() { |
| 1489 | return Some(self.quarantine_rejected_authority(error)); |
| 1490 | } |
| 1491 | } |
| 1492 | RemoteEvent::RuntimeCursor { run_id, cursor } => { |
| 1493 | self.reconcile_runtime_cursor(run_id, *cursor); |
| 1494 | if let Err(error) = self.schedule_runtime_chat_refresh_if_ready() { |
| 1495 | return Some(self.quarantine_rejected_authority(error)); |
| 1496 | } |
| 1497 | } |
| 1498 | RemoteEvent::RuntimeChatHostReleased => { |
| 1499 | if let Err(error) = self.complete_runtime_chat_refresh() { |
| 1500 | return Some(self.quarantine_rejected_authority(error)); |
| 1501 | } |
| 1502 | } |
| 1503 | RemoteEvent::RuntimeChatProjection(projection) => { |
| 1504 | if let Err(error) = self.upload_runtime_chat_projection(projection) { |
| 1505 | if let Some(host) = &self.runtime_chat { |
| 1506 | host.release_projection( |
| 1507 | &projection.native_thread_id, |
| 1508 | projection.native_seq, |
| 1509 | ); |
| 1510 | } |
| 1511 | self.stop_worker(); |
| 1512 | self.status = Status::Failed; |
| 1513 | self.status_detail = error.clone(); |
| 1514 | self.ownership_blocked_until = |
| 1515 | Some(Instant::now() + OWNERSHIP_LOCK_AFTER_FAILURE); |
| 1516 | return Some(RemoteEvent::Failed(error)); |
| 1517 | } |
| 1518 | } |
| 1519 | RemoteEvent::FailedPreLease(reason) => { |
| 1520 | // No server-confirmed lease ever existed, so nothing is |
| 1521 | // locked: no reconnect blackout, no approval handoff to |
| 1522 | // undo, and `/rc` can retry immediately. |
| 1523 | self.status = Status::Failed; |
| 1524 | self.status_detail = reason.clone(); |
| 1525 | self.ownership_blocked_until = None; |
| 1526 | if !self.has_durable_classic_lease() { |
| 1527 | self.active_run = None; |
| 1528 | self.pending_local_turn_run = None; |
| 1529 | } |
| 1530 | } |
| 1531 | RemoteEvent::Failed(reason) => { |
| 1532 | self.status = Status::Failed; |
| 1533 | self.status_detail = |
| 1534 | format!("{reason}; waiting for the last server lease to expire safely"); |
| 1535 | self.ownership_blocked_until = Some(Instant::now() + OWNERSHIP_LOCK_AFTER_FAILURE); |
| 1536 | // Exact unacknowledged runtime envelopes remain owned by this |
| 1537 | // controller (and its journal). Keep any classic provider-turn |
| 1538 | // association live as well: the engine can still deliver its |
| 1539 | // exact terminal status/usage after the transport fails. |
| 1540 | } |
| 1541 | RemoteEvent::Stopped => { |
| 1542 | self.status = Status::Off; |
| 1543 | self.status_detail = "off".to_string(); |
| 1544 | self.active_run = None; |
| 1545 | self.pending_local_turn_run = None; |
| 1546 | self.attached_run_id = None; |
| 1547 | self.attached_workspace_id = None; |
| 1548 | self.runtime_chat_attachment = None; |
| 1549 | self.links = RemoteLinks::default(); |
| 1550 | self.ownership_blocked_until = None; |
| 1551 | // The worker only reports Stopped after draining through the |
| 1552 | // server-confirmed cursor and posting the offline heartbeat, |
| 1553 | // so an empty pending set means the journal is spent. |
| 1554 | if self.pending_runtime_events.values().all(BTreeMap::is_empty) |
| 1555 | && let Some(journal) = &self.journal |
| 1556 | { |
| 1557 | journal.remove(); |
| 1558 | } |
| 1559 | if !self.pending_approvals.is_empty() { |
| 1560 | let approvals = self |
| 1561 | .pending_approvals |
| 1562 | .drain() |
| 1563 | .map(|(_, value)| value) |
| 1564 | .collect(); |
| 1565 | return Some(RemoteEvent::OwnershipRestored { approvals }); |
| 1566 | } |
| 1567 | } |
| 1568 | RemoteEvent::Notice(_) |
| 1569 | | RemoteEvent::Command { .. } |
| 1570 | | RemoteEvent::OwnershipRestored { .. } => {} |
| 1571 | } |
| 1572 | Some(event) |
| 1573 | } |
| 1574 | |
| 1575 | /// The validated web link for the live session, once the control plane |
| 1576 | /// has advertised one. |
| 1577 | pub fn run_url(&self) -> Option<&str> { |
| 1578 | if matches!(self.status, Status::Connected | Status::Stopping) { |
| 1579 | self.links.run_url.as_deref() |
| 1580 | } else { |
| 1581 | None |
| 1582 | } |
| 1583 | } |
| 1584 | |
| 1585 | /// The validated web link for this computer's settings row, if advertised. |
| 1586 | pub fn computer_url(&self) -> Option<&str> { |
| 1587 | if matches!(self.status, Status::Connected | Status::Stopping) { |
| 1588 | self.links.computer_url.as_deref() |
| 1589 | } else { |
| 1590 | None |
| 1591 | } |
| 1592 | } |
| 1593 | |
| 1594 | /// One word for the posture bar's right slot, `None` while remote |
| 1595 | /// control is off so the bar paints nothing rather than "off". |
| 1596 | #[must_use] |
| 1597 | pub fn status_word(&self) -> Option<&'static str> { |
| 1598 | match self.status { |
| 1599 | Status::Off => None, |
| 1600 | Status::Connecting => Some("connecting"), |
| 1601 | Status::Connected => Some("connected"), |
| 1602 | Status::Stopping => Some("stopping"), |
| 1603 | Status::Failed => Some("failed"), |
| 1604 | } |
| 1605 | } |
| 1606 | |
| 1607 | /// Test-only: advertise a validated live session link without a control |
| 1608 | /// plane. Mirrors the shape the runner-lease parser installs (`Connected` |
| 1609 | /// plus validated links) so `/rc link`/`/rc open` routing can be exercised |
| 1610 | /// deterministically offline. |
| 1611 | #[cfg(test)] |
| 1612 | pub(crate) fn install_live_link_for_test(&mut self, run_url: &str, computer_url: Option<&str>) { |
| 1613 | self.status = Status::Connected; |
| 1614 | self.status_detail = "test link".to_string(); |
| 1615 | self.links = RemoteLinks { |
| 1616 | run_url: Some(run_url.to_string()), |
| 1617 | computer_url: computer_url.map(str::to_string), |
| 1618 | }; |
| 1619 | } |
| 1620 | |
| 1621 | pub fn status_line(&self) -> String { |
| 1622 | match self.status { |
| 1623 | Status::Off => "Remote control: off".to_string(), |
| 1624 | Status::Connecting => format!("Remote control: connecting · {}", self.status_detail), |
| 1625 | Status::Connected => match self.links.run_url.as_deref() { |
| 1626 | Some(url) => format!( |
| 1627 | "Remote control: connected · account {} · {} · open {url}", |
| 1628 | self.account_ref.as_deref().unwrap_or("account"), |
| 1629 | self.status_detail |
| 1630 | ), |
| 1631 | None => format!( |
| 1632 | "Remote control: connected · account {} · {}", |
| 1633 | self.account_ref.as_deref().unwrap_or("account"), |
| 1634 | self.status_detail |
| 1635 | ), |
| 1636 | }, |
| 1637 | Status::Stopping => { |
| 1638 | "Remote control: stopping · confirming the runner is offline".to_string() |
| 1639 | } |
| 1640 | Status::Failed => { |
| 1641 | if self |
| 1642 | .ownership_blocked_until |
| 1643 | .is_some_and(|deadline| Instant::now() < deadline) |
| 1644 | { |
| 1645 | format!( |
| 1646 | "Remote control: lost after connecting · {} · reconnect waits for the server lease to drain", |
| 1647 | self.status_detail |
| 1648 | ) |
| 1649 | } else { |
| 1650 | format!( |
| 1651 | "Remote control: failed before connecting · {} · /rc to retry", |
| 1652 | self.status_detail |
| 1653 | ) |
| 1654 | } |
| 1655 | } |
| 1656 | } |
| 1657 | } |
| 1658 | |
| 1659 | /// Whether the web mirror can carry an approval decision right now. |
| 1660 | /// |
| 1661 | /// `Connecting` deliberately does not qualify: there is not yet an |
| 1662 | /// attachment/run cursor able to carry a typed approval, so the local |
| 1663 | /// card stays the only actionable surface until `Connected`. A |
| 1664 | /// transport failure also disqualifies — a dead relay cannot deliver a |
| 1665 | /// decision, and the local card remains the source of truth either way. |
| 1666 | /// |
| 1667 | /// Mirror semantics: this never gates *local* input. It only decides |
| 1668 | /// whether the approval card is *also* shared with the web. |
| 1669 | pub fn can_share_approval_with_web(&self) -> bool { |
| 1670 | let attached = match self.status { |
| 1671 | // A connection alone is not enough. Until a concrete typed turn |
| 1672 | // id is bound, `record_remote_approval` has nowhere safe to send |
| 1673 | // a decision, so the card stays local-only. |
| 1674 | Status::Connected | Status::Stopping => self.active_run.is_some(), |
| 1675 | Status::Off | Status::Connecting | Status::Failed => false, |
| 1676 | }; |
| 1677 | attached && !self.applying_remote_command |
| 1678 | } |
| 1679 | |
| 1680 | /// Record that the *local* surface answered an approval, so a late web |
| 1681 | /// decision for the same tool is acknowledged as "no longer pending" |
| 1682 | /// instead of double-answering the engine. First decision wins; the |
| 1683 | /// other surface is told. |
| 1684 | pub fn resolve_pending_approval(&mut self, tool_id: &str, approved: bool) -> bool { |
| 1685 | let gate = projected_approval_id(tool_id); |
| 1686 | if self.pending_approvals.remove(&gate).is_none() { |
| 1687 | return false; |
| 1688 | } |
| 1689 | if let Some(active) = self.active_run.clone() { |
| 1690 | self.upload_envelope( |
| 1691 | &active.run_id, |
| 1692 | "approval.resolved", |
| 1693 | Some(&active.turn_id), |
| 1694 | json!({ |
| 1695 | "id": gate, |
| 1696 | "approval_id": gate, |
| 1697 | "decision": if approved { "approved" } else { "denied" }, |
| 1698 | "decided_by": "terminal", |
| 1699 | }), |
| 1700 | ); |
| 1701 | } |
| 1702 | true |
| 1703 | } |
| 1704 | |
| 1705 | pub fn set_applying_remote_command(&mut self, value: bool) { |
| 1706 | self.applying_remote_command = value; |
| 1707 | } |
| 1708 | |
| 1709 | /// Test-only: put the controller into the exact state a live connected |
| 1710 | /// mirror with an attached run would be in, without a relay worker. |
| 1711 | #[cfg(test)] |
| 1712 | pub(crate) fn force_mirror_connected_for_tests(&mut self, run_id: &str, turn_id: &str) { |
| 1713 | self.status = Status::Connected; |
| 1714 | self.status_detail = "web mirror connected".to_string(); |
| 1715 | self.attached_run_id = Some(run_id.to_string()); |
| 1716 | self.active_run = Some(ActiveRelayRun { |
| 1717 | run_id: run_id.to_string(), |
| 1718 | turn_id: turn_id.to_string(), |
| 1719 | }); |
| 1720 | } |
| 1721 | |
| 1722 | #[cfg(test)] |
| 1723 | pub(crate) fn queue_remote_event_for_tests(&mut self, event: RemoteEvent) { |
| 1724 | let (event_tx, event_rx) = mpsc::unbounded_channel(); |
| 1725 | event_tx.send(event).expect("test event receiver"); |
| 1726 | self.event_rx = Some(event_rx); |
| 1727 | if self.worker_tx.is_none() { |
| 1728 | let (worker_tx, _worker_rx) = mpsc::unbounded_channel(); |
| 1729 | self.worker_tx = Some(worker_tx); |
| 1730 | } |
| 1731 | } |
| 1732 | |
| 1733 | pub fn claim_command( |
| 1734 | &mut self, |
| 1735 | run_id: &str, |
| 1736 | seq: u64, |
| 1737 | command: &RemoteCommand, |
| 1738 | ) -> Result<bool, String> { |
| 1739 | if self.status != Status::Connected || self.attached_run_id.as_deref() != Some(run_id) { |
| 1740 | return Err( |
| 1741 | "The remote command does not belong to the current authorized attachment." |
| 1742 | .to_string(), |
| 1743 | ); |
| 1744 | } |
| 1745 | let fingerprint = command_fingerprint(command); |
| 1746 | let key = (run_id.to_string(), seq); |
| 1747 | if let Some(existing) = self.command_fingerprints.get(&key) { |
| 1748 | if existing == &fingerprint { |
| 1749 | // Runtime Chat's native operation key is idempotent. The UI |
| 1750 | // bridge deliberately re-enters that path for an exact replay |
| 1751 | // so a failed acknowledgement POST is issued again; admission |
| 1752 | // gates below describe *new* work and must not false-fail it. |
| 1753 | return Ok(false); |
| 1754 | } |
| 1755 | return Err( |
| 1756 | "The control plane reused a command sequence with different content.".to_string(), |
| 1757 | ); |
| 1758 | } |
| 1759 | if let RemoteCommand::RuntimeChatPrompt(prompt) = command |
| 1760 | && let Some(host) = &self.runtime_chat |
| 1761 | && host.is_exact_prompt_replay(prompt)? |
| 1762 | { |
| 1763 | // The controller's command-sequence cache is intentionally |
| 1764 | // process-local, while the native operation/turn binding is |
| 1765 | // durable. After a crash or deferred provider-config refresh, |
| 1766 | // recognize that durable exact replay before new-work gates and |
| 1767 | // reissue its acknowledgement through the idempotent host path. |
| 1768 | self.command_fingerprints.insert(key, fingerprint); |
| 1769 | return Ok(false); |
| 1770 | } |
| 1771 | match command { |
| 1772 | RemoteCommand::Prompt { .. } | RemoteCommand::RuntimeChatPrompt(_) |
| 1773 | if self.pending_runtime_chat_configuration.is_some() => |
| 1774 | { |
| 1775 | return Err( |
| 1776 | "Wait for Runtime Chat to finish refreshing its local model configuration." |
| 1777 | .to_string(), |
| 1778 | ); |
| 1779 | } |
| 1780 | RemoteCommand::RuntimeChatPrompt(_) if self.has_active_run() => { |
| 1781 | return Err("Finish the active Work turn before starting Runtime Chat.".to_string()); |
| 1782 | } |
| 1783 | RemoteCommand::RuntimeChatPrompt(_) if self.has_unacknowledged_integrity_events() => { |
| 1784 | return Err( |
| 1785 | "Wait for the current Runtime Chat catalog or terminal event to be acknowledged before starting another turn." |
| 1786 | .to_string(), |
| 1787 | ); |
| 1788 | } |
| 1789 | RemoteCommand::Prompt { .. } |
| 1790 | if self |
| 1791 | .runtime_chat |
| 1792 | .as_ref() |
| 1793 | .is_some_and(RuntimeChatRelayHost::has_any_unsettled_turns) |
| 1794 | || self |
| 1795 | .pending_runtime_events |
| 1796 | .get(run_id) |
| 1797 | .is_some_and(|events| events.values().any(|entry| entry.integrity)) => |
| 1798 | { |
| 1799 | return Err( |
| 1800 | "Finish and acknowledge the active Runtime Chat turn before starting Work." |
| 1801 | .to_string(), |
| 1802 | ); |
| 1803 | } |
| 1804 | _ => {} |
| 1805 | } |
| 1806 | self.command_fingerprints.insert(key, fingerprint); |
| 1807 | Ok(true) |
| 1808 | } |
| 1809 | |
| 1810 | pub fn activate_prompt(&mut self, run_id: &str, turn_id: &str) -> Result<(), String> { |
| 1811 | self.activate_prompt_with_lease_id(run_id, turn_id, String::new()) |
| 1812 | } |
| 1813 | |
| 1814 | fn activate_prompt_with_lease_id( |
| 1815 | &mut self, |
| 1816 | run_id: &str, |
| 1817 | turn_id: &str, |
| 1818 | lease_id: String, |
| 1819 | ) -> Result<(), String> { |
| 1820 | let active = ActiveRelayRun { |
| 1821 | run_id: run_id.to_string(), |
| 1822 | turn_id: turn_id.to_string(), |
| 1823 | }; |
| 1824 | self.begin_classic_lease(ClassicRunLease { |
| 1825 | run_id: active.run_id.clone(), |
| 1826 | turn_id: Some(active.turn_id.clone()), |
| 1827 | lease_id, |
| 1828 | seq_floor: self.runtime_seq_floor(&active.run_id), |
| 1829 | })?; |
| 1830 | self.active_run = Some(active); |
| 1831 | Ok(()) |
| 1832 | } |
| 1833 | |
| 1834 | fn promote_pending_local_turn(&mut self, run_id: &str, turn_id: &str) -> Result<(), String> { |
| 1835 | let lease_id = self |
| 1836 | .journal |
| 1837 | .as_ref() |
| 1838 | .and_then(RuntimeEventJournal::classic_lease) |
| 1839 | .filter(|lease| lease.run_id == run_id && lease.turn_id.is_none()) |
| 1840 | .map(|lease| lease.lease_id) |
| 1841 | .ok_or_else(|| { |
| 1842 | "Remote control lost its durable pre-dispatch lease generation.".to_string() |
| 1843 | })?; |
| 1844 | self.activate_prompt_with_lease_id(run_id, turn_id, lease_id) |
| 1845 | } |
| 1846 | |
| 1847 | #[cfg(test)] |
| 1848 | pub fn active_run_matches(&self, run_id: &str) -> bool { |
| 1849 | self.active_run |
| 1850 | .as_ref() |
| 1851 | .is_some_and(|active| active.run_id == run_id) |
| 1852 | || self.pending_local_turn_run.as_deref() == Some(run_id) |
| 1853 | } |
| 1854 | |
| 1855 | /// Legacy Work cancellation is exact-turn scoped. A delayed command for |
| 1856 | /// an older turn in the same run must never cancel the current provider |
| 1857 | /// lifecycle. |
| 1858 | pub fn active_turn_matches(&self, run_id: &str, turn_id: &str) -> bool { |
| 1859 | self.active_run |
| 1860 | .as_ref() |
| 1861 | .is_some_and(|active| active.run_id == run_id && active.turn_id == turn_id) |
| 1862 | || self |
| 1863 | .journal |
| 1864 | .as_ref() |
| 1865 | .and_then(RuntimeEventJournal::classic_lease) |
| 1866 | .is_some_and(|lease| { |
| 1867 | lease.run_id == run_id && lease.turn_id.as_deref() == Some(turn_id) |
| 1868 | }) |
| 1869 | } |
| 1870 | |
| 1871 | /// A remotely-owned turn must reach a terminal engine event before the |
| 1872 | /// user can release the relay lease. Dropping the worker earlier would |
| 1873 | /// discard the run binding and strand the control-plane ledger while the |
| 1874 | /// local engine continued producing results. |
| 1875 | pub fn has_active_run(&self) -> bool { |
| 1876 | self.active_run.is_some() |
| 1877 | || self.pending_local_turn_run.is_some() |
| 1878 | || self.has_durable_classic_lease() |
| 1879 | } |
| 1880 | |
| 1881 | fn has_durable_classic_lease(&self) -> bool { |
| 1882 | self.journal |
| 1883 | .as_ref() |
| 1884 | .and_then(RuntimeEventJournal::classic_lease) |
| 1885 | .is_some() |
| 1886 | } |
| 1887 | |
| 1888 | /// Bind the server-confirmed attachment to the local turn that was |
| 1889 | /// already running when `/rc` was invoked. |
| 1890 | /// |
| 1891 | /// With a typed runtime turn id the binding is immediate. During the |
| 1892 | /// narrow dispatch-before-`TurnStarted` window, the run is parked and the |
| 1893 | /// first typed start event completes the binding. Replays are idempotent: |
| 1894 | /// an existing binding is never replaced and no runtime envelope is |
| 1895 | /// emitted by this method itself. |
| 1896 | pub fn attach_current_local_turn(&mut self, turn_id: Option<&str>) -> bool { |
| 1897 | if self.status != Status::Connected || self.has_active_run() { |
| 1898 | return false; |
| 1899 | } |
| 1900 | let Some(run_id) = self.attached_run_id.clone() else { |
| 1901 | return false; |
| 1902 | }; |
| 1903 | match turn_id.map(str::trim).filter(|turn_id| !turn_id.is_empty()) { |
| 1904 | Some(turn_id) => { |
| 1905 | if self.activate_prompt(&run_id, turn_id).is_err() { |
| 1906 | return false; |
| 1907 | } |
| 1908 | } |
| 1909 | None => { |
| 1910 | if self |
| 1911 | .begin_classic_lease(ClassicRunLease { |
| 1912 | run_id: run_id.clone(), |
| 1913 | turn_id: None, |
| 1914 | lease_id: String::new(), |
| 1915 | seq_floor: self.runtime_seq_floor(&run_id), |
| 1916 | }) |
| 1917 | .is_err() |
| 1918 | { |
| 1919 | return false; |
| 1920 | } |
| 1921 | self.pending_local_turn_run = Some(run_id); |
| 1922 | } |
| 1923 | } |
| 1924 | true |
| 1925 | } |
| 1926 | |
| 1927 | /// Release a dispatch-window binding when the local dispatcher becomes |
| 1928 | /// idle without ever producing a typed `TurnStarted`. The account-owned |
| 1929 | /// attachment remains connected and can accept a later web prompt; only |
| 1930 | /// the nonexistent turn lease is removed. |
| 1931 | pub fn release_unstarted_local_turn(&mut self) -> bool { |
| 1932 | if self.pending_local_turn_run.is_none() { |
| 1933 | return false; |
| 1934 | } |
| 1935 | if self.finish_classic_lease().is_err() { |
| 1936 | return false; |
| 1937 | } |
| 1938 | self.pending_local_turn_run = None; |
| 1939 | true |
| 1940 | } |
| 1941 | |
| 1942 | fn begin_classic_lease(&mut self, lease: ClassicRunLease) -> Result<(), String> { |
| 1943 | let mut lease = lease; |
| 1944 | lease.seq_floor = lease.seq_floor.max(self.runtime_seq_floor(&lease.run_id)); |
| 1945 | if let Some(previous) = self |
| 1946 | .journal |
| 1947 | .as_ref() |
| 1948 | .and_then(RuntimeEventJournal::classic_lease) |
| 1949 | .filter(|previous| previous.run_id == lease.run_id) |
| 1950 | { |
| 1951 | lease.seq_floor = lease.seq_floor.max(previous.seq_floor); |
| 1952 | } |
| 1953 | // A blank id always starts a fresh generation. The dispatch-window |
| 1954 | // promotion above is the only valid reuse and supplies its durable id |
| 1955 | // explicitly so a stale pre-dispatch lease cannot be inherited here. |
| 1956 | if lease.lease_id.is_empty() { |
| 1957 | lease.lease_id = format!("classic_lease_{}", uuid::Uuid::new_v4().simple()); |
| 1958 | } |
| 1959 | if !lease.valid() { |
| 1960 | return Err("The remote Work turn identity is invalid.".to_string()); |
| 1961 | } |
| 1962 | let Some(journal) = &self.journal else { |
| 1963 | #[cfg(test)] |
| 1964 | return Ok(()); |
| 1965 | #[cfg(not(test))] |
| 1966 | { |
| 1967 | self.status = Status::Failed; |
| 1968 | self.status_detail = |
| 1969 | "Remote control requires a durable account-delivery journal.".to_string(); |
| 1970 | return Err(self.status_detail.clone()); |
| 1971 | } |
| 1972 | }; |
| 1973 | if let Err(error) = journal.set_classic_lease(Some(lease), &self.pending_runtime_events) { |
| 1974 | self.status = Status::Failed; |
| 1975 | self.status_detail = error.clone(); |
| 1976 | self.ownership_blocked_until = Some(Instant::now() + OWNERSHIP_LOCK_AFTER_FAILURE); |
| 1977 | return Err(error); |
| 1978 | } |
| 1979 | Ok(()) |
| 1980 | } |
| 1981 | |
| 1982 | fn finish_classic_lease(&mut self) -> Result<(), String> { |
| 1983 | let Some(journal) = &self.journal else { |
| 1984 | #[cfg(test)] |
| 1985 | return Ok(()); |
| 1986 | #[cfg(not(test))] |
| 1987 | return Err("Remote control requires a durable account-delivery journal.".to_string()); |
| 1988 | }; |
| 1989 | if let Err(error) = journal.set_classic_lease(None, &self.pending_runtime_events) { |
| 1990 | self.status = Status::Failed; |
| 1991 | self.status_detail = error.clone(); |
| 1992 | self.ownership_blocked_until = Some(Instant::now() + OWNERSHIP_LOCK_AFTER_FAILURE); |
| 1993 | return Err(error); |
| 1994 | } |
| 1995 | Ok(()) |
| 1996 | } |
| 1997 | |
| 1998 | /// A remote prompt can fail during local route preparation before the |
| 1999 | /// engine owns a turn and therefore before it can emit `EngineEvent::Error`. |
| 2000 | /// That failure is still terminal for the account-owned run. |
| 2001 | pub fn fail_active_dispatch(&mut self, error: &str) { |
| 2002 | self.fail_active_run("dispatch_failed", error); |
| 2003 | } |
| 2004 | |
| 2005 | fn apply_attachment(&mut self, attachment: &RemoteAttachment) { |
| 2006 | self.attached_run_id = Some(attachment.run_id.clone()); |
| 2007 | self.attached_workspace_id = Some(attachment.workspace_id.clone()); |
| 2008 | self.runtime_chat_attachment = Some(attachment.clone()); |
| 2009 | self.reconcile_runtime_cursor(&attachment.run_id, attachment.runtime_cursor); |
| 2010 | let local_cursor = self |
| 2011 | .pending_runtime_events |
| 2012 | .get(&attachment.run_id) |
| 2013 | .and_then(|events| events.last_key_value().map(|(seq, _)| *seq)) |
| 2014 | .unwrap_or(0); |
| 2015 | let cursor = self.event_seq.entry(attachment.run_id.clone()).or_insert(0); |
| 2016 | *cursor = (*cursor).max(attachment.runtime_cursor).max(local_cursor); |
| 2017 | self.flush_pending_runtime_events(&attachment.run_id); |
| 2018 | // `snapshot_present` is server history, not proof that this freshly |
| 2019 | // loaded TUI process has uploaded its current saved history. The local |
| 2020 | // marker below prevents ordinary same-process reconnect duplication. |
| 2021 | } |
| 2022 | |
| 2023 | fn upload_runtime_chat_catalog(&mut self, attachment: &RemoteAttachment) -> Result<(), String> { |
| 2024 | if self.pending_runtime_chat_configuration.is_some() { |
| 2025 | return Err( |
| 2026 | "Runtime Chat must refresh its local model configuration before advertising a catalog." |
| 2027 | .to_string(), |
| 2028 | ); |
| 2029 | } |
| 2030 | let host = self |
| 2031 | .runtime_chat |
| 2032 | .clone() |
| 2033 | .ok_or_else(|| "This Runtime cannot advertise an isolated Chat relay.".to_string())?; |
| 2034 | let payload = host.catalog(&attachment.runtime_chat_relay_challenge)?; |
| 2035 | let payload_fingerprint = RuntimeChatRelayHost::catalog_payload_fingerprint(&payload)?; |
| 2036 | let receipt_key = format!( |
| 2037 | "{}:{}", |
| 2038 | attachment.runtime_chat_relay_challenge, payload_fingerprint |
| 2039 | ); |
| 2040 | if self |
| 2041 | .uploaded_runtime_catalog_receipt |
| 2042 | .get(&attachment.run_id) |
| 2043 | .is_some_and(|receipt| receipt == &receipt_key) |
| 2044 | { |
| 2045 | return Ok(()); |
| 2046 | } |
| 2047 | let source_event_id = |
| 2048 | runtime_chat_catalog_source_event_id(&attachment.run_id, &payload_fingerprint); |
| 2049 | if self.pending_runtime_chat_catalog_matches(&attachment.run_id, &source_event_id) { |
| 2050 | self.uploaded_runtime_catalog_receipt |
| 2051 | .insert(attachment.run_id.clone(), receipt_key); |
| 2052 | return Ok(()); |
| 2053 | } |
| 2054 | // A local receipt proves only that the catalog was journaled, not that |
| 2055 | // the server cursor acknowledged it. After restart, resend the stable |
| 2056 | // semantic source id whenever no matching pending entry remains. CWC |
| 2057 | // deduplicates an exact replay and rejects conflicting reuse, while a |
| 2058 | // changed safe catalog gets a different payload fingerprint/source id. |
| 2059 | let seq = self.next_runtime_seq(&attachment.run_id); |
| 2060 | let envelope = runtime_chat_envelope( |
| 2061 | seq, |
| 2062 | "runtime.catalog", |
| 2063 | None, |
| 2064 | None, |
| 2065 | Some(&source_event_id), |
| 2066 | RUNTIME_CHAT_CATALOG_TIMESTAMP.to_string(), |
| 2067 | payload, |
| 2068 | ); |
| 2069 | if !self.queue_runtime_chat_envelope(&attachment.run_id, envelope) { |
| 2070 | return Err("Runtime Chat could not journal its challenge-bound catalog.".to_string()); |
| 2071 | } |
| 2072 | self.uploaded_runtime_catalog_receipt |
| 2073 | .insert(attachment.run_id.clone(), receipt_key); |
| 2074 | Ok(()) |
| 2075 | } |
| 2076 | |
| 2077 | fn pending_runtime_chat_catalog_matches(&self, run_id: &str, source_event_id: &str) -> bool { |
| 2078 | self.pending_runtime_events |
| 2079 | .get(run_id) |
| 2080 | .is_some_and(|events| { |
| 2081 | events.values().any(|event| { |
| 2082 | event.envelope.get("event").and_then(Value::as_str) == Some("runtime.catalog") |
| 2083 | && event |
| 2084 | .envelope |
| 2085 | .get("source_event_id") |
| 2086 | .and_then(Value::as_str) |
| 2087 | == Some(source_event_id) |
| 2088 | }) |
| 2089 | }) |
| 2090 | } |
| 2091 | |
| 2092 | fn upload_runtime_chat_projection( |
| 2093 | &mut self, |
| 2094 | projection: &RuntimeChatProjection, |
| 2095 | ) -> Result<(), String> { |
| 2096 | if self.attached_run_id.as_deref() != Some(projection.run_id.as_str()) { |
| 2097 | // A reconnect can change the single attached run after the worker |
| 2098 | // claimed an event from the previous run. Leave that durable |
| 2099 | // backlog with its rightful run instead of delivering it across |
| 2100 | // authority or poisoning the new attachment. |
| 2101 | self.runtime_chat |
| 2102 | .as_ref() |
| 2103 | .ok_or_else(|| "This Runtime Chat relay is not configured.".to_string())? |
| 2104 | .release_projection(&projection.native_thread_id, projection.native_seq); |
| 2105 | return Ok(()); |
| 2106 | } |
| 2107 | let seq = self.next_runtime_seq(&projection.run_id); |
| 2108 | let envelope = runtime_chat_envelope( |
| 2109 | seq, |
| 2110 | projection.event, |
| 2111 | Some(&projection.virtual_thread_id), |
| 2112 | Some(&projection.virtual_turn_id), |
| 2113 | Some(&projection.source_event_id), |
| 2114 | projection.timestamp.clone(), |
| 2115 | projection.payload.clone(), |
| 2116 | ); |
| 2117 | if !self.queue_runtime_chat_envelope(&projection.run_id, envelope) { |
| 2118 | return Err("Runtime Chat could not journal its native turn event.".to_string()); |
| 2119 | } |
| 2120 | self.runtime_chat |
| 2121 | .as_ref() |
| 2122 | .ok_or_else(|| "This Runtime Chat relay is not configured.".to_string())? |
| 2123 | .mark_projected( |
| 2124 | &projection.native_thread_id, |
| 2125 | projection.native_seq, |
| 2126 | &projection.virtual_turn_id, |
| 2127 | projection.event, |
| 2128 | ) |
| 2129 | } |
| 2130 | |
| 2131 | pub fn upload_snapshot(&mut self, run_id: &str, messages: &[Message]) { |
| 2132 | if self.uploaded_snapshots.contains(run_id) { |
| 2133 | return; |
| 2134 | } |
| 2135 | let seq = self.next_runtime_seq(run_id); |
| 2136 | let envelope = bounded_session_snapshot_envelope(seq, messages); |
| 2137 | if self.queue_runtime_envelope(run_id, envelope) { |
| 2138 | self.uploaded_snapshots.insert(run_id.to_string()); |
| 2139 | } |
| 2140 | } |
| 2141 | |
| 2142 | pub fn acknowledge( |
| 2143 | &self, |
| 2144 | run_id: &str, |
| 2145 | seq: u64, |
| 2146 | command: &RemoteCommand, |
| 2147 | status: &str, |
| 2148 | error: Option<String>, |
| 2149 | ) { |
| 2150 | let Some(tx) = &self.worker_tx else { |
| 2151 | return; |
| 2152 | }; |
| 2153 | let _ = tx.send(WorkerCommand::Upload { |
| 2154 | run_id: run_id.to_string(), |
| 2155 | acknowledgements: vec![CommandAcknowledgement { |
| 2156 | command_seq: seq, |
| 2157 | command_type: command.kind().to_string(), |
| 2158 | status: status.to_string(), |
| 2159 | turn_id: command.turn_id().map(ToString::to_string), |
| 2160 | error: error.map(|value| value.chars().take(800).collect()), |
| 2161 | }], |
| 2162 | envelopes: Vec::new(), |
| 2163 | }); |
| 2164 | } |
| 2165 | |
| 2166 | pub(crate) async fn apply_runtime_chat_prompt( |
| 2167 | &self, |
| 2168 | command: &RuntimeChatPrompt, |
| 2169 | ) -> Result<(), String> { |
| 2170 | if self.status != Status::Connected |
| 2171 | || self.attached_run_id.as_deref() != Some(command.run_id.as_str()) |
| 2172 | || self.attached_workspace_id.as_deref() != Some(command.workspace.id.as_str()) |
| 2173 | || self.target_ref.as_deref() != Some(command.workspace.target_ref.as_str()) |
| 2174 | { |
| 2175 | return Err("The Runtime Chat command does not match the attached run.".to_string()); |
| 2176 | } |
| 2177 | let host = self |
| 2178 | .runtime_chat |
| 2179 | .clone() |
| 2180 | .ok_or_else(|| "This Runtime Chat relay is not configured.".to_string())?; |
| 2181 | if self.pending_runtime_chat_configuration.is_some() |
| 2182 | && !host.is_exact_prompt_replay(command)? |
| 2183 | { |
| 2184 | return Err("Runtime Chat is refreshing its local model configuration.".to_string()); |
| 2185 | } |
| 2186 | host.apply_prompt(command).await |
| 2187 | } |
| 2188 | |
| 2189 | pub(crate) async fn interrupt_runtime_chat( |
| 2190 | &self, |
| 2191 | run_id: &str, |
| 2192 | scope: &RuntimeChatControlScope, |
| 2193 | virtual_turn_id: &str, |
| 2194 | ) -> Result<(), String> { |
| 2195 | if self.status != Status::Connected || self.attached_run_id.as_deref() != Some(run_id) { |
| 2196 | return Err("The Runtime Chat interrupt does not match the attached run.".to_string()); |
| 2197 | } |
| 2198 | let host = self |
| 2199 | .runtime_chat |
| 2200 | .clone() |
| 2201 | .ok_or_else(|| "This Runtime Chat relay is not configured.".to_string())?; |
| 2202 | host.interrupt(run_id, scope, virtual_turn_id).await |
| 2203 | } |
| 2204 | |
| 2205 | pub fn record_remote_approval( |
| 2206 | &mut self, |
| 2207 | tool_id: &str, |
| 2208 | tool_name: &str, |
| 2209 | description: &str, |
| 2210 | _input: &Value, |
| 2211 | _approval_key: &str, |
| 2212 | _intent_summary: Option<&str>, |
| 2213 | ) -> String { |
| 2214 | let gate = projected_approval_id(tool_id); |
| 2215 | self.pending_approvals.insert( |
| 2216 | gate.clone(), |
| 2217 | PendingRemoteApproval { |
| 2218 | tool_id: tool_id.to_string(), |
| 2219 | }, |
| 2220 | ); |
| 2221 | if let Some(active) = self.active_run.clone() { |
| 2222 | self.upload_envelope( |
| 2223 | &active.run_id, |
| 2224 | "approval.required", |
| 2225 | Some(&active.turn_id), |
| 2226 | json!({ |
| 2227 | "id": gate, |
| 2228 | "approval_id": gate, |
| 2229 | "tool_name": tool_name, |
| 2230 | "description": description, |
| 2231 | }), |
| 2232 | ); |
| 2233 | } |
| 2234 | gate |
| 2235 | } |
| 2236 | |
| 2237 | pub fn take_pending_approval(&mut self, gate: &str) -> Option<String> { |
| 2238 | self.pending_approvals |
| 2239 | .remove(gate) |
| 2240 | .map(|approval| approval.tool_id) |
| 2241 | } |
| 2242 | |
| 2243 | pub fn observe_engine_event(&mut self, event: &EngineEvent) { |
| 2244 | // `/rc` can attach while the host is still preparing a turn. The |
| 2245 | // server run is known first; `TurnStarted` supplies the authoritative |
| 2246 | // runtime turn id later. Promote exactly once, before the ordinary |
| 2247 | // projection below observes the event. |
| 2248 | if let EngineEvent::TurnStarted { turn_id, .. } = event |
| 2249 | && self.active_run.is_none() |
| 2250 | && let Some(run_id) = self.pending_local_turn_run.take() |
| 2251 | && self.promote_pending_local_turn(&run_id, turn_id).is_err() |
| 2252 | { |
| 2253 | self.pending_local_turn_run = Some(run_id); |
| 2254 | return; |
| 2255 | } |
| 2256 | let Some(active) = self.active_run.clone() else { |
| 2257 | return; |
| 2258 | }; |
| 2259 | match event { |
| 2260 | EngineEvent::MessageDelta { content, .. } => { |
| 2261 | self.upload_delta(&active.run_id, &active.turn_id, content); |
| 2262 | } |
| 2263 | EngineEvent::ToolCallStarted { id, name, .. } => { |
| 2264 | self.upload_envelope( |
| 2265 | &active.run_id, |
| 2266 | "item.started", |
| 2267 | Some(&active.turn_id), |
| 2268 | json!({ "tool": { "id": id, "name": name, "input": {} } }), |
| 2269 | ); |
| 2270 | } |
| 2271 | EngineEvent::ToolCallComplete { id, result, .. } => { |
| 2272 | let (event_name, status) = if result.is_ok() { |
| 2273 | ("item.completed", "completed") |
| 2274 | } else { |
| 2275 | ("item.failed", "failed") |
| 2276 | }; |
| 2277 | self.upload_envelope( |
| 2278 | &active.run_id, |
| 2279 | event_name, |
| 2280 | Some(&active.turn_id), |
| 2281 | json!({ |
| 2282 | "item": { |
| 2283 | "id": id, |
| 2284 | "kind": "tool_call", |
| 2285 | "status": status, |
| 2286 | "summary": "", |
| 2287 | "detail": "", |
| 2288 | } |
| 2289 | }), |
| 2290 | ); |
| 2291 | } |
| 2292 | EngineEvent::TurnStarted { turn_id, route, .. } => { |
| 2293 | self.active_run = Some(ActiveRelayRun { |
| 2294 | run_id: active.run_id.clone(), |
| 2295 | turn_id: turn_id.clone(), |
| 2296 | }); |
| 2297 | self.upload_envelope( |
| 2298 | &active.run_id, |
| 2299 | "turn.started", |
| 2300 | Some(turn_id), |
| 2301 | json!({ |
| 2302 | "turn": { |
| 2303 | "model": route.as_ref().map(|value| value.model.as_str()).unwrap_or(""), |
| 2304 | "mode": "", |
| 2305 | } |
| 2306 | }), |
| 2307 | ); |
| 2308 | } |
| 2309 | EngineEvent::TurnComplete { usage, status, .. } => { |
| 2310 | let status = match status { |
| 2311 | TurnOutcomeStatus::Completed => "completed", |
| 2312 | TurnOutcomeStatus::Interrupted => "interrupted", |
| 2313 | TurnOutcomeStatus::Failed => "failed", |
| 2314 | }; |
| 2315 | let terminal_durable = self.upload_envelope( |
| 2316 | &active.run_id, |
| 2317 | "turn.completed", |
| 2318 | Some(&active.turn_id), |
| 2319 | json!({ "turn": { "status": status, "usage": usage } }), |
| 2320 | ); |
| 2321 | if terminal_durable && self.finish_classic_lease().is_ok() { |
| 2322 | if self.resync_required.remove(&active.run_id) { |
| 2323 | // Deltas were shed under pressure during this turn; the UI |
| 2324 | // must now upload a bounded current snapshot so account |
| 2325 | // truth is restored at the terminal boundary. |
| 2326 | self.resync_ready.push(active.run_id.clone()); |
| 2327 | } |
| 2328 | self.active_run = None; |
| 2329 | self.pending_local_turn_run = None; |
| 2330 | } |
| 2331 | } |
| 2332 | EngineEvent::Error { |
| 2333 | envelope, |
| 2334 | recoverable, |
| 2335 | } if !recoverable => { |
| 2336 | self.fail_active_run(&envelope.code, &envelope.message); |
| 2337 | } |
| 2338 | _ => {} |
| 2339 | } |
| 2340 | } |
| 2341 | |
| 2342 | fn fail_active_run(&mut self, code: &str, error: &str) { |
| 2343 | let Some(active) = self.active_run.clone() else { |
| 2344 | return; |
| 2345 | }; |
| 2346 | let message = bounded_remote_error_message(error); |
| 2347 | let item_id = projected_error_item_id(&active.run_id, &active.turn_id, code); |
| 2348 | let item_durable = self.upload_envelope( |
| 2349 | &active.run_id, |
| 2350 | "item.failed", |
| 2351 | Some(&active.turn_id), |
| 2352 | json!({ |
| 2353 | "item": { |
| 2354 | "id": item_id, |
| 2355 | "kind": "error", |
| 2356 | "status": "failed", |
| 2357 | "summary": message, |
| 2358 | "detail": message, |
| 2359 | } |
| 2360 | }), |
| 2361 | ); |
| 2362 | let terminal_durable = self.upload_envelope( |
| 2363 | &active.run_id, |
| 2364 | "turn.completed", |
| 2365 | Some(&active.turn_id), |
| 2366 | json!({ "turn": { "status": "failed", "usage": {} } }), |
| 2367 | ); |
| 2368 | if item_durable && terminal_durable && self.finish_classic_lease().is_ok() { |
| 2369 | self.active_run = None; |
| 2370 | self.pending_local_turn_run = None; |
| 2371 | } |
| 2372 | } |
| 2373 | |
| 2374 | fn upload_envelope( |
| 2375 | &mut self, |
| 2376 | run_id: &str, |
| 2377 | event: &str, |
| 2378 | turn_id: Option<&str>, |
| 2379 | payload: Value, |
| 2380 | ) -> bool { |
| 2381 | let prior_status = self.status; |
| 2382 | let prior_status_detail = self.status_detail.clone(); |
| 2383 | let prior_ownership_blocked_until = self.ownership_blocked_until; |
| 2384 | let seq = self.next_runtime_seq(run_id); |
| 2385 | let envelope = runtime_envelope( |
| 2386 | seq, |
| 2387 | event, |
| 2388 | turn_id, |
| 2389 | chrono::Utc::now().to_rfc3339(), |
| 2390 | payload, |
| 2391 | ); |
| 2392 | if self.queue_runtime_envelope(run_id, envelope) { |
| 2393 | return true; |
| 2394 | } |
| 2395 | if integrity_critical_event(event) |
| 2396 | && self |
| 2397 | .journal |
| 2398 | .as_ref() |
| 2399 | .is_some_and(|journal| journal.persist(&self.pending_runtime_events).is_ok()) |
| 2400 | { |
| 2401 | // Atomic replacement can fail transiently (sharing/AV/power-loss |
| 2402 | // boundary). The exact envelope is still retained in memory; |
| 2403 | // make one immediate durable retry before returning to UI code |
| 2404 | // that may clear an approval or terminal lease. Only after that |
| 2405 | // retry succeeds may transport handoff occur. |
| 2406 | if self.status == Status::Failed |
| 2407 | && self.status_detail |
| 2408 | == "Remote control could not durably journal its account event." |
| 2409 | { |
| 2410 | self.status = prior_status; |
| 2411 | self.status_detail = prior_status_detail; |
| 2412 | self.ownership_blocked_until = prior_ownership_blocked_until; |
| 2413 | } |
| 2414 | self.flush_pending_runtime_events(run_id); |
| 2415 | return true; |
| 2416 | } |
| 2417 | false |
| 2418 | } |
| 2419 | |
| 2420 | fn next_runtime_seq(&self, run_id: &str) -> u64 { |
| 2421 | let acknowledged = self.event_seq.get(run_id).copied().unwrap_or(0); |
| 2422 | let pending = self |
| 2423 | .pending_runtime_events |
| 2424 | .get(run_id) |
| 2425 | .and_then(|events| events.last_key_value().map(|(seq, _)| *seq)) |
| 2426 | .unwrap_or(0); |
| 2427 | acknowledged.max(pending).saturating_add(1) |
| 2428 | } |
| 2429 | |
| 2430 | fn runtime_seq_floor(&self, run_id: &str) -> u64 { |
| 2431 | self.next_runtime_seq(run_id).saturating_sub(1) |
| 2432 | } |
| 2433 | |
| 2434 | fn queue_runtime_envelope(&mut self, run_id: &str, envelope: Value) -> bool { |
| 2435 | // Per-run sequence order must reach the transport in order, so any |
| 2436 | // deferred delta is handed off before a later envelope is queued. |
| 2437 | // Persist the new sequence before transport handoff. This gives the |
| 2438 | // ordinary mirrored Work path the same crash boundary as Runtime Chat: |
| 2439 | // a send may be replayed after an ambiguous crash, but an accepted |
| 2440 | // terminal/approval/failure row is never send-first and then lost. |
| 2441 | self.hand_off_deferred(run_id); |
| 2442 | self.queue_runtime_envelope_inner(run_id, envelope, false, true, false) |
| 2443 | } |
| 2444 | |
| 2445 | fn queue_runtime_chat_envelope(&mut self, run_id: &str, envelope: Value) -> bool { |
| 2446 | // Native Chat advances its own durable event cursor only after this |
| 2447 | // account-delivery journal is durable. Persist before handing the |
| 2448 | // envelope to transport; a resend after an ambiguous send is safe via |
| 2449 | // the stable source_event_id contract. |
| 2450 | self.hand_off_deferred(run_id); |
| 2451 | self.queue_runtime_envelope_inner(run_id, envelope, false, true, true) |
| 2452 | } |
| 2453 | |
| 2454 | fn queue_runtime_envelope_inner( |
| 2455 | &mut self, |
| 2456 | run_id: &str, |
| 2457 | envelope: Value, |
| 2458 | defer: bool, |
| 2459 | require_durable_journal: bool, |
| 2460 | rollback_on_journal_failure: bool, |
| 2461 | ) -> bool { |
| 2462 | if require_durable_journal && self.journal.is_none() { |
| 2463 | self.status_detail = |
| 2464 | "Remote control requires a durable account-delivery journal.".to_string(); |
| 2465 | self.status = Status::Failed; |
| 2466 | self.ownership_blocked_until = Some(Instant::now() + OWNERSHIP_LOCK_AFTER_FAILURE); |
| 2467 | return false; |
| 2468 | } |
| 2469 | let Some(seq) = runtime_envelope_seq(&envelope) else { |
| 2470 | self.status_detail = "a local runtime event had no valid sequence".to_string(); |
| 2471 | return false; |
| 2472 | }; |
| 2473 | let encoded_len = serde_json::to_vec(&envelope) |
| 2474 | .map(|body| body.len()) |
| 2475 | .unwrap_or(usize::MAX); |
| 2476 | if encoded_len > MAX_RUNTIME_ENVELOPE_BYTES { |
| 2477 | self.status_detail = "a local runtime event exceeded the safe relay limit".to_string(); |
| 2478 | return false; |
| 2479 | } |
| 2480 | let integrity = runtime_envelope_event(&envelope).is_some_and(integrity_critical_event); |
| 2481 | let already_pending = self |
| 2482 | .pending_runtime_events |
| 2483 | .get(run_id) |
| 2484 | .and_then(|events| events.get(&seq)) |
| 2485 | .is_some(); |
| 2486 | let previous_event_seq = self.event_seq.get(run_id).copied(); |
| 2487 | if already_pending { |
| 2488 | let entry = self |
| 2489 | .pending_runtime_events |
| 2490 | .get(run_id) |
| 2491 | .and_then(|events| events.get(&seq)) |
| 2492 | .expect("checked above"); |
| 2493 | if entry.envelope != envelope { |
| 2494 | self.status_detail = |
| 2495 | "a local runtime sequence changed before acknowledgement".to_string(); |
| 2496 | return false; |
| 2497 | } |
| 2498 | } else { |
| 2499 | if !self.reserve_capacity(run_id, encoded_len, integrity) { |
| 2500 | return false; |
| 2501 | } |
| 2502 | self.pending_runtime_events |
| 2503 | .entry(run_id.to_string()) |
| 2504 | .or_default() |
| 2505 | .insert( |
| 2506 | seq, |
| 2507 | PendingRuntimeEnvelope { |
| 2508 | envelope: envelope.clone(), |
| 2509 | encoded_len, |
| 2510 | integrity, |
| 2511 | handed_off: false, |
| 2512 | }, |
| 2513 | ); |
| 2514 | self.pending_event_count += 1; |
| 2515 | self.pending_encoded_bytes = self.pending_encoded_bytes.saturating_add(encoded_len); |
| 2516 | } |
| 2517 | self.event_seq |
| 2518 | .entry(run_id.to_string()) |
| 2519 | .and_modify(|cursor| *cursor = (*cursor).max(seq)) |
| 2520 | .or_insert(seq); |
| 2521 | if require_durable_journal |
| 2522 | && self |
| 2523 | .journal |
| 2524 | .as_ref() |
| 2525 | .expect("checked above") |
| 2526 | .persist(&self.pending_runtime_events) |
| 2527 | .is_err() |
| 2528 | { |
| 2529 | if rollback_on_journal_failure && !already_pending { |
| 2530 | let remove_run = if let Some(events) = self.pending_runtime_events.get_mut(run_id) { |
| 2531 | events.remove(&seq); |
| 2532 | events.is_empty() |
| 2533 | } else { |
| 2534 | false |
| 2535 | }; |
| 2536 | if remove_run { |
| 2537 | self.pending_runtime_events.remove(run_id); |
| 2538 | } |
| 2539 | self.pending_event_count = self.pending_event_count.saturating_sub(1); |
| 2540 | self.pending_encoded_bytes = self.pending_encoded_bytes.saturating_sub(encoded_len); |
| 2541 | } |
| 2542 | if rollback_on_journal_failure { |
| 2543 | match previous_event_seq { |
| 2544 | Some(cursor) => { |
| 2545 | self.event_seq.insert(run_id.to_string(), cursor); |
| 2546 | } |
| 2547 | None => { |
| 2548 | self.event_seq.remove(run_id); |
| 2549 | } |
| 2550 | } |
| 2551 | } |
| 2552 | self.status_detail = |
| 2553 | "Remote control could not durably journal its account event.".to_string(); |
| 2554 | self.status = Status::Failed; |
| 2555 | self.ownership_blocked_until = Some(Instant::now() + OWNERSHIP_LOCK_AFTER_FAILURE); |
| 2556 | return false; |
| 2557 | } |
| 2558 | if defer { |
| 2559 | self.deferred_delta.insert(run_id.to_string(), seq); |
| 2560 | } else { |
| 2561 | if let Some(entry) = self |
| 2562 | .pending_runtime_events |
| 2563 | .get_mut(run_id) |
| 2564 | .and_then(|events| events.get_mut(&seq)) |
| 2565 | { |
| 2566 | entry.handed_off = true; |
| 2567 | } |
| 2568 | self.send_runtime_envelope(run_id, envelope); |
| 2569 | if !require_durable_journal { |
| 2570 | self.persist_journal(); |
| 2571 | } |
| 2572 | } |
| 2573 | true |
| 2574 | } |
| 2575 | |
| 2576 | /// Bounded-journal admission control. |
| 2577 | /// |
| 2578 | /// Integrity-critical envelopes may use the full budget, ordinary deltas |
| 2579 | /// only the unreserved share. A shed delta marks the run for terminal- |
| 2580 | /// boundary resynchronization; a shed integrity envelope can never happen |
| 2581 | /// silently — the relay fails closed and local input stays locked through |
| 2582 | /// the server lease expiry. |
| 2583 | fn reserve_capacity(&mut self, run_id: &str, encoded_len: usize, integrity: bool) -> bool { |
| 2584 | let (event_budget, byte_budget) = if integrity { |
| 2585 | (MAX_JOURNAL_EVENTS, MAX_JOURNAL_ENCODED_BYTES) |
| 2586 | } else { |
| 2587 | ( |
| 2588 | MAX_JOURNAL_EVENTS - JOURNAL_RESERVED_INTEGRITY_EVENTS, |
| 2589 | MAX_JOURNAL_ENCODED_BYTES - JOURNAL_RESERVED_INTEGRITY_BYTES, |
| 2590 | ) |
| 2591 | }; |
| 2592 | if self.pending_event_count < event_budget |
| 2593 | && self.pending_encoded_bytes.saturating_add(encoded_len) <= byte_budget |
| 2594 | { |
| 2595 | return true; |
| 2596 | } |
| 2597 | if integrity { |
| 2598 | self.status = Status::Failed; |
| 2599 | self.status_detail = |
| 2600 | "the runtime delivery buffer overflowed; waiting for the last server lease to expire safely" |
| 2601 | .to_string(); |
| 2602 | self.ownership_blocked_until = Some(Instant::now() + OWNERSHIP_LOCK_AFTER_FAILURE); |
| 2603 | } else { |
| 2604 | self.resync_required.insert(run_id.to_string()); |
| 2605 | } |
| 2606 | false |
| 2607 | } |
| 2608 | |
| 2609 | /// Streams a message delta, coalescing into the run's deferred envelope |
| 2610 | /// while that envelope has provably never been handed to the transport. |
| 2611 | fn upload_delta(&mut self, run_id: &str, turn_id: &str, content: &str) { |
| 2612 | if let Some(seq) = self.deferred_delta.get(run_id).copied() { |
| 2613 | if self.try_coalesce_delta(run_id, seq, turn_id, content) { |
| 2614 | return; |
| 2615 | } |
| 2616 | self.hand_off_deferred(run_id); |
| 2617 | } |
| 2618 | let seq = self.next_runtime_seq(run_id); |
| 2619 | let envelope = runtime_envelope( |
| 2620 | seq, |
| 2621 | "item.delta", |
| 2622 | Some(turn_id), |
| 2623 | chrono::Utc::now().to_rfc3339(), |
| 2624 | json!({ "kind": "agent_message", "delta": content }), |
| 2625 | ); |
| 2626 | self.queue_runtime_envelope_inner(run_id, envelope, true, false, false); |
| 2627 | } |
| 2628 | |
| 2629 | fn try_coalesce_delta(&mut self, run_id: &str, seq: u64, turn_id: &str, content: &str) -> bool { |
| 2630 | let Some(entry) = self |
| 2631 | .pending_runtime_events |
| 2632 | .get_mut(run_id) |
| 2633 | .and_then(|events| events.get_mut(&seq)) |
| 2634 | else { |
| 2635 | return false; |
| 2636 | }; |
| 2637 | if entry.handed_off |
| 2638 | || entry.envelope.get("turn_id").and_then(Value::as_str) != Some(turn_id) |
| 2639 | { |
| 2640 | return false; |
| 2641 | } |
| 2642 | let Some(existing) = entry |
| 2643 | .envelope |
| 2644 | .pointer("/payload/delta") |
| 2645 | .and_then(Value::as_str) |
| 2646 | else { |
| 2647 | return false; |
| 2648 | }; |
| 2649 | let merged = format!("{existing}{content}"); |
| 2650 | let mut candidate = entry.envelope.clone(); |
| 2651 | candidate["payload"]["delta"] = Value::String(merged); |
| 2652 | let encoded_len = serde_json::to_vec(&candidate) |
| 2653 | .map(|body| body.len()) |
| 2654 | .unwrap_or(usize::MAX); |
| 2655 | if encoded_len > DELTA_COALESCE_BYTE_CAP { |
| 2656 | return false; |
| 2657 | } |
| 2658 | let old_len = entry.encoded_len; |
| 2659 | entry.envelope = candidate; |
| 2660 | entry.encoded_len = encoded_len; |
| 2661 | self.pending_encoded_bytes = self |
| 2662 | .pending_encoded_bytes |
| 2663 | .saturating_sub(old_len) |
| 2664 | .saturating_add(encoded_len); |
| 2665 | true |
| 2666 | } |
| 2667 | |
| 2668 | /// Hands the run's deferred delta to the transport. From this point the |
| 2669 | /// envelope may have reached the server and becomes immutable. |
| 2670 | fn hand_off_deferred(&mut self, run_id: &str) { |
| 2671 | let Some(seq) = self.deferred_delta.get(run_id).copied() else { |
| 2672 | return; |
| 2673 | }; |
| 2674 | let Some(journal) = &self.journal else { |
| 2675 | self.status = Status::Failed; |
| 2676 | self.status_detail = |
| 2677 | "Remote control requires a durable account-delivery journal.".to_string(); |
| 2678 | self.ownership_blocked_until = Some(Instant::now() + OWNERSHIP_LOCK_AFTER_FAILURE); |
| 2679 | return; |
| 2680 | }; |
| 2681 | if journal.persist(&self.pending_runtime_events).is_err() { |
| 2682 | self.status = Status::Failed; |
| 2683 | self.status_detail = |
| 2684 | "Remote control could not durably journal its account event.".to_string(); |
| 2685 | self.ownership_blocked_until = Some(Instant::now() + OWNERSHIP_LOCK_AFTER_FAILURE); |
| 2686 | return; |
| 2687 | } |
| 2688 | self.deferred_delta.remove(run_id); |
| 2689 | let Some(envelope) = self |
| 2690 | .pending_runtime_events |
| 2691 | .get_mut(run_id) |
| 2692 | .and_then(|events| events.get_mut(&seq)) |
| 2693 | .map(|entry| { |
| 2694 | entry.handed_off = true; |
| 2695 | entry.envelope.clone() |
| 2696 | }) |
| 2697 | else { |
| 2698 | return; |
| 2699 | }; |
| 2700 | self.send_runtime_envelope(run_id, envelope); |
| 2701 | } |
| 2702 | |
| 2703 | fn hand_off_all_deferred(&mut self) { |
| 2704 | let runs: Vec<String> = self.deferred_delta.keys().cloned().collect(); |
| 2705 | for run_id in runs { |
| 2706 | self.hand_off_deferred(&run_id); |
| 2707 | } |
| 2708 | } |
| 2709 | |
| 2710 | /// The UI drains this after each engine event batch and answers with |
| 2711 | /// `upload_resync_snapshot` for the returned run. |
| 2712 | pub fn take_pending_resync(&mut self) -> Option<String> { |
| 2713 | self.resync_ready.pop() |
| 2714 | } |
| 2715 | |
| 2716 | /// Uploads a bounded current-history snapshot to repair account truth |
| 2717 | /// after deltas were shed under pressure. |
| 2718 | pub fn upload_resync_snapshot(&mut self, run_id: &str, messages: &[Message]) { |
| 2719 | let seq = self.next_runtime_seq(run_id); |
| 2720 | let envelope = bounded_session_snapshot_envelope(seq, messages); |
| 2721 | self.queue_runtime_envelope(run_id, envelope); |
| 2722 | } |
| 2723 | |
| 2724 | fn persist_journal(&mut self) { |
| 2725 | let Some(journal) = &self.journal else { |
| 2726 | return; |
| 2727 | }; |
| 2728 | if journal.persist(&self.pending_runtime_events).is_err() { |
| 2729 | // Crash durability is degraded, but nothing is lost silently: the |
| 2730 | // live relay keeps every envelope in memory and `/rc stop` still |
| 2731 | // requires the server-confirmed drain. |
| 2732 | self.status_detail = |
| 2733 | "the delivery journal could not be written; stop waits for server confirmation" |
| 2734 | .to_string(); |
| 2735 | } |
| 2736 | } |
| 2737 | |
| 2738 | /// Replaces the in-memory pending set from a verified journal load. |
| 2739 | fn reset_pending_from(&mut self, restored: HashMap<String, BTreeMap<u64, Value>>) { |
| 2740 | self.pending_runtime_events.clear(); |
| 2741 | self.deferred_delta.clear(); |
| 2742 | self.pending_event_count = 0; |
| 2743 | self.pending_encoded_bytes = 0; |
| 2744 | for (run_id, events) in restored { |
| 2745 | let mut pending = BTreeMap::new(); |
| 2746 | for (seq, envelope) in events { |
| 2747 | let encoded_len = serde_json::to_vec(&envelope) |
| 2748 | .map(|body| body.len()) |
| 2749 | .unwrap_or(usize::MAX); |
| 2750 | let integrity = |
| 2751 | runtime_envelope_event(&envelope).is_some_and(integrity_critical_event); |
| 2752 | self.pending_event_count += 1; |
| 2753 | self.pending_encoded_bytes = self.pending_encoded_bytes.saturating_add(encoded_len); |
| 2754 | pending.insert( |
| 2755 | seq, |
| 2756 | PendingRuntimeEnvelope { |
| 2757 | envelope, |
| 2758 | encoded_len, |
| 2759 | integrity, |
| 2760 | handed_off: false, |
| 2761 | }, |
| 2762 | ); |
| 2763 | } |
| 2764 | if let Some((top, _)) = pending.last_key_value() { |
| 2765 | let top = *top; |
| 2766 | self.event_seq |
| 2767 | .entry(run_id.clone()) |
| 2768 | .and_modify(|cursor| *cursor = (*cursor).max(top)) |
| 2769 | .or_insert(top); |
| 2770 | } |
| 2771 | if !pending.is_empty() { |
| 2772 | self.pending_runtime_events.insert(run_id, pending); |
| 2773 | } |
| 2774 | } |
| 2775 | } |
| 2776 | |
| 2777 | fn send_runtime_envelope(&self, run_id: &str, envelope: Value) { |
| 2778 | let Some(tx) = &self.worker_tx else { |
| 2779 | return; |
| 2780 | }; |
| 2781 | let _ = tx.send(WorkerCommand::Upload { |
| 2782 | run_id: run_id.to_string(), |
| 2783 | acknowledgements: Vec::new(), |
| 2784 | envelopes: vec![envelope], |
| 2785 | }); |
| 2786 | } |
| 2787 | |
| 2788 | fn flush_pending_runtime_events(&mut self, run_id: &str) { |
| 2789 | // A reconnect resend covers everything, deferred deltas included; |
| 2790 | // after this every envelope may have reached the server. |
| 2791 | self.deferred_delta.remove(run_id); |
| 2792 | let mut to_send = Vec::new(); |
| 2793 | if let Some(events) = self.pending_runtime_events.get_mut(run_id) { |
| 2794 | for entry in events.values_mut() { |
| 2795 | entry.handed_off = true; |
| 2796 | to_send.push(entry.envelope.clone()); |
| 2797 | } |
| 2798 | } |
| 2799 | if to_send.is_empty() { |
| 2800 | return; |
| 2801 | } |
| 2802 | for envelope in to_send { |
| 2803 | self.send_runtime_envelope(run_id, envelope); |
| 2804 | } |
| 2805 | self.persist_journal(); |
| 2806 | } |
| 2807 | |
| 2808 | fn flush_all_pending(&mut self) { |
| 2809 | let runs: Vec<String> = self.pending_runtime_events.keys().cloned().collect(); |
| 2810 | for run_id in runs { |
| 2811 | self.flush_pending_runtime_events(&run_id); |
| 2812 | } |
| 2813 | } |
| 2814 | |
| 2815 | fn reconcile_runtime_cursor(&mut self, run_id: &str, cursor: u64) { |
| 2816 | if cursor > JS_MAX_SAFE_INTEGER { |
| 2817 | self.status_detail = "the server returned an unsafe runtime cursor".to_string(); |
| 2818 | return; |
| 2819 | } |
| 2820 | let covered_classic_terminal = self |
| 2821 | .journal |
| 2822 | .as_ref() |
| 2823 | .and_then(RuntimeEventJournal::classic_lease) |
| 2824 | .filter(|lease| lease.run_id == run_id) |
| 2825 | .is_some_and(|lease| { |
| 2826 | let turn_id = lease |
| 2827 | .turn_id |
| 2828 | .clone() |
| 2829 | .unwrap_or_else(|| classic_recovery_turn_id(&lease.run_id, &lease.lease_id)); |
| 2830 | self.pending_runtime_events |
| 2831 | .get(run_id) |
| 2832 | .is_some_and(|events| { |
| 2833 | events.range(..=cursor).any(|(seq, entry)| { |
| 2834 | *seq > lease.seq_floor |
| 2835 | && runtime_envelope_event(&entry.envelope) == Some("turn.completed") |
| 2836 | && entry.envelope.get("turn_id").and_then(Value::as_str) |
| 2837 | == Some(turn_id.as_str()) |
| 2838 | }) |
| 2839 | }) |
| 2840 | }); |
| 2841 | let classic_state_result = if covered_classic_terminal { |
| 2842 | self.journal |
| 2843 | .as_ref() |
| 2844 | .expect("lease came from this journal") |
| 2845 | .set_classic_lease(None, &self.pending_runtime_events) |
| 2846 | } else if let Some(journal) = &self.journal { |
| 2847 | journal.advance_classic_seq_floor(run_id, cursor) |
| 2848 | } else { |
| 2849 | Ok(()) |
| 2850 | }; |
| 2851 | if let Err(error) = classic_state_result { |
| 2852 | // Never compact the only durable evidence of the run's sequence |
| 2853 | // history unless the canonical active lease first records the |
| 2854 | // acknowledged floor. A reconnect can retry this exact cursor. |
| 2855 | self.status = Status::Failed; |
| 2856 | self.status_detail = error; |
| 2857 | self.ownership_blocked_until = Some(Instant::now() + OWNERSHIP_LOCK_AFTER_FAILURE); |
| 2858 | return; |
| 2859 | } |
| 2860 | if covered_classic_terminal { |
| 2861 | self.active_run = None; |
| 2862 | self.pending_local_turn_run = None; |
| 2863 | } |
| 2864 | let mut empty = false; |
| 2865 | let mut retired_any = false; |
| 2866 | if let Some(events) = self.pending_runtime_events.get_mut(run_id) { |
| 2867 | let retired: Vec<u64> = events.range(..=cursor).map(|(seq, _)| *seq).collect(); |
| 2868 | for seq in retired { |
| 2869 | if let Some(entry) = events.remove(&seq) { |
| 2870 | retired_any = true; |
| 2871 | self.pending_event_count = self.pending_event_count.saturating_sub(1); |
| 2872 | self.pending_encoded_bytes = |
| 2873 | self.pending_encoded_bytes.saturating_sub(entry.encoded_len); |
| 2874 | } |
| 2875 | } |
| 2876 | empty = events.is_empty(); |
| 2877 | } |
| 2878 | if empty { |
| 2879 | self.pending_runtime_events.remove(run_id); |
| 2880 | } |
| 2881 | if self |
| 2882 | .deferred_delta |
| 2883 | .get(run_id) |
| 2884 | .is_some_and(|seq| *seq <= cursor) |
| 2885 | { |
| 2886 | self.deferred_delta.remove(run_id); |
| 2887 | } |
| 2888 | self.event_seq |
| 2889 | .entry(run_id.to_string()) |
| 2890 | .and_modify(|known| *known = (*known).max(cursor)) |
| 2891 | .or_insert(cursor); |
| 2892 | if retired_any { |
| 2893 | // Compact the acknowledged prefix out of the journal promptly. |
| 2894 | self.persist_journal(); |
| 2895 | } |
| 2896 | // Runtime Chat retains the exclusive provider-request lease through |
| 2897 | // local terminal projection *and* this server-confirmed cursor. Only |
| 2898 | // then may interactive/autonomous provider work resume for the |
| 2899 | // attached CWC run. Catalog/approval integrity rows conservatively |
| 2900 | // keep the same gate closed until their own acknowledgement as well. |
| 2901 | if !self.has_unacknowledged_integrity_events() |
| 2902 | && let Some(host) = &self.runtime_chat |
| 2903 | { |
| 2904 | host.release_inference_ownership_if_settled(); |
| 2905 | } |
| 2906 | } |
| 2907 | } |
| 2908 | |
| 2909 | impl Drop for RemoteControlController { |
| 2910 | fn drop(&mut self) { |
| 2911 | self.stop_worker(); |
| 2912 | } |
| 2913 | } |
| 2914 | |
| 2915 | impl RemoteCommand { |
| 2916 | fn kind(&self) -> &'static str { |
| 2917 | match self { |
| 2918 | Self::Prompt { .. } | Self::RuntimeChatPrompt(_) => "prompt.request", |
| 2919 | Self::Approval { .. } => "approval.decision", |
| 2920 | Self::Control { .. } => "run.control", |
| 2921 | } |
| 2922 | } |
| 2923 | |
| 2924 | fn turn_id(&self) -> Option<&str> { |
| 2925 | match self { |
| 2926 | Self::Prompt { turn_id, .. } => Some(turn_id), |
| 2927 | Self::RuntimeChatPrompt(prompt) => Some(&prompt.turn_id), |
| 2928 | Self::Control { turn_id, .. } => turn_id.as_deref(), |
| 2929 | Self::Approval { .. } => None, |
| 2930 | } |
| 2931 | } |
| 2932 | |
| 2933 | fn is_runtime_chat(&self) -> bool { |
| 2934 | matches!( |
| 2935 | self, |
| 2936 | Self::RuntimeChatPrompt(_) |
| 2937 | | Self::Control { |
| 2938 | runtime_chat: Some(_), |
| 2939 | .. |
| 2940 | } |
| 2941 | ) |
| 2942 | } |
| 2943 | } |
| 2944 | |
| 2945 | /// Opaque identity for the enrolled folder. It is a hash of the workspace |
| 2946 | /// path only: every session opened in the same folder shares one target, so |
| 2947 | /// the control plane sees one grant per folder rather than one per session |
| 2948 | /// (the session itself travels separately as `sessionRef`). |
| 2949 | pub fn target_ref(workspace: &Path) -> String { |
| 2950 | let mut hasher = Sha256::new(); |
| 2951 | hasher.update(b"codewhale-remote-target:v2\0"); |
| 2952 | hasher.update(workspace.to_string_lossy().as_bytes()); |
| 2953 | format!("target_{}", &bytes_to_hex(&hasher.finalize())[..32]) |
| 2954 | } |
| 2955 | |
| 2956 | /// Status-bar banner shown while the web owns this session. When the control |
| 2957 | /// plane advertised a session link the banner leads with it; otherwise it |
| 2958 | /// falls back to the opaque account and runner receipts. |
| 2959 | pub fn remote_control_banner(account_ref: &str, runner_id: &str, run_url: Option<&str>) -> String { |
| 2960 | match run_url { |
| 2961 | Some(url) => format!("WEB MIRROR · {url} · /rc stop"), |
| 2962 | None => format!("WEB MIRROR · account {account_ref} · runner {runner_id} · /rc stop"), |
| 2963 | } |
| 2964 | } |
| 2965 | |
| 2966 | /// Transcript note announcing where the live session can be followed. |
| 2967 | pub fn remote_control_link_notice(run_url: &str) -> String { |
| 2968 | format!( |
| 2969 | "Remote control is live at {run_url} — run /rc open to open it in your browser, or /rc link to print it. Both surfaces stay usable; one turn runs at a time." |
| 2970 | ) |
| 2971 | } |
| 2972 | |
| 2973 | fn runtime_envelope( |
| 2974 | seq: u64, |
| 2975 | event: &str, |
| 2976 | turn_id: Option<&str>, |
| 2977 | timestamp: String, |
| 2978 | payload: Value, |
| 2979 | ) -> Value { |
| 2980 | json!({ |
| 2981 | "schema_version": 1, |
| 2982 | "seq": seq, |
| 2983 | "event": event, |
| 2984 | "kind": event, |
| 2985 | "turn_id": turn_id, |
| 2986 | "timestamp": timestamp, |
| 2987 | "payload": payload, |
| 2988 | }) |
| 2989 | } |
| 2990 | |
| 2991 | fn runtime_chat_envelope( |
| 2992 | seq: u64, |
| 2993 | event: &str, |
| 2994 | thread_id: Option<&str>, |
| 2995 | turn_id: Option<&str>, |
| 2996 | source_event_id: Option<&str>, |
| 2997 | timestamp: String, |
| 2998 | payload: Value, |
| 2999 | ) -> Value { |
| 3000 | let mut envelope = json!({ |
| 3001 | "schema_version": 2, |
| 3002 | "seq": seq, |
| 3003 | "event": event, |
| 3004 | "kind": event, |
| 3005 | "timestamp": timestamp, |
| 3006 | "payload": payload, |
| 3007 | }); |
| 3008 | if let Some(thread_id) = thread_id { |
| 3009 | envelope["thread_id"] = json!(thread_id); |
| 3010 | } |
| 3011 | if let Some(turn_id) = turn_id { |
| 3012 | envelope["turn_id"] = json!(turn_id); |
| 3013 | } |
| 3014 | if let Some(source_event_id) = source_event_id { |
| 3015 | envelope["source_event_id"] = json!(source_event_id); |
| 3016 | } |
| 3017 | envelope |
| 3018 | } |
| 3019 | |
| 3020 | fn runtime_chat_catalog_source_event_id(run_id: &str, payload_fingerprint: &str) -> String { |
| 3021 | let mut hasher = Sha256::new(); |
| 3022 | hasher.update(b"codewhale.runtime-chat-catalog.v1\0"); |
| 3023 | hasher.update(run_id.as_bytes()); |
| 3024 | hasher.update(b"\0"); |
| 3025 | hasher.update(payload_fingerprint.as_bytes()); |
| 3026 | format!("catalog_{}", bytes_to_hex(&hasher.finalize())) |
| 3027 | } |
| 3028 | |
| 3029 | fn runtime_envelope_seq(envelope: &Value) -> Option<u64> { |
| 3030 | envelope |
| 3031 | .get("seq") |
| 3032 | .and_then(Value::as_u64) |
| 3033 | .filter(|seq| (1..=JS_MAX_SAFE_INTEGER).contains(seq)) |
| 3034 | } |
| 3035 | |
| 3036 | fn bounded_session_snapshot_envelope(seq: u64, messages: &[Message]) -> Value { |
| 3037 | let timestamp = chrono::Utc::now().to_rfc3339(); |
| 3038 | let candidates = messages |
| 3039 | .iter() |
| 3040 | .rev() |
| 3041 | .filter_map(project_session_message) |
| 3042 | .take(MAX_SNAPSHOT_MESSAGES) |
| 3043 | .collect::<Vec<_>>(); |
| 3044 | let mut kept = Vec::<Value>::new(); |
| 3045 | for (role, text) in candidates { |
| 3046 | let full = json!({ "role": role, "text": text }); |
| 3047 | kept.insert(0, full); |
| 3048 | if snapshot_envelope_len(seq, ×tamp, &kept) <= SNAPSHOT_ENVELOPE_BYTE_BUDGET { |
| 3049 | continue; |
| 3050 | } |
| 3051 | kept.remove(0); |
| 3052 | let max_chars = text.chars().count(); |
| 3053 | let mut low = 0usize; |
| 3054 | let mut high = max_chars; |
| 3055 | while low < high { |
| 3056 | let mid = low + (high - low).div_ceil(2); |
| 3057 | let prefix = unicode_prefix(&text, mid); |
| 3058 | kept.insert(0, json!({ "role": role, "text": prefix })); |
| 3059 | let fits = |
| 3060 | snapshot_envelope_len(seq, ×tamp, &kept) <= SNAPSHOT_ENVELOPE_BYTE_BUDGET; |
| 3061 | kept.remove(0); |
| 3062 | if fits { |
| 3063 | low = mid; |
| 3064 | } else { |
| 3065 | high = mid - 1; |
| 3066 | } |
| 3067 | } |
| 3068 | if low >= MIN_TRUNCATED_MESSAGE_CHARS || (kept.is_empty() && low > 0) { |
| 3069 | kept.insert( |
| 3070 | 0, |
| 3071 | json!({ "role": role, "text": unicode_prefix(&text, low) }), |
| 3072 | ); |
| 3073 | } |
| 3074 | break; |
| 3075 | } |
| 3076 | let envelope = runtime_envelope( |
| 3077 | seq, |
| 3078 | "session.snapshot", |
| 3079 | None, |
| 3080 | timestamp, |
| 3081 | json!({ "messages": kept }), |
| 3082 | ); |
| 3083 | debug_assert!( |
| 3084 | serde_json::to_vec(&envelope).is_ok_and(|body| body.len() <= SNAPSHOT_ENVELOPE_BYTE_BUDGET) |
| 3085 | ); |
| 3086 | envelope |
| 3087 | } |
| 3088 | |
| 3089 | fn project_session_message(message: &Message) -> Option<(String, String)> { |
| 3090 | let role = match message.role.as_str() { |
| 3091 | "user" => "user", |
| 3092 | "assistant" => "assistant", |
| 3093 | _ => return None, |
| 3094 | }; |
| 3095 | let text = message |
| 3096 | .content |
| 3097 | .iter() |
| 3098 | .filter_map(|block| match block { |
| 3099 | ContentBlock::Text { text, .. } => Some(text.as_str()), |
| 3100 | _ => None, |
| 3101 | }) |
| 3102 | .collect::<Vec<_>>() |
| 3103 | .join("\n"); |
| 3104 | if text.trim().is_empty() { |
| 3105 | return None; |
| 3106 | } |
| 3107 | Some(( |
| 3108 | role.to_string(), |
| 3109 | text.chars() |
| 3110 | .take(MAX_SNAPSHOT_MESSAGE_CHARS) |
| 3111 | .collect::<String>(), |
| 3112 | )) |
| 3113 | } |
| 3114 | |
| 3115 | fn snapshot_envelope_len(seq: u64, timestamp: &str, messages: &[Value]) -> usize { |
| 3116 | serde_json::to_vec(&runtime_envelope( |
| 3117 | seq, |
| 3118 | "session.snapshot", |
| 3119 | None, |
| 3120 | timestamp.to_string(), |
| 3121 | json!({ "messages": messages }), |
| 3122 | )) |
| 3123 | .map(|body| body.len()) |
| 3124 | .unwrap_or(usize::MAX) |
| 3125 | } |
| 3126 | |
| 3127 | fn unicode_prefix(value: &str, chars: usize) -> String { |
| 3128 | value.chars().take(chars).collect() |
| 3129 | } |
| 3130 | |
| 3131 | fn projected_approval_id(raw: &str) -> String { |
| 3132 | let mut hasher = Sha256::new(); |
| 3133 | hasher.update(b"local-runtime:approval\0"); |
| 3134 | hasher.update(raw.as_bytes()); |
| 3135 | format!("local_approval_{}", &bytes_to_hex(&hasher.finalize())[..24]) |
| 3136 | } |
| 3137 | |
| 3138 | /// Whether this view is the approval card for exactly `gate` (the projected |
| 3139 | /// approval id). Used by the web mirror to dismiss the matching card — never |
| 3140 | /// an unrelated approval that happens to be on top. |
| 3141 | pub(crate) fn view_is_approval_for_gate( |
| 3142 | view: &dyn crate::tui::views::ModalView, |
| 3143 | gate: &str, |
| 3144 | ) -> bool { |
| 3145 | view.kind() == crate::tui::views::ModalKind::Approval |
| 3146 | && view |
| 3147 | .approval_request_id() |
| 3148 | .is_some_and(|tool_id| projected_approval_id(tool_id) == gate) |
| 3149 | } |
| 3150 | |
| 3151 | fn projected_error_item_id(run_id: &str, turn_id: &str, code: &str) -> String { |
| 3152 | let mut hasher = Sha256::new(); |
| 3153 | hasher.update(b"local-runtime:error\0"); |
| 3154 | hasher.update(run_id.as_bytes()); |
| 3155 | hasher.update(b"\0"); |
| 3156 | hasher.update(turn_id.as_bytes()); |
| 3157 | hasher.update(b"\0"); |
| 3158 | hasher.update(code.as_bytes()); |
| 3159 | format!("local_item_{}", &bytes_to_hex(&hasher.finalize())[..24]) |
| 3160 | } |
| 3161 | |
| 3162 | fn bounded_remote_error_message(error: &str) -> String { |
| 3163 | let without_nul = error.replace('\0', " "); |
| 3164 | let redacted = codewhale_config::persistence::redact_secrets(&without_nul); |
| 3165 | let message = redacted.trim(); |
| 3166 | if message.is_empty() { |
| 3167 | return "The local model turn failed.".to_string(); |
| 3168 | } |
| 3169 | crate::utils::truncate_with_ellipsis(message, MAX_REMOTE_ERROR_MESSAGE_BYTES, "…") |
| 3170 | } |
| 3171 | |
| 3172 | fn command_fingerprint(command: &RemoteCommand) -> String { |
| 3173 | let canonical = format!("{command:?}"); |
| 3174 | bytes_to_hex(&Sha256::digest(canonical.as_bytes())) |
| 3175 | } |
| 3176 | |
| 3177 | fn bytes_to_hex(bytes: &[u8]) -> String { |
| 3178 | bytes.iter().map(|byte| format!("{byte:02x}")).collect() |
| 3179 | } |
| 3180 | |
| 3181 | async fn relay_worker( |
| 3182 | start: RemoteStart, |
| 3183 | mut runtime_chat: Option<RuntimeChatRelayHost>, |
| 3184 | mut worker_rx: mpsc::UnboundedReceiver<WorkerCommand>, |
| 3185 | event_tx: mpsc::UnboundedSender<RemoteEvent>, |
| 3186 | phase: &mut RelayPhase, |
| 3187 | ) -> Result<(), String> { |
| 3188 | let base = runner_control_plane_base()?; |
| 3189 | let client = crate::tls::reqwest_client_builder() |
| 3190 | .https_only(!cfg!(debug_assertions)) |
| 3191 | .redirect(reqwest::redirect::Policy::none()) |
| 3192 | .timeout(Duration::from_secs(20)) |
| 3193 | .build() |
| 3194 | .map_err(|_| "Remote control could not initialize secure networking.".to_string())?; |
| 3195 | |
| 3196 | let saved_enrollment = load_persisted_enrollment()?; |
| 3197 | // The device id is stable across enrollments: it is what lets the control |
| 3198 | // plane fold every folder enrolled from this terminal into one computer. |
| 3199 | let device_id = stable_device_id( |
| 3200 | saved_enrollment |
| 3201 | .as_ref() |
| 3202 | .map(|saved| saved.device_id.as_str()), |
| 3203 | )?; |
| 3204 | let mut enrollment = match saved_enrollment { |
| 3205 | Some(saved) if saved.matches(&start, &base) => { |
| 3206 | match refresh_enrollment(&client, saved).await { |
| 3207 | Ok(enrollment) => enrollment, |
| 3208 | Err(error) if error == "runner_enrollment_revoked" => { |
| 3209 | delete_persisted_enrollment(); |
| 3210 | enroll_device(&client, &base, &start, &device_id, &event_tx).await? |
| 3211 | } |
| 3212 | Err(error) => return Err(error), |
| 3213 | } |
| 3214 | } |
| 3215 | Some(_) => { |
| 3216 | delete_persisted_enrollment(); |
| 3217 | enroll_device(&client, &base, &start, &device_id, &event_tx).await? |
| 3218 | } |
| 3219 | None => enroll_device(&client, &base, &start, &device_id, &event_tx).await?, |
| 3220 | }; |
| 3221 | |
| 3222 | let connection = connect_runner(&client, &enrollment, &start).await?; |
| 3223 | // connect_runner answered with a server-confirmed attachment: a lease |
| 3224 | // exists from here on, and every later failure is a lost-after-lease |
| 3225 | // disconnect that must stay fail-closed. |
| 3226 | *phase = RelayPhase::Leased; |
| 3227 | let mut runner_id = connection.runner_id.clone(); |
| 3228 | event_tx |
| 3229 | .send(RemoteEvent::Connected { |
| 3230 | account_ref: enrollment.persisted.account_ref.clone(), |
| 3231 | runner_id: runner_id.clone(), |
| 3232 | target_ref: start.target_ref.clone(), |
| 3233 | attachment: connection.attachment, |
| 3234 | links: connection.links, |
| 3235 | }) |
| 3236 | .map_err(|_| "The terminal remote-control owner stopped.".to_string())?; |
| 3237 | let mut last_heartbeat = Instant::now() - HEARTBEAT_INTERVAL; |
| 3238 | let mut command_cursor: HashMap<String, u64> = HashMap::new(); |
| 3239 | let mut delivered: HashMap<(String, u64), String> = HashMap::new(); |
| 3240 | let mut runtime_outbox = RuntimeTransportOutbox::default(); |
| 3241 | let mut runtime_upload_tick = tokio::time::interval(RUNTIME_UPLOAD_RETRY_INTERVAL); |
| 3242 | runtime_upload_tick.set_missed_tick_behavior(tokio::time::MissedTickBehavior::Skip); |
| 3243 | let mut runtime_chat_tick = tokio::time::interval(RUNTIME_UPLOAD_RETRY_INTERVAL); |
| 3244 | runtime_chat_tick.set_missed_tick_behavior(tokio::time::MissedTickBehavior::Skip); |
| 3245 | // One deadline across loop iterations. Reconstructing `sleep(SYNC_INTERVAL)` |
| 3246 | // lets the 250ms Runtime Chat tick reset poll/heartbeat/token refresh. |
| 3247 | let mut sync_tick = |
| 3248 | tokio::time::interval_at(tokio::time::Instant::now() + SYNC_INTERVAL, SYNC_INTERVAL); |
| 3249 | sync_tick.set_missed_tick_behavior(tokio::time::MissedTickBehavior::Skip); |
| 3250 | let mut runtime_retry_delay = RUNTIME_UPLOAD_RETRY_INTERVAL; |
| 3251 | let mut runtime_retry_not_before = Instant::now(); |
| 3252 | |
| 3253 | loop { |
| 3254 | tokio::select! { |
| 3255 | command = worker_rx.recv() => { |
| 3256 | match command { |
| 3257 | Some(WorkerCommand::Upload { run_id, acknowledgements, envelopes }) => { |
| 3258 | if !envelopes.is_empty() { |
| 3259 | if !acknowledgements.is_empty() || envelopes.len() != 1 { |
| 3260 | return Err("The local runtime queued an invalid event batch.".to_string()); |
| 3261 | } |
| 3262 | runtime_outbox.enqueue(&run_id, envelopes[0].clone())?; |
| 3263 | continue; |
| 3264 | } |
| 3265 | let body = Some(json!({ "acknowledgements": acknowledgements, "envelopes": envelopes })); |
| 3266 | let result = runner_request( |
| 3267 | &client, |
| 3268 | &enrollment, |
| 3269 | Method::POST, |
| 3270 | &["api", "local-runners", &runner_id, "runs", &run_id, "events"], |
| 3271 | &[], |
| 3272 | body.clone(), |
| 3273 | ) |
| 3274 | .await; |
| 3275 | if let Err(err) = result { |
| 3276 | if err == "runner_access_token_expired" { |
| 3277 | refresh_enrollment_and_reconnect( |
| 3278 | &client, |
| 3279 | &mut enrollment, |
| 3280 | &mut runner_id, |
| 3281 | &start, |
| 3282 | &event_tx, |
| 3283 | ) |
| 3284 | .await?; |
| 3285 | runner_request( |
| 3286 | &client, |
| 3287 | &enrollment, |
| 3288 | Method::POST, |
| 3289 | &["api", "local-runners", &runner_id, "runs", &run_id, "events"], |
| 3290 | &[], |
| 3291 | body, |
| 3292 | ) |
| 3293 | .await?; |
| 3294 | } else { |
| 3295 | return Err(err); |
| 3296 | } |
| 3297 | } |
| 3298 | } |
| 3299 | Some(WorkerCommand::ReleaseRuntimeChatHost) => { |
| 3300 | runtime_chat.take(); |
| 3301 | event_tx |
| 3302 | .send(RemoteEvent::RuntimeChatHostReleased) |
| 3303 | .map_err(|_| "The terminal remote-control owner stopped.".to_string())?; |
| 3304 | } |
| 3305 | Some(WorkerCommand::InstallRuntimeChatHost(host)) => { |
| 3306 | if runtime_chat.is_some() { |
| 3307 | return Err( |
| 3308 | "Runtime Chat received an invalid provider-refresh install." |
| 3309 | .to_string(), |
| 3310 | ); |
| 3311 | } |
| 3312 | runtime_chat = Some(host); |
| 3313 | } |
| 3314 | Some(WorkerCommand::Stop) | None => { |
| 3315 | // Do not return local input until the control plane has |
| 3316 | // durably released this lease. Every queued runtime |
| 3317 | // envelope must first drain behind the server-confirmed |
| 3318 | // cursor; only then may the offline heartbeat be |
| 3319 | // posted. If either cannot be confirmed, this worker |
| 3320 | // errors out and the UI keeps ownership locked through |
| 3321 | // the server-side lease expiry instead. |
| 3322 | drain_runtime_outbox_for_stop( |
| 3323 | &client, |
| 3324 | &mut enrollment, |
| 3325 | &mut runner_id, |
| 3326 | &start, |
| 3327 | &event_tx, |
| 3328 | &mut runtime_outbox, |
| 3329 | Instant::now() + STOP_DRAIN_DEADLINE, |
| 3330 | ) |
| 3331 | .await?; |
| 3332 | let hb = post_heartbeat(&client, &enrollment, &runner_id, &start, "offline").await; |
| 3333 | if let Err(err) = hb { |
| 3334 | if err == "runner_access_token_expired" { |
| 3335 | refresh_enrollment_and_reconnect( |
| 3336 | &client, |
| 3337 | &mut enrollment, |
| 3338 | &mut runner_id, |
| 3339 | &start, |
| 3340 | &event_tx, |
| 3341 | ) |
| 3342 | .await?; |
| 3343 | post_heartbeat(&client, &enrollment, &runner_id, &start, "offline").await?; |
| 3344 | } else { |
| 3345 | return Err(err); |
| 3346 | } |
| 3347 | } |
| 3348 | let _ = event_tx.send(RemoteEvent::Stopped); |
| 3349 | return Ok(()); |
| 3350 | } |
| 3351 | } |
| 3352 | } |
| 3353 | _ = runtime_upload_tick.tick(), if !runtime_outbox.events.is_empty() => { |
| 3354 | if Instant::now() < runtime_retry_not_before { |
| 3355 | continue; |
| 3356 | } |
| 3357 | match runtime_outbox |
| 3358 | .try_flush_one(&client, &enrollment, &runner_id) |
| 3359 | .await? |
| 3360 | { |
| 3361 | RuntimeFlushOutcome::Idle => { |
| 3362 | runtime_retry_delay = RUNTIME_UPLOAD_RETRY_INTERVAL; |
| 3363 | runtime_retry_not_before = Instant::now(); |
| 3364 | } |
| 3365 | RuntimeFlushOutcome::Retryable => { |
| 3366 | runtime_retry_not_before = Instant::now() + runtime_retry_delay; |
| 3367 | runtime_retry_delay = runtime_retry_delay |
| 3368 | .saturating_mul(2) |
| 3369 | .min(RUNTIME_UPLOAD_MAX_BACKOFF); |
| 3370 | } |
| 3371 | RuntimeFlushOutcome::Accepted { run_id, cursor } => { |
| 3372 | runtime_retry_delay = RUNTIME_UPLOAD_RETRY_INTERVAL; |
| 3373 | runtime_retry_not_before = Instant::now(); |
| 3374 | event_tx |
| 3375 | .send(RemoteEvent::RuntimeCursor { run_id, cursor }) |
| 3376 | .map_err(|_| "The terminal remote-control owner stopped.".to_string())?; |
| 3377 | } |
| 3378 | RuntimeFlushOutcome::AccessTokenExpired => { |
| 3379 | refresh_enrollment_and_reconnect( |
| 3380 | &client, |
| 3381 | &mut enrollment, |
| 3382 | &mut runner_id, |
| 3383 | &start, |
| 3384 | &event_tx, |
| 3385 | ) |
| 3386 | .await?; |
| 3387 | runtime_retry_delay = RUNTIME_UPLOAD_RETRY_INTERVAL; |
| 3388 | runtime_retry_not_before = Instant::now(); |
| 3389 | } |
| 3390 | } |
| 3391 | } |
| 3392 | _ = runtime_chat_tick.tick(), if runtime_chat.is_some() => { |
| 3393 | let host = runtime_chat.as_ref().expect("select guard requires Runtime Chat host"); |
| 3394 | for projection in host.pending_projections().await? { |
| 3395 | event_tx |
| 3396 | .send(RemoteEvent::RuntimeChatProjection(projection)) |
| 3397 | .map_err(|_| "The terminal remote-control owner stopped.".to_string())?; |
| 3398 | } |
| 3399 | } |
| 3400 | _ = sync_tick.tick() => { |
| 3401 | if enrollment_needs_refresh(&enrollment) { |
| 3402 | // Proactive refresh before expiry; reconnect to keep runner lease valid. |
| 3403 | match refresh_enrollment(&client, enrollment.persisted.clone()).await { |
| 3404 | Ok(new_enrollment) => { |
| 3405 | install_reconnected_enrollment( |
| 3406 | &mut enrollment, |
| 3407 | new_enrollment, |
| 3408 | &start, |
| 3409 | )?; |
| 3410 | reconnect_runner( |
| 3411 | &client, |
| 3412 | &enrollment, |
| 3413 | &mut runner_id, |
| 3414 | &start, |
| 3415 | &event_tx, |
| 3416 | ) |
| 3417 | .await?; |
| 3418 | } |
| 3419 | Err(err) if err == "runner_enrollment_revoked" => { |
| 3420 | delete_persisted_enrollment(); |
| 3421 | let device_id = enrollment.persisted.device_id.clone(); |
| 3422 | let candidate = |
| 3423 | enroll_device(&client, &base, &start, &device_id, &event_tx).await?; |
| 3424 | if let Err(error) = install_reconnected_enrollment( |
| 3425 | &mut enrollment, |
| 3426 | candidate, |
| 3427 | &start, |
| 3428 | ) { |
| 3429 | // enroll_device persists its candidate. Never retain a |
| 3430 | // credential for authority the live worker rejected. |
| 3431 | delete_persisted_enrollment(); |
| 3432 | return Err(error); |
| 3433 | } |
| 3434 | reconnect_runner( |
| 3435 | &client, |
| 3436 | &enrollment, |
| 3437 | &mut runner_id, |
| 3438 | &start, |
| 3439 | &event_tx, |
| 3440 | ) |
| 3441 | .await?; |
| 3442 | } |
| 3443 | Err(err) => return Err(err), |
| 3444 | } |
| 3445 | } |
| 3446 | if last_heartbeat.elapsed() >= HEARTBEAT_INTERVAL { |
| 3447 | let hb = post_heartbeat(&client, &enrollment, &runner_id, &start, "active").await; |
| 3448 | if let Err(err) = hb { |
| 3449 | if err == "runner_access_token_expired" { |
| 3450 | refresh_enrollment_and_reconnect( |
| 3451 | &client, |
| 3452 | &mut enrollment, |
| 3453 | &mut runner_id, |
| 3454 | &start, |
| 3455 | &event_tx, |
| 3456 | ) |
| 3457 | .await?; |
| 3458 | post_heartbeat(&client, &enrollment, &runner_id, &start, "active").await?; |
| 3459 | } else { |
| 3460 | return Err(err); |
| 3461 | } |
| 3462 | } |
| 3463 | last_heartbeat = Instant::now(); |
| 3464 | } |
| 3465 | let runs = match list_runs(&client, &enrollment, &runner_id).await { |
| 3466 | Ok(v) => v, |
| 3467 | Err(err) if err == "runner_access_token_expired" => { |
| 3468 | refresh_enrollment_and_reconnect( |
| 3469 | &client, |
| 3470 | &mut enrollment, |
| 3471 | &mut runner_id, |
| 3472 | &start, |
| 3473 | &event_tx, |
| 3474 | ) |
| 3475 | .await?; |
| 3476 | list_runs(&client, &enrollment, &runner_id).await? |
| 3477 | } |
| 3478 | Err(err) => return Err(err), |
| 3479 | }; |
| 3480 | for run_id in runs { |
| 3481 | let since = command_cursor.get(&run_id).copied().unwrap_or(0); |
| 3482 | let listed_commands = match list_commands( |
| 3483 | &client, |
| 3484 | &enrollment, |
| 3485 | &runner_id, |
| 3486 | &run_id, |
| 3487 | since, |
| 3488 | ) |
| 3489 | .await |
| 3490 | { |
| 3491 | Ok(v) => v, |
| 3492 | Err(err) if err == "runner_access_token_expired" => { |
| 3493 | refresh_enrollment_and_reconnect( |
| 3494 | &client, |
| 3495 | &mut enrollment, |
| 3496 | &mut runner_id, |
| 3497 | &start, |
| 3498 | &event_tx, |
| 3499 | ) |
| 3500 | .await?; |
| 3501 | list_commands(&client, &enrollment, &runner_id, &run_id, since).await? |
| 3502 | } |
| 3503 | Err(err) => return Err(err), |
| 3504 | }; |
| 3505 | for listed in listed_commands { |
| 3506 | let seq = listed.seq; |
| 3507 | if !listed.ack_status.is_empty() { |
| 3508 | if listed.ack_status == "accepted" { |
| 3509 | let rr = recover_run( |
| 3510 | &client, |
| 3511 | &enrollment, |
| 3512 | &runner_id, |
| 3513 | &run_id, |
| 3514 | "accepted command has no terminal acknowledgement after runner restart", |
| 3515 | ) |
| 3516 | .await; |
| 3517 | if let Err(err) = rr { |
| 3518 | if err == "runner_access_token_expired" { |
| 3519 | refresh_enrollment_and_reconnect( |
| 3520 | &client, |
| 3521 | &mut enrollment, |
| 3522 | &mut runner_id, |
| 3523 | &start, |
| 3524 | &event_tx, |
| 3525 | ) |
| 3526 | .await?; |
| 3527 | recover_run( |
| 3528 | &client, |
| 3529 | &enrollment, |
| 3530 | &runner_id, |
| 3531 | &run_id, |
| 3532 | "accepted command has no terminal acknowledgement after runner restart", |
| 3533 | ) |
| 3534 | .await?; |
| 3535 | } else { |
| 3536 | return Err(err); |
| 3537 | } |
| 3538 | } |
| 3539 | } |
| 3540 | command_cursor.insert(run_id.clone(), seq); |
| 3541 | continue; |
| 3542 | } |
| 3543 | let command = parse_remote_command(&listed.command, &run_id)?; |
| 3544 | let fingerprint = command_fingerprint(&command); |
| 3545 | let key = (run_id.clone(), seq); |
| 3546 | if let Some(existing) = delivered.get(&key) { |
| 3547 | if existing != &fingerprint { |
| 3548 | return Err("The control plane replayed a changed command sequence.".to_string()); |
| 3549 | } |
| 3550 | } else { |
| 3551 | delivered.insert(key, fingerprint); |
| 3552 | // Runtime Chat reports `applied` only after the |
| 3553 | // durable native thread manager has accepted the |
| 3554 | // exact operation key. Leaving the command |
| 3555 | // unacknowledged until then makes crash replay |
| 3556 | // naturally re-enter the idempotent native path. |
| 3557 | if !command.is_runtime_chat() { |
| 3558 | let up = upload_command_accepted( |
| 3559 | &client, |
| 3560 | &enrollment, |
| 3561 | &runner_id, |
| 3562 | &run_id, |
| 3563 | seq, |
| 3564 | &command, |
| 3565 | ) |
| 3566 | .await; |
| 3567 | if let Err(err) = up { |
| 3568 | if err == "runner_access_token_expired" { |
| 3569 | refresh_enrollment_and_reconnect( |
| 3570 | &client, |
| 3571 | &mut enrollment, |
| 3572 | &mut runner_id, |
| 3573 | &start, |
| 3574 | &event_tx, |
| 3575 | ) |
| 3576 | .await?; |
| 3577 | upload_command_accepted( |
| 3578 | &client, |
| 3579 | &enrollment, |
| 3580 | &runner_id, |
| 3581 | &run_id, |
| 3582 | seq, |
| 3583 | &command, |
| 3584 | ) |
| 3585 | .await?; |
| 3586 | } else { |
| 3587 | return Err(err); |
| 3588 | } |
| 3589 | } |
| 3590 | } |
| 3591 | event_tx.send(RemoteEvent::Command { |
| 3592 | run_id: run_id.clone(), |
| 3593 | seq, |
| 3594 | command, |
| 3595 | }).map_err(|_| "The terminal remote-control owner stopped.".to_string())?; |
| 3596 | } |
| 3597 | command_cursor.insert(run_id.clone(), seq.max(since)); |
| 3598 | } |
| 3599 | } |
| 3600 | } |
| 3601 | } |
| 3602 | } |
| 3603 | } |
| 3604 | |
| 3605 | impl RuntimeTransportOutbox { |
| 3606 | fn enqueue(&mut self, run_id: &str, envelope: Value) -> Result<(), String> { |
| 3607 | if !valid_opaque_ref(run_id) { |
| 3608 | return Err("The local runtime queued an invalid run id.".to_string()); |
| 3609 | } |
| 3610 | let seq = runtime_envelope_seq(&envelope) |
| 3611 | .ok_or_else(|| "The local runtime queued an invalid event sequence.".to_string())?; |
| 3612 | let encoded = serde_json::to_vec(&envelope) |
| 3613 | .map_err(|_| "The local runtime could not encode an event.".to_string())?; |
| 3614 | if encoded.len() > MAX_RUNTIME_ENVELOPE_BYTES { |
| 3615 | return Err("The local runtime queued an oversized event.".to_string()); |
| 3616 | } |
| 3617 | let key = (run_id.to_string(), seq); |
| 3618 | if let Some(existing) = self.events.get(&key) { |
| 3619 | if existing != &envelope { |
| 3620 | return Err( |
| 3621 | "The local runtime changed an unacknowledged event sequence.".to_string(), |
| 3622 | ); |
| 3623 | } |
| 3624 | return Ok(()); |
| 3625 | } |
| 3626 | self.events.insert(key, envelope); |
| 3627 | Ok(()) |
| 3628 | } |
| 3629 | |
| 3630 | async fn try_flush_one( |
| 3631 | &mut self, |
| 3632 | client: &Client, |
| 3633 | enrollment: &LiveEnrollment, |
| 3634 | runner_id: &str, |
| 3635 | ) -> Result<RuntimeFlushOutcome, String> { |
| 3636 | let Some(((run_id, seq), envelope)) = self |
| 3637 | .events |
| 3638 | .first_key_value() |
| 3639 | .map(|(key, value)| (key.clone(), value.clone())) |
| 3640 | else { |
| 3641 | return Ok(RuntimeFlushOutcome::Idle); |
| 3642 | }; |
| 3643 | match post_runtime_event(client, enrollment, runner_id, &run_id, seq, &envelope).await? { |
| 3644 | RuntimePostOutcome::Retryable => Ok(RuntimeFlushOutcome::Retryable), |
| 3645 | RuntimePostOutcome::AccessTokenExpired => Ok(RuntimeFlushOutcome::AccessTokenExpired), |
| 3646 | RuntimePostOutcome::Accepted(cursor) => { |
| 3647 | self.events.retain(|(pending_run, pending_seq), _| { |
| 3648 | pending_run != &run_id || *pending_seq > cursor |
| 3649 | }); |
| 3650 | Ok(RuntimeFlushOutcome::Accepted { run_id, cursor }) |
| 3651 | } |
| 3652 | } |
| 3653 | } |
| 3654 | } |
| 3655 | |
| 3656 | /// Flushes every queued runtime envelope through the server-confirmed cursor |
| 3657 | /// before a stop may be acknowledged. Emits `RuntimeCursor` events so the |
| 3658 | /// controller compacts its journal as acknowledgements land. Failing to drain |
| 3659 | /// by `deadline` is a hard error: the stop is *not* confirmed and the caller |
| 3660 | /// must leave ownership locked. |
| 3661 | #[allow(clippy::too_many_arguments)] |
| 3662 | async fn drain_runtime_outbox_for_stop( |
| 3663 | client: &Client, |
| 3664 | enrollment: &mut LiveEnrollment, |
| 3665 | runner_id: &mut String, |
| 3666 | start: &RemoteStart, |
| 3667 | event_tx: &mpsc::UnboundedSender<RemoteEvent>, |
| 3668 | outbox: &mut RuntimeTransportOutbox, |
| 3669 | deadline: Instant, |
| 3670 | ) -> Result<(), String> { |
| 3671 | let mut delay = RUNTIME_UPLOAD_RETRY_INTERVAL; |
| 3672 | while !outbox.events.is_empty() { |
| 3673 | if Instant::now() >= deadline { |
| 3674 | return Err( |
| 3675 | "queued runtime events were not server-acknowledged in time; the stop was not confirmed" |
| 3676 | .to_string(), |
| 3677 | ); |
| 3678 | } |
| 3679 | match outbox.try_flush_one(client, enrollment, runner_id).await? { |
| 3680 | RuntimeFlushOutcome::Idle => break, |
| 3681 | RuntimeFlushOutcome::Accepted { run_id, cursor } => { |
| 3682 | delay = RUNTIME_UPLOAD_RETRY_INTERVAL; |
| 3683 | let _ = event_tx.send(RemoteEvent::RuntimeCursor { run_id, cursor }); |
| 3684 | } |
| 3685 | RuntimeFlushOutcome::Retryable => { |
| 3686 | tokio::time::sleep(delay).await; |
| 3687 | delay = delay.saturating_mul(2).min(RUNTIME_UPLOAD_MAX_BACKOFF); |
| 3688 | } |
| 3689 | RuntimeFlushOutcome::AccessTokenExpired => { |
| 3690 | refresh_enrollment_and_reconnect(client, enrollment, runner_id, start, event_tx) |
| 3691 | .await?; |
| 3692 | } |
| 3693 | } |
| 3694 | } |
| 3695 | Ok(()) |
| 3696 | } |
| 3697 | |
| 3698 | async fn post_runtime_event( |
| 3699 | client: &Client, |
| 3700 | enrollment: &LiveEnrollment, |
| 3701 | runner_id: &str, |
| 3702 | run_id: &str, |
| 3703 | seq: u64, |
| 3704 | envelope: &Value, |
| 3705 | ) -> Result<RuntimePostOutcome, String> { |
| 3706 | let url = control_plane_url( |
| 3707 | &enrollment.persisted.control_plane_base, |
| 3708 | &["api", "local-runners", runner_id, "runs", run_id, "events"], |
| 3709 | &[], |
| 3710 | )?; |
| 3711 | let response = match client |
| 3712 | .post(url) |
| 3713 | .bearer_auth(&enrollment.access_token) |
| 3714 | .json(&json!({ |
| 3715 | "acknowledgements": [], |
| 3716 | "envelopes": [envelope], |
| 3717 | })) |
| 3718 | .send() |
| 3719 | .await |
| 3720 | { |
| 3721 | Ok(response) => response, |
| 3722 | Err(_) => return Ok(RuntimePostOutcome::Retryable), |
| 3723 | }; |
| 3724 | let status = response.status(); |
| 3725 | if matches!(status, StatusCode::UNAUTHORIZED | StatusCode::FORBIDDEN) { |
| 3726 | return Ok(RuntimePostOutcome::AccessTokenExpired); |
| 3727 | } |
| 3728 | if matches!( |
| 3729 | status, |
| 3730 | StatusCode::REQUEST_TIMEOUT | StatusCode::TOO_EARLY | StatusCode::TOO_MANY_REQUESTS |
| 3731 | ) || status.is_server_error() |
| 3732 | { |
| 3733 | return Ok(RuntimePostOutcome::Retryable); |
| 3734 | } |
| 3735 | if !status.is_success() { |
| 3736 | return Err(format!( |
| 3737 | "The remote-control server rejected runtime event {seq} ({status})." |
| 3738 | )); |
| 3739 | } |
| 3740 | let value = match read_bounded_json(response).await { |
| 3741 | Ok(value) => value, |
| 3742 | Err(_) => return Ok(RuntimePostOutcome::Retryable), |
| 3743 | }; |
| 3744 | let Some(cursor) = value |
| 3745 | .get("cursor") |
| 3746 | .and_then(Value::as_u64) |
| 3747 | .filter(|cursor| *cursor >= seq && *cursor <= JS_MAX_SAFE_INTEGER) |
| 3748 | else { |
| 3749 | // A success without a durable cursor is indistinguishable from a lost |
| 3750 | // response. Retain and retry the exact same event body. |
| 3751 | return Ok(RuntimePostOutcome::Retryable); |
| 3752 | }; |
| 3753 | Ok(RuntimePostOutcome::Accepted(cursor)) |
| 3754 | } |
| 3755 | |
| 3756 | impl PersistedEnrollment { |
| 3757 | fn matches(&self, start: &RemoteStart, base: &str) -> bool { |
| 3758 | self.schema_version == 1 |
| 3759 | && self.control_plane_base == base |
| 3760 | && self.target_ref == start.target_ref |
| 3761 | && self.runtime_version == start.runtime_version |
| 3762 | && self.runtime_commit == start.runtime_commit |
| 3763 | && valid_opaque_ref(&self.runner_enrollment_id) |
| 3764 | && valid_opaque_ref(&self.account_ref) |
| 3765 | && valid_opaque_ref(&self.device_id) |
| 3766 | && valid_opaque_ref(&self.target_grant_ref) |
| 3767 | && valid_secret(&self.bootstrap_secret) |
| 3768 | } |
| 3769 | } |
| 3770 | |
| 3771 | async fn enroll_device( |
| 3772 | client: &Client, |
| 3773 | base: &str, |
| 3774 | start: &RemoteStart, |
| 3775 | device_id: &str, |
| 3776 | event_tx: &mpsc::UnboundedSender<RemoteEvent>, |
| 3777 | ) -> Result<LiveEnrollment, String> { |
| 3778 | let value = public_request( |
| 3779 | client, |
| 3780 | Method::POST, |
| 3781 | control_plane_url(base, &["api", "runner", "device", "start"], &[])?, |
| 3782 | json!({ |
| 3783 | "deviceId": device_id, |
| 3784 | "deviceLabel": "Codewhale terminal", |
| 3785 | "targetRef": start.target_ref, |
| 3786 | "targetLabel": start.workspace_label, |
| 3787 | "runtimeVersion": start.runtime_version, |
| 3788 | "runtimeCommit": start.runtime_commit, |
| 3789 | "capabilities": CAPABILITIES, |
| 3790 | }), |
| 3791 | ) |
| 3792 | .await?; |
| 3793 | let device_code = secret_field(&value, "deviceCode")?; |
| 3794 | let user_code = string_field(&value, "userCode")?; |
| 3795 | let verification_uri = string_field(&value, "verificationUriComplete")?; |
| 3796 | let interval = value |
| 3797 | .get("interval") |
| 3798 | .and_then(Value::as_u64) |
| 3799 | .filter(|value| (1..=30).contains(value)) |
| 3800 | .ok_or_else(|| { |
| 3801 | "Codewhale returned an invalid device authorization interval.".to_string() |
| 3802 | })?; |
| 3803 | let expires_in = value |
| 3804 | .get("expiresIn") |
| 3805 | .and_then(Value::as_u64) |
| 3806 | .filter(|value| (60..=1800).contains(value)) |
| 3807 | .ok_or_else(|| "Codewhale returned an invalid device authorization expiry.".to_string())?; |
| 3808 | validate_authorization_url(&verification_uri, &user_code)?; |
| 3809 | let _ = event_tx.send(RemoteEvent::Notice(format!( |
| 3810 | "Authorize this terminal at {verification_uri} (code {user_code})." |
| 3811 | ))); |
| 3812 | let _ = webbrowser::open(&verification_uri); |
| 3813 | let deadline = Instant::now() + Duration::from_secs(expires_in); |
| 3814 | loop { |
| 3815 | if Instant::now() >= deadline { |
| 3816 | return Err("Remote-control authorization expired; run /rc again.".to_string()); |
| 3817 | } |
| 3818 | tokio::time::sleep(Duration::from_secs(interval)).await; |
| 3819 | let response = client |
| 3820 | .post(control_plane_url( |
| 3821 | base, |
| 3822 | &["api", "runner", "device", "token"], |
| 3823 | &[], |
| 3824 | )?) |
| 3825 | .json(&json!({ "deviceCode": device_code })) |
| 3826 | .send() |
| 3827 | .await |
| 3828 | .map_err(|_| "Remote-control authorization could not reach Codewhale.".to_string())?; |
| 3829 | if response.status() == StatusCode::ACCEPTED { |
| 3830 | continue; |
| 3831 | } |
| 3832 | if !response.status().is_success() { |
| 3833 | return Err("Remote-control authorization was rejected.".to_string()); |
| 3834 | } |
| 3835 | let exchange = read_bounded_json(response).await?; |
| 3836 | let enrollment = enrollment_from_exchange(exchange, base, device_id, start)?; |
| 3837 | save_persisted_enrollment(&enrollment.persisted)?; |
| 3838 | return Ok(enrollment); |
| 3839 | } |
| 3840 | } |
| 3841 | |
| 3842 | fn enrollment_from_exchange( |
| 3843 | value: Value, |
| 3844 | base: &str, |
| 3845 | device_id: &str, |
| 3846 | start: &RemoteStart, |
| 3847 | ) -> Result<LiveEnrollment, String> { |
| 3848 | if value.get("status").and_then(Value::as_str) != Some("approved") { |
| 3849 | return Err("Codewhale returned an invalid runner credential.".to_string()); |
| 3850 | } |
| 3851 | let record = value |
| 3852 | .get("enrollment") |
| 3853 | .filter(|value| value.is_object()) |
| 3854 | .ok_or_else(|| "Codewhale returned an invalid runner credential.".to_string())?; |
| 3855 | let enrollment_id = opaque_field(record, "id")?; |
| 3856 | let account_ref = opaque_field(record, "userId")?; |
| 3857 | let returned_device = opaque_field(record, "deviceId")?; |
| 3858 | if returned_device != device_id |
| 3859 | || record.get("runtimeVersion").and_then(Value::as_str) |
| 3860 | != Some(start.runtime_version.as_str()) |
| 3861 | || record.get("runtimeCommit").and_then(Value::as_str) |
| 3862 | != Some(start.runtime_commit.as_str()) |
| 3863 | || !exact_capabilities(record.get("capabilities")) |
| 3864 | { |
| 3865 | return Err("The runner credential does not match this terminal.".to_string()); |
| 3866 | } |
| 3867 | let target_grant_ref = record |
| 3868 | .get("targetGrants") |
| 3869 | .and_then(Value::as_array) |
| 3870 | .and_then(|grants| { |
| 3871 | grants.iter().find(|grant| { |
| 3872 | grant.get("targetRef").and_then(Value::as_str) == Some(start.target_ref.as_str()) |
| 3873 | && grant |
| 3874 | .get("revokedAt") |
| 3875 | .and_then(Value::as_str) |
| 3876 | .unwrap_or_default() |
| 3877 | .is_empty() |
| 3878 | }) |
| 3879 | }) |
| 3880 | .and_then(|grant| grant.get("grantId")) |
| 3881 | .and_then(Value::as_str) |
| 3882 | .filter(|value| valid_opaque_ref(value)) |
| 3883 | .ok_or_else(|| "Codewhale returned no grant for this session.".to_string())? |
| 3884 | .to_string(); |
| 3885 | Ok(LiveEnrollment { |
| 3886 | persisted: PersistedEnrollment { |
| 3887 | schema_version: 1, |
| 3888 | control_plane_base: base.to_string(), |
| 3889 | runner_enrollment_id: enrollment_id, |
| 3890 | account_ref, |
| 3891 | device_id: returned_device, |
| 3892 | target_ref: start.target_ref.clone(), |
| 3893 | target_grant_ref, |
| 3894 | runtime_version: start.runtime_version.clone(), |
| 3895 | runtime_commit: start.runtime_commit.clone(), |
| 3896 | bootstrap_secret: secret_field(&value, "bootstrapSecret")?, |
| 3897 | }, |
| 3898 | access_token: access_token(&value)?, |
| 3899 | }) |
| 3900 | } |
| 3901 | |
| 3902 | async fn refresh_enrollment( |
| 3903 | client: &Client, |
| 3904 | persisted: PersistedEnrollment, |
| 3905 | ) -> Result<LiveEnrollment, String> { |
| 3906 | let url = control_plane_url( |
| 3907 | &persisted.control_plane_base, |
| 3908 | &["api", "runner", "enrollments", "token"], |
| 3909 | &[], |
| 3910 | )?; |
| 3911 | let response = client |
| 3912 | .post(url) |
| 3913 | .json(&json!({ |
| 3914 | "enrollmentId": persisted.runner_enrollment_id, |
| 3915 | "bootstrapSecret": persisted.bootstrap_secret, |
| 3916 | })) |
| 3917 | .send() |
| 3918 | .await |
| 3919 | .map_err(|_| "Remote-control credential refresh could not reach Codewhale.".to_string())?; |
| 3920 | if matches!( |
| 3921 | response.status(), |
| 3922 | StatusCode::UNAUTHORIZED | StatusCode::FORBIDDEN |
| 3923 | ) { |
| 3924 | return Err("runner_enrollment_revoked".to_string()); |
| 3925 | } |
| 3926 | if !response.status().is_success() { |
| 3927 | return Err("Remote-control credential refresh was rejected.".to_string()); |
| 3928 | } |
| 3929 | let value = read_bounded_json(response).await?; |
| 3930 | let record = value |
| 3931 | .get("enrollment") |
| 3932 | .filter(|value| value.is_object()) |
| 3933 | .ok_or_else(|| "Codewhale returned an invalid refreshed credential.".to_string())?; |
| 3934 | if record.get("id").and_then(Value::as_str) != Some(persisted.runner_enrollment_id.as_str()) |
| 3935 | || record.get("userId").and_then(Value::as_str) != Some(persisted.account_ref.as_str()) |
| 3936 | || record.get("deviceId").and_then(Value::as_str) != Some(persisted.device_id.as_str()) |
| 3937 | || record.get("runtimeVersion").and_then(Value::as_str) |
| 3938 | != Some(persisted.runtime_version.as_str()) |
| 3939 | || record.get("runtimeCommit").and_then(Value::as_str) |
| 3940 | != Some(persisted.runtime_commit.as_str()) |
| 3941 | || !exact_capabilities(record.get("capabilities")) |
| 3942 | { |
| 3943 | return Err("Codewhale returned a mismatched refreshed credential.".to_string()); |
| 3944 | } |
| 3945 | Ok(LiveEnrollment { |
| 3946 | persisted, |
| 3947 | access_token: access_token(&value)?, |
| 3948 | }) |
| 3949 | } |
| 3950 | |
| 3951 | async fn connect_runner( |
| 3952 | client: &Client, |
| 3953 | enrollment: &LiveEnrollment, |
| 3954 | start: &RemoteStart, |
| 3955 | ) -> Result<RunnerConnection, String> { |
| 3956 | let value = runner_request( |
| 3957 | client, |
| 3958 | enrollment, |
| 3959 | Method::POST, |
| 3960 | &["api", "local-runners", "connect"], |
| 3961 | &[], |
| 3962 | Some(connect_runner_body(enrollment, start)), |
| 3963 | ) |
| 3964 | .await?; |
| 3965 | parse_runner_connection(&value, enrollment, start) |
| 3966 | } |
| 3967 | |
| 3968 | fn connect_runner_body(enrollment: &LiveEnrollment, start: &RemoteStart) -> Value { |
| 3969 | let mut body = json!({ |
| 3970 | "deviceId": enrollment.persisted.device_id, |
| 3971 | "targetRef": start.target_ref, |
| 3972 | "displayLabel": start.workspace_label, |
| 3973 | "runtimeVersion": start.runtime_version, |
| 3974 | "runtimeCommit": start.runtime_commit, |
| 3975 | "capabilities": CAPABILITIES, |
| 3976 | "status": "active", |
| 3977 | // This is the only session attachment input. It is an opaque runtime |
| 3978 | // id, never a workspace path, prompt, environment, or credential. |
| 3979 | "sessionRef": start.session_id, |
| 3980 | "runtimeChatRelayProtocol": RUNTIME_CHAT_RELAY_PROTOCOL, |
| 3981 | }); |
| 3982 | if let Some(repo) = start |
| 3983 | .git_remote |
| 3984 | .as_deref() |
| 3985 | .and_then(normalize_observed_git_repo) |
| 3986 | { |
| 3987 | body["gitRemote"] = json!(repo); |
| 3988 | } |
| 3989 | body |
| 3990 | } |
| 3991 | |
| 3992 | /// Collapse a git remote to `owner/name`. Paths, credentials, and unknown |
| 3993 | /// hosts are dropped so the control plane never receives a folder identity. |
| 3994 | pub fn normalize_observed_git_repo(input: &str) -> Option<String> { |
| 3995 | let raw = input.trim(); |
| 3996 | if raw.is_empty() { |
| 3997 | return None; |
| 3998 | } |
| 3999 | let allowed_host = |host: &str| { |
| 4000 | matches!( |
| 4001 | host.to_ascii_lowercase().as_str(), |
| 4002 | "github.com" | "www.github.com" | "gitee.com" | "cnb.cool" |
| 4003 | ) |
| 4004 | }; |
| 4005 | let path = if let Some((authority, path)) = raw.split_once(':') |
| 4006 | && !raw.contains("://") |
| 4007 | && authority.starts_with("git@") |
| 4008 | && allowed_host(authority.trim_start_matches("git@")) |
| 4009 | { |
| 4010 | path.to_string() |
| 4011 | } else { |
| 4012 | let url = Url::parse(raw).ok()?; |
| 4013 | if url.scheme() != "https" |
| 4014 | || !url.username().is_empty() |
| 4015 | || url.password().is_some() |
| 4016 | || url.port().is_some() |
| 4017 | || url.query().is_some() |
| 4018 | || url.fragment().is_some() |
| 4019 | || !allowed_host(url.host_str()?) |
| 4020 | { |
| 4021 | return None; |
| 4022 | } |
| 4023 | url.path().trim_start_matches('/').to_string() |
| 4024 | }; |
| 4025 | let path = path.trim_end_matches('/').trim_end_matches(".git"); |
| 4026 | let mut parts = path.split('/'); |
| 4027 | let owner = parts.next()?; |
| 4028 | let name = parts.next()?; |
| 4029 | if parts.next().is_some() { |
| 4030 | return None; |
| 4031 | } |
| 4032 | if owner.len() > 80 || name.len() > 80 { |
| 4033 | return None; |
| 4034 | } |
| 4035 | if !owner |
| 4036 | .chars() |
| 4037 | .all(|ch| ch.is_ascii_alphanumeric() || matches!(ch, '.' | '_' | '-')) |
| 4038 | || !name |
| 4039 | .chars() |
| 4040 | .all(|ch| ch.is_ascii_alphanumeric() || matches!(ch, '.' | '_' | '-')) |
| 4041 | { |
| 4042 | return None; |
| 4043 | } |
| 4044 | if matches!(owner, "." | "..") || matches!(name, "." | "..") { |
| 4045 | return None; |
| 4046 | } |
| 4047 | Some(format!("{owner}/{name}")) |
| 4048 | } |
| 4049 | |
| 4050 | pub fn observed_git_repo(workspace: &Path) -> Option<String> { |
| 4051 | let output = std::process::Command::new("git") |
| 4052 | .arg("-C") |
| 4053 | .arg(workspace) |
| 4054 | .args(["remote", "get-url", "origin"]) |
| 4055 | .output() |
| 4056 | .ok()?; |
| 4057 | if !output.status.success() { |
| 4058 | return None; |
| 4059 | } |
| 4060 | normalize_observed_git_repo(std::str::from_utf8(&output.stdout).ok()?) |
| 4061 | } |
| 4062 | |
| 4063 | fn parse_runner_connection( |
| 4064 | value: &Value, |
| 4065 | enrollment: &LiveEnrollment, |
| 4066 | start: &RemoteStart, |
| 4067 | ) -> Result<RunnerConnection, String> { |
| 4068 | let response = value |
| 4069 | .as_object() |
| 4070 | .filter(|record| { |
| 4071 | record.len() == 2 && record.contains_key("runner") && record.contains_key("attachment") |
| 4072 | }) |
| 4073 | .ok_or_else(|| "Codewhale returned an invalid runner attachment response.".to_string())?; |
| 4074 | let runner = response |
| 4075 | .get("runner") |
| 4076 | .and_then(Value::as_object) |
| 4077 | .ok_or_else(|| "Codewhale returned an invalid runner lease.".to_string())?; |
| 4078 | let runner_id = runner |
| 4079 | .get("id") |
| 4080 | .and_then(Value::as_str) |
| 4081 | .filter(|value| valid_opaque_ref(value)) |
| 4082 | .map(ToString::to_string) |
| 4083 | .ok_or_else(|| "Codewhale returned an invalid runner lease.".to_string())?; |
| 4084 | let runner_binding_matches = runner.get("userId").and_then(Value::as_str) |
| 4085 | == Some(enrollment.persisted.account_ref.as_str()) |
| 4086 | && runner.get("deviceId").and_then(Value::as_str) |
| 4087 | == Some(enrollment.persisted.device_id.as_str()) |
| 4088 | && runner.get("targetRef").and_then(Value::as_str) == Some(start.target_ref.as_str()) |
| 4089 | && runner.get("runtimeVersion").and_then(Value::as_str) |
| 4090 | == Some(start.runtime_version.as_str()) |
| 4091 | && runner.get("runtimeCommit").and_then(Value::as_str) |
| 4092 | == Some(start.runtime_commit.as_str()) |
| 4093 | && runner.get("controlPath").and_then(Value::as_str) == Some("outbound_relay") |
| 4094 | && runner.get("status").and_then(Value::as_str) == Some("active") |
| 4095 | && runner.get("active").and_then(Value::as_bool) == Some(true) |
| 4096 | && exact_capabilities(runner.get("capabilities")); |
| 4097 | if !runner_binding_matches { |
| 4098 | return Err("Codewhale returned a runner lease for a different session.".to_string()); |
| 4099 | } |
| 4100 | |
| 4101 | let attachment = response |
| 4102 | .get("attachment") |
| 4103 | .and_then(Value::as_object) |
| 4104 | .filter(|record| { |
| 4105 | record.len() == 6 |
| 4106 | && record.contains_key("runId") |
| 4107 | && record.contains_key("workspaceId") |
| 4108 | && record.contains_key("runtimeCursor") |
| 4109 | && record.contains_key("snapshotPresent") |
| 4110 | && record.contains_key("runtimeChatRelayProtocol") |
| 4111 | && record.contains_key("runtimeChatRelayChallenge") |
| 4112 | }) |
| 4113 | .ok_or_else(|| "Codewhale returned an invalid session attachment.".to_string())?; |
| 4114 | let run_id = attachment |
| 4115 | .get("runId") |
| 4116 | .and_then(Value::as_str) |
| 4117 | .filter(|value| valid_opaque_ref(value)) |
| 4118 | .map(ToString::to_string) |
| 4119 | .ok_or_else(|| "Codewhale returned an invalid attached run.".to_string())?; |
| 4120 | let workspace_id = attachment |
| 4121 | .get("workspaceId") |
| 4122 | .and_then(Value::as_str) |
| 4123 | .filter(|value| valid_opaque_ref(value)) |
| 4124 | .map(ToString::to_string) |
| 4125 | .ok_or_else(|| "Codewhale returned an invalid attached workspace.".to_string())?; |
| 4126 | let runtime_cursor = attachment |
| 4127 | .get("runtimeCursor") |
| 4128 | .and_then(Value::as_u64) |
| 4129 | .filter(|value| *value <= JS_MAX_SAFE_INTEGER) |
| 4130 | .ok_or_else(|| "Codewhale returned an invalid runtime event cursor.".to_string())?; |
| 4131 | let snapshot_present = attachment |
| 4132 | .get("snapshotPresent") |
| 4133 | .and_then(Value::as_bool) |
| 4134 | .ok_or_else(|| "Codewhale returned an invalid snapshot receipt.".to_string())?; |
| 4135 | let runtime_chat_relay_protocol = attachment |
| 4136 | .get("runtimeChatRelayProtocol") |
| 4137 | .and_then(Value::as_str) |
| 4138 | .filter(|value| *value == RUNTIME_CHAT_RELAY_PROTOCOL) |
| 4139 | .map(ToString::to_string) |
| 4140 | .ok_or_else(|| { |
| 4141 | "Codewhale returned an unsupported Runtime Chat relay protocol.".to_string() |
| 4142 | })?; |
| 4143 | let runtime_chat_relay_challenge = attachment |
| 4144 | .get("runtimeChatRelayChallenge") |
| 4145 | .and_then(Value::as_str) |
| 4146 | .filter(|value| { |
| 4147 | (32..=128).contains(&value.len()) |
| 4148 | && value |
| 4149 | .bytes() |
| 4150 | .all(|byte| byte.is_ascii_alphanumeric() || matches!(byte, b'_' | b'-')) |
| 4151 | }) |
| 4152 | .map(ToString::to_string) |
| 4153 | .ok_or_else(|| "Codewhale returned an invalid Runtime Chat relay challenge.".to_string())?; |
| 4154 | |
| 4155 | let links = parse_remote_links(runner, &run_id); |
| 4156 | |
| 4157 | Ok(RunnerConnection { |
| 4158 | runner_id, |
| 4159 | attachment: RemoteAttachment { |
| 4160 | run_id, |
| 4161 | workspace_id, |
| 4162 | runtime_cursor, |
| 4163 | snapshot_present, |
| 4164 | runtime_chat_relay_protocol, |
| 4165 | runtime_chat_relay_challenge, |
| 4166 | }, |
| 4167 | links, |
| 4168 | }) |
| 4169 | } |
| 4170 | |
| 4171 | /// Read the optional `runUrl` / `computerUrl` advertised on the runner lease. |
| 4172 | /// Absent fields yield `None`; present-but-invalid values are dropped (never |
| 4173 | /// displayed, never opened) rather than failing the attachment, because a |
| 4174 | /// link is a convenience receipt and not part of the ownership contract. |
| 4175 | fn parse_remote_links(runner: &serde_json::Map<String, Value>, run_id: &str) -> RemoteLinks { |
| 4176 | let run_url = runner |
| 4177 | .get("runUrl") |
| 4178 | .and_then(Value::as_str) |
| 4179 | .and_then(|value| validate_run_url(value, run_id)); |
| 4180 | let computer_url = runner |
| 4181 | .get("computerUrl") |
| 4182 | .and_then(Value::as_str) |
| 4183 | .and_then(validate_computer_url); |
| 4184 | RemoteLinks { |
| 4185 | run_url, |
| 4186 | computer_url, |
| 4187 | } |
| 4188 | } |
| 4189 | |
| 4190 | /// Parse a web link and accept it only on the Codewhale app origin (or, in |
| 4191 | /// debug builds only, a loopback development origin) with no credentials, |
| 4192 | /// port, or fragment — the same shape `validate_authorization_url` enforces. |
| 4193 | fn app_link_url(value: &str) -> Option<Url> { |
| 4194 | let trimmed = value.trim(); |
| 4195 | if trimmed.is_empty() || trimmed.len() > 2048 { |
| 4196 | return None; |
| 4197 | } |
| 4198 | let url = Url::parse(trimmed).ok()?; |
| 4199 | let production_origin = |
| 4200 | url.scheme() == "https" && url.host_str() == Some(APP_ORIGIN_HOST) && url.port().is_none(); |
| 4201 | let debug_loopback = cfg!(debug_assertions) |
| 4202 | && url.scheme() == "http" |
| 4203 | && matches!(url.host_str(), Some("127.0.0.1" | "localhost")); |
| 4204 | if !(production_origin || debug_loopback) |
| 4205 | || !url.username().is_empty() |
| 4206 | || url.password().is_some() |
| 4207 | || url.fragment().is_some() |
| 4208 | { |
| 4209 | return None; |
| 4210 | } |
| 4211 | Some(url) |
| 4212 | } |
| 4213 | |
| 4214 | /// `/session?run=<runId>` for exactly the attached run; anything else is |
| 4215 | /// dropped so the terminal can never send the user to a different session. |
| 4216 | fn validate_run_url(value: &str, run_id: &str) -> Option<String> { |
| 4217 | let url = app_link_url(value)?; |
| 4218 | let pairs = url.query_pairs().collect::<Vec<_>>(); |
| 4219 | if url.path() != "/session" || pairs.len() != 1 || pairs[0].0 != "run" || pairs[0].1 != run_id { |
| 4220 | return None; |
| 4221 | } |
| 4222 | Some(url.to_string()) |
| 4223 | } |
| 4224 | |
| 4225 | /// `/settings` (optionally `?section=…`) on the app origin. |
| 4226 | fn validate_computer_url(value: &str) -> Option<String> { |
| 4227 | let url = app_link_url(value)?; |
| 4228 | let pairs = url.query_pairs().collect::<Vec<_>>(); |
| 4229 | if url.path() != "/settings" |
| 4230 | || pairs.len() > 1 |
| 4231 | || pairs.iter().any(|(key, _)| *key != "section") |
| 4232 | { |
| 4233 | return None; |
| 4234 | } |
| 4235 | Some(url.to_string()) |
| 4236 | } |
| 4237 | |
| 4238 | async fn post_heartbeat( |
| 4239 | client: &Client, |
| 4240 | enrollment: &LiveEnrollment, |
| 4241 | runner_id: &str, |
| 4242 | start: &RemoteStart, |
| 4243 | status: &str, |
| 4244 | ) -> Result<(), String> { |
| 4245 | runner_request( |
| 4246 | client, |
| 4247 | enrollment, |
| 4248 | Method::POST, |
| 4249 | &["api", "local-runners", runner_id, "heartbeat"], |
| 4250 | &[], |
| 4251 | Some(json!({ |
| 4252 | "runtimeVersion": start.runtime_version, |
| 4253 | "runtimeCommit": start.runtime_commit, |
| 4254 | "capabilities": CAPABILITIES, |
| 4255 | "status": status, |
| 4256 | })), |
| 4257 | ) |
| 4258 | .await |
| 4259 | .map(|_| ()) |
| 4260 | } |
| 4261 | |
| 4262 | async fn list_runs( |
| 4263 | client: &Client, |
| 4264 | enrollment: &LiveEnrollment, |
| 4265 | runner_id: &str, |
| 4266 | ) -> Result<Vec<String>, String> { |
| 4267 | let value = runner_request( |
| 4268 | client, |
| 4269 | enrollment, |
| 4270 | Method::GET, |
| 4271 | &["api", "local-runners", runner_id, "runs"], |
| 4272 | &[], |
| 4273 | None, |
| 4274 | ) |
| 4275 | .await?; |
| 4276 | let runs = value |
| 4277 | .get("runs") |
| 4278 | .and_then(Value::as_array) |
| 4279 | .filter(|runs| runs.len() <= MAX_RUNS) |
| 4280 | .ok_or_else(|| "Codewhale returned an invalid runner run list.".to_string())?; |
| 4281 | runs.iter() |
| 4282 | .map(|run| { |
| 4283 | run.get("id") |
| 4284 | .and_then(Value::as_str) |
| 4285 | .filter(|value| valid_opaque_ref(value)) |
| 4286 | .map(ToString::to_string) |
| 4287 | .ok_or_else(|| "Codewhale returned an invalid runner run.".to_string()) |
| 4288 | }) |
| 4289 | .collect() |
| 4290 | } |
| 4291 | |
| 4292 | async fn list_commands( |
| 4293 | client: &Client, |
| 4294 | enrollment: &LiveEnrollment, |
| 4295 | runner_id: &str, |
| 4296 | run_id: &str, |
| 4297 | since: u64, |
| 4298 | ) -> Result<Vec<ListedCommand>, String> { |
| 4299 | let value = runner_request( |
| 4300 | client, |
| 4301 | enrollment, |
| 4302 | Method::GET, |
| 4303 | &[ |
| 4304 | "api", |
| 4305 | "local-runners", |
| 4306 | runner_id, |
| 4307 | "runs", |
| 4308 | run_id, |
| 4309 | "commands", |
| 4310 | ], |
| 4311 | &[ |
| 4312 | ("since_seq", since.to_string()), |
| 4313 | ("include_accepted", "1".to_string()), |
| 4314 | ], |
| 4315 | None, |
| 4316 | ) |
| 4317 | .await?; |
| 4318 | let commands = value |
| 4319 | .get("commands") |
| 4320 | .and_then(Value::as_array) |
| 4321 | .filter(|commands| commands.len() <= MAX_COMMANDS) |
| 4322 | .ok_or_else(|| "Codewhale returned an invalid command list.".to_string())?; |
| 4323 | commands |
| 4324 | .iter() |
| 4325 | .map(|item| { |
| 4326 | let seq = item |
| 4327 | .get("seq") |
| 4328 | .and_then(Value::as_u64) |
| 4329 | .filter(|value| *value > since) |
| 4330 | .ok_or_else(|| "Codewhale returned an invalid command sequence.".to_string())?; |
| 4331 | let command = item |
| 4332 | .get("command") |
| 4333 | .filter(|value| value.is_object()) |
| 4334 | .cloned() |
| 4335 | .ok_or_else(|| "Codewhale returned an invalid typed command.".to_string())?; |
| 4336 | Ok(ListedCommand { |
| 4337 | seq, |
| 4338 | command, |
| 4339 | ack_status: item |
| 4340 | .get("ackStatus") |
| 4341 | .and_then(Value::as_str) |
| 4342 | .unwrap_or_default() |
| 4343 | .to_string(), |
| 4344 | }) |
| 4345 | }) |
| 4346 | .collect() |
| 4347 | } |
| 4348 | |
| 4349 | struct ListedCommand { |
| 4350 | seq: u64, |
| 4351 | command: Value, |
| 4352 | ack_status: String, |
| 4353 | } |
| 4354 | |
| 4355 | async fn upload_command_accepted( |
| 4356 | client: &Client, |
| 4357 | enrollment: &LiveEnrollment, |
| 4358 | runner_id: &str, |
| 4359 | run_id: &str, |
| 4360 | seq: u64, |
| 4361 | command: &RemoteCommand, |
| 4362 | ) -> Result<(), String> { |
| 4363 | runner_request( |
| 4364 | client, |
| 4365 | enrollment, |
| 4366 | Method::POST, |
| 4367 | &["api", "local-runners", runner_id, "runs", run_id, "events"], |
| 4368 | &[], |
| 4369 | Some(json!({ |
| 4370 | "acknowledgements": [{ |
| 4371 | "commandSeq": seq, |
| 4372 | "commandType": command.kind(), |
| 4373 | "status": "accepted", |
| 4374 | "turnId": command.turn_id(), |
| 4375 | }], |
| 4376 | "envelopes": [], |
| 4377 | })), |
| 4378 | ) |
| 4379 | .await |
| 4380 | .map(|_| ()) |
| 4381 | } |
| 4382 | |
| 4383 | async fn recover_run( |
| 4384 | client: &Client, |
| 4385 | enrollment: &LiveEnrollment, |
| 4386 | runner_id: &str, |
| 4387 | run_id: &str, |
| 4388 | reason: &str, |
| 4389 | ) -> Result<(), String> { |
| 4390 | runner_request( |
| 4391 | client, |
| 4392 | enrollment, |
| 4393 | Method::POST, |
| 4394 | &[ |
| 4395 | "api", |
| 4396 | "local-runners", |
| 4397 | runner_id, |
| 4398 | "runs", |
| 4399 | run_id, |
| 4400 | "recovery", |
| 4401 | ], |
| 4402 | &[], |
| 4403 | Some(json!({ "reason": reason })), |
| 4404 | ) |
| 4405 | .await |
| 4406 | .map(|_| ()) |
| 4407 | } |
| 4408 | |
| 4409 | fn parse_remote_command(value: &Value, expected_run_id: &str) -> Result<RemoteCommand, String> { |
| 4410 | if value.get("runId").and_then(Value::as_str) != Some(expected_run_id) { |
| 4411 | return Err("A remote command targeted a different run.".to_string()); |
| 4412 | } |
| 4413 | match value.get("type").and_then(Value::as_str) { |
| 4414 | Some("prompt.request") => { |
| 4415 | if value.get("images").is_some() && value.get("runtimeBindingId").is_none() { |
| 4416 | return Err("Image input is unavailable for legacy remote Work; use native Runtime or Runtime Chat.".to_string()); |
| 4417 | } |
| 4418 | let exact_legacy_prompt = value.as_object().is_some_and(|record| { |
| 4419 | record.len() == 4 |
| 4420 | && ["type", "runId", "turnId", "prompt"] |
| 4421 | .iter() |
| 4422 | .all(|key| record.contains_key(*key)) |
| 4423 | }); |
| 4424 | if !exact_legacy_prompt { |
| 4425 | // A managed Chat prompt has a rich deny-unknown schema. Any |
| 4426 | // field beyond the exact legacy Work shape must satisfy that |
| 4427 | // schema in full; missing binding/version/route/tool fields can |
| 4428 | // never downgrade into the permissive Work parser. |
| 4429 | let prompt: RuntimeChatPrompt = |
| 4430 | serde_json::from_value(value.clone()).map_err(|_| { |
| 4431 | "Codewhale sent an invalid Runtime Chat prompt contract.".to_string() |
| 4432 | })?; |
| 4433 | prompt.validate_shape()?; |
| 4434 | if prompt.run_id != expected_run_id { |
| 4435 | return Err("A Runtime Chat prompt targeted a different run.".to_string()); |
| 4436 | } |
| 4437 | return Ok(RemoteCommand::RuntimeChatPrompt(Box::new(prompt))); |
| 4438 | } |
| 4439 | let turn_id = value |
| 4440 | .get("turnId") |
| 4441 | .and_then(Value::as_str) |
| 4442 | .filter(|value| valid_opaque_ref(value)) |
| 4443 | .ok_or_else(|| "A remote prompt had no valid turn id.".to_string())?; |
| 4444 | let prompt = value |
| 4445 | .get("prompt") |
| 4446 | .and_then(Value::as_str) |
| 4447 | .map(str::trim) |
| 4448 | .filter(|value| !value.is_empty() && value.len() <= 128 * 1024) |
| 4449 | .ok_or_else(|| "A remote prompt was empty or oversized.".to_string())?; |
| 4450 | Ok(RemoteCommand::Prompt { |
| 4451 | turn_id: turn_id.to_string(), |
| 4452 | prompt: prompt.to_string(), |
| 4453 | }) |
| 4454 | } |
| 4455 | Some("approval.decision") => { |
| 4456 | let exact_approval = value.as_object().is_some_and(|record| { |
| 4457 | record.len() == 4 |
| 4458 | && ["type", "runId", "gate", "decision"] |
| 4459 | .iter() |
| 4460 | .all(|key| record.contains_key(*key)) |
| 4461 | }); |
| 4462 | if !exact_approval { |
| 4463 | return Err("Codewhale sent an invalid approval contract.".to_string()); |
| 4464 | } |
| 4465 | let gate = value |
| 4466 | .get("gate") |
| 4467 | .and_then(Value::as_str) |
| 4468 | .filter(|value| valid_opaque_ref(value)) |
| 4469 | .ok_or_else(|| "A remote approval had no valid gate id.".to_string())?; |
| 4470 | let approved = match value.get("decision").and_then(Value::as_str) { |
| 4471 | Some("approved") => true, |
| 4472 | Some("denied") => false, |
| 4473 | _ => return Err("A remote approval had an invalid decision.".to_string()), |
| 4474 | }; |
| 4475 | Ok(RemoteCommand::Approval { |
| 4476 | gate: gate.to_string(), |
| 4477 | approved, |
| 4478 | }) |
| 4479 | } |
| 4480 | Some("run.control") => { |
| 4481 | let rich_runtime_control = |
| 4482 | value.get("runtimeBindingId").is_some() || value.get("runtimeThreadId").is_some(); |
| 4483 | if rich_runtime_control { |
| 4484 | let record = value |
| 4485 | .as_object() |
| 4486 | .filter(|record| { |
| 4487 | record.len() == 7 |
| 4488 | && record.contains_key("type") |
| 4489 | && record.contains_key("runId") |
| 4490 | && record.contains_key("action") |
| 4491 | && record.contains_key("reason") |
| 4492 | && record.contains_key("turnId") |
| 4493 | && record.contains_key("runtimeBindingId") |
| 4494 | && record.contains_key("runtimeThreadId") |
| 4495 | }) |
| 4496 | .ok_or_else(|| { |
| 4497 | "Codewhale sent an invalid Runtime Chat interrupt contract.".to_string() |
| 4498 | })?; |
| 4499 | if record.get("action").and_then(Value::as_str) != Some("interrupt") { |
| 4500 | return Err("Runtime Chat supports only an exact turn interrupt.".to_string()); |
| 4501 | } |
| 4502 | let reason = record |
| 4503 | .get("reason") |
| 4504 | .and_then(Value::as_str) |
| 4505 | .filter(|reason| { |
| 4506 | !reason.trim().is_empty() && reason.len() <= 800 && !reason.contains('\0') |
| 4507 | }) |
| 4508 | .ok_or_else(|| "A Runtime Chat interrupt reason is invalid.".to_string())?; |
| 4509 | let _ = reason; |
| 4510 | let turn_id = record |
| 4511 | .get("turnId") |
| 4512 | .and_then(Value::as_str) |
| 4513 | .ok_or_else(|| "A Runtime Chat interrupt turn is invalid.".to_string())? |
| 4514 | .to_string(); |
| 4515 | let scope = RuntimeChatControlScope { |
| 4516 | runtime_binding_id: record |
| 4517 | .get("runtimeBindingId") |
| 4518 | .and_then(Value::as_str) |
| 4519 | .ok_or_else(|| "A Runtime Chat interrupt binding is invalid.".to_string())? |
| 4520 | .to_string(), |
| 4521 | runtime_thread_id: record |
| 4522 | .get("runtimeThreadId") |
| 4523 | .and_then(Value::as_str) |
| 4524 | .ok_or_else(|| "A Runtime Chat interrupt thread is invalid.".to_string())? |
| 4525 | .to_string(), |
| 4526 | }; |
| 4527 | scope.validate_for_turn(&turn_id)?; |
| 4528 | return Ok(RemoteCommand::Control { |
| 4529 | action: RemoteControlRequest::Interrupt, |
| 4530 | turn_id: Some(turn_id), |
| 4531 | runtime_chat: Some(scope), |
| 4532 | }); |
| 4533 | } |
| 4534 | let exact_legacy_control = value.as_object().is_some_and(|record| { |
| 4535 | record.len() == 4 |
| 4536 | && ["type", "runId", "action", "turnId"] |
| 4537 | .iter() |
| 4538 | .all(|key| record.contains_key(*key)) |
| 4539 | }); |
| 4540 | if !exact_legacy_control { |
| 4541 | return Err("Codewhale sent an invalid run-control contract.".to_string()); |
| 4542 | } |
| 4543 | let action = match value.get("action").and_then(Value::as_str) { |
| 4544 | Some("interrupt") => RemoteControlRequest::Interrupt, |
| 4545 | Some("cancel") => RemoteControlRequest::Cancel, |
| 4546 | _ => return Err("A remote run-control command had an invalid action.".to_string()), |
| 4547 | }; |
| 4548 | let turn_id = value |
| 4549 | .get("turnId") |
| 4550 | .and_then(Value::as_str) |
| 4551 | .filter(|turn_id| valid_opaque_ref(turn_id)) |
| 4552 | .map(ToString::to_string) |
| 4553 | .ok_or_else(|| "A remote run-control command had an invalid turn.".to_string())?; |
| 4554 | Ok(RemoteCommand::Control { |
| 4555 | action, |
| 4556 | turn_id: Some(turn_id), |
| 4557 | runtime_chat: None, |
| 4558 | }) |
| 4559 | } |
| 4560 | _ => Err("Codewhale sent an unsupported remote command.".to_string()), |
| 4561 | } |
| 4562 | } |
| 4563 | |
| 4564 | async fn runner_request( |
| 4565 | client: &Client, |
| 4566 | enrollment: &LiveEnrollment, |
| 4567 | method: Method, |
| 4568 | segments: &[&str], |
| 4569 | query: &[(&str, String)], |
| 4570 | body: Option<Value>, |
| 4571 | ) -> Result<Value, String> { |
| 4572 | let url = control_plane_url(&enrollment.persisted.control_plane_base, segments, query)?; |
| 4573 | let mut request = client |
| 4574 | .request(method, url) |
| 4575 | .bearer_auth(&enrollment.access_token); |
| 4576 | if let Some(body) = body { |
| 4577 | request = request.json(&body); |
| 4578 | } |
| 4579 | let response = request |
| 4580 | .send() |
| 4581 | .await |
| 4582 | .map_err(|_| "Remote control lost its secure connection.".to_string())?; |
| 4583 | if matches!( |
| 4584 | response.status(), |
| 4585 | StatusCode::UNAUTHORIZED | StatusCode::FORBIDDEN |
| 4586 | ) { |
| 4587 | return Err("runner_access_token_expired".to_string()); |
| 4588 | } |
| 4589 | if !response.status().is_success() { |
| 4590 | let status = response.status(); |
| 4591 | let excerpt = rejection_excerpt(response).await; |
| 4592 | return Err(match excerpt { |
| 4593 | Some(reason) => { |
| 4594 | format!("The remote-control server rejected a request ({status}): {reason}.") |
| 4595 | } |
| 4596 | None => format!("The remote-control server rejected a request ({status})."), |
| 4597 | }); |
| 4598 | } |
| 4599 | let limit = if segments.last() == Some(&"commands") { |
| 4600 | codewhale_protocol::runtime::MAX_RUNTIME_IMAGE_BODY_BYTES |
| 4601 | } else { |
| 4602 | MAX_RESPONSE_BYTES |
| 4603 | }; |
| 4604 | read_bounded_json_with_limit(response, limit).await |
| 4605 | } |
| 4606 | |
| 4607 | async fn public_request( |
| 4608 | client: &Client, |
| 4609 | method: Method, |
| 4610 | url: Url, |
| 4611 | body: Value, |
| 4612 | ) -> Result<Value, String> { |
| 4613 | let response = client |
| 4614 | .request(method, url) |
| 4615 | .json(&body) |
| 4616 | .send() |
| 4617 | .await |
| 4618 | .map_err(|_| "Remote control could not reach Codewhale.".to_string())?; |
| 4619 | if !response.status().is_success() { |
| 4620 | let status = response.status(); |
| 4621 | let excerpt = rejection_excerpt(response).await; |
| 4622 | return Err(match excerpt { |
| 4623 | Some(reason) => { |
| 4624 | format!("Codewhale rejected remote-control enrollment ({status}): {reason}.") |
| 4625 | } |
| 4626 | None => format!("Codewhale rejected remote-control enrollment ({status})."), |
| 4627 | }); |
| 4628 | } |
| 4629 | read_bounded_json(response).await |
| 4630 | } |
| 4631 | |
| 4632 | /// Bounded error-body read: at most MAX_RESPONSE_BYTES, so a misbehaving |
| 4633 | /// control plane cannot force an unbounded in-memory read through the |
| 4634 | /// rejection-excerpt path. |
| 4635 | async fn rejection_excerpt(response: reqwest::Response) -> Option<String> { |
| 4636 | if response |
| 4637 | .content_length() |
| 4638 | .is_some_and(|length| length > MAX_RESPONSE_BYTES as u64) |
| 4639 | { |
| 4640 | return None; |
| 4641 | } |
| 4642 | let mut body = Vec::new(); |
| 4643 | let mut response = response; |
| 4644 | while let Some(chunk) = response.chunk().await.ok()? { |
| 4645 | body.extend_from_slice(&chunk); |
| 4646 | if body.len() > MAX_RESPONSE_BYTES { |
| 4647 | return None; |
| 4648 | } |
| 4649 | } |
| 4650 | sanitized_rejection_excerpt(&body) |
| 4651 | } |
| 4652 | |
| 4653 | /// Sanitized, bounded reason excerpt from a rejection body. Only the |
| 4654 | /// conventional error fields are read, control characters are stripped, and |
| 4655 | /// the excerpt is capped — raw server bytes are never echoed further. |
| 4656 | fn sanitized_rejection_excerpt(body: &[u8]) -> Option<String> { |
| 4657 | let parsed: Value = serde_json::from_slice(body).ok()?; |
| 4658 | for field in ["error", "message", "title"] { |
| 4659 | if let Some(text) = parsed.get(field).and_then(Value::as_str) { |
| 4660 | let cleaned: String = text |
| 4661 | .chars() |
| 4662 | .filter(|character| !character.is_control()) |
| 4663 | .take(140) |
| 4664 | .collect(); |
| 4665 | let trimmed = cleaned.trim(); |
| 4666 | if !trimmed.is_empty() { |
| 4667 | return Some(trimmed.to_string()); |
| 4668 | } |
| 4669 | } |
| 4670 | } |
| 4671 | None |
| 4672 | } |
| 4673 | |
| 4674 | async fn read_bounded_json(response: reqwest::Response) -> Result<Value, String> { |
| 4675 | read_bounded_json_with_limit(response, MAX_RESPONSE_BYTES).await |
| 4676 | } |
| 4677 | |
| 4678 | async fn read_bounded_json_with_limit( |
| 4679 | mut response: reqwest::Response, |
| 4680 | limit: usize, |
| 4681 | ) -> Result<Value, String> { |
| 4682 | if response |
| 4683 | .content_length() |
| 4684 | .is_some_and(|length| length > limit as u64) |
| 4685 | { |
| 4686 | return Err("Codewhale returned an oversized remote-control response.".to_string()); |
| 4687 | } |
| 4688 | let mut bytes = Vec::new(); |
| 4689 | while let Some(chunk) = response |
| 4690 | .chunk() |
| 4691 | .await |
| 4692 | .map_err(|_| "Codewhale returned an unreadable response.".to_string())? |
| 4693 | { |
| 4694 | if bytes.len().saturating_add(chunk.len()) > limit { |
| 4695 | return Err("Codewhale returned an oversized remote-control response.".to_string()); |
| 4696 | } |
| 4697 | bytes.extend_from_slice(&chunk); |
| 4698 | } |
| 4699 | serde_json::from_slice(&bytes) |
| 4700 | .map_err(|_| "Codewhale returned an invalid remote-control response.".to_string()) |
| 4701 | } |
| 4702 | |
| 4703 | fn runner_control_plane_base() -> Result<String, String> { |
| 4704 | if cfg!(debug_assertions) |
| 4705 | && let Ok(value) = std::env::var("CWC_RUNNER_CONTROL_PLANE_BASE") |
| 4706 | { |
| 4707 | let parsed = |
| 4708 | Url::parse(&value).map_err(|_| "The runner control plane is invalid.".to_string())?; |
| 4709 | let loopback = parsed.scheme() == "http" |
| 4710 | && matches!(parsed.host_str(), Some("127.0.0.1" | "localhost")) |
| 4711 | && parsed.path() == "/" |
| 4712 | && parsed.query().is_none() |
| 4713 | && parsed.fragment().is_none(); |
| 4714 | if loopback { |
| 4715 | return Ok(parsed.to_string()); |
| 4716 | } |
| 4717 | return Err( |
| 4718 | "Debug remote control only accepts an explicit loopback control plane.".to_string(), |
| 4719 | ); |
| 4720 | } |
| 4721 | Ok(PRODUCTION_CONTROL_PLANE.to_string()) |
| 4722 | } |
| 4723 | |
| 4724 | fn control_plane_url( |
| 4725 | base: &str, |
| 4726 | segments: &[&str], |
| 4727 | query: &[(&str, String)], |
| 4728 | ) -> Result<Url, String> { |
| 4729 | let mut url = |
| 4730 | Url::parse(base).map_err(|_| "The runner control plane is invalid.".to_string())?; |
| 4731 | { |
| 4732 | let mut path = url |
| 4733 | .path_segments_mut() |
| 4734 | .map_err(|_| "The runner control plane is invalid.".to_string())?; |
| 4735 | path.pop_if_empty(); |
| 4736 | for segment in segments { |
| 4737 | path.push(segment); |
| 4738 | } |
| 4739 | } |
| 4740 | if !query.is_empty() { |
| 4741 | let mut pairs = url.query_pairs_mut(); |
| 4742 | for (key, value) in query { |
| 4743 | pairs.append_pair(key, value); |
| 4744 | } |
| 4745 | } |
| 4746 | Ok(url) |
| 4747 | } |
| 4748 | |
| 4749 | fn load_persisted_enrollment() -> Result<Option<PersistedEnrollment>, String> { |
| 4750 | let Some(raw) = remote_control_secrets() |
| 4751 | .get(ENROLLMENT_SECRET_SLOT) |
| 4752 | .map_err(|error| format!("Could not read the saved remote-control enrollment: {error}"))? |
| 4753 | else { |
| 4754 | return Ok(None); |
| 4755 | }; |
| 4756 | serde_json::from_str(&raw) |
| 4757 | .map(Some) |
| 4758 | .map_err(|_| "The saved remote-control enrollment is invalid.".to_string()) |
| 4759 | } |
| 4760 | |
| 4761 | fn save_persisted_enrollment(enrollment: &PersistedEnrollment) -> Result<(), String> { |
| 4762 | let raw = serde_json::to_string(enrollment) |
| 4763 | .map_err(|_| "Could not encode the remote-control enrollment.".to_string())?; |
| 4764 | remote_control_secrets() |
| 4765 | .set(ENROLLMENT_SECRET_SLOT, &raw) |
| 4766 | .map_err(|error| format!("Could not securely save the remote-control enrollment: {error}")) |
| 4767 | } |
| 4768 | |
| 4769 | fn load_persisted_device_identity() -> Result<Option<PersistedDeviceIdentity>, String> { |
| 4770 | let Some(raw) = remote_control_secrets() |
| 4771 | .get(DEVICE_IDENTITY_SECRET_SLOT) |
| 4772 | .map_err(|error| format!("Could not read the saved remote-control device id: {error}"))? |
| 4773 | else { |
| 4774 | return Ok(None); |
| 4775 | }; |
| 4776 | // An unreadable identity is replaced rather than fatal: the worst case is |
| 4777 | // one extra computer row, never a lost session. |
| 4778 | Ok(serde_json::from_str(&raw).ok()) |
| 4779 | } |
| 4780 | |
| 4781 | fn save_persisted_device_identity(identity: &PersistedDeviceIdentity) -> Result<(), String> { |
| 4782 | let raw = serde_json::to_string(identity) |
| 4783 | .map_err(|_| "Could not encode the remote-control device id.".to_string())?; |
| 4784 | remote_control_secrets() |
| 4785 | .set(DEVICE_IDENTITY_SECRET_SLOT, &raw) |
| 4786 | .map_err(|error| format!("Could not securely save the remote-control device id: {error}")) |
| 4787 | } |
| 4788 | |
| 4789 | /// Load (or mint and persist) the machine-stable device id. |
| 4790 | fn stable_device_id(enrollment_device_id: Option<&str>) -> Result<String, String> { |
| 4791 | let saved = load_persisted_device_identity()?; |
| 4792 | let (device_id, needs_save) = resolve_device_identity(saved, enrollment_device_id); |
| 4793 | if needs_save { |
| 4794 | save_persisted_device_identity(&PersistedDeviceIdentity { |
| 4795 | schema_version: 1, |
| 4796 | device_id: device_id.clone(), |
| 4797 | })?; |
| 4798 | } |
| 4799 | Ok(device_id) |
| 4800 | } |
| 4801 | |
| 4802 | fn delete_persisted_enrollment() { |
| 4803 | if let Err(error) = remote_control_secrets().delete(ENROLLMENT_SECRET_SLOT) { |
| 4804 | tracing::warn!("could not delete revoked remote-control enrollment: {error}"); |
| 4805 | } |
| 4806 | } |
| 4807 | |
| 4808 | /// Remote-control enrollment/device receipts deliberately use Codewhale's |
| 4809 | /// permission-bounded file store even when the operator chose the OS keyring |
| 4810 | /// for model-provider credentials. This keeps `/rc` and desktop dogfood from |
| 4811 | /// triggering a Keychain access dialog while preserving provider credential |
| 4812 | /// custody and its explicit backend selection unchanged. |
| 4813 | fn remote_control_secrets() -> codewhale_secrets::Secrets { |
| 4814 | codewhale_secrets::Secrets::file_backed() |
| 4815 | } |
| 4816 | |
| 4817 | fn install_reconnected_enrollment( |
| 4818 | current: &mut LiveEnrollment, |
| 4819 | candidate: LiveEnrollment, |
| 4820 | start: &RemoteStart, |
| 4821 | ) -> Result<(), String> { |
| 4822 | if candidate.persisted.account_ref != current.persisted.account_ref |
| 4823 | || candidate.persisted.target_ref != current.persisted.target_ref |
| 4824 | || candidate.persisted.target_ref != start.target_ref |
| 4825 | || candidate.persisted.device_id != current.persisted.device_id |
| 4826 | { |
| 4827 | return Err( |
| 4828 | "Remote control stopped because re-authorization selected a different account or folder. Start a new local session to switch authority." |
| 4829 | .to_string(), |
| 4830 | ); |
| 4831 | } |
| 4832 | *current = candidate; |
| 4833 | Ok(()) |
| 4834 | } |
| 4835 | |
| 4836 | async fn refresh_enrollment_and_reconnect( |
| 4837 | client: &Client, |
| 4838 | enrollment: &mut LiveEnrollment, |
| 4839 | runner_id: &mut String, |
| 4840 | start: &RemoteStart, |
| 4841 | event_tx: &mpsc::UnboundedSender<RemoteEvent>, |
| 4842 | ) -> Result<(), String> { |
| 4843 | let base = enrollment.persisted.control_plane_base.clone(); |
| 4844 | match refresh_enrollment(client, enrollment.persisted.clone()).await { |
| 4845 | Ok(new_enrollment) => { |
| 4846 | install_reconnected_enrollment(enrollment, new_enrollment, start)?; |
| 4847 | reconnect_runner(client, enrollment, runner_id, start, event_tx).await |
| 4848 | } |
| 4849 | Err(err) if err == "runner_enrollment_revoked" => { |
| 4850 | delete_persisted_enrollment(); |
| 4851 | let device_id = enrollment.persisted.device_id.clone(); |
| 4852 | let candidate = enroll_device(client, &base, start, &device_id, event_tx).await?; |
| 4853 | if let Err(error) = install_reconnected_enrollment(enrollment, candidate, start) { |
| 4854 | delete_persisted_enrollment(); |
| 4855 | return Err(error); |
| 4856 | } |
| 4857 | reconnect_runner(client, enrollment, runner_id, start, event_tx).await |
| 4858 | } |
| 4859 | Err(err) => Err(err), |
| 4860 | } |
| 4861 | } |
| 4862 | |
| 4863 | async fn reconnect_runner( |
| 4864 | client: &Client, |
| 4865 | enrollment: &LiveEnrollment, |
| 4866 | runner_id: &mut String, |
| 4867 | start: &RemoteStart, |
| 4868 | event_tx: &mpsc::UnboundedSender<RemoteEvent>, |
| 4869 | ) -> Result<(), String> { |
| 4870 | let connection = connect_runner(client, enrollment, start).await?; |
| 4871 | *runner_id = connection.runner_id; |
| 4872 | event_tx |
| 4873 | .send(RemoteEvent::Attachment { |
| 4874 | account_ref: enrollment.persisted.account_ref.clone(), |
| 4875 | target_ref: enrollment.persisted.target_ref.clone(), |
| 4876 | attachment: connection.attachment, |
| 4877 | links: connection.links, |
| 4878 | }) |
| 4879 | .map_err(|_| "The terminal remote-control owner stopped.".to_string()) |
| 4880 | } |
| 4881 | |
| 4882 | fn enrollment_needs_refresh(enrollment: &LiveEnrollment) -> bool { |
| 4883 | jwt_expiry(&enrollment.access_token) |
| 4884 | .is_none_or(|expiry| expiry <= epoch_seconds().saturating_add(60)) |
| 4885 | } |
| 4886 | |
| 4887 | fn jwt_expiry(token: &str) -> Option<u64> { |
| 4888 | use base64::{Engine as _, engine::general_purpose::URL_SAFE_NO_PAD}; |
| 4889 | let payload = URL_SAFE_NO_PAD.decode(token.split('.').nth(1)?).ok()?; |
| 4890 | serde_json::from_slice::<Value>(&payload) |
| 4891 | .ok()? |
| 4892 | .get("exp")? |
| 4893 | .as_u64() |
| 4894 | } |
| 4895 | |
| 4896 | fn access_token(value: &Value) -> Result<String, String> { |
| 4897 | let token = value |
| 4898 | .get("credential") |
| 4899 | .and_then(|value| value.get("accessToken")) |
| 4900 | .and_then(Value::as_str) |
| 4901 | .filter(|value| { |
| 4902 | (64..=8192).contains(&value.len()) && !value.chars().any(char::is_whitespace) |
| 4903 | }) |
| 4904 | .ok_or_else(|| "Codewhale returned an invalid runner access token.".to_string())? |
| 4905 | .to_string(); |
| 4906 | if jwt_expiry(&token).is_none_or(|expiry| expiry <= epoch_seconds()) { |
| 4907 | return Err("Codewhale returned an expired runner access token.".to_string()); |
| 4908 | } |
| 4909 | Ok(token) |
| 4910 | } |
| 4911 | |
| 4912 | fn exact_capabilities(value: Option<&Value>) -> bool { |
| 4913 | let Some(items) = value.and_then(Value::as_array) else { |
| 4914 | return false; |
| 4915 | }; |
| 4916 | let mut actual = items.iter().filter_map(Value::as_str).collect::<Vec<_>>(); |
| 4917 | actual.sort_unstable(); |
| 4918 | actual == CAPABILITIES |
| 4919 | } |
| 4920 | |
| 4921 | fn validate_authorization_url(value: &str, user_code: &str) -> Result<(), String> { |
| 4922 | let url = Url::parse(value) |
| 4923 | .map_err(|_| "Codewhale returned an invalid authorization URL.".to_string())?; |
| 4924 | let pairs = url.query_pairs().collect::<Vec<_>>(); |
| 4925 | if url.scheme() != "https" |
| 4926 | || url.host_str() != Some("app.codewhale.net") |
| 4927 | || url.path() != "/runner/authorize" |
| 4928 | || url.port().is_some() |
| 4929 | || !url.username().is_empty() |
| 4930 | || url.password().is_some() |
| 4931 | || url.fragment().is_some() |
| 4932 | || pairs.len() != 1 |
| 4933 | || pairs[0].0 != "user_code" |
| 4934 | || pairs[0].1 != user_code |
| 4935 | { |
| 4936 | return Err("Codewhale returned an invalid authorization URL.".to_string()); |
| 4937 | } |
| 4938 | Ok(()) |
| 4939 | } |
| 4940 | |
| 4941 | fn string_field(value: &Value, field: &str) -> Result<String, String> { |
| 4942 | value |
| 4943 | .get(field) |
| 4944 | .and_then(Value::as_str) |
| 4945 | .map(str::trim) |
| 4946 | .filter(|value| !value.is_empty() && value.len() <= 2048) |
| 4947 | .map(ToString::to_string) |
| 4948 | .ok_or_else(|| format!("Codewhale returned an invalid {field}.")) |
| 4949 | } |
| 4950 | |
| 4951 | fn secret_field(value: &Value, field: &str) -> Result<String, String> { |
| 4952 | value |
| 4953 | .get(field) |
| 4954 | .and_then(Value::as_str) |
| 4955 | .filter(|value| valid_secret(value)) |
| 4956 | .map(ToString::to_string) |
| 4957 | .ok_or_else(|| format!("Codewhale returned an invalid {field}.")) |
| 4958 | } |
| 4959 | |
| 4960 | fn opaque_field(value: &Value, field: &str) -> Result<String, String> { |
| 4961 | value |
| 4962 | .get(field) |
| 4963 | .and_then(Value::as_str) |
| 4964 | .filter(|value| valid_opaque_ref(value)) |
| 4965 | .map(ToString::to_string) |
| 4966 | .ok_or_else(|| format!("Codewhale returned an invalid {field}.")) |
| 4967 | } |
| 4968 | |
| 4969 | fn valid_opaque_ref(value: &str) -> bool { |
| 4970 | (3..=160).contains(&value.len()) |
| 4971 | && value |
| 4972 | .bytes() |
| 4973 | .all(|byte| byte.is_ascii_alphanumeric() || matches!(byte, b'_' | b'-' | b'.')) |
| 4974 | } |
| 4975 | |
| 4976 | fn valid_session_ref(value: &str) -> bool { |
| 4977 | (1..=160).contains(&value.len()) |
| 4978 | && value |
| 4979 | .bytes() |
| 4980 | .next() |
| 4981 | .is_some_and(|byte| byte.is_ascii_alphanumeric()) |
| 4982 | && value.bytes().all(|byte| { |
| 4983 | byte.is_ascii_alphanumeric() || matches!(byte, b'_' | b'-' | b'.' | b':' | b'@') |
| 4984 | }) |
| 4985 | && !value.contains("..") |
| 4986 | } |
| 4987 | |
| 4988 | fn valid_secret(value: &str) -> bool { |
| 4989 | (32..=8192).contains(&value.len()) && !value.chars().any(char::is_whitespace) |
| 4990 | } |
| 4991 | |
| 4992 | fn valid_runtime_version(value: &str) -> bool { |
| 4993 | semver::Version::parse(value).is_ok() && value.len() <= 64 |
| 4994 | } |
| 4995 | |
| 4996 | fn valid_runtime_commit(value: &str) -> bool { |
| 4997 | value.len() == 40 && value.bytes().all(|byte| byte.is_ascii_hexdigit()) |
| 4998 | } |
| 4999 | |
| 5000 | fn epoch_seconds() -> u64 { |
| 5001 | SystemTime::now() |
| 5002 | .duration_since(UNIX_EPOCH) |
| 5003 | .unwrap_or_default() |
| 5004 | .as_secs() |
| 5005 | } |
| 5006 | |
| 5007 | #[cfg(test)] |
| 5008 | mod tests { |
| 5009 | use super::*; |
| 5010 | use codewhale_models::Role; |
| 5011 | use std::sync::{ |
| 5012 | Arc, Mutex, |
| 5013 | atomic::{AtomicUsize, Ordering}, |
| 5014 | }; |
| 5015 | use wiremock::{ |
| 5016 | Mock, MockServer, Request, Respond, ResponseTemplate, |
| 5017 | matchers::{body_json, method, path, query_param}, |
| 5018 | }; |
| 5019 | |
| 5020 | #[derive(Clone, Default)] |
| 5021 | struct AmbiguousRuntimeResponder { |
| 5022 | bodies: Arc<Mutex<Vec<Value>>>, |
| 5023 | } |
| 5024 | |
| 5025 | #[derive(Clone, Default)] |
| 5026 | struct ControlPollCounter { |
| 5027 | hits: Arc<AtomicUsize>, |
| 5028 | } |
| 5029 | |
| 5030 | impl Respond for ControlPollCounter { |
| 5031 | fn respond(&self, _request: &Request) -> ResponseTemplate { |
| 5032 | self.hits.fetch_add(1, Ordering::SeqCst); |
| 5033 | ResponseTemplate::new(200).set_body_json(json!({ "runs": [] })) |
| 5034 | } |
| 5035 | } |
| 5036 | |
| 5037 | impl Respond for AmbiguousRuntimeResponder { |
| 5038 | fn respond(&self, request: &Request) -> ResponseTemplate { |
| 5039 | let body: Value = serde_json::from_slice(&request.body).expect("runtime request JSON"); |
| 5040 | let mut bodies = self.bodies.lock().expect("runtime request bodies"); |
| 5041 | bodies.push(body); |
| 5042 | if bodies.len() == 1 { |
| 5043 | // Model a committed request whose response was truncated in |
| 5044 | // transit. The client must keep the exact body and retry. |
| 5045 | ResponseTemplate::new(200).set_body_raw("{", "application/json") |
| 5046 | } else { |
| 5047 | ResponseTemplate::new(200).set_body_json(json!({ |
| 5048 | "accepted": [], |
| 5049 | "count": 0, |
| 5050 | "cursor": 1 |
| 5051 | })) |
| 5052 | } |
| 5053 | } |
| 5054 | } |
| 5055 | |
| 5056 | fn text_message(role: &str, text: impl Into<String>) -> Message { |
| 5057 | Message { |
| 5058 | role: Role::from(role), |
| 5059 | content: vec![ContentBlock::Text { |
| 5060 | text: text.into(), |
| 5061 | cache_control: None, |
| 5062 | }], |
| 5063 | } |
| 5064 | } |
| 5065 | |
| 5066 | fn fixture_start() -> RemoteStart { |
| 5067 | RemoteStart { |
| 5068 | workspace_label: "private-project".to_string(), |
| 5069 | target_ref: "target_fixture".to_string(), |
| 5070 | session_id: "session:fixture@01".to_string(), |
| 5071 | runtime_version: "0.9.6".to_string(), |
| 5072 | runtime_commit: "a".repeat(40), |
| 5073 | journal_dir: None, |
| 5074 | git_remote: None, |
| 5075 | } |
| 5076 | } |
| 5077 | |
| 5078 | fn fixture_enrollment(base: &str) -> LiveEnrollment { |
| 5079 | LiveEnrollment { |
| 5080 | persisted: PersistedEnrollment { |
| 5081 | schema_version: 1, |
| 5082 | control_plane_base: base.to_string(), |
| 5083 | runner_enrollment_id: "enrollment_fixture".to_string(), |
| 5084 | account_ref: "account_fixture".to_string(), |
| 5085 | device_id: "device_fixture".to_string(), |
| 5086 | target_ref: "target_fixture".to_string(), |
| 5087 | target_grant_ref: "grant_fixture".to_string(), |
| 5088 | runtime_version: "0.9.6".to_string(), |
| 5089 | runtime_commit: "a".repeat(40), |
| 5090 | bootstrap_secret: "b".repeat(43), |
| 5091 | }, |
| 5092 | access_token: "fixture-runner-access-token".to_string(), |
| 5093 | } |
| 5094 | } |
| 5095 | |
| 5096 | fn fixture_connection_response() -> Value { |
| 5097 | json!({ |
| 5098 | "runner": { |
| 5099 | "id": "runner_fixture", |
| 5100 | "userId": "account_fixture", |
| 5101 | "deviceId": "device_fixture", |
| 5102 | "targetRef": "target_fixture", |
| 5103 | "displayLabel": "private-project", |
| 5104 | "runtimeVersion": "0.9.6", |
| 5105 | "runtimeCommit": "a".repeat(40), |
| 5106 | "capabilities": CAPABILITIES, |
| 5107 | "controlPath": "outbound_relay", |
| 5108 | "status": "active", |
| 5109 | "active": true, |
| 5110 | "capacity": 1, |
| 5111 | "lastHeartbeatAt": "2026-08-08T12:00:00.000Z", |
| 5112 | "expiresAt": "2026-08-08T12:01:30.000Z", |
| 5113 | "revokedAt": "", |
| 5114 | "createdAt": "2026-08-08T12:00:00.000Z", |
| 5115 | "updatedAt": "2026-08-08T12:00:00.000Z" |
| 5116 | }, |
| 5117 | "attachment": { |
| 5118 | "runId": "run_fixture", |
| 5119 | "workspaceId": "workspace_fixture", |
| 5120 | "runtimeCursor": 41, |
| 5121 | "snapshotPresent": false, |
| 5122 | "runtimeChatRelayProtocol": RUNTIME_CHAT_RELAY_PROTOCOL, |
| 5123 | "runtimeChatRelayChallenge": "a".repeat(32) |
| 5124 | } |
| 5125 | }) |
| 5126 | } |
| 5127 | |
| 5128 | #[test] |
| 5129 | fn observed_git_repo_is_owner_name_not_a_path() { |
| 5130 | assert_eq!( |
| 5131 | normalize_observed_git_repo("git@github.com:Hmbown/CodeWhale.git").as_deref(), |
| 5132 | Some("Hmbown/CodeWhale") |
| 5133 | ); |
| 5134 | assert_eq!( |
| 5135 | normalize_observed_git_repo("https://github.com/Hmbown/cwc.git").as_deref(), |
| 5136 | Some("Hmbown/cwc") |
| 5137 | ); |
| 5138 | assert_eq!( |
| 5139 | normalize_observed_git_repo("/Volumes/VIXinSSD/CW/codewhale"), |
| 5140 | None |
| 5141 | ); |
| 5142 | for private_or_untrusted in [ |
| 5143 | "file:///Users/alice/PrivateProject.git", |
| 5144 | "https://internal.example/acme/secret-repo.git", |
| 5145 | "https://alice:password@github.com/acme/repo.git", |
| 5146 | "https://github.com/acme/repo.git?token=secret", |
| 5147 | "ssh://git@github.com/acme/repo.git", |
| 5148 | "acme/secret-repo", |
| 5149 | ] { |
| 5150 | assert_eq!( |
| 5151 | normalize_observed_git_repo(private_or_untrusted), |
| 5152 | None, |
| 5153 | "untrusted git identity must not cross: {private_or_untrusted}" |
| 5154 | ); |
| 5155 | } |
| 5156 | } |
| 5157 | |
| 5158 | #[test] |
| 5159 | fn connect_body_can_carry_an_observed_repo_without_a_path() { |
| 5160 | let enrollment = fixture_enrollment("https://api.codewhale.net/"); |
| 5161 | let mut start = fixture_start(); |
| 5162 | start.git_remote = Some("git@github.com:Hmbown/CodeWhale.git".to_string()); |
| 5163 | let body = connect_runner_body(&enrollment, &start); |
| 5164 | assert_eq!(body["gitRemote"], "Hmbown/CodeWhale"); |
| 5165 | assert!(body.get("workspacePath").is_none()); |
| 5166 | assert!(body.get("path").is_none()); |
| 5167 | } |
| 5168 | |
| 5169 | #[test] |
| 5170 | fn live_worker_rejects_reenrollment_authority_switch_before_assignment() { |
| 5171 | let start = fixture_start(); |
| 5172 | let mut current = fixture_enrollment("https://api.codewhale.net/"); |
| 5173 | let original_access_token = current.access_token.clone(); |
| 5174 | let mut other_account = current.clone(); |
| 5175 | other_account.persisted.account_ref = "account_other".to_string(); |
| 5176 | other_account.access_token = "other-account-access-token".to_string(); |
| 5177 | let error = install_reconnected_enrollment(&mut current, other_account, &start) |
| 5178 | .expect_err("a live worker must never change tenant authority"); |
| 5179 | assert!(error.contains("different account or folder")); |
| 5180 | assert_eq!(current.persisted.account_ref, "account_fixture"); |
| 5181 | assert_eq!(current.persisted.target_ref, "target_fixture"); |
| 5182 | assert_eq!(current.access_token, original_access_token); |
| 5183 | |
| 5184 | let mut other_target = current.clone(); |
| 5185 | other_target.persisted.target_ref = "target_other".to_string(); |
| 5186 | assert!(install_reconnected_enrollment(&mut current, other_target, &start).is_err()); |
| 5187 | assert_eq!(current.persisted.target_ref, "target_fixture"); |
| 5188 | |
| 5189 | let mut refreshed = current.clone(); |
| 5190 | refreshed.access_token = "same-authority-refreshed-access-token".to_string(); |
| 5191 | install_reconnected_enrollment(&mut current, refreshed, &start).unwrap(); |
| 5192 | assert_eq!( |
| 5193 | current.access_token, |
| 5194 | "same-authority-refreshed-access-token" |
| 5195 | ); |
| 5196 | } |
| 5197 | |
| 5198 | #[test] |
| 5199 | fn target_identity_is_stable_without_exposing_the_path() { |
| 5200 | let target = target_ref(Path::new("/Users/alice/private/project")); |
| 5201 | assert!(target.starts_with("target_")); |
| 5202 | assert_eq!(target.len(), 39); |
| 5203 | assert!(!target.contains("alice")); |
| 5204 | // Every session opened in the same folder shares one target, so the |
| 5205 | // control plane keeps one grant per folder rather than one per `/rc`. |
| 5206 | assert_eq!( |
| 5207 | target, |
| 5208 | target_ref(Path::new("/Users/alice/private/project")) |
| 5209 | ); |
| 5210 | assert_ne!(target, target_ref(Path::new("/Users/alice/private/other"))); |
| 5211 | } |
| 5212 | |
| 5213 | #[test] |
| 5214 | fn runtime_chat_same_session_reconfigure_releases_then_reacquires_owner_lock() { |
| 5215 | let root = tempfile::tempdir().unwrap(); |
| 5216 | let registry = Arc::new(crate::plugins::PluginRegistry::empty(root.path())); |
| 5217 | let mut controller = RemoteControlController::default(); |
| 5218 | let old_config = crate::config::Config { |
| 5219 | provider: Some("ollama".to_string()), |
| 5220 | default_text_model: Some("old-safe-model".to_string()), |
| 5221 | ..crate::config::Config::default() |
| 5222 | }; |
| 5223 | controller |
| 5224 | .configure_runtime_chat( |
| 5225 | old_config, |
| 5226 | Arc::clone(®istry), |
| 5227 | root.path().join("runtime-chat"), |
| 5228 | "target_fixture".to_string(), |
| 5229 | "session_fixture".to_string(), |
| 5230 | ) |
| 5231 | .unwrap(); |
| 5232 | assert_eq!( |
| 5233 | controller |
| 5234 | .runtime_chat |
| 5235 | .as_ref() |
| 5236 | .unwrap() |
| 5237 | .configured_default_model_for_tests(), |
| 5238 | "old-safe-model" |
| 5239 | ); |
| 5240 | |
| 5241 | let mut competing = RemoteControlController::default(); |
| 5242 | assert!( |
| 5243 | competing |
| 5244 | .configure_runtime_chat( |
| 5245 | crate::config::Config::default(), |
| 5246 | Arc::clone(®istry), |
| 5247 | root.path().join("runtime-chat"), |
| 5248 | "target_fixture".to_string(), |
| 5249 | "session_fixture".to_string(), |
| 5250 | ) |
| 5251 | .is_err() |
| 5252 | ); |
| 5253 | |
| 5254 | let new_config = crate::config::Config { |
| 5255 | provider: Some("ollama".to_string()), |
| 5256 | default_text_model: Some("new-safe-model".to_string()), |
| 5257 | ..crate::config::Config::default() |
| 5258 | }; |
| 5259 | controller |
| 5260 | .configure_runtime_chat( |
| 5261 | new_config, |
| 5262 | Arc::clone(®istry), |
| 5263 | root.path().join("runtime-chat"), |
| 5264 | "target_fixture".to_string(), |
| 5265 | "session_fixture".to_string(), |
| 5266 | ) |
| 5267 | .unwrap(); |
| 5268 | assert_eq!( |
| 5269 | controller |
| 5270 | .runtime_chat |
| 5271 | .as_ref() |
| 5272 | .unwrap() |
| 5273 | .configured_default_model_for_tests(), |
| 5274 | "new-safe-model", |
| 5275 | "a settled same-scope restart must refresh immutable provider configuration" |
| 5276 | ); |
| 5277 | drop(controller); |
| 5278 | competing |
| 5279 | .configure_runtime_chat( |
| 5280 | crate::config::Config::default(), |
| 5281 | registry, |
| 5282 | root.path().join("runtime-chat"), |
| 5283 | "target_fixture".to_string(), |
| 5284 | "session_fixture".to_string(), |
| 5285 | ) |
| 5286 | .unwrap(); |
| 5287 | } |
| 5288 | |
| 5289 | #[test] |
| 5290 | fn runtime_chat_backlog_refreshes_immutable_config_before_new_catalog_or_prompt() { |
| 5291 | let root = tempfile::tempdir().unwrap(); |
| 5292 | let registry = Arc::new(crate::plugins::PluginRegistry::empty(root.path())); |
| 5293 | let mut controller = RemoteControlController::default(); |
| 5294 | let old_config = crate::config::Config { |
| 5295 | provider: Some("ollama".to_string()), |
| 5296 | default_text_model: Some("old-safe-model".to_string()), |
| 5297 | ..crate::config::Config::default() |
| 5298 | }; |
| 5299 | controller |
| 5300 | .configure_runtime_chat( |
| 5301 | old_config, |
| 5302 | Arc::clone(®istry), |
| 5303 | root.path().join("runtime-chat"), |
| 5304 | "target-1".to_string(), |
| 5305 | "session-1".to_string(), |
| 5306 | ) |
| 5307 | .unwrap(); |
| 5308 | controller |
| 5309 | .runtime_chat |
| 5310 | .as_ref() |
| 5311 | .unwrap() |
| 5312 | .bind_account("account-1", "target-1") |
| 5313 | .unwrap(); |
| 5314 | controller.journal = Some( |
| 5315 | RuntimeEventJournal::open(root.path(), "target-1", "session-1") |
| 5316 | .expect("runtime chat delivery journal"), |
| 5317 | ); |
| 5318 | |
| 5319 | let pending_seq = controller.next_runtime_seq("run-1"); |
| 5320 | assert!(controller.queue_runtime_chat_envelope( |
| 5321 | "run-1", |
| 5322 | runtime_chat_envelope( |
| 5323 | pending_seq, |
| 5324 | "runtime.catalog", |
| 5325 | None, |
| 5326 | None, |
| 5327 | Some("catalog_old_fixture"), |
| 5328 | RUNTIME_CHAT_CATALOG_TIMESTAMP.to_string(), |
| 5329 | json!({ "old": true }), |
| 5330 | ), |
| 5331 | )); |
| 5332 | let new_config = crate::config::Config { |
| 5333 | provider: Some("ollama".to_string()), |
| 5334 | default_text_model: Some("new-safe-model".to_string()), |
| 5335 | ..crate::config::Config::default() |
| 5336 | }; |
| 5337 | controller |
| 5338 | .configure_runtime_chat( |
| 5339 | new_config, |
| 5340 | Arc::clone(®istry), |
| 5341 | root.path().join("runtime-chat"), |
| 5342 | "target-1".to_string(), |
| 5343 | "session-1".to_string(), |
| 5344 | ) |
| 5345 | .unwrap(); |
| 5346 | assert!(controller.pending_runtime_chat_configuration.is_some()); |
| 5347 | assert_eq!( |
| 5348 | controller |
| 5349 | .runtime_chat |
| 5350 | .as_ref() |
| 5351 | .unwrap() |
| 5352 | .configured_default_model_for_tests(), |
| 5353 | "old-safe-model", |
| 5354 | "the old host remains only as the durable backlog owner" |
| 5355 | ); |
| 5356 | |
| 5357 | let (worker_tx, mut worker_rx) = mpsc::unbounded_channel(); |
| 5358 | let (event_tx, event_rx) = mpsc::unbounded_channel(); |
| 5359 | controller.worker_tx = Some(worker_tx); |
| 5360 | controller.event_rx = Some(event_rx); |
| 5361 | controller.status = Status::Connected; |
| 5362 | controller.account_ref = Some("account-1".to_string()); |
| 5363 | controller.target_ref = Some("target-1".to_string()); |
| 5364 | controller.attached_run_id = Some("run-1".to_string()); |
| 5365 | controller.attached_workspace_id = Some("workspace-1".to_string()); |
| 5366 | controller.runtime_chat_attachment = Some(RemoteAttachment { |
| 5367 | run_id: "run-1".to_string(), |
| 5368 | workspace_id: "workspace-1".to_string(), |
| 5369 | runtime_cursor: pending_seq, |
| 5370 | snapshot_present: false, |
| 5371 | runtime_chat_relay_protocol: RUNTIME_CHAT_RELAY_PROTOCOL.to_string(), |
| 5372 | runtime_chat_relay_challenge: "c".repeat(32), |
| 5373 | }); |
| 5374 | let command = parse_remote_command(&runtime_chat_prompt_fixture(), "run-1").unwrap(); |
| 5375 | assert!( |
| 5376 | controller.claim_command("run-1", 1, &command).is_err(), |
| 5377 | "no newly configured prompt may reach the stale host while backlog drains" |
| 5378 | ); |
| 5379 | |
| 5380 | event_tx |
| 5381 | .send(RemoteEvent::RuntimeCursor { |
| 5382 | run_id: "run-1".to_string(), |
| 5383 | cursor: pending_seq, |
| 5384 | }) |
| 5385 | .unwrap(); |
| 5386 | assert!(matches!( |
| 5387 | controller.try_next_event(), |
| 5388 | Some(RemoteEvent::RuntimeCursor { .. }) |
| 5389 | )); |
| 5390 | assert!(matches!( |
| 5391 | worker_rx.try_recv().unwrap(), |
| 5392 | WorkerCommand::ReleaseRuntimeChatHost |
| 5393 | )); |
| 5394 | |
| 5395 | event_tx.send(RemoteEvent::RuntimeChatHostReleased).unwrap(); |
| 5396 | assert!(matches!( |
| 5397 | controller.try_next_event(), |
| 5398 | Some(RemoteEvent::RuntimeChatHostReleased) |
| 5399 | )); |
| 5400 | assert_eq!( |
| 5401 | controller |
| 5402 | .runtime_chat |
| 5403 | .as_ref() |
| 5404 | .unwrap() |
| 5405 | .configured_default_model_for_tests(), |
| 5406 | "new-safe-model" |
| 5407 | ); |
| 5408 | assert!(matches!( |
| 5409 | worker_rx.try_recv().unwrap(), |
| 5410 | WorkerCommand::InstallRuntimeChatHost(_) |
| 5411 | )); |
| 5412 | let WorkerCommand::Upload { envelopes, .. } = worker_rx.try_recv().unwrap() else { |
| 5413 | panic!("the refreshed host must upload a new challenge-bound catalog"); |
| 5414 | }; |
| 5415 | assert_eq!(envelopes.len(), 1); |
| 5416 | assert_eq!(envelopes[0]["event"], "runtime.catalog"); |
| 5417 | let serialized = envelopes[0].to_string(); |
| 5418 | assert!(serialized.contains("new-safe-model")); |
| 5419 | assert!(!serialized.contains("old-safe-model")); |
| 5420 | assert!(controller.pending_runtime_chat_configuration.is_none()); |
| 5421 | } |
| 5422 | |
| 5423 | #[tokio::test] |
| 5424 | async fn runtime_chat_provider_ownership_releases_only_after_terminal_server_cursor() { |
| 5425 | let root = tempfile::tempdir().unwrap(); |
| 5426 | let mut controller = RemoteControlController::default(); |
| 5427 | controller |
| 5428 | .configure_runtime_chat( |
| 5429 | crate::config::Config::default(), |
| 5430 | Arc::new(crate::plugins::PluginRegistry::empty(root.path())), |
| 5431 | root.path().join("runtime-chat"), |
| 5432 | "target_fixture".to_string(), |
| 5433 | "session_fixture".to_string(), |
| 5434 | ) |
| 5435 | .unwrap(); |
| 5436 | let host = controller.runtime_chat.as_ref().unwrap().clone(); |
| 5437 | controller.journal = Some( |
| 5438 | RuntimeEventJournal::open(root.path(), "target_fixture", "session_fixture") |
| 5439 | .expect("runtime chat delivery journal"), |
| 5440 | ); |
| 5441 | host.bind_account("account_fixture", "target_fixture") |
| 5442 | .unwrap(); |
| 5443 | let native_thread_id = "thr_terminal_gate_fixture"; |
| 5444 | let virtual_thread_id = format!("local_thread_{}", "a".repeat(24)); |
| 5445 | let virtual_turn_id = format!("local_turn_{}", "b".repeat(24)); |
| 5446 | host.install_unsettled_turn_for_tests( |
| 5447 | "run_fixture", |
| 5448 | native_thread_id, |
| 5449 | &virtual_thread_id, |
| 5450 | &virtual_turn_id, |
| 5451 | ) |
| 5452 | .unwrap(); |
| 5453 | host.acquire_inference_ownership_for_tests().await; |
| 5454 | assert!(host.inference_ownership_is_held_for_tests()); |
| 5455 | |
| 5456 | let seq = controller.next_runtime_seq("run_fixture"); |
| 5457 | assert!(controller.queue_runtime_chat_envelope( |
| 5458 | "run_fixture", |
| 5459 | runtime_chat_envelope( |
| 5460 | seq, |
| 5461 | "turn.completed", |
| 5462 | Some(&virtual_thread_id), |
| 5463 | Some(&virtual_turn_id), |
| 5464 | Some("native_event_terminal_gate_fixture"), |
| 5465 | "2026-08-23T00:00:00Z".to_string(), |
| 5466 | json!({ "status": "completed" }), |
| 5467 | ), |
| 5468 | )); |
| 5469 | host.mark_projected(native_thread_id, 1, &virtual_turn_id, "turn.completed") |
| 5470 | .unwrap(); |
| 5471 | assert!(!host.has_any_unsettled_turns()); |
| 5472 | controller.reconcile_runtime_cursor("run_fixture", seq.saturating_sub(1)); |
| 5473 | assert!( |
| 5474 | host.inference_ownership_is_held_for_tests(), |
| 5475 | "local terminal projection is not a server acknowledgement" |
| 5476 | ); |
| 5477 | controller.reconcile_runtime_cursor("run_fixture", seq); |
| 5478 | assert!( |
| 5479 | !host.inference_ownership_is_held_for_tests(), |
| 5480 | "the accepted server cursor releases attached-run provider ownership" |
| 5481 | ); |
| 5482 | } |
| 5483 | |
| 5484 | #[tokio::test] |
| 5485 | async fn restart_rehydrates_runtime_chat_ownership_until_terminal_cursor() { |
| 5486 | let root = tempfile::tempdir().unwrap(); |
| 5487 | let runtime_root = root.path().join("runtime-chat"); |
| 5488 | let journal_root = root.path().join("journal"); |
| 5489 | std::fs::create_dir_all(&journal_root).unwrap(); |
| 5490 | let terminal_seq; |
| 5491 | { |
| 5492 | let mut first = RemoteControlController::default(); |
| 5493 | first |
| 5494 | .configure_runtime_chat( |
| 5495 | crate::config::Config::default(), |
| 5496 | Arc::new(crate::plugins::PluginRegistry::empty(root.path())), |
| 5497 | runtime_root.clone(), |
| 5498 | "target_fixture".to_string(), |
| 5499 | "session_fixture".to_string(), |
| 5500 | ) |
| 5501 | .unwrap(); |
| 5502 | first.journal = Some( |
| 5503 | RuntimeEventJournal::open(&journal_root, "target_fixture", "session_fixture") |
| 5504 | .unwrap(), |
| 5505 | ); |
| 5506 | let host = first.runtime_chat.as_ref().unwrap().clone(); |
| 5507 | host.bind_account("account_fixture", "target_fixture") |
| 5508 | .unwrap(); |
| 5509 | let native_thread_id = "thr_restart_terminal_gate"; |
| 5510 | let virtual_thread_id = format!("local_thread_{}", "c".repeat(24)); |
| 5511 | let virtual_turn_id = format!("local_turn_{}", "d".repeat(24)); |
| 5512 | host.install_unsettled_turn_for_tests( |
| 5513 | "run_fixture", |
| 5514 | native_thread_id, |
| 5515 | &virtual_thread_id, |
| 5516 | &virtual_turn_id, |
| 5517 | ) |
| 5518 | .unwrap(); |
| 5519 | host.acquire_inference_ownership_for_tests().await; |
| 5520 | terminal_seq = first.next_runtime_seq("run_fixture"); |
| 5521 | assert!(first.queue_runtime_chat_envelope( |
| 5522 | "run_fixture", |
| 5523 | runtime_chat_envelope( |
| 5524 | terminal_seq, |
| 5525 | "turn.completed", |
| 5526 | Some(&virtual_thread_id), |
| 5527 | Some(&virtual_turn_id), |
| 5528 | Some("native_event_restart_terminal_gate"), |
| 5529 | "2026-08-23T00:00:00Z".to_string(), |
| 5530 | json!({ "turn": { "status": "completed" } }), |
| 5531 | ), |
| 5532 | )); |
| 5533 | host.mark_projected(native_thread_id, 1, &virtual_turn_id, "turn.completed") |
| 5534 | .unwrap(); |
| 5535 | assert!(!host.has_any_unsettled_turns()); |
| 5536 | } |
| 5537 | |
| 5538 | let mut reopened = RemoteControlController::default(); |
| 5539 | reopened |
| 5540 | .configure_runtime_chat( |
| 5541 | crate::config::Config::default(), |
| 5542 | Arc::new(crate::plugins::PluginRegistry::empty(root.path())), |
| 5543 | runtime_root, |
| 5544 | "target_fixture".to_string(), |
| 5545 | "session_fixture".to_string(), |
| 5546 | ) |
| 5547 | .unwrap(); |
| 5548 | let journal = |
| 5549 | RuntimeEventJournal::open(&journal_root, "target_fixture", "session_fixture").unwrap(); |
| 5550 | reopened.reset_pending_from(journal.load().unwrap()); |
| 5551 | reopened.journal = Some(journal); |
| 5552 | reopened |
| 5553 | .recover_runtime_chat_ownership_before_worker() |
| 5554 | .unwrap(); |
| 5555 | |
| 5556 | let mut participant = tokio::spawn(async { |
| 5557 | crate::client::acquire_remote_control_inference_participant().await |
| 5558 | }); |
| 5559 | assert!( |
| 5560 | tokio::time::timeout(Duration::from_millis(40), &mut participant) |
| 5561 | .await |
| 5562 | .is_err(), |
| 5563 | "restart recovery must retain the provider writer before CWC cursor ack" |
| 5564 | ); |
| 5565 | reopened.reconcile_runtime_cursor("run_fixture", terminal_seq); |
| 5566 | let permit = tokio::time::timeout(Duration::from_secs(1), participant) |
| 5567 | .await |
| 5568 | .expect("participant resumes after cursor") |
| 5569 | .expect("participant task"); |
| 5570 | drop(permit); |
| 5571 | } |
| 5572 | |
| 5573 | #[tokio::test] |
| 5574 | async fn restart_reclaims_unsettled_runtime_chat_before_worker_or_fails_closed() { |
| 5575 | let root = tempfile::tempdir().unwrap(); |
| 5576 | let runtime_root = root.path().join("runtime-chat"); |
| 5577 | let native_thread_id = "thr_restart_unsettled_gate"; |
| 5578 | let virtual_thread_id = format!("local_thread_{}", "e".repeat(24)); |
| 5579 | let virtual_turn_id = format!("local_turn_{}", "f".repeat(24)); |
| 5580 | { |
| 5581 | let mut first = RemoteControlController::default(); |
| 5582 | first |
| 5583 | .configure_runtime_chat( |
| 5584 | crate::config::Config::default(), |
| 5585 | Arc::new(crate::plugins::PluginRegistry::empty(root.path())), |
| 5586 | runtime_root.clone(), |
| 5587 | "target_fixture".to_string(), |
| 5588 | "session_fixture".to_string(), |
| 5589 | ) |
| 5590 | .unwrap(); |
| 5591 | let host = first.runtime_chat.as_ref().unwrap(); |
| 5592 | host.bind_account("account_fixture", "target_fixture") |
| 5593 | .unwrap(); |
| 5594 | host.install_unsettled_turn_for_tests( |
| 5595 | "run_fixture", |
| 5596 | native_thread_id, |
| 5597 | &virtual_thread_id, |
| 5598 | &virtual_turn_id, |
| 5599 | ) |
| 5600 | .unwrap(); |
| 5601 | } |
| 5602 | |
| 5603 | let mut reopened = RemoteControlController::default(); |
| 5604 | reopened |
| 5605 | .configure_runtime_chat( |
| 5606 | crate::config::Config::default(), |
| 5607 | Arc::new(crate::plugins::PluginRegistry::empty(root.path())), |
| 5608 | runtime_root, |
| 5609 | "target_fixture".to_string(), |
| 5610 | "session_fixture".to_string(), |
| 5611 | ) |
| 5612 | .unwrap(); |
| 5613 | let host = reopened.runtime_chat.as_ref().unwrap().clone(); |
| 5614 | assert!(host.has_any_unsettled_turns()); |
| 5615 | |
| 5616 | let participant = crate::client::acquire_remote_control_inference_participant().await; |
| 5617 | let error = reopened |
| 5618 | .recover_runtime_chat_ownership_before_worker() |
| 5619 | .unwrap_err(); |
| 5620 | assert!(error.contains("active local turn"), "{error}"); |
| 5621 | assert!(!host.inference_ownership_is_held_for_tests()); |
| 5622 | drop(participant); |
| 5623 | |
| 5624 | reopened |
| 5625 | .recover_runtime_chat_ownership_before_worker() |
| 5626 | .unwrap(); |
| 5627 | let mut blocked = tokio::spawn(async { |
| 5628 | crate::client::acquire_remote_control_inference_participant().await |
| 5629 | }); |
| 5630 | assert!( |
| 5631 | tokio::time::timeout(Duration::from_millis(40), &mut blocked) |
| 5632 | .await |
| 5633 | .is_err(), |
| 5634 | "an unsettled recovered turn must own the gate before Connected" |
| 5635 | ); |
| 5636 | host.mark_projected(native_thread_id, 1, &virtual_turn_id, "turn.completed") |
| 5637 | .unwrap(); |
| 5638 | host.release_inference_ownership_if_settled(); |
| 5639 | let permit = tokio::time::timeout(Duration::from_secs(1), blocked) |
| 5640 | .await |
| 5641 | .expect("participant resumes after durable terminal settlement") |
| 5642 | .expect("participant task"); |
| 5643 | drop(permit); |
| 5644 | } |
| 5645 | |
| 5646 | #[tokio::test(flavor = "current_thread")] |
| 5647 | async fn actual_start_reclaims_runtime_chat_writer_before_worker_spawn() { |
| 5648 | let root = tempfile::tempdir().unwrap(); |
| 5649 | let runtime_root = root.path().join("runtime-chat"); |
| 5650 | let journal_root = root.path().join("journal"); |
| 5651 | let native_thread_id = "thr_actual_start_unsettled"; |
| 5652 | let virtual_thread_id = format!("local_thread_{}", "c".repeat(24)); |
| 5653 | let virtual_turn_id = format!("local_turn_{}", "d".repeat(24)); |
| 5654 | { |
| 5655 | let mut first = RemoteControlController::default(); |
| 5656 | first |
| 5657 | .prepare_remote_control_session_journal( |
| 5658 | &journal_root, |
| 5659 | "target_fixture", |
| 5660 | "session_fixture", |
| 5661 | ) |
| 5662 | .unwrap(); |
| 5663 | first |
| 5664 | .configure_runtime_chat( |
| 5665 | crate::config::Config::default(), |
| 5666 | Arc::new(crate::plugins::PluginRegistry::empty(root.path())), |
| 5667 | runtime_root.clone(), |
| 5668 | "target_fixture".to_string(), |
| 5669 | "session_fixture".to_string(), |
| 5670 | ) |
| 5671 | .unwrap(); |
| 5672 | first |
| 5673 | .runtime_chat |
| 5674 | .as_ref() |
| 5675 | .unwrap() |
| 5676 | .bind_account("account_fixture", "target_fixture") |
| 5677 | .unwrap(); |
| 5678 | first |
| 5679 | .runtime_chat |
| 5680 | .as_ref() |
| 5681 | .unwrap() |
| 5682 | .install_unsettled_turn_for_tests( |
| 5683 | "run_fixture", |
| 5684 | native_thread_id, |
| 5685 | &virtual_thread_id, |
| 5686 | &virtual_turn_id, |
| 5687 | ) |
| 5688 | .unwrap(); |
| 5689 | } |
| 5690 | |
| 5691 | let mut reopened = RemoteControlController::default(); |
| 5692 | reopened |
| 5693 | .prepare_remote_control_session_journal( |
| 5694 | &journal_root, |
| 5695 | "target_fixture", |
| 5696 | "session_fixture", |
| 5697 | ) |
| 5698 | .unwrap(); |
| 5699 | reopened |
| 5700 | .configure_runtime_chat( |
| 5701 | crate::config::Config::default(), |
| 5702 | Arc::new(crate::plugins::PluginRegistry::empty(root.path())), |
| 5703 | runtime_root, |
| 5704 | "target_fixture".to_string(), |
| 5705 | "session_fixture".to_string(), |
| 5706 | ) |
| 5707 | .unwrap(); |
| 5708 | let host = reopened.runtime_chat.as_ref().unwrap().clone(); |
| 5709 | let start = RemoteStart { |
| 5710 | session_id: "session_fixture".to_string(), |
| 5711 | journal_dir: Some(journal_root), |
| 5712 | ..fixture_start() |
| 5713 | }; |
| 5714 | |
| 5715 | let participant = crate::client::acquire_remote_control_inference_participant().await; |
| 5716 | let error = reopened.start(start.clone()).unwrap_err(); |
| 5717 | assert!(error.contains("active local turn"), "{error}"); |
| 5718 | assert!(reopened.worker.is_none()); |
| 5719 | assert_eq!(reopened.status, Status::Off); |
| 5720 | drop(participant); |
| 5721 | |
| 5722 | reopened.start(start).unwrap(); |
| 5723 | assert!(host.inference_ownership_is_held_for_tests()); |
| 5724 | assert_eq!(reopened.status, Status::Connecting); |
| 5725 | assert!(reopened.worker.is_some()); |
| 5726 | // This current-thread test has not yielded since spawning the worker, |
| 5727 | // so abort it before enrollment can perform any network I/O. |
| 5728 | reopened.stop_worker(); |
| 5729 | host.mark_projected(native_thread_id, 1, &virtual_turn_id, "turn.completed") |
| 5730 | .unwrap(); |
| 5731 | host.release_inference_ownership_if_settled(); |
| 5732 | assert!(!host.inference_ownership_is_held_for_tests()); |
| 5733 | } |
| 5734 | |
| 5735 | #[test] |
| 5736 | fn aborting_a_relay_worker_releases_unjournaled_projection_claims() { |
| 5737 | let root = tempfile::tempdir().unwrap(); |
| 5738 | let mut controller = RemoteControlController::default(); |
| 5739 | controller |
| 5740 | .configure_runtime_chat( |
| 5741 | crate::config::Config::default(), |
| 5742 | Arc::new(crate::plugins::PluginRegistry::empty(root.path())), |
| 5743 | root.path().join("runtime-chat"), |
| 5744 | "target_fixture".to_string(), |
| 5745 | "session_fixture".to_string(), |
| 5746 | ) |
| 5747 | .unwrap(); |
| 5748 | let host = controller.runtime_chat.as_ref().unwrap().clone(); |
| 5749 | host.install_projection_claim_for_tests("thr_claimed", 7); |
| 5750 | assert!(host.projection_is_claimed_for_tests("thr_claimed", 7)); |
| 5751 | |
| 5752 | controller.stop_worker(); |
| 5753 | |
| 5754 | assert!(!host.projection_is_claimed_for_tests("thr_claimed", 7)); |
| 5755 | } |
| 5756 | |
| 5757 | #[tokio::test] |
| 5758 | async fn idle_runtime_chat_ticks_cannot_starve_the_control_poll() { |
| 5759 | if !cfg!(debug_assertions) { |
| 5760 | return; |
| 5761 | } |
| 5762 | let _env = crate::test_support::lock_test_env(); |
| 5763 | let secrets_root = tempfile::tempdir().expect("isolated remote-control secrets"); |
| 5764 | let _codewhale_home = |
| 5765 | crate::test_support::EnvVarGuard::set("CODEWHALE_HOME", secrets_root.path()); |
| 5766 | let server = MockServer::start().await; |
| 5767 | let plane = format!("{}/", server.uri().trim_end_matches('/')); |
| 5768 | let _control_plane = |
| 5769 | crate::test_support::EnvVarGuard::set("CWC_RUNNER_CONTROL_PLANE_BASE", &plane); |
| 5770 | let base = runner_control_plane_base().expect("loopback control plane"); |
| 5771 | save_persisted_enrollment(&fixture_enrollment(&base).persisted) |
| 5772 | .expect("persist matching enrollment"); |
| 5773 | |
| 5774 | let access_token = crate::test_support::future_test_jwt(&"a".repeat(40)); |
| 5775 | Mock::given(method("POST")) |
| 5776 | .and(path("/api/runner/enrollments/token")) |
| 5777 | .respond_with(ResponseTemplate::new(200).set_body_json(json!({ |
| 5778 | "enrollment": { |
| 5779 | "id": "enrollment_fixture", |
| 5780 | "userId": "account_fixture", |
| 5781 | "deviceId": "device_fixture", |
| 5782 | "runtimeVersion": "0.9.6", |
| 5783 | "runtimeCommit": "a".repeat(40), |
| 5784 | "capabilities": CAPABILITIES, |
| 5785 | }, |
| 5786 | "credential": { "accessToken": access_token }, |
| 5787 | }))) |
| 5788 | .mount(&server) |
| 5789 | .await; |
| 5790 | Mock::given(method("POST")) |
| 5791 | .and(path("/api/local-runners/connect")) |
| 5792 | .respond_with(ResponseTemplate::new(200).set_body_json(fixture_connection_response())) |
| 5793 | .mount(&server) |
| 5794 | .await; |
| 5795 | Mock::given(method("POST")) |
| 5796 | .and(path("/api/local-runners/runner_fixture/heartbeat")) |
| 5797 | .respond_with(ResponseTemplate::new(200).set_body_json(json!({}))) |
| 5798 | .mount(&server) |
| 5799 | .await; |
| 5800 | let polls = ControlPollCounter::default(); |
| 5801 | Mock::given(method("GET")) |
| 5802 | .and(path("/api/local-runners/runner_fixture/runs")) |
| 5803 | .respond_with(polls.clone()) |
| 5804 | .mount(&server) |
| 5805 | .await; |
| 5806 | |
| 5807 | let runtime_root = tempfile::tempdir().expect("idle Runtime Chat host"); |
| 5808 | let host = RuntimeChatRelayHost::open( |
| 5809 | crate::config::Config::default(), |
| 5810 | Arc::new(crate::plugins::PluginRegistry::empty(runtime_root.path())), |
| 5811 | runtime_root.path().to_path_buf(), |
| 5812 | "target_fixture".to_string(), |
| 5813 | "session_fixture".to_string(), |
| 5814 | ) |
| 5815 | .expect("open idle Runtime Chat host"); |
| 5816 | let start = fixture_start(); |
| 5817 | let (_worker_tx, worker_rx) = mpsc::unbounded_channel(); |
| 5818 | let (event_tx, mut event_rx) = mpsc::unbounded_channel(); |
| 5819 | let worker = tokio::spawn(async move { |
| 5820 | let mut phase = RelayPhase::Enrolling; |
| 5821 | relay_worker(start, Some(host), worker_rx, event_tx, &mut phase).await |
| 5822 | }); |
| 5823 | |
| 5824 | tokio::time::timeout(Duration::from_secs(5), async { |
| 5825 | loop { |
| 5826 | match event_rx.recv().await { |
| 5827 | Some(RemoteEvent::Connected { .. }) => break, |
| 5828 | Some(RemoteEvent::Notice(_)) => {} |
| 5829 | Some(RemoteEvent::Failed(error)) => { |
| 5830 | panic!("relay worker failed before attach: {error}"); |
| 5831 | } |
| 5832 | Some(RemoteEvent::FailedPreLease(error)) => { |
| 5833 | panic!("relay worker failed before attach: {error}"); |
| 5834 | } |
| 5835 | Some(_) => panic!("relay worker emitted an unexpected event before attach"), |
| 5836 | None => panic!("relay worker stopped before attach"), |
| 5837 | } |
| 5838 | } |
| 5839 | }) |
| 5840 | .await |
| 5841 | .expect("relay worker should attach"); |
| 5842 | |
| 5843 | tokio::time::sleep(RUNTIME_UPLOAD_RETRY_INTERVAL.saturating_mul(2)).await; |
| 5844 | assert_eq!( |
| 5845 | polls.hits.load(Ordering::SeqCst), |
| 5846 | 0, |
| 5847 | "the first control poll must still wait SYNC_INTERVAL while idle chat ticks fire" |
| 5848 | ); |
| 5849 | tokio::time::sleep(SYNC_INTERVAL).await; |
| 5850 | assert!( |
| 5851 | polls.hits.load(Ordering::SeqCst) >= 1, |
| 5852 | "idle Runtime Chat ticks must not reset the control poll" |
| 5853 | ); |
| 5854 | worker.abort(); |
| 5855 | let _ = worker.await; |
| 5856 | } |
| 5857 | |
| 5858 | #[test] |
| 5859 | fn preattachment_failure_cannot_release_a_recovered_unsettled_turn() { |
| 5860 | let root = tempfile::tempdir().unwrap(); |
| 5861 | let registry = Arc::new(crate::plugins::PluginRegistry::empty(root.path())); |
| 5862 | let mut controller = RemoteControlController::default(); |
| 5863 | controller |
| 5864 | .configure_runtime_chat( |
| 5865 | crate::config::Config::default(), |
| 5866 | Arc::clone(®istry), |
| 5867 | root.path().join("runtime-chat"), |
| 5868 | "target_fixture".to_string(), |
| 5869 | "session_fixture".to_string(), |
| 5870 | ) |
| 5871 | .unwrap(); |
| 5872 | let host = controller.runtime_chat.as_ref().unwrap(); |
| 5873 | host.bind_account("account_fixture", "target_fixture") |
| 5874 | .unwrap(); |
| 5875 | host.install_unsettled_turn_for_tests( |
| 5876 | "run_old", |
| 5877 | "thr_recovered", |
| 5878 | &format!("local_thread_{}", "a".repeat(24)), |
| 5879 | &format!("local_turn_{}", "b".repeat(24)), |
| 5880 | ) |
| 5881 | .unwrap(); |
| 5882 | let retained_model = host.configured_default_model_for_tests(); |
| 5883 | controller |
| 5884 | .configure_runtime_chat( |
| 5885 | crate::config::Config { |
| 5886 | provider: Some("ollama".to_string()), |
| 5887 | default_text_model: Some("must-not-replace-live-host".to_string()), |
| 5888 | ..crate::config::Config::default() |
| 5889 | }, |
| 5890 | Arc::clone(®istry), |
| 5891 | root.path().join("runtime-chat"), |
| 5892 | "target_fixture".to_string(), |
| 5893 | "session_fixture".to_string(), |
| 5894 | ) |
| 5895 | .unwrap(); |
| 5896 | assert_eq!( |
| 5897 | controller |
| 5898 | .runtime_chat |
| 5899 | .as_ref() |
| 5900 | .unwrap() |
| 5901 | .configured_default_model_for_tests(), |
| 5902 | retained_model, |
| 5903 | "an unsettled same-scope turn must retain its lifetime manager" |
| 5904 | ); |
| 5905 | controller.status = Status::Failed; |
| 5906 | controller.ownership_blocked_until = Some(Instant::now() - Duration::from_millis(1)); |
| 5907 | |
| 5908 | assert!(controller.try_next_event().is_none()); |
| 5909 | assert_eq!(controller.status, Status::Failed); |
| 5910 | assert!(controller.runtime_chat.is_some()); |
| 5911 | assert!( |
| 5912 | controller |
| 5913 | .configure_runtime_chat( |
| 5914 | crate::config::Config::default(), |
| 5915 | registry, |
| 5916 | root.path().join("runtime-chat"), |
| 5917 | "target_other".to_string(), |
| 5918 | "session_other".to_string(), |
| 5919 | ) |
| 5920 | .is_err() |
| 5921 | ); |
| 5922 | } |
| 5923 | |
| 5924 | #[test] |
| 5925 | fn runtime_chat_reenrollment_rejects_attachment_from_a_different_account() { |
| 5926 | let root = tempfile::tempdir().unwrap(); |
| 5927 | let registry = Arc::new(crate::plugins::PluginRegistry::empty(root.path())); |
| 5928 | let (mut controller, _worker_rx, event_tx, _journal_root) = wired_controller(); |
| 5929 | controller |
| 5930 | .configure_runtime_chat( |
| 5931 | crate::config::Config::default(), |
| 5932 | registry, |
| 5933 | root.path().join("runtime-chat"), |
| 5934 | "target_fixture".to_string(), |
| 5935 | "session_fixture".to_string(), |
| 5936 | ) |
| 5937 | .unwrap(); |
| 5938 | controller |
| 5939 | .runtime_chat |
| 5940 | .as_ref() |
| 5941 | .unwrap() |
| 5942 | .bind_account("account_a", "target_fixture") |
| 5943 | .unwrap(); |
| 5944 | controller.account_ref = Some("account_a".to_string()); |
| 5945 | controller.target_ref = Some("target_fixture".to_string()); |
| 5946 | controller.attached_run_id = Some("run_account_a".to_string()); |
| 5947 | |
| 5948 | event_tx |
| 5949 | .send(RemoteEvent::Attachment { |
| 5950 | account_ref: "account_b".to_string(), |
| 5951 | target_ref: "target_fixture".to_string(), |
| 5952 | attachment: RemoteAttachment { |
| 5953 | run_id: "run_account_b".to_string(), |
| 5954 | workspace_id: "workspace_account_b".to_string(), |
| 5955 | runtime_cursor: 0, |
| 5956 | snapshot_present: false, |
| 5957 | runtime_chat_relay_protocol: RUNTIME_CHAT_RELAY_PROTOCOL.to_string(), |
| 5958 | runtime_chat_relay_challenge: "b".repeat(32), |
| 5959 | }, |
| 5960 | links: RemoteLinks::default(), |
| 5961 | }) |
| 5962 | .unwrap(); |
| 5963 | |
| 5964 | let event = controller.try_next_event().unwrap(); |
| 5965 | let RemoteEvent::Failed(error) = event else { |
| 5966 | panic!("an account-switch attachment must fail closed"); |
| 5967 | }; |
| 5968 | assert!(error.contains("belongs to another account"), "{error}"); |
| 5969 | assert_eq!(controller.account_ref.as_deref(), Some("account_a")); |
| 5970 | assert_eq!(controller.attached_run_id.as_deref(), Some("run_account_a")); |
| 5971 | assert_eq!(controller.status, Status::Failed); |
| 5972 | assert!(controller.worker_tx.is_none()); |
| 5973 | assert!(controller.event_rx.is_none()); |
| 5974 | for command in [ |
| 5975 | RemoteCommand::Prompt { |
| 5976 | turn_id: "turn_rejected".to_string(), |
| 5977 | prompt: "must not execute".to_string(), |
| 5978 | }, |
| 5979 | RemoteCommand::Approval { |
| 5980 | gate: "local_approval_rejected".to_string(), |
| 5981 | approved: true, |
| 5982 | }, |
| 5983 | ] { |
| 5984 | assert!( |
| 5985 | controller |
| 5986 | .claim_command("run_account_b", 1, &command) |
| 5987 | .is_err(), |
| 5988 | "rejected account command must not be claimable: {command:?}" |
| 5989 | ); |
| 5990 | } |
| 5991 | } |
| 5992 | |
| 5993 | #[tokio::test] |
| 5994 | async fn connect_request_opts_into_runtime_chat_without_exposing_local_state() { |
| 5995 | let server = MockServer::start().await; |
| 5996 | let enrollment = fixture_enrollment(&format!("{}/", server.uri())); |
| 5997 | let start = fixture_start(); |
| 5998 | let expected = json!({ |
| 5999 | "deviceId": "device_fixture", |
| 6000 | "targetRef": "target_fixture", |
| 6001 | "displayLabel": "private-project", |
| 6002 | "runtimeVersion": "0.9.6", |
| 6003 | "runtimeCommit": "a".repeat(40), |
| 6004 | "capabilities": CAPABILITIES, |
| 6005 | "status": "active", |
| 6006 | "sessionRef": "session:fixture@01", |
| 6007 | "runtimeChatRelayProtocol": RUNTIME_CHAT_RELAY_PROTOCOL |
| 6008 | }); |
| 6009 | assert_eq!(connect_runner_body(&enrollment, &start), expected); |
| 6010 | for forbidden in [ |
| 6011 | "sessionId", |
| 6012 | "workspacePath", |
| 6013 | "path", |
| 6014 | "prompt", |
| 6015 | "environment", |
| 6016 | "env", |
| 6017 | "token", |
| 6018 | "credential", |
| 6019 | ] { |
| 6020 | assert!(expected.get(forbidden).is_none(), "leaked {forbidden}"); |
| 6021 | } |
| 6022 | Mock::given(method("POST")) |
| 6023 | .and(path("/api/local-runners/connect")) |
| 6024 | .and(body_json(expected)) |
| 6025 | .respond_with(ResponseTemplate::new(200).set_body_json(fixture_connection_response())) |
| 6026 | .expect(1) |
| 6027 | .mount(&server) |
| 6028 | .await; |
| 6029 | let client = crate::tls::reqwest_client_builder() |
| 6030 | .redirect(reqwest::redirect::Policy::none()) |
| 6031 | .build() |
| 6032 | .expect("fixture client"); |
| 6033 | |
| 6034 | let connection = connect_runner(&client, &enrollment, &start) |
| 6035 | .await |
| 6036 | .expect("strict runner attachment"); |
| 6037 | |
| 6038 | assert_eq!(connection.runner_id, "runner_fixture"); |
| 6039 | assert_eq!(connection.attachment.run_id, "run_fixture"); |
| 6040 | assert_eq!(connection.attachment.runtime_cursor, 41); |
| 6041 | assert!(!connection.attachment.snapshot_present); |
| 6042 | assert_eq!( |
| 6043 | connection.attachment.runtime_chat_relay_protocol, |
| 6044 | RUNTIME_CHAT_RELAY_PROTOCOL |
| 6045 | ); |
| 6046 | assert_eq!( |
| 6047 | connection.attachment.runtime_chat_relay_challenge, |
| 6048 | "a".repeat(32) |
| 6049 | ); |
| 6050 | } |
| 6051 | |
| 6052 | #[test] |
| 6053 | fn connection_without_links_yields_no_urls() { |
| 6054 | let enrollment = fixture_enrollment("https://api.codewhale.net/"); |
| 6055 | let start = fixture_start(); |
| 6056 | let connection = |
| 6057 | parse_runner_connection(&fixture_connection_response(), &enrollment, &start) |
| 6058 | .expect("legacy lease without links still attaches"); |
| 6059 | assert_eq!(connection.links, RemoteLinks::default()); |
| 6060 | assert!(connection.links.run_url.is_none()); |
| 6061 | assert!(connection.links.computer_url.is_none()); |
| 6062 | } |
| 6063 | |
| 6064 | #[test] |
| 6065 | fn connection_links_are_parsed_from_the_runner_lease() { |
| 6066 | let enrollment = fixture_enrollment("https://api.codewhale.net/"); |
| 6067 | let start = fixture_start(); |
| 6068 | let mut value = fixture_connection_response(); |
| 6069 | value["runner"]["runUrl"] = json!("https://app.codewhale.net/session?run=run_fixture"); |
| 6070 | value["runner"]["computerUrl"] = |
| 6071 | json!("https://app.codewhale.net/settings?section=workspaces"); |
| 6072 | let connection = parse_runner_connection(&value, &enrollment, &start) |
| 6073 | .expect("lease with links attaches"); |
| 6074 | assert_eq!( |
| 6075 | connection.links.run_url.as_deref(), |
| 6076 | Some("https://app.codewhale.net/session?run=run_fixture") |
| 6077 | ); |
| 6078 | assert_eq!( |
| 6079 | connection.links.computer_url.as_deref(), |
| 6080 | Some("https://app.codewhale.net/settings?section=workspaces") |
| 6081 | ); |
| 6082 | } |
| 6083 | |
| 6084 | #[test] |
| 6085 | fn connection_links_off_origin_or_for_another_run_are_dropped_not_fatal() { |
| 6086 | let enrollment = fixture_enrollment("https://api.codewhale.net/"); |
| 6087 | let start = fixture_start(); |
| 6088 | for spoofed in [ |
| 6089 | "http://app.codewhale.net/session?run=run_fixture", |
| 6090 | "https://app.codewhale.net.evil.example/session?run=run_fixture", |
| 6091 | "https://evil.example/session?run=run_fixture", |
| 6092 | "https://user:pw@app.codewhale.net/session?run=run_fixture", |
| 6093 | "https://app.codewhale.net:8443/session?run=run_fixture", |
| 6094 | "https://app.codewhale.net/session?run=run_fixture#token=abc", |
| 6095 | "https://app.codewhale.net/session?run=run_other", |
| 6096 | "https://app.codewhale.net/session?run=run_fixture&next=https://evil.example", |
| 6097 | "https://app.codewhale.net/logout?run=run_fixture", |
| 6098 | "", |
| 6099 | "not a url", |
| 6100 | ] { |
| 6101 | let mut value = fixture_connection_response(); |
| 6102 | value["runner"]["runUrl"] = json!(spoofed); |
| 6103 | value["runner"]["computerUrl"] = json!(spoofed); |
| 6104 | let connection = parse_runner_connection(&value, &enrollment, &start) |
| 6105 | .expect("a bad link must not break the attachment"); |
| 6106 | assert!( |
| 6107 | connection.links.run_url.is_none(), |
| 6108 | "run link accepted: {spoofed}" |
| 6109 | ); |
| 6110 | assert!( |
| 6111 | connection.links.computer_url.is_none(), |
| 6112 | "computer link accepted: {spoofed}" |
| 6113 | ); |
| 6114 | } |
| 6115 | // A non-string value is treated as absent. |
| 6116 | let mut value = fixture_connection_response(); |
| 6117 | value["runner"]["runUrl"] = json!(42); |
| 6118 | let connection = parse_runner_connection(&value, &enrollment, &start).unwrap(); |
| 6119 | assert!(connection.links.run_url.is_none()); |
| 6120 | } |
| 6121 | |
| 6122 | #[test] |
| 6123 | fn banner_leads_with_the_session_link_when_present() { |
| 6124 | let with_link = remote_control_banner( |
| 6125 | "account_fixture", |
| 6126 | "runner_fixture", |
| 6127 | Some("https://app.codewhale.net/session?run=run_fixture"), |
| 6128 | ); |
| 6129 | assert_eq!( |
| 6130 | with_link, |
| 6131 | "WEB MIRROR · https://app.codewhale.net/session?run=run_fixture · /rc stop" |
| 6132 | ); |
| 6133 | let without_link = remote_control_banner("account_fixture", "runner_fixture", None); |
| 6134 | assert_eq!( |
| 6135 | without_link, |
| 6136 | "WEB MIRROR · account account_fixture · runner runner_fixture · /rc stop" |
| 6137 | ); |
| 6138 | let notice = |
| 6139 | remote_control_link_notice("https://app.codewhale.net/session?run=run_fixture"); |
| 6140 | assert!(notice.starts_with( |
| 6141 | "Remote control is live at https://app.codewhale.net/session?run=run_fixture" |
| 6142 | )); |
| 6143 | assert!(notice.contains("/rc open")); |
| 6144 | assert!(notice.contains("/rc link")); |
| 6145 | } |
| 6146 | |
| 6147 | #[test] |
| 6148 | fn controller_exposes_links_only_while_connected() { |
| 6149 | let mut controller = RemoteControlController::default(); |
| 6150 | assert!(controller.run_url().is_none()); |
| 6151 | let (worker_tx, _worker_rx) = mpsc::unbounded_channel(); |
| 6152 | let (event_tx, event_rx) = mpsc::unbounded_channel(); |
| 6153 | controller.worker_tx = Some(worker_tx); |
| 6154 | controller.event_rx = Some(event_rx); |
| 6155 | event_tx |
| 6156 | .send(RemoteEvent::Connected { |
| 6157 | account_ref: "account_fixture".to_string(), |
| 6158 | runner_id: "runner_fixture".to_string(), |
| 6159 | target_ref: "target_fixture".to_string(), |
| 6160 | attachment: RemoteAttachment { |
| 6161 | run_id: "run_fixture".to_string(), |
| 6162 | workspace_id: "workspace_fixture".to_string(), |
| 6163 | runtime_cursor: 0, |
| 6164 | snapshot_present: false, |
| 6165 | runtime_chat_relay_protocol: RUNTIME_CHAT_RELAY_PROTOCOL.to_string(), |
| 6166 | runtime_chat_relay_challenge: "a".repeat(32), |
| 6167 | }, |
| 6168 | links: RemoteLinks { |
| 6169 | run_url: Some("https://app.codewhale.net/session?run=run_fixture".to_string()), |
| 6170 | computer_url: Some( |
| 6171 | "https://app.codewhale.net/settings?section=workspaces".to_string(), |
| 6172 | ), |
| 6173 | }, |
| 6174 | }) |
| 6175 | .unwrap(); |
| 6176 | controller.try_next_event().unwrap(); |
| 6177 | assert_eq!( |
| 6178 | controller.run_url(), |
| 6179 | Some("https://app.codewhale.net/session?run=run_fixture") |
| 6180 | ); |
| 6181 | assert_eq!( |
| 6182 | controller.computer_url(), |
| 6183 | Some("https://app.codewhale.net/settings?section=workspaces") |
| 6184 | ); |
| 6185 | assert!( |
| 6186 | controller |
| 6187 | .status_line() |
| 6188 | .contains("open https://app.codewhale.net/session?run=run_fixture") |
| 6189 | ); |
| 6190 | |
| 6191 | event_tx.send(RemoteEvent::Stopped).unwrap(); |
| 6192 | controller.try_next_event().unwrap(); |
| 6193 | assert!(controller.run_url().is_none()); |
| 6194 | assert!(controller.computer_url().is_none()); |
| 6195 | } |
| 6196 | |
| 6197 | #[test] |
| 6198 | fn connecting_never_blocks_local_prompts_and_shares_approvals_only_once_attached() { |
| 6199 | let mut controller = RemoteControlController::default(); |
| 6200 | controller.status = Status::Connecting; |
| 6201 | controller.status_detail = "waiting for account authorization".to_string(); |
| 6202 | // Mirror semantics: there is no local-input gate at all. The only |
| 6203 | // shared-decision surface is the approval card, and only once a |
| 6204 | // typed turn is bound. |
| 6205 | assert!( |
| 6206 | !controller.can_share_approval_with_web(), |
| 6207 | "without a confirmed run cursor the web cannot receive an approval" |
| 6208 | ); |
| 6209 | |
| 6210 | controller.status = Status::Connected; |
| 6211 | controller.attached_run_id = Some("run_fixture".to_string()); |
| 6212 | assert!( |
| 6213 | !controller.can_share_approval_with_web(), |
| 6214 | "a connected idle session has no typed turn to receive approvals" |
| 6215 | ); |
| 6216 | assert!(controller.attach_current_local_turn(Some("turn_fixture"))); |
| 6217 | assert!(controller.can_share_approval_with_web()); |
| 6218 | } |
| 6219 | |
| 6220 | #[test] |
| 6221 | fn persisted_device_identity_survives_a_reload_and_outlives_enrollments() { |
| 6222 | let (minted, needs_save) = resolve_device_identity(None, None); |
| 6223 | assert!(needs_save); |
| 6224 | assert!(minted.starts_with("device_")); |
| 6225 | assert!(valid_opaque_ref(&minted)); |
| 6226 | |
| 6227 | let identity = PersistedDeviceIdentity { |
| 6228 | schema_version: 1, |
| 6229 | device_id: minted.clone(), |
| 6230 | }; |
| 6231 | let raw = serde_json::to_string(&identity).expect("encode device identity"); |
| 6232 | let reloaded: PersistedDeviceIdentity = |
| 6233 | serde_json::from_str(&raw).expect("decode device identity"); |
| 6234 | assert_eq!(reloaded, identity); |
| 6235 | |
| 6236 | // A saved identity wins over any enrollment's id and needs no re-save. |
| 6237 | let (resolved, needs_save) = |
| 6238 | resolve_device_identity(Some(reloaded.clone()), Some("device_enrolled")); |
| 6239 | assert_eq!(resolved, minted); |
| 6240 | assert!(!needs_save); |
| 6241 | |
| 6242 | // Without a saved identity, an existing enrollment's device id is |
| 6243 | // adopted (upgrading terminals keep their computer row) and persisted. |
| 6244 | let (adopted, needs_save) = resolve_device_identity(None, Some("device_enrolled")); |
| 6245 | assert_eq!(adopted, "device_enrolled"); |
| 6246 | assert!(needs_save); |
| 6247 | |
| 6248 | // An unreadable identity is replaced, never fatal. |
| 6249 | let broken = PersistedDeviceIdentity { |
| 6250 | schema_version: 2, |
| 6251 | device_id: "x".to_string(), |
| 6252 | }; |
| 6253 | let (replaced, needs_save) = resolve_device_identity(Some(broken), None); |
| 6254 | assert_ne!(replaced, "x"); |
| 6255 | assert!(needs_save); |
| 6256 | assert!(serde_json::from_str::<PersistedDeviceIdentity>("{\"deviceId\":\"a\"}").is_err()); |
| 6257 | } |
| 6258 | |
| 6259 | #[test] |
| 6260 | fn remote_control_receipts_never_select_the_system_keychain() { |
| 6261 | assert!( |
| 6262 | remote_control_secrets() |
| 6263 | .backend_name() |
| 6264 | .starts_with("file-based"), |
| 6265 | "/rc enrollment must not trigger an OS keychain prompt" |
| 6266 | ); |
| 6267 | } |
| 6268 | |
| 6269 | #[test] |
| 6270 | fn attachment_response_validation_fails_closed() { |
| 6271 | let enrollment = fixture_enrollment("https://api.codewhale.net/"); |
| 6272 | let start = fixture_start(); |
| 6273 | let valid = fixture_connection_response(); |
| 6274 | assert!(parse_runner_connection(&valid, &enrollment, &start).is_ok()); |
| 6275 | |
| 6276 | let mut missing = valid.clone(); |
| 6277 | missing.as_object_mut().unwrap().remove("attachment"); |
| 6278 | assert!(parse_runner_connection(&missing, &enrollment, &start).is_err()); |
| 6279 | |
| 6280 | let mut oversized_cursor = valid.clone(); |
| 6281 | oversized_cursor["attachment"]["runtimeCursor"] = json!(JS_MAX_SAFE_INTEGER + 1); |
| 6282 | assert!(parse_runner_connection(&oversized_cursor, &enrollment, &start).is_err()); |
| 6283 | |
| 6284 | let mut false_receipt = valid.clone(); |
| 6285 | false_receipt["attachment"]["snapshotPresent"] = json!("false"); |
| 6286 | assert!(parse_runner_connection(&false_receipt, &enrollment, &start).is_err()); |
| 6287 | |
| 6288 | let mut extra_authority = valid.clone(); |
| 6289 | extra_authority["attachment"]["workspacePath"] = json!("/private/project"); |
| 6290 | assert!(parse_runner_connection(&extra_authority, &enrollment, &start).is_err()); |
| 6291 | |
| 6292 | let mut wrong_protocol = valid.clone(); |
| 6293 | wrong_protocol["attachment"]["runtimeChatRelayProtocol"] = json!("legacy"); |
| 6294 | assert!(parse_runner_connection(&wrong_protocol, &enrollment, &start).is_err()); |
| 6295 | |
| 6296 | let mut invalid_challenge = valid.clone(); |
| 6297 | invalid_challenge["attachment"]["runtimeChatRelayChallenge"] = json!("too-short"); |
| 6298 | assert!(parse_runner_connection(&invalid_challenge, &enrollment, &start).is_err()); |
| 6299 | |
| 6300 | let mut wrong_control_path = valid; |
| 6301 | wrong_control_path["runner"]["controlPath"] = json!("direct_native"); |
| 6302 | assert!(parse_runner_connection(&wrong_control_path, &enrollment, &start).is_err()); |
| 6303 | } |
| 6304 | |
| 6305 | #[test] |
| 6306 | fn attachment_cursor_seeds_the_first_runtime_event_sequence() { |
| 6307 | let (mut controller, mut worker_rx, event_tx, _journal_root) = wired_controller(); |
| 6308 | event_tx |
| 6309 | .send(RemoteEvent::Connected { |
| 6310 | account_ref: "account_fixture".to_string(), |
| 6311 | runner_id: "runner_fixture".to_string(), |
| 6312 | target_ref: "target_fixture".to_string(), |
| 6313 | attachment: RemoteAttachment { |
| 6314 | run_id: "run_fixture".to_string(), |
| 6315 | workspace_id: "workspace_fixture".to_string(), |
| 6316 | runtime_cursor: 41, |
| 6317 | snapshot_present: false, |
| 6318 | runtime_chat_relay_protocol: RUNTIME_CHAT_RELAY_PROTOCOL.to_string(), |
| 6319 | runtime_chat_relay_challenge: "a".repeat(32), |
| 6320 | }, |
| 6321 | links: RemoteLinks::default(), |
| 6322 | }) |
| 6323 | .unwrap(); |
| 6324 | |
| 6325 | assert!(matches!( |
| 6326 | controller.try_next_event(), |
| 6327 | Some(RemoteEvent::Connected { .. }) |
| 6328 | )); |
| 6329 | controller.upload_snapshot("run_fixture", &[]); |
| 6330 | |
| 6331 | let WorkerCommand::Upload { envelopes, .. } = worker_rx.try_recv().unwrap() else { |
| 6332 | panic!("expected snapshot upload"); |
| 6333 | }; |
| 6334 | assert_eq!(envelopes.len(), 1); |
| 6335 | assert_eq!(envelopes[0]["event"], "session.snapshot"); |
| 6336 | assert_eq!(envelopes[0]["seq"], 42); |
| 6337 | } |
| 6338 | |
| 6339 | #[test] |
| 6340 | fn connected_attachment_adopts_the_existing_turn_and_streams_typed_state_once() { |
| 6341 | let (mut controller, mut worker_rx, event_tx, _journal_root) = wired_controller(); |
| 6342 | event_tx |
| 6343 | .send(RemoteEvent::Connected { |
| 6344 | account_ref: "account_fixture".to_string(), |
| 6345 | runner_id: "runner_fixture".to_string(), |
| 6346 | target_ref: "target_fixture".to_string(), |
| 6347 | attachment: RemoteAttachment { |
| 6348 | run_id: "run_fixture".to_string(), |
| 6349 | workspace_id: "workspace_fixture".to_string(), |
| 6350 | runtime_cursor: 11, |
| 6351 | snapshot_present: false, |
| 6352 | runtime_chat_relay_protocol: RUNTIME_CHAT_RELAY_PROTOCOL.to_string(), |
| 6353 | runtime_chat_relay_challenge: "a".repeat(32), |
| 6354 | }, |
| 6355 | links: RemoteLinks::default(), |
| 6356 | }) |
| 6357 | .unwrap(); |
| 6358 | assert!(matches!( |
| 6359 | controller.try_next_event(), |
| 6360 | Some(RemoteEvent::Connected { .. }) |
| 6361 | )); |
| 6362 | assert!( |
| 6363 | !controller.can_share_approval_with_web(), |
| 6364 | "a connected attachment without a bound typed turn keeps approvals local" |
| 6365 | ); |
| 6366 | assert!(controller.attach_current_local_turn(Some("turn_existing"))); |
| 6367 | assert!( |
| 6368 | controller.can_share_approval_with_web(), |
| 6369 | "the bound active turn can carry typed approvals to the web" |
| 6370 | ); |
| 6371 | assert!(controller.has_active_run()); |
| 6372 | assert!(controller.active_run_matches("run_fixture")); |
| 6373 | assert!( |
| 6374 | !controller.attach_current_local_turn(Some("turn_duplicate")), |
| 6375 | "replaying the attachment must not replace or duplicate the turn" |
| 6376 | ); |
| 6377 | |
| 6378 | controller.observe_engine_event(&EngineEvent::MessageDelta { |
| 6379 | index: 0, |
| 6380 | content: "existing turn output".to_string(), |
| 6381 | }); |
| 6382 | controller.observe_engine_event(&EngineEvent::ToolCallStarted { |
| 6383 | id: "tool_existing".to_string(), |
| 6384 | name: "shell".to_string(), |
| 6385 | input: json!({ "never": "relayed" }), |
| 6386 | }); |
| 6387 | let gate = controller.record_remote_approval( |
| 6388 | "tool_existing", |
| 6389 | "shell", |
| 6390 | "approve bounded fixture", |
| 6391 | &json!({ "credential": "must-not-cross" }), |
| 6392 | "approval-key", |
| 6393 | None, |
| 6394 | ); |
| 6395 | |
| 6396 | let mut envelopes = Vec::new(); |
| 6397 | for _ in 0..3 { |
| 6398 | let WorkerCommand::Upload { |
| 6399 | envelopes: batch, .. |
| 6400 | } = worker_rx.try_recv().expect("typed active-turn upload") |
| 6401 | else { |
| 6402 | panic!("active-turn state must use the runtime envelope channel"); |
| 6403 | }; |
| 6404 | envelopes.extend(batch); |
| 6405 | } |
| 6406 | assert_eq!( |
| 6407 | envelopes |
| 6408 | .iter() |
| 6409 | .map(|event| event["event"].as_str().unwrap()) |
| 6410 | .collect::<Vec<_>>(), |
| 6411 | vec!["item.delta", "item.started", "approval.required"] |
| 6412 | ); |
| 6413 | assert!( |
| 6414 | envelopes |
| 6415 | .iter() |
| 6416 | .all(|event| event["turn_id"] == "turn_existing") |
| 6417 | ); |
| 6418 | assert_eq!( |
| 6419 | envelopes[2]["payload"]["approval_id"].as_str(), |
| 6420 | Some(gate.as_str()) |
| 6421 | ); |
| 6422 | let projected = serde_json::to_string(&envelopes).unwrap(); |
| 6423 | assert!(!projected.contains("must-not-cross")); |
| 6424 | assert!(!projected.contains("approval-key")); |
| 6425 | assert!(worker_rx.try_recv().is_err()); |
| 6426 | } |
| 6427 | |
| 6428 | #[test] |
| 6429 | fn dispatch_window_attachment_promotes_on_typed_start_and_reconnect_is_idempotent() { |
| 6430 | let (mut controller, mut worker_rx, event_tx, _journal_root) = wired_controller(); |
| 6431 | let attachment = RemoteAttachment { |
| 6432 | run_id: "run_fixture".to_string(), |
| 6433 | workspace_id: "workspace_fixture".to_string(), |
| 6434 | runtime_cursor: 3, |
| 6435 | snapshot_present: false, |
| 6436 | runtime_chat_relay_protocol: RUNTIME_CHAT_RELAY_PROTOCOL.to_string(), |
| 6437 | runtime_chat_relay_challenge: "a".repeat(32), |
| 6438 | }; |
| 6439 | event_tx |
| 6440 | .send(RemoteEvent::Connected { |
| 6441 | account_ref: "account_fixture".to_string(), |
| 6442 | runner_id: "runner_fixture".to_string(), |
| 6443 | target_ref: "target_fixture".to_string(), |
| 6444 | attachment: attachment.clone(), |
| 6445 | links: RemoteLinks::default(), |
| 6446 | }) |
| 6447 | .unwrap(); |
| 6448 | controller.try_next_event().unwrap(); |
| 6449 | |
| 6450 | assert!(controller.attach_current_local_turn(None)); |
| 6451 | let pending_lease = controller |
| 6452 | .journal |
| 6453 | .as_ref() |
| 6454 | .and_then(RuntimeEventJournal::classic_lease) |
| 6455 | .expect("dispatch-window lease is durable"); |
| 6456 | assert!(pending_lease.turn_id.is_none()); |
| 6457 | assert!( |
| 6458 | controller.has_active_run(), |
| 6459 | "the dispatch window gates stop" |
| 6460 | ); |
| 6461 | assert!(controller.active_run_matches("run_fixture")); |
| 6462 | assert!( |
| 6463 | !controller.can_share_approval_with_web(), |
| 6464 | "a pending dispatch has no typed turn id, so approvals stay local" |
| 6465 | ); |
| 6466 | assert!(worker_rx.try_recv().is_err()); |
| 6467 | |
| 6468 | controller.observe_engine_event(&EngineEvent::TurnStarted { |
| 6469 | turn_id: "turn_started_later".to_string(), |
| 6470 | created_at: chrono::Utc::now(), |
| 6471 | route: None, |
| 6472 | }); |
| 6473 | let WorkerCommand::Upload { envelopes, .. } = |
| 6474 | worker_rx.try_recv().expect("one typed turn start") |
| 6475 | else { |
| 6476 | panic!("turn start must use the runtime envelope channel"); |
| 6477 | }; |
| 6478 | assert_eq!(envelopes.len(), 1); |
| 6479 | assert_eq!(envelopes[0]["seq"], 4); |
| 6480 | assert_eq!(envelopes[0]["event"], "turn.started"); |
| 6481 | assert_eq!(envelopes[0]["turn_id"], "turn_started_later"); |
| 6482 | let promoted_lease = controller |
| 6483 | .journal |
| 6484 | .as_ref() |
| 6485 | .and_then(RuntimeEventJournal::classic_lease) |
| 6486 | .expect("typed start keeps the durable lease"); |
| 6487 | assert_eq!( |
| 6488 | promoted_lease.turn_id.as_deref(), |
| 6489 | Some("turn_started_later") |
| 6490 | ); |
| 6491 | assert_eq!( |
| 6492 | promoted_lease.lease_id, pending_lease.lease_id, |
| 6493 | "typed start promotes the same dispatch generation" |
| 6494 | ); |
| 6495 | let started_envelope = envelopes[0].clone(); |
| 6496 | assert!( |
| 6497 | controller.can_share_approval_with_web(), |
| 6498 | "typed TurnStarted promotes the dispatch into a web-owned active turn" |
| 6499 | ); |
| 6500 | |
| 6501 | event_tx |
| 6502 | .send(RemoteEvent::Attachment { |
| 6503 | account_ref: "account_fixture".to_string(), |
| 6504 | target_ref: "target_fixture".to_string(), |
| 6505 | attachment, |
| 6506 | links: RemoteLinks::default(), |
| 6507 | }) |
| 6508 | .unwrap(); |
| 6509 | controller.try_next_event().unwrap(); |
| 6510 | assert!( |
| 6511 | !controller.attach_current_local_turn(Some("turn_replayed")), |
| 6512 | "a reconnect must not rebind the live turn" |
| 6513 | ); |
| 6514 | let WorkerCommand::Upload { envelopes, .. } = worker_rx |
| 6515 | .try_recv() |
| 6516 | .expect("the unacknowledged start is replayed unchanged") |
| 6517 | else { |
| 6518 | panic!("runtime replay must use the envelope channel"); |
| 6519 | }; |
| 6520 | assert_eq!( |
| 6521 | envelopes, |
| 6522 | vec![started_envelope], |
| 6523 | "retry safety preserves the original sequence and payload" |
| 6524 | ); |
| 6525 | assert!(worker_rx.try_recv().is_err()); |
| 6526 | |
| 6527 | controller.observe_engine_event(&turn_complete_event()); |
| 6528 | let WorkerCommand::Upload { envelopes, .. } = |
| 6529 | worker_rx.try_recv().expect("one terminal receipt") |
| 6530 | else { |
| 6531 | panic!("turn completion must use the runtime envelope channel"); |
| 6532 | }; |
| 6533 | assert_eq!(envelopes.len(), 1); |
| 6534 | assert_eq!(envelopes[0]["seq"], 5); |
| 6535 | assert_eq!(envelopes[0]["event"], "turn.completed"); |
| 6536 | assert_eq!(envelopes[0]["turn_id"], "turn_started_later"); |
| 6537 | assert!(!controller.has_active_run()); |
| 6538 | assert!(worker_rx.try_recv().is_err()); |
| 6539 | } |
| 6540 | |
| 6541 | #[test] |
| 6542 | fn dispatch_window_that_ends_before_typed_start_returns_to_idle_attachment() { |
| 6543 | let (mut controller, mut worker_rx, event_tx, _journal_root) = wired_controller(); |
| 6544 | event_tx |
| 6545 | .send(RemoteEvent::Connected { |
| 6546 | account_ref: "account_fixture".to_string(), |
| 6547 | runner_id: "runner_fixture".to_string(), |
| 6548 | target_ref: "target_fixture".to_string(), |
| 6549 | attachment: RemoteAttachment { |
| 6550 | run_id: "run_fixture".to_string(), |
| 6551 | workspace_id: "workspace_fixture".to_string(), |
| 6552 | runtime_cursor: 8, |
| 6553 | snapshot_present: false, |
| 6554 | runtime_chat_relay_protocol: RUNTIME_CHAT_RELAY_PROTOCOL.to_string(), |
| 6555 | runtime_chat_relay_challenge: "a".repeat(32), |
| 6556 | }, |
| 6557 | links: RemoteLinks::default(), |
| 6558 | }) |
| 6559 | .unwrap(); |
| 6560 | controller.try_next_event().unwrap(); |
| 6561 | |
| 6562 | assert!(controller.attach_current_local_turn(None)); |
| 6563 | assert!(controller.has_active_run()); |
| 6564 | assert!(controller.release_unstarted_local_turn()); |
| 6565 | assert!(!controller.has_active_run()); |
| 6566 | assert!(!controller.can_share_approval_with_web()); |
| 6567 | assert!( |
| 6568 | !controller.release_unstarted_local_turn(), |
| 6569 | "reconciling the same idle boundary is idempotent" |
| 6570 | ); |
| 6571 | assert!(worker_rx.try_recv().is_err()); |
| 6572 | } |
| 6573 | |
| 6574 | #[test] |
| 6575 | fn fresh_controller_refreshes_old_server_snapshot_then_deduplicates_reconnects() { |
| 6576 | let (mut controller, mut worker_rx, event_tx, _journal_root) = wired_controller(); |
| 6577 | event_tx |
| 6578 | .send(RemoteEvent::Connected { |
| 6579 | account_ref: "account_fixture".to_string(), |
| 6580 | runner_id: "runner_fixture".to_string(), |
| 6581 | target_ref: "target_fixture".to_string(), |
| 6582 | attachment: RemoteAttachment { |
| 6583 | run_id: "run_fixture".to_string(), |
| 6584 | workspace_id: "workspace_fixture".to_string(), |
| 6585 | runtime_cursor: 7, |
| 6586 | snapshot_present: true, |
| 6587 | runtime_chat_relay_protocol: RUNTIME_CHAT_RELAY_PROTOCOL.to_string(), |
| 6588 | runtime_chat_relay_challenge: "a".repeat(32), |
| 6589 | }, |
| 6590 | links: RemoteLinks::default(), |
| 6591 | }) |
| 6592 | .unwrap(); |
| 6593 | controller.try_next_event().unwrap(); |
| 6594 | |
| 6595 | controller.upload_snapshot("run_fixture", &[]); |
| 6596 | let WorkerCommand::Upload { envelopes, .. } = worker_rx.try_recv().unwrap() else { |
| 6597 | panic!("fresh controller must refresh saved history"); |
| 6598 | }; |
| 6599 | assert_eq!(envelopes[0]["seq"], 8); |
| 6600 | |
| 6601 | event_tx |
| 6602 | .send(RemoteEvent::Attachment { |
| 6603 | account_ref: "account_fixture".to_string(), |
| 6604 | target_ref: "target_fixture".to_string(), |
| 6605 | attachment: RemoteAttachment { |
| 6606 | run_id: "run_fixture".to_string(), |
| 6607 | workspace_id: "workspace_fixture".to_string(), |
| 6608 | runtime_cursor: 7, |
| 6609 | snapshot_present: false, |
| 6610 | runtime_chat_relay_protocol: RUNTIME_CHAT_RELAY_PROTOCOL.to_string(), |
| 6611 | runtime_chat_relay_challenge: "a".repeat(32), |
| 6612 | }, |
| 6613 | links: RemoteLinks::default(), |
| 6614 | }) |
| 6615 | .unwrap(); |
| 6616 | controller.try_next_event().unwrap(); |
| 6617 | controller.upload_snapshot("run_fixture", &[]); |
| 6618 | let WorkerCommand::Upload { envelopes, .. } = worker_rx.try_recv().unwrap() else { |
| 6619 | panic!("unacknowledged snapshot must be retried"); |
| 6620 | }; |
| 6621 | assert_eq!(envelopes[0]["seq"], 8); |
| 6622 | controller.upload_snapshot("run_fixture", &[]); |
| 6623 | assert!(worker_rx.try_recv().is_err()); |
| 6624 | |
| 6625 | event_tx |
| 6626 | .send(RemoteEvent::Attachment { |
| 6627 | account_ref: "account_fixture".to_string(), |
| 6628 | target_ref: "target_fixture".to_string(), |
| 6629 | attachment: RemoteAttachment { |
| 6630 | run_id: "run_fixture".to_string(), |
| 6631 | workspace_id: "workspace_fixture".to_string(), |
| 6632 | runtime_cursor: 8, |
| 6633 | snapshot_present: true, |
| 6634 | runtime_chat_relay_protocol: RUNTIME_CHAT_RELAY_PROTOCOL.to_string(), |
| 6635 | runtime_chat_relay_challenge: "a".repeat(32), |
| 6636 | }, |
| 6637 | links: RemoteLinks::default(), |
| 6638 | }) |
| 6639 | .unwrap(); |
| 6640 | controller.try_next_event().unwrap(); |
| 6641 | controller.upload_snapshot("run_fixture", &[]); |
| 6642 | assert!(worker_rx.try_recv().is_err()); |
| 6643 | } |
| 6644 | |
| 6645 | #[test] |
| 6646 | fn reconnect_cursor_retires_only_the_acknowledged_prefix() { |
| 6647 | let (mut controller, mut worker_rx, event_tx, _journal_root) = wired_controller(); |
| 6648 | controller.event_seq.insert("run_fixture".to_string(), 6); |
| 6649 | controller.upload_envelope( |
| 6650 | "run_fixture", |
| 6651 | "item.delta", |
| 6652 | None, |
| 6653 | json!({ "delta": "seven" }), |
| 6654 | ); |
| 6655 | controller.upload_envelope( |
| 6656 | "run_fixture", |
| 6657 | "item.delta", |
| 6658 | None, |
| 6659 | json!({ "delta": "eight" }), |
| 6660 | ); |
| 6661 | worker_rx.try_recv().unwrap(); |
| 6662 | worker_rx.try_recv().unwrap(); |
| 6663 | |
| 6664 | event_tx |
| 6665 | .send(RemoteEvent::Attachment { |
| 6666 | account_ref: "account_fixture".to_string(), |
| 6667 | target_ref: "target_fixture".to_string(), |
| 6668 | attachment: RemoteAttachment { |
| 6669 | run_id: "run_fixture".to_string(), |
| 6670 | workspace_id: "workspace_fixture".to_string(), |
| 6671 | runtime_cursor: 7, |
| 6672 | snapshot_present: false, |
| 6673 | runtime_chat_relay_protocol: RUNTIME_CHAT_RELAY_PROTOCOL.to_string(), |
| 6674 | runtime_chat_relay_challenge: "a".repeat(32), |
| 6675 | }, |
| 6676 | links: RemoteLinks::default(), |
| 6677 | }) |
| 6678 | .unwrap(); |
| 6679 | controller.try_next_event().unwrap(); |
| 6680 | |
| 6681 | let WorkerCommand::Upload { envelopes, .. } = worker_rx.try_recv().unwrap() else { |
| 6682 | panic!("seq 8 must remain pending"); |
| 6683 | }; |
| 6684 | assert_eq!(envelopes.len(), 1); |
| 6685 | assert_eq!(envelopes[0]["seq"], 8); |
| 6686 | assert_eq!( |
| 6687 | controller |
| 6688 | .pending_runtime_events |
| 6689 | .get("run_fixture") |
| 6690 | .unwrap() |
| 6691 | .keys() |
| 6692 | .copied() |
| 6693 | .collect::<Vec<_>>(), |
| 6694 | vec![8] |
| 6695 | ); |
| 6696 | |
| 6697 | event_tx |
| 6698 | .send(RemoteEvent::Attachment { |
| 6699 | account_ref: "account_fixture".to_string(), |
| 6700 | target_ref: "target_fixture".to_string(), |
| 6701 | attachment: RemoteAttachment { |
| 6702 | run_id: "run_fixture".to_string(), |
| 6703 | workspace_id: "workspace_fixture".to_string(), |
| 6704 | runtime_cursor: 6, |
| 6705 | snapshot_present: false, |
| 6706 | runtime_chat_relay_protocol: RUNTIME_CHAT_RELAY_PROTOCOL.to_string(), |
| 6707 | runtime_chat_relay_challenge: "a".repeat(32), |
| 6708 | }, |
| 6709 | links: RemoteLinks::default(), |
| 6710 | }) |
| 6711 | .unwrap(); |
| 6712 | controller.try_next_event().unwrap(); |
| 6713 | let WorkerCommand::Upload { envelopes, .. } = worker_rx.try_recv().unwrap() else { |
| 6714 | panic!("older cursor cannot discard seq 8"); |
| 6715 | }; |
| 6716 | assert_eq!(envelopes[0]["seq"], 8); |
| 6717 | |
| 6718 | event_tx |
| 6719 | .send(RemoteEvent::RuntimeCursor { |
| 6720 | run_id: "run_fixture".to_string(), |
| 6721 | cursor: 8, |
| 6722 | }) |
| 6723 | .unwrap(); |
| 6724 | controller.try_next_event().unwrap(); |
| 6725 | assert!( |
| 6726 | !controller |
| 6727 | .pending_runtime_events |
| 6728 | .contains_key("run_fixture") |
| 6729 | ); |
| 6730 | } |
| 6731 | |
| 6732 | #[tokio::test] |
| 6733 | async fn ambiguous_success_retries_the_identical_runtime_event_until_cursor_acceptance() { |
| 6734 | let server = MockServer::start().await; |
| 6735 | let responder = AmbiguousRuntimeResponder::default(); |
| 6736 | Mock::given(method("POST")) |
| 6737 | .and(path( |
| 6738 | "/api/local-runners/runner_fixture/runs/run_fixture/events", |
| 6739 | )) |
| 6740 | .respond_with(responder.clone()) |
| 6741 | .expect(2) |
| 6742 | .mount(&server) |
| 6743 | .await; |
| 6744 | let enrollment = fixture_enrollment(&format!("{}/", server.uri())); |
| 6745 | let client = crate::tls::reqwest_client_builder() |
| 6746 | .redirect(reqwest::redirect::Policy::none()) |
| 6747 | .build() |
| 6748 | .expect("fixture client"); |
| 6749 | let envelope = runtime_envelope( |
| 6750 | 1, |
| 6751 | "item.delta", |
| 6752 | None, |
| 6753 | "2026-08-08T12:00:00Z".to_string(), |
| 6754 | json!({ "delta": "exact body" }), |
| 6755 | ); |
| 6756 | let mut outbox = RuntimeTransportOutbox::default(); |
| 6757 | outbox |
| 6758 | .enqueue("run_fixture", envelope) |
| 6759 | .expect("queue runtime event"); |
| 6760 | |
| 6761 | assert_eq!( |
| 6762 | outbox |
| 6763 | .try_flush_one(&client, &enrollment, "runner_fixture") |
| 6764 | .await |
| 6765 | .unwrap(), |
| 6766 | RuntimeFlushOutcome::Retryable |
| 6767 | ); |
| 6768 | assert_eq!(outbox.events.len(), 1); |
| 6769 | assert_eq!( |
| 6770 | outbox |
| 6771 | .try_flush_one(&client, &enrollment, "runner_fixture") |
| 6772 | .await |
| 6773 | .unwrap(), |
| 6774 | RuntimeFlushOutcome::Accepted { |
| 6775 | run_id: "run_fixture".to_string(), |
| 6776 | cursor: 1, |
| 6777 | } |
| 6778 | ); |
| 6779 | assert!(outbox.events.is_empty()); |
| 6780 | let bodies = responder.bodies.lock().unwrap(); |
| 6781 | assert_eq!(bodies.len(), 2); |
| 6782 | assert_eq!(bodies[0], bodies[1]); |
| 6783 | } |
| 6784 | |
| 6785 | #[test] |
| 6786 | fn snapshot_envelope_is_unicode_safe_and_keeps_newest_history() { |
| 6787 | let messages = (0..80) |
| 6788 | .map(|index| { |
| 6789 | let marker = format!("message-{index:02}-"); |
| 6790 | text_message( |
| 6791 | if index % 2 == 0 { "user" } else { "assistant" }, |
| 6792 | marker + &"🫧\"\\\n".repeat(1_500), |
| 6793 | ) |
| 6794 | }) |
| 6795 | .collect::<Vec<_>>(); |
| 6796 | |
| 6797 | let envelope = bounded_session_snapshot_envelope(1, &messages); |
| 6798 | let encoded = serde_json::to_vec(&envelope).unwrap(); |
| 6799 | let retained = envelope["payload"]["messages"].as_array().unwrap(); |
| 6800 | |
| 6801 | assert!(encoded.len() <= SNAPSHOT_ENVELOPE_BYTE_BUDGET); |
| 6802 | assert!(encoded.len() < MAX_RUNTIME_ENVELOPE_BYTES); |
| 6803 | assert!(!retained.is_empty()); |
| 6804 | assert!(retained.len() <= MAX_SNAPSHOT_MESSAGES); |
| 6805 | assert!( |
| 6806 | retained.last().unwrap()["text"] |
| 6807 | .as_str() |
| 6808 | .unwrap() |
| 6809 | .starts_with("message-79-") |
| 6810 | ); |
| 6811 | for message in retained { |
| 6812 | let text = message["text"].as_str().unwrap(); |
| 6813 | assert!(!text.contains('\u{FFFD}')); |
| 6814 | assert!(text.is_char_boundary(text.len())); |
| 6815 | } |
| 6816 | } |
| 6817 | |
| 6818 | #[test] |
| 6819 | fn snapshot_truncation_pins_the_exact_encoded_byte_boundary() { |
| 6820 | let source = "🫧\"\\\n".repeat(40_000); |
| 6821 | let message = text_message("assistant", source); |
| 6822 | let envelope = bounded_session_snapshot_envelope(9, std::slice::from_ref(&message)); |
| 6823 | let encoded = serde_json::to_vec(&envelope).unwrap(); |
| 6824 | assert!(encoded.len() <= SNAPSHOT_ENVELOPE_BYTE_BUDGET); |
| 6825 | |
| 6826 | let retained = envelope["payload"]["messages"][0]["text"].as_str().unwrap(); |
| 6827 | let projected = project_session_message(&message).unwrap().1; |
| 6828 | let retained_chars = retained.chars().count(); |
| 6829 | let next = projected.chars().nth(retained_chars).unwrap(); |
| 6830 | let mut one_more = retained.to_string(); |
| 6831 | one_more.push(next); |
| 6832 | let timestamp = envelope["timestamp"].as_str().unwrap(); |
| 6833 | let expanded = vec![json!({ "role": "assistant", "text": one_more })]; |
| 6834 | assert!( |
| 6835 | snapshot_envelope_len(9, timestamp, &expanded) > SNAPSHOT_ENVELOPE_BYTE_BUDGET, |
| 6836 | "one more Unicode scalar must cross the chosen encoded boundary" |
| 6837 | ); |
| 6838 | } |
| 6839 | |
| 6840 | #[test] |
| 6841 | fn fatal_engine_error_projects_failure_and_releases_the_remote_run() { |
| 6842 | let (mut controller, mut worker_rx, _event_tx, _journal_root) = wired_controller(); |
| 6843 | controller |
| 6844 | .activate_prompt("run_fixture", "turn_fixture") |
| 6845 | .unwrap(); |
| 6846 | let secret = "sk-runtime-secret-that-must-not-cross-the-relay"; |
| 6847 | let message = format!( |
| 6848 | "DeepSeek API key: {secret}\n{}", |
| 6849 | "🫧".repeat(MAX_REMOTE_ERROR_MESSAGE_BYTES) |
| 6850 | ); |
| 6851 | |
| 6852 | controller.observe_engine_event(&EngineEvent::Error { |
| 6853 | envelope: crate::error_taxonomy::ErrorEnvelope::new( |
| 6854 | crate::error_taxonomy::ErrorCategory::Authentication, |
| 6855 | crate::error_taxonomy::ErrorSeverity::Critical, |
| 6856 | false, |
| 6857 | "llm_auth_error", |
| 6858 | message, |
| 6859 | ), |
| 6860 | recoverable: false, |
| 6861 | }); |
| 6862 | |
| 6863 | assert!(!controller.has_active_run()); |
| 6864 | let WorkerCommand::Upload { |
| 6865 | envelopes: failed, .. |
| 6866 | } = worker_rx.try_recv().expect("fatal item upload") |
| 6867 | else { |
| 6868 | panic!("fatal error must upload an item.failed envelope"); |
| 6869 | }; |
| 6870 | let WorkerCommand::Upload { |
| 6871 | envelopes: completed, |
| 6872 | .. |
| 6873 | } = worker_rx.try_recv().expect("fatal turn upload") |
| 6874 | else { |
| 6875 | panic!("fatal error must upload a terminal turn envelope"); |
| 6876 | }; |
| 6877 | assert_eq!(failed.len(), 1); |
| 6878 | assert_eq!(failed[0]["seq"], 1); |
| 6879 | assert_eq!(failed[0]["event"], "item.failed"); |
| 6880 | assert_eq!(failed[0]["turn_id"], "turn_fixture"); |
| 6881 | assert_eq!(failed[0]["payload"]["item"]["kind"], "error"); |
| 6882 | assert_eq!(failed[0]["payload"]["item"]["status"], "failed"); |
| 6883 | let projected = failed[0]["payload"]["item"]["detail"] |
| 6884 | .as_str() |
| 6885 | .expect("bounded error detail"); |
| 6886 | assert!(projected.len() <= MAX_REMOTE_ERROR_MESSAGE_BYTES); |
| 6887 | assert!(projected.is_char_boundary(projected.len())); |
| 6888 | assert!(!projected.contains(secret)); |
| 6889 | assert!(projected.contains("[redacted]")); |
| 6890 | |
| 6891 | assert_eq!(completed.len(), 1); |
| 6892 | assert_eq!(completed[0]["seq"], 2); |
| 6893 | assert_eq!(completed[0]["event"], "turn.completed"); |
| 6894 | assert_eq!(completed[0]["turn_id"], "turn_fixture"); |
| 6895 | assert_eq!(completed[0]["payload"]["turn"]["status"], "failed"); |
| 6896 | assert_eq!( |
| 6897 | controller.pending_runtime_events["run_fixture"] |
| 6898 | .keys() |
| 6899 | .copied() |
| 6900 | .collect::<Vec<_>>(), |
| 6901 | vec![1, 2] |
| 6902 | ); |
| 6903 | assert!(worker_rx.try_recv().is_err()); |
| 6904 | } |
| 6905 | |
| 6906 | #[test] |
| 6907 | fn recoverable_engine_error_stays_nonterminal_for_provider_fallback() { |
| 6908 | let mut controller = RemoteControlController::default(); |
| 6909 | let (worker_tx, mut worker_rx) = mpsc::unbounded_channel(); |
| 6910 | controller.worker_tx = Some(worker_tx); |
| 6911 | controller |
| 6912 | .activate_prompt("run_fixture", "turn_fixture") |
| 6913 | .unwrap(); |
| 6914 | |
| 6915 | controller.observe_engine_event(&EngineEvent::Error { |
| 6916 | envelope: crate::error_taxonomy::ErrorEnvelope::network( |
| 6917 | "temporary provider connection failure", |
| 6918 | ), |
| 6919 | recoverable: true, |
| 6920 | }); |
| 6921 | |
| 6922 | assert!(controller.active_run_matches("run_fixture")); |
| 6923 | assert!(controller.pending_runtime_events.is_empty()); |
| 6924 | assert!(worker_rx.try_recv().is_err()); |
| 6925 | } |
| 6926 | |
| 6927 | #[test] |
| 6928 | fn terminal_pre_dispatch_error_uses_the_same_failure_projection() { |
| 6929 | let (mut controller, mut worker_rx, _event_tx, _journal_root) = wired_controller(); |
| 6930 | controller |
| 6931 | .activate_prompt("run_fixture", "turn_fixture") |
| 6932 | .unwrap(); |
| 6933 | |
| 6934 | controller.fail_active_dispatch( |
| 6935 | "DeepSeek API key: sk-preflight-secret-that-must-not-cross-the-relay", |
| 6936 | ); |
| 6937 | |
| 6938 | assert!(!controller.has_active_run()); |
| 6939 | let WorkerCommand::Upload { envelopes, .. } = |
| 6940 | worker_rx.try_recv().expect("pre-dispatch item upload") |
| 6941 | else { |
| 6942 | panic!("pre-dispatch error must upload item.failed"); |
| 6943 | }; |
| 6944 | assert_eq!(envelopes[0]["event"], "item.failed"); |
| 6945 | assert!(!envelopes[0].to_string().contains("sk-preflight-secret")); |
| 6946 | let WorkerCommand::Upload { envelopes, .. } = |
| 6947 | worker_rx.try_recv().expect("pre-dispatch turn upload") |
| 6948 | else { |
| 6949 | panic!("pre-dispatch error must upload turn.completed"); |
| 6950 | }; |
| 6951 | assert_eq!(envelopes[0]["event"], "turn.completed"); |
| 6952 | assert_eq!(envelopes[0]["payload"]["turn"]["status"], "failed"); |
| 6953 | assert!(worker_rx.try_recv().is_err()); |
| 6954 | } |
| 6955 | |
| 6956 | #[test] |
| 6957 | fn typed_command_parser_rejects_shell_and_cross_run_content() { |
| 6958 | let prompt = parse_remote_command( |
| 6959 | &json!({ |
| 6960 | "type": "prompt.request", |
| 6961 | "runId": "run-1", |
| 6962 | "turnId": "turn-1", |
| 6963 | "prompt": "Continue", |
| 6964 | }), |
| 6965 | "run-1", |
| 6966 | ) |
| 6967 | .unwrap(); |
| 6968 | assert_eq!( |
| 6969 | prompt, |
| 6970 | RemoteCommand::Prompt { |
| 6971 | turn_id: "turn-1".to_string(), |
| 6972 | prompt: "Continue".to_string(), |
| 6973 | } |
| 6974 | ); |
| 6975 | assert!( |
| 6976 | parse_remote_command( |
| 6977 | &json!({ |
| 6978 | "type": "shell", |
| 6979 | "runId": "run-1", |
| 6980 | "command": "rm -rf /", |
| 6981 | }), |
| 6982 | "run-1" |
| 6983 | ) |
| 6984 | .is_err() |
| 6985 | ); |
| 6986 | assert!( |
| 6987 | parse_remote_command( |
| 6988 | &json!({ |
| 6989 | "type": "prompt.request", |
| 6990 | "runId": "run-other", |
| 6991 | "turnId": "turn-1", |
| 6992 | "prompt": "Continue", |
| 6993 | }), |
| 6994 | "run-1" |
| 6995 | ) |
| 6996 | .is_err() |
| 6997 | ); |
| 6998 | } |
| 6999 | |
| 7000 | fn runtime_chat_prompt_fixture() -> Value { |
| 7001 | json!({ |
| 7002 | "type": "prompt.request", |
| 7003 | "runId": "run-1", |
| 7004 | "turnId": format!("local_turn_{}", "b".repeat(24)), |
| 7005 | "operationKey": "operation-1", |
| 7006 | "runtimeBindingId": "binding-1", |
| 7007 | "runtimeThreadId": format!("local_thread_{}", "a".repeat(24)), |
| 7008 | "prompt": "Hello from account Chat", |
| 7009 | "systemPrompt": "Answer directly.", |
| 7010 | "model": "model-1", |
| 7011 | "modelProvider": "ollama", |
| 7012 | "modelProviderId": "ollama", |
| 7013 | "reasoningEffort": "high", |
| 7014 | "allowedTools": [], |
| 7015 | "mode": "chat", |
| 7016 | "requestedMode": "chat", |
| 7017 | "workspace": { |
| 7018 | "id": "workspace-1", |
| 7019 | "targetRef": "target-1" |
| 7020 | } |
| 7021 | }) |
| 7022 | } |
| 7023 | |
| 7024 | #[test] |
| 7025 | fn runtime_chat_parser_requires_exact_route_tools_mode_and_interrupt_scope() { |
| 7026 | let command = parse_remote_command(&runtime_chat_prompt_fixture(), "run-1").unwrap(); |
| 7027 | assert!(matches!(command, RemoteCommand::RuntimeChatPrompt(_))); |
| 7028 | |
| 7029 | for mutation in [ |
| 7030 | ("allowedTools", json!(["bash"])), |
| 7031 | ("mode", json!("work")), |
| 7032 | ("requestedMode", json!("operate")), |
| 7033 | ("modelProvider", json!("https://provider.invalid")), |
| 7034 | ] { |
| 7035 | let mut value = runtime_chat_prompt_fixture(); |
| 7036 | value[mutation.0] = mutation.1; |
| 7037 | assert!(parse_remote_command(&value, "run-1").is_err()); |
| 7038 | } |
| 7039 | let mut unknown = runtime_chat_prompt_fixture(); |
| 7040 | unknown["workspacePath"] = json!("/private/workspace"); |
| 7041 | assert!(parse_remote_command(&unknown, "run-1").is_err()); |
| 7042 | let mut missing_markers = runtime_chat_prompt_fixture(); |
| 7043 | let record = missing_markers.as_object_mut().unwrap(); |
| 7044 | record.remove("operationKey"); |
| 7045 | record.remove("runtimeBindingId"); |
| 7046 | record.remove("runtimeThreadId"); |
| 7047 | assert!( |
| 7048 | parse_remote_command(&missing_markers, "run-1").is_err(), |
| 7049 | "a rich Chat-shaped payload cannot downgrade into a legacy Work prompt" |
| 7050 | ); |
| 7051 | |
| 7052 | let interrupt = json!({ |
| 7053 | "type": "run.control", |
| 7054 | "runId": "run-1", |
| 7055 | "action": "interrupt", |
| 7056 | "reason": "Stop this turn", |
| 7057 | "turnId": format!("local_turn_{}", "b".repeat(24)), |
| 7058 | "runtimeBindingId": "binding-1", |
| 7059 | "runtimeThreadId": format!("local_thread_{}", "a".repeat(24)) |
| 7060 | }); |
| 7061 | assert!(matches!( |
| 7062 | parse_remote_command(&interrupt, "run-1").unwrap(), |
| 7063 | RemoteCommand::Control { |
| 7064 | runtime_chat: Some(_), |
| 7065 | .. |
| 7066 | } |
| 7067 | )); |
| 7068 | let mut drifted = interrupt; |
| 7069 | drifted["runtimeThreadId"] = json!(format!("local_thread_{}", "F".repeat(24))); |
| 7070 | assert!(parse_remote_command(&drifted, "run-1").is_err()); |
| 7071 | } |
| 7072 | |
| 7073 | #[test] |
| 7074 | fn legacy_control_requires_and_matches_the_exact_active_turn() { |
| 7075 | assert!( |
| 7076 | parse_remote_command( |
| 7077 | &json!({ |
| 7078 | "type": "run.control", |
| 7079 | "runId": "run-1", |
| 7080 | "action": "interrupt" |
| 7081 | }), |
| 7082 | "run-1" |
| 7083 | ) |
| 7084 | .is_err(), |
| 7085 | "run-only cancellation can target a newer turn in the same run" |
| 7086 | ); |
| 7087 | assert!( |
| 7088 | parse_remote_command( |
| 7089 | &json!({ |
| 7090 | "type": "run.control", |
| 7091 | "runId": "run-1", |
| 7092 | "action": "interrupt", |
| 7093 | "turnId": "../../stale" |
| 7094 | }), |
| 7095 | "run-1" |
| 7096 | ) |
| 7097 | .is_err() |
| 7098 | ); |
| 7099 | let parsed = parse_remote_command( |
| 7100 | &json!({ |
| 7101 | "type": "run.control", |
| 7102 | "runId": "run-1", |
| 7103 | "action": "interrupt", |
| 7104 | "turnId": "turn-current" |
| 7105 | }), |
| 7106 | "run-1", |
| 7107 | ) |
| 7108 | .unwrap(); |
| 7109 | assert!(matches!( |
| 7110 | parsed, |
| 7111 | RemoteCommand::Control { |
| 7112 | turn_id: Some(ref turn_id), |
| 7113 | runtime_chat: None, |
| 7114 | .. |
| 7115 | } if turn_id == "turn-current" |
| 7116 | )); |
| 7117 | |
| 7118 | let mut controller = RemoteControlController::default(); |
| 7119 | controller.activate_prompt("run-1", "turn-current").unwrap(); |
| 7120 | assert!(controller.active_turn_matches("run-1", "turn-current")); |
| 7121 | assert!(!controller.active_turn_matches("run-1", "turn-stale")); |
| 7122 | assert!(!controller.active_turn_matches("run-other", "turn-current")); |
| 7123 | } |
| 7124 | |
| 7125 | #[test] |
| 7126 | fn runtime_chat_semantic_envelope_carries_stable_native_source_id() { |
| 7127 | let source_event_id = format!("native_event_{}", "a".repeat(64)); |
| 7128 | let semantic = runtime_chat_envelope( |
| 7129 | 9, |
| 7130 | "item.delta", |
| 7131 | Some("local_thread_fixture"), |
| 7132 | Some("local_turn_fixture"), |
| 7133 | Some(&source_event_id), |
| 7134 | "2026-08-23T00:00:00Z".to_string(), |
| 7135 | json!({ "delta": "hello" }), |
| 7136 | ); |
| 7137 | assert_eq!(semantic["schema_version"], 2); |
| 7138 | assert_eq!(semantic["source_event_id"], source_event_id); |
| 7139 | |
| 7140 | let catalog_payload = json!({ |
| 7141 | "schemaVersion": 2, |
| 7142 | "challenge": "a".repeat(32), |
| 7143 | "providers": [{ "id": "safe", "models": [{ "id": "model-1" }] }] |
| 7144 | }); |
| 7145 | let fingerprint = |
| 7146 | RuntimeChatRelayHost::catalog_payload_fingerprint(&catalog_payload).unwrap(); |
| 7147 | let catalog_source_id = runtime_chat_catalog_source_event_id("run_fixture", &fingerprint); |
| 7148 | let catalog = runtime_chat_envelope( |
| 7149 | 1, |
| 7150 | "runtime.catalog", |
| 7151 | None, |
| 7152 | None, |
| 7153 | Some(&catalog_source_id), |
| 7154 | RUNTIME_CHAT_CATALOG_TIMESTAMP.to_string(), |
| 7155 | catalog_payload.clone(), |
| 7156 | ); |
| 7157 | let replay = runtime_chat_envelope( |
| 7158 | 99, |
| 7159 | "runtime.catalog", |
| 7160 | None, |
| 7161 | None, |
| 7162 | Some(&catalog_source_id), |
| 7163 | RUNTIME_CHAT_CATALOG_TIMESTAMP.to_string(), |
| 7164 | catalog_payload, |
| 7165 | ); |
| 7166 | assert_eq!(catalog["source_event_id"], catalog_source_id); |
| 7167 | let mut catalog_semantic = catalog; |
| 7168 | let mut replay_semantic = replay; |
| 7169 | catalog_semantic.as_object_mut().unwrap().remove("seq"); |
| 7170 | replay_semantic.as_object_mut().unwrap().remove("seq"); |
| 7171 | assert_eq!(catalog_semantic, replay_semantic); |
| 7172 | |
| 7173 | let changed_fingerprint = RuntimeChatRelayHost::catalog_payload_fingerprint( |
| 7174 | &json!({ "schemaVersion": 2, "challenge": "a".repeat(32), "providers": [] }), |
| 7175 | ) |
| 7176 | .unwrap(); |
| 7177 | assert_ne!( |
| 7178 | runtime_chat_catalog_source_event_id("run_fixture", &changed_fingerprint), |
| 7179 | catalog_source_id |
| 7180 | ); |
| 7181 | } |
| 7182 | |
| 7183 | #[test] |
| 7184 | fn approval_projection_matches_control_plane_namespace() { |
| 7185 | assert_eq!(projected_approval_id("tool-call-1").len(), 39); |
| 7186 | assert!(projected_approval_id("tool-call-1").starts_with("local_approval_")); |
| 7187 | assert_ne!( |
| 7188 | projected_approval_id("tool-call-1"), |
| 7189 | projected_approval_id("tool-call-2") |
| 7190 | ); |
| 7191 | } |
| 7192 | |
| 7193 | #[test] |
| 7194 | fn authorization_url_is_exact_and_cannot_redirect_or_add_parameters() { |
| 7195 | assert!( |
| 7196 | validate_authorization_url( |
| 7197 | "https://app.codewhale.net/runner/authorize?user_code=ABCD-EFGH-JKLM", |
| 7198 | "ABCD-EFGH-JKLM", |
| 7199 | ) |
| 7200 | .is_ok() |
| 7201 | ); |
| 7202 | for spoofed in [ |
| 7203 | "http://app.codewhale.net/runner/authorize?user_code=ABCD-EFGH-JKLM", |
| 7204 | "https://app.codewhale.net.evil.example/runner/authorize?user_code=ABCD-EFGH-JKLM", |
| 7205 | "https://app.codewhale.net/runner/authorize?user_code=ABCD-EFGH-JKLM&next=https://evil.example", |
| 7206 | "https://app.codewhale.net/runner/authorize?user_code=WRONG-CODE", |
| 7207 | ] { |
| 7208 | assert!(validate_authorization_url(spoofed, "ABCD-EFGH-JKLM").is_err()); |
| 7209 | } |
| 7210 | } |
| 7211 | |
| 7212 | #[test] |
| 7213 | fn command_sequences_are_content_bound_and_replay_safe() { |
| 7214 | let mut controller = RemoteControlController::default(); |
| 7215 | controller.status = Status::Connected; |
| 7216 | controller.attached_run_id = Some("run-1".to_string()); |
| 7217 | let prompt = RemoteCommand::Prompt { |
| 7218 | turn_id: "turn-1".to_string(), |
| 7219 | prompt: "Continue".to_string(), |
| 7220 | }; |
| 7221 | assert_eq!(controller.claim_command("run-1", 1, &prompt), Ok(true)); |
| 7222 | assert_eq!(controller.claim_command("run-1", 1, &prompt), Ok(false)); |
| 7223 | assert!( |
| 7224 | controller |
| 7225 | .claim_command( |
| 7226 | "run-1", |
| 7227 | 1, |
| 7228 | &RemoteCommand::Prompt { |
| 7229 | turn_id: "turn-1".to_string(), |
| 7230 | prompt: "Changed".to_string(), |
| 7231 | }, |
| 7232 | ) |
| 7233 | .is_err() |
| 7234 | ); |
| 7235 | } |
| 7236 | |
| 7237 | #[tokio::test] |
| 7238 | async fn durable_runtime_chat_replay_bypasses_only_new_work_recovery_gates() { |
| 7239 | let root = tempfile::tempdir().unwrap(); |
| 7240 | let runtime_root = root.path().join("runtime-chat"); |
| 7241 | let command = parse_remote_command(&runtime_chat_prompt_fixture(), "run-1").unwrap(); |
| 7242 | let RemoteCommand::RuntimeChatPrompt(prompt) = &command else { |
| 7243 | panic!("fixture must parse as Runtime Chat"); |
| 7244 | }; |
| 7245 | { |
| 7246 | let mut first = RemoteControlController::default(); |
| 7247 | first |
| 7248 | .configure_runtime_chat( |
| 7249 | crate::config::Config::default(), |
| 7250 | Arc::new(crate::plugins::PluginRegistry::empty(root.path())), |
| 7251 | runtime_root.clone(), |
| 7252 | "target-1".to_string(), |
| 7253 | "session_fixture".to_string(), |
| 7254 | ) |
| 7255 | .unwrap(); |
| 7256 | let host = first.runtime_chat.as_ref().unwrap(); |
| 7257 | host.bind_account("account_fixture", "target-1").unwrap(); |
| 7258 | host.install_prompt_replay_for_tests(prompt, true) |
| 7259 | .await |
| 7260 | .unwrap(); |
| 7261 | } |
| 7262 | |
| 7263 | let mut controller = RemoteControlController::default(); |
| 7264 | controller |
| 7265 | .configure_runtime_chat( |
| 7266 | crate::config::Config::default(), |
| 7267 | Arc::new(crate::plugins::PluginRegistry::empty(root.path())), |
| 7268 | runtime_root.clone(), |
| 7269 | "target-1".to_string(), |
| 7270 | "session_fixture".to_string(), |
| 7271 | ) |
| 7272 | .unwrap(); |
| 7273 | controller |
| 7274 | .runtime_chat |
| 7275 | .as_ref() |
| 7276 | .unwrap() |
| 7277 | .bind_account("account_fixture", "target-1") |
| 7278 | .unwrap(); |
| 7279 | controller |
| 7280 | .runtime_chat |
| 7281 | .as_ref() |
| 7282 | .unwrap() |
| 7283 | .authorize_run("run-1") |
| 7284 | .unwrap(); |
| 7285 | let envelope = runtime_chat_envelope( |
| 7286 | 9, |
| 7287 | "turn.completed", |
| 7288 | Some(&prompt.runtime_thread_id), |
| 7289 | Some(&prompt.turn_id), |
| 7290 | Some("native_event_exact_replay_gate"), |
| 7291 | "2026-08-23T00:00:00Z".to_string(), |
| 7292 | json!({ "turn": { "status": "completed" } }), |
| 7293 | ); |
| 7294 | controller.pending_runtime_events.insert( |
| 7295 | "run-1".to_string(), |
| 7296 | BTreeMap::from([( |
| 7297 | 9, |
| 7298 | PendingRuntimeEnvelope { |
| 7299 | encoded_len: serde_json::to_vec(&envelope).unwrap().len(), |
| 7300 | envelope, |
| 7301 | integrity: true, |
| 7302 | handed_off: true, |
| 7303 | }, |
| 7304 | )]), |
| 7305 | ); |
| 7306 | let changed_config = crate::config::Config { |
| 7307 | default_text_model: Some("changed-after-restart".to_string()), |
| 7308 | ..crate::config::Config::default() |
| 7309 | }; |
| 7310 | controller |
| 7311 | .configure_runtime_chat( |
| 7312 | changed_config, |
| 7313 | Arc::new(crate::plugins::PluginRegistry::empty(root.path())), |
| 7314 | runtime_root, |
| 7315 | "target-1".to_string(), |
| 7316 | "session_fixture".to_string(), |
| 7317 | ) |
| 7318 | .unwrap(); |
| 7319 | assert!(controller.pending_runtime_chat_configuration.is_some()); |
| 7320 | controller.status = Status::Connected; |
| 7321 | controller.attached_run_id = Some("run-1".to_string()); |
| 7322 | controller.attached_workspace_id = Some(prompt.workspace.id.clone()); |
| 7323 | controller.target_ref = Some(prompt.workspace.target_ref.clone()); |
| 7324 | |
| 7325 | assert_eq!(controller.claim_command("run-1", 17, &command), Ok(false)); |
| 7326 | controller.apply_runtime_chat_prompt(prompt).await.unwrap(); |
| 7327 | let mut changed = runtime_chat_prompt_fixture(); |
| 7328 | changed["prompt"] = json!("changed replay body"); |
| 7329 | let changed = parse_remote_command(&changed, "run-1").unwrap(); |
| 7330 | assert!(controller.claim_command("run-1", 17, &changed).is_err()); |
| 7331 | } |
| 7332 | |
| 7333 | #[tokio::test] |
| 7334 | async fn pre_lease_failure_never_locks_and_allows_immediate_retry() { |
| 7335 | let (mut controller, _worker_rx, event_tx, _journal_root) = wired_controller(); |
| 7336 | controller.status = Status::Connecting; |
| 7337 | event_tx |
| 7338 | .send(RemoteEvent::FailedPreLease( |
| 7339 | "Codewhale rejected remote-control enrollment (403): client version not accepted." |
| 7340 | .to_string(), |
| 7341 | )) |
| 7342 | .unwrap(); |
| 7343 | assert!(matches!( |
| 7344 | controller.try_next_event(), |
| 7345 | Some(RemoteEvent::FailedPreLease(_)) |
| 7346 | )); |
| 7347 | assert_eq!(controller.status, Status::Failed); |
| 7348 | assert!( |
| 7349 | controller.ownership_blocked_until.is_none(), |
| 7350 | "a rejection before any lease must never start the reconnect blackout" |
| 7351 | ); |
| 7352 | let line = controller.status_line(); |
| 7353 | assert!(line.contains("failed before connecting"), "{line}"); |
| 7354 | assert!(line.contains("/rc to retry"), "{line}"); |
| 7355 | assert!( |
| 7356 | line.contains("403"), |
| 7357 | "the sanitized HTTP status must surface: {line}" |
| 7358 | ); |
| 7359 | assert!(line.contains("client version not accepted"), "{line}"); |
| 7360 | |
| 7361 | // Stopping after a pre-lease failure is an ordinary reset (no lease |
| 7362 | // to drain, no blackout to honor). |
| 7363 | controller.stop(); |
| 7364 | assert_eq!(controller.status, Status::Off); |
| 7365 | |
| 7366 | // Immediate retry is allowed — no lease drain wait. |
| 7367 | let result = controller.start(RemoteStart { |
| 7368 | workspace_label: "fixture".to_string(), |
| 7369 | target_ref: "target_fixture".to_string(), |
| 7370 | session_id: "session_fixture".to_string(), |
| 7371 | runtime_version: "0.9.1".to_string(), |
| 7372 | runtime_commit: "a".repeat(40), |
| 7373 | journal_dir: None, |
| 7374 | git_remote: None, |
| 7375 | }); |
| 7376 | assert!(result.is_ok(), "{result:?}"); |
| 7377 | } |
| 7378 | |
| 7379 | #[test] |
| 7380 | fn sanitized_rejection_excerpt_reads_only_bounded_error_fields() { |
| 7381 | assert_eq!( |
| 7382 | sanitized_rejection_excerpt( |
| 7383 | br#"{"error":"client version not accepted","details":"noise"}"# |
| 7384 | ) |
| 7385 | .as_deref(), |
| 7386 | Some("client version not accepted") |
| 7387 | ); |
| 7388 | assert_eq!( |
| 7389 | sanitized_rejection_excerpt(br#"{"message":"enrollment closed"}"#).as_deref(), |
| 7390 | Some("enrollment closed") |
| 7391 | ); |
| 7392 | // Control characters are stripped, never echoed. A JSON-escaped NUL |
| 7393 | // parses into the value; the strip must remove it. A raw NUL byte |
| 7394 | // makes serde_json reject the body outright, which is also safe |
| 7395 | // (no excerpt) — assert both directions. |
| 7396 | let with_control = "{\"error\":\"bad\\u0000opaque\"}".to_string(); |
| 7397 | assert_eq!( |
| 7398 | sanitized_rejection_excerpt(with_control.as_bytes()).as_deref(), |
| 7399 | Some("badopaque") |
| 7400 | ); |
| 7401 | let raw_nul = "{\"error\":\"bad\u{0}opaque\"}".to_string(); |
| 7402 | assert_eq!(sanitized_rejection_excerpt(raw_nul.as_bytes()), None); |
| 7403 | // Non-JSON bodies yield no excerpt. |
| 7404 | assert_eq!(sanitized_rejection_excerpt(b"<html>403</html>"), None); |
| 7405 | // Overlong reasons are capped. |
| 7406 | let long = format!("{{\"error\":\"{}\"}}", "x".repeat(400)); |
| 7407 | let excerpt = sanitized_rejection_excerpt(long.as_bytes()).expect("capped excerpt"); |
| 7408 | assert!(excerpt.chars().count() <= 140); |
| 7409 | } |
| 7410 | |
| 7411 | #[test] |
| 7412 | fn view_gate_matching_never_confuses_two_approval_cards() { |
| 7413 | use crate::tui::approval::ApprovalRequest; |
| 7414 | use crate::tui::approval::ApprovalView; |
| 7415 | use crate::tui::views::ViewStack; |
| 7416 | |
| 7417 | let request = ApprovalRequest::new( |
| 7418 | "tool_A", |
| 7419 | "edit", |
| 7420 | "Edit A", |
| 7421 | &serde_json::json!({ "file": "a" }), |
| 7422 | "approval_key_A", |
| 7423 | ); |
| 7424 | let card = ApprovalView::new(request); |
| 7425 | let gate_a = projected_approval_id("tool_A"); |
| 7426 | let gate_b = projected_approval_id("tool_B"); |
| 7427 | |
| 7428 | let mut stack = ViewStack::new(); |
| 7429 | stack.push(card); |
| 7430 | assert!( |
| 7431 | stack.top_matches_approval_gate(&gate_a), |
| 7432 | "the matching gate must match" |
| 7433 | ); |
| 7434 | assert!( |
| 7435 | !stack.top_matches_approval_gate(&gate_b), |
| 7436 | "a different gate must NEVER match this card — the whole point of identity-aware dismissal" |
| 7437 | ); |
| 7438 | assert!(!stack.top_matches_approval_gate("local_approval_missing")); |
| 7439 | } |
| 7440 | |
| 7441 | #[tokio::test] |
| 7442 | async fn enrollment_rejection_carries_a_sanitized_actionable_reason() { |
| 7443 | let server = MockServer::start().await; |
| 7444 | Mock::given(method("POST")) |
| 7445 | .and(path("/oauth/device")) |
| 7446 | .respond_with(ResponseTemplate::new(403).set_body_json(json!({ |
| 7447 | "error": "client version not accepted", |
| 7448 | "documentation": "https://example.test/docs" |
| 7449 | }))) |
| 7450 | .mount(&server) |
| 7451 | .await; |
| 7452 | let client = crate::tls::reqwest_client_builder() |
| 7453 | .https_only(false) |
| 7454 | .redirect(reqwest::redirect::Policy::none()) |
| 7455 | .timeout(Duration::from_secs(5)) |
| 7456 | .build() |
| 7457 | .expect("test client"); |
| 7458 | let error = public_request( |
| 7459 | &client, |
| 7460 | Method::POST, |
| 7461 | Url::parse(&format!( |
| 7462 | "{}/oauth/device", |
| 7463 | server.uri().trim_end_matches('/') |
| 7464 | )) |
| 7465 | .expect("server url"), |
| 7466 | json!({ "audience": "codewhale-runner" }), |
| 7467 | ) |
| 7468 | .await |
| 7469 | .expect_err("403 must fail"); |
| 7470 | assert!(error.contains("403"), "{error}"); |
| 7471 | assert!(error.contains("client version not accepted"), "{error}"); |
| 7472 | assert!( |
| 7473 | !error.contains("documentation"), |
| 7474 | "only the conventional error fields may surface: {error}" |
| 7475 | ); |
| 7476 | } |
| 7477 | |
| 7478 | #[tokio::test] |
| 7479 | async fn failed_relay_keeps_reconnect_blocked_until_lease_expiry() { |
| 7480 | let mut controller = RemoteControlController::default(); |
| 7481 | controller.status = Status::Failed; |
| 7482 | controller.ownership_blocked_until = Some(Instant::now() + Duration::from_secs(90)); |
| 7483 | // Mirror semantics: local input is never locked, but reconnecting |
| 7484 | // while the server lease may still be live is refused — the web must |
| 7485 | // not see two runners for the same session. |
| 7486 | let start = RemoteStart { |
| 7487 | workspace_label: "fixture".to_string(), |
| 7488 | target_ref: "target_fixture".to_string(), |
| 7489 | session_id: "session_fixture".to_string(), |
| 7490 | runtime_version: "0.9.1".to_string(), |
| 7491 | runtime_commit: "a".repeat(40), |
| 7492 | journal_dir: None, |
| 7493 | git_remote: None, |
| 7494 | }; |
| 7495 | assert!(controller.start(start.clone()).is_err()); |
| 7496 | // The web cannot answer shared approvals while the relay is failed. |
| 7497 | assert!(!controller.can_share_approval_with_web()); |
| 7498 | controller.ownership_blocked_until = Some(Instant::now() - Duration::from_secs(1)); |
| 7499 | assert!(controller.start(start).is_ok()); |
| 7500 | } |
| 7501 | |
| 7502 | #[test] |
| 7503 | fn stop_after_lease_expiry_preserves_pending_approvals_for_restoration() { |
| 7504 | let mut controller = RemoteControlController::default(); |
| 7505 | controller.status = Status::Failed; |
| 7506 | controller.ownership_blocked_until = Some(Instant::now() - Duration::from_secs(1)); |
| 7507 | controller.pending_approvals.insert( |
| 7508 | "approval_fixture".to_string(), |
| 7509 | PendingRemoteApproval { |
| 7510 | tool_id: "tool_fixture".to_string(), |
| 7511 | }, |
| 7512 | ); |
| 7513 | |
| 7514 | controller.stop(); |
| 7515 | assert_eq!(controller.status, Status::Failed); |
| 7516 | assert_eq!(controller.pending_approvals.len(), 1); |
| 7517 | |
| 7518 | let event = controller.try_next_event(); |
| 7519 | assert!(matches!( |
| 7520 | event, |
| 7521 | Some(RemoteEvent::OwnershipRestored { approvals }) |
| 7522 | if approvals.len() == 1 && approvals[0].tool_id == "tool_fixture" |
| 7523 | )); |
| 7524 | assert_eq!(controller.status, Status::Off); |
| 7525 | assert!(controller.pending_approvals.is_empty()); |
| 7526 | } |
| 7527 | |
| 7528 | #[test] |
| 7529 | fn cancelling_a_connect_keeps_reconnect_blocked_until_lease_drain() { |
| 7530 | let mut controller = RemoteControlController::default(); |
| 7531 | controller.status = Status::Connecting; |
| 7532 | controller.stop(); |
| 7533 | |
| 7534 | assert_eq!(controller.status, Status::Failed); |
| 7535 | // Mirror semantics: nothing about a cancelled connect locks local |
| 7536 | // input; reconnect stays blocked until the possible lease drains. |
| 7537 | let result = controller.start(RemoteStart { |
| 7538 | workspace_label: "fixture".to_string(), |
| 7539 | target_ref: "target_fixture".to_string(), |
| 7540 | session_id: "session_fixture".to_string(), |
| 7541 | runtime_version: "0.9.1".to_string(), |
| 7542 | runtime_commit: "a".repeat(40), |
| 7543 | journal_dir: None, |
| 7544 | git_remote: None, |
| 7545 | }); |
| 7546 | assert!(result.is_err()); |
| 7547 | assert!(result.unwrap_err().contains("previous remote lease")); |
| 7548 | } |
| 7549 | |
| 7550 | #[test] |
| 7551 | fn failed_worker_retains_snapshot_marker_and_exact_unacked_event() { |
| 7552 | let (mut controller, mut worker_rx, event_tx, _journal_root) = wired_controller(); |
| 7553 | controller.status = Status::Connected; |
| 7554 | controller.upload_snapshot("run-1", &[]); |
| 7555 | let WorkerCommand::Upload { envelopes, .. } = worker_rx.try_recv().unwrap() else { |
| 7556 | panic!("snapshot queued"); |
| 7557 | }; |
| 7558 | let exact = envelopes[0].clone(); |
| 7559 | event_tx |
| 7560 | .send(RemoteEvent::Failed("fixture disconnect".to_string())) |
| 7561 | .unwrap(); |
| 7562 | |
| 7563 | assert!(matches!( |
| 7564 | controller.try_next_event(), |
| 7565 | Some(RemoteEvent::Failed(_)) |
| 7566 | )); |
| 7567 | assert!(controller.uploaded_snapshots.contains("run-1")); |
| 7568 | assert_eq!( |
| 7569 | controller.pending_runtime_events["run-1"] |
| 7570 | .values() |
| 7571 | .next() |
| 7572 | .map(|entry| &entry.envelope), |
| 7573 | Some(&exact) |
| 7574 | ); |
| 7575 | // Fail-closed: the web can no longer answer shared approvals, and |
| 7576 | // reconnecting waits out the possible server lease. |
| 7577 | assert!(!controller.can_share_approval_with_web()); |
| 7578 | assert!(controller.ownership_blocked_until.is_some()); |
| 7579 | } |
| 7580 | |
| 7581 | #[tokio::test] |
| 7582 | async fn cwc_runner_wire_contract_preserves_pending_and_recovery_commands() { |
| 7583 | let server = MockServer::start().await; |
| 7584 | Mock::given(method("GET")) |
| 7585 | .and(path("/api/local-runners/runner-1/runs/run-1/commands")) |
| 7586 | .and(query_param("since_seq", "0")) |
| 7587 | .and(query_param("include_accepted", "1")) |
| 7588 | .respond_with(ResponseTemplate::new(200).set_body_json(json!({ |
| 7589 | "commands": [{ |
| 7590 | "seq": 1, |
| 7591 | "deliveryStatus": "pending", |
| 7592 | "ackStatus": "", |
| 7593 | "command": { |
| 7594 | "type": "prompt.request", |
| 7595 | "runId": "run-1", |
| 7596 | "turnId": "turn-1", |
| 7597 | "prompt": "Continue from the web." |
| 7598 | } |
| 7599 | }, { |
| 7600 | "seq": 2, |
| 7601 | "deliveryStatus": "acknowledged", |
| 7602 | "ackStatus": "accepted", |
| 7603 | "command": { |
| 7604 | "type": "run.control", |
| 7605 | "runId": "run-1", |
| 7606 | "action": "interrupt" |
| 7607 | } |
| 7608 | }] |
| 7609 | }))) |
| 7610 | .expect(1) |
| 7611 | .mount(&server) |
| 7612 | .await; |
| 7613 | Mock::given(method("POST")) |
| 7614 | .and(path("/api/local-runners/runner-1/runs/run-1/events")) |
| 7615 | .and(body_json(json!({ |
| 7616 | "acknowledgements": [{ |
| 7617 | "commandSeq": 1, |
| 7618 | "commandType": "prompt.request", |
| 7619 | "status": "accepted", |
| 7620 | "turnId": "turn-1" |
| 7621 | }], |
| 7622 | "envelopes": [] |
| 7623 | }))) |
| 7624 | .respond_with(ResponseTemplate::new(200).set_body_json(json!({ |
| 7625 | "accepted": [], |
| 7626 | "count": 1, |
| 7627 | "cursor": 0 |
| 7628 | }))) |
| 7629 | .expect(1) |
| 7630 | .mount(&server) |
| 7631 | .await; |
| 7632 | |
| 7633 | let enrollment = LiveEnrollment { |
| 7634 | persisted: PersistedEnrollment { |
| 7635 | schema_version: 1, |
| 7636 | control_plane_base: format!("{}/", server.uri()), |
| 7637 | runner_enrollment_id: "enrollment-1".to_string(), |
| 7638 | account_ref: "account-1".to_string(), |
| 7639 | device_id: "device-1".to_string(), |
| 7640 | target_ref: "target-1".to_string(), |
| 7641 | target_grant_ref: "grant-1".to_string(), |
| 7642 | runtime_version: "0.9.1".to_string(), |
| 7643 | runtime_commit: "a".repeat(40), |
| 7644 | bootstrap_secret: "b".repeat(43), |
| 7645 | }, |
| 7646 | access_token: "fixture-runner-access-token".to_string(), |
| 7647 | }; |
| 7648 | let client = crate::tls::reqwest_client_builder() |
| 7649 | .redirect(reqwest::redirect::Policy::none()) |
| 7650 | .build() |
| 7651 | .expect("fixture client"); |
| 7652 | |
| 7653 | let listed = list_commands(&client, &enrollment, "runner-1", "run-1", 0) |
| 7654 | .await |
| 7655 | .expect("CWC command list"); |
| 7656 | assert_eq!(listed.len(), 2); |
| 7657 | assert_eq!(listed[0].ack_status, ""); |
| 7658 | assert_eq!(listed[1].ack_status, "accepted"); |
| 7659 | let prompt = |
| 7660 | parse_remote_command(&listed[0].command, "run-1").expect("typed prompt command"); |
| 7661 | upload_command_accepted( |
| 7662 | &client, |
| 7663 | &enrollment, |
| 7664 | "runner-1", |
| 7665 | "run-1", |
| 7666 | listed[0].seq, |
| 7667 | &prompt, |
| 7668 | ) |
| 7669 | .await |
| 7670 | .expect("durable accepted acknowledgement"); |
| 7671 | } |
| 7672 | |
| 7673 | fn wired_controller() -> ( |
| 7674 | RemoteControlController, |
| 7675 | mpsc::UnboundedReceiver<WorkerCommand>, |
| 7676 | mpsc::UnboundedSender<RemoteEvent>, |
| 7677 | tempfile::TempDir, |
| 7678 | ) { |
| 7679 | let mut controller = RemoteControlController::default(); |
| 7680 | let (worker_tx, worker_rx) = mpsc::unbounded_channel(); |
| 7681 | let (event_tx, event_rx) = mpsc::unbounded_channel(); |
| 7682 | let journal_root = tempfile::tempdir().expect("wired controller journal root"); |
| 7683 | controller.journal = Some( |
| 7684 | RuntimeEventJournal::open( |
| 7685 | journal_root.path(), |
| 7686 | "target_wired_fixture", |
| 7687 | "session_wired_fixture", |
| 7688 | ) |
| 7689 | .expect("wired controller journal"), |
| 7690 | ); |
| 7691 | controller.worker_tx = Some(worker_tx); |
| 7692 | controller.event_rx = Some(event_rx); |
| 7693 | (controller, worker_rx, event_tx, journal_root) |
| 7694 | } |
| 7695 | |
| 7696 | fn turn_complete_event() -> EngineEvent { |
| 7697 | EngineEvent::TurnComplete { |
| 7698 | usage: codewhale_models::Usage::default(), |
| 7699 | parent_route_usage: codewhale_models::Usage::default(), |
| 7700 | routed_usage_dropped_records: 0, |
| 7701 | status: TurnOutcomeStatus::Completed, |
| 7702 | error: None, |
| 7703 | tool_catalog: None, |
| 7704 | base_url: None, |
| 7705 | } |
| 7706 | } |
| 7707 | |
| 7708 | #[test] |
| 7709 | fn stop_refusal_holds_until_terminal_event_is_acknowledged() { |
| 7710 | let (mut controller, mut worker_rx, event_tx, _journal_root) = wired_controller(); |
| 7711 | controller |
| 7712 | .activate_prompt("run_fixture", "turn_fixture") |
| 7713 | .unwrap(); |
| 7714 | let refusal = controller.stop_refusal().expect("active turn blocks stop"); |
| 7715 | assert!(refusal.contains("active remote turn"), "{refusal}"); |
| 7716 | |
| 7717 | controller.observe_engine_event(&turn_complete_event()); |
| 7718 | assert!( |
| 7719 | !controller.has_active_run(), |
| 7720 | "the terminal event releases the run binding" |
| 7721 | ); |
| 7722 | let refusal = controller |
| 7723 | .stop_refusal() |
| 7724 | .expect("a queued but unacknowledged terminal event must still block stop"); |
| 7725 | assert!(refusal.contains("acknowledged"), "{refusal}"); |
| 7726 | let WorkerCommand::Upload { envelopes, .. } = worker_rx.try_recv().unwrap() else { |
| 7727 | panic!("the terminal envelope must be handed to the transport"); |
| 7728 | }; |
| 7729 | assert_eq!(envelopes[0]["event"], "turn.completed"); |
| 7730 | let seq = envelopes[0]["seq"].as_u64().expect("terminal seq"); |
| 7731 | |
| 7732 | event_tx |
| 7733 | .send(RemoteEvent::RuntimeCursor { |
| 7734 | run_id: "run_fixture".to_string(), |
| 7735 | cursor: seq, |
| 7736 | }) |
| 7737 | .unwrap(); |
| 7738 | controller.try_next_event().unwrap(); |
| 7739 | assert_eq!( |
| 7740 | controller.stop_refusal(), |
| 7741 | None, |
| 7742 | "a server-acknowledged terminal event unblocks stop" |
| 7743 | ); |
| 7744 | } |
| 7745 | |
| 7746 | #[test] |
| 7747 | fn different_run_attachment_is_quarantined_during_active_classic_work() { |
| 7748 | let (mut controller, _worker_rx, event_tx, _journal_root) = wired_controller(); |
| 7749 | let attachment = |run_id: &str| RemoteAttachment { |
| 7750 | run_id: run_id.to_string(), |
| 7751 | workspace_id: "workspace_fixture".to_string(), |
| 7752 | runtime_cursor: 0, |
| 7753 | snapshot_present: false, |
| 7754 | runtime_chat_relay_protocol: RUNTIME_CHAT_RELAY_PROTOCOL.to_string(), |
| 7755 | runtime_chat_relay_challenge: "a".repeat(32), |
| 7756 | }; |
| 7757 | event_tx |
| 7758 | .send(RemoteEvent::Connected { |
| 7759 | account_ref: "account_fixture".to_string(), |
| 7760 | runner_id: "runner_fixture".to_string(), |
| 7761 | target_ref: "target_wired_fixture".to_string(), |
| 7762 | attachment: attachment("run_a"), |
| 7763 | links: RemoteLinks::default(), |
| 7764 | }) |
| 7765 | .unwrap(); |
| 7766 | controller.try_next_event().unwrap(); |
| 7767 | controller.activate_prompt("run_a", "turn_a").unwrap(); |
| 7768 | |
| 7769 | event_tx |
| 7770 | .send(RemoteEvent::Attachment { |
| 7771 | account_ref: "account_fixture".to_string(), |
| 7772 | target_ref: "target_wired_fixture".to_string(), |
| 7773 | attachment: attachment("run_b"), |
| 7774 | links: RemoteLinks::default(), |
| 7775 | }) |
| 7776 | .unwrap(); |
| 7777 | let rejected = controller |
| 7778 | .try_next_event() |
| 7779 | .expect("rejected attachment event"); |
| 7780 | assert!(matches!(rejected, RemoteEvent::Failed(_))); |
| 7781 | assert_eq!(controller.attached_run_id.as_deref(), Some("run_a")); |
| 7782 | assert!(controller.active_turn_matches("run_a", "turn_a")); |
| 7783 | assert_eq!(controller.status, Status::Failed); |
| 7784 | assert!( |
| 7785 | controller.worker_tx.is_none(), |
| 7786 | "rejected authority is quarantined" |
| 7787 | ); |
| 7788 | assert!( |
| 7789 | controller |
| 7790 | .claim_command( |
| 7791 | "run_b", |
| 7792 | 1, |
| 7793 | &RemoteCommand::Prompt { |
| 7794 | turn_id: "turn_b".to_string(), |
| 7795 | prompt: "must not run".to_string(), |
| 7796 | }, |
| 7797 | ) |
| 7798 | .is_err(), |
| 7799 | "no command from the rejected run may execute" |
| 7800 | ); |
| 7801 | } |
| 7802 | |
| 7803 | #[test] |
| 7804 | fn failed_stop_stays_fail_closed_with_no_dual_ownership() { |
| 7805 | let (mut controller, _worker_rx, event_tx, _journal_root) = wired_controller(); |
| 7806 | controller.status = Status::Connected; |
| 7807 | controller.stop(); |
| 7808 | assert_eq!(controller.status, Status::Stopping); |
| 7809 | assert!( |
| 7810 | !controller.can_share_approval_with_web(), |
| 7811 | "stopping must close the shared-decision channel before confirmation" |
| 7812 | ); |
| 7813 | |
| 7814 | // The worker could not confirm the drain or the offline heartbeat. |
| 7815 | event_tx |
| 7816 | .send(RemoteEvent::Failed( |
| 7817 | "the offline heartbeat could not be delivered".to_string(), |
| 7818 | )) |
| 7819 | .unwrap(); |
| 7820 | let event = controller.try_next_event().unwrap(); |
| 7821 | assert!(matches!(event, RemoteEvent::Failed(_))); |
| 7822 | assert!( |
| 7823 | !controller.can_share_approval_with_web(), |
| 7824 | "an unconfirmed stop must stay fail-closed through the lease expiry" |
| 7825 | ); |
| 7826 | assert!(controller.ownership_blocked_until.is_some()); |
| 7827 | assert!( |
| 7828 | controller.status_line().contains("lost after connecting"), |
| 7829 | "{}", |
| 7830 | controller.status_line() |
| 7831 | ); |
| 7832 | assert!( |
| 7833 | controller.try_next_event().is_none(), |
| 7834 | "ownership must not be restored while the lease could still be live" |
| 7835 | ); |
| 7836 | } |
| 7837 | |
| 7838 | #[tokio::test] |
| 7839 | async fn stop_drain_flushes_runtime_outbox_with_byte_identical_retries() { |
| 7840 | let server = MockServer::start().await; |
| 7841 | let responder = AmbiguousRuntimeResponder::default(); |
| 7842 | Mock::given(method("POST")) |
| 7843 | .and(path( |
| 7844 | "/api/local-runners/runner_fixture/runs/run_fixture/events", |
| 7845 | )) |
| 7846 | .respond_with(responder.clone()) |
| 7847 | .expect(2) |
| 7848 | .mount(&server) |
| 7849 | .await; |
| 7850 | let mut enrollment = fixture_enrollment(&format!("{}/", server.uri())); |
| 7851 | let mut runner_id = "runner_fixture".to_string(); |
| 7852 | let client = crate::tls::reqwest_client_builder() |
| 7853 | .redirect(reqwest::redirect::Policy::none()) |
| 7854 | .build() |
| 7855 | .expect("fixture client"); |
| 7856 | let (event_tx, mut event_rx) = mpsc::unbounded_channel(); |
| 7857 | let mut outbox = RuntimeTransportOutbox::default(); |
| 7858 | outbox |
| 7859 | .enqueue( |
| 7860 | "run_fixture", |
| 7861 | runtime_envelope( |
| 7862 | 1, |
| 7863 | "turn.completed", |
| 7864 | Some("turn_fixture"), |
| 7865 | "2026-08-08T12:00:00Z".to_string(), |
| 7866 | json!({ "turn": { "status": "completed", "usage": {} } }), |
| 7867 | ), |
| 7868 | ) |
| 7869 | .expect("queue terminal envelope"); |
| 7870 | |
| 7871 | drain_runtime_outbox_for_stop( |
| 7872 | &client, |
| 7873 | &mut enrollment, |
| 7874 | &mut runner_id, |
| 7875 | &fixture_start(), |
| 7876 | &event_tx, |
| 7877 | &mut outbox, |
| 7878 | Instant::now() + Duration::from_secs(10), |
| 7879 | ) |
| 7880 | .await |
| 7881 | .expect("the drain must complete before stop is confirmed"); |
| 7882 | |
| 7883 | assert!(outbox.events.is_empty(), "the outbox must drain fully"); |
| 7884 | let RemoteEvent::RuntimeCursor { run_id, cursor } = event_rx |
| 7885 | .try_recv() |
| 7886 | .expect("cursor event for journal compaction") |
| 7887 | else { |
| 7888 | panic!("drain must surface the server cursor"); |
| 7889 | }; |
| 7890 | assert_eq!(run_id, "run_fixture"); |
| 7891 | assert_eq!(cursor, 1); |
| 7892 | let bodies = responder.bodies.lock().unwrap(); |
| 7893 | assert_eq!(bodies.len(), 2, "ambiguous response must be retried"); |
| 7894 | assert_eq!(bodies[0], bodies[1], "retries must be byte-identical"); |
| 7895 | } |
| 7896 | |
| 7897 | #[tokio::test] |
| 7898 | async fn stop_drain_deadline_failure_refuses_to_confirm_stop() { |
| 7899 | let server = MockServer::start().await; |
| 7900 | Mock::given(method("POST")) |
| 7901 | .and(path( |
| 7902 | "/api/local-runners/runner_fixture/runs/run_fixture/events", |
| 7903 | )) |
| 7904 | .respond_with(ResponseTemplate::new(500)) |
| 7905 | .mount(&server) |
| 7906 | .await; |
| 7907 | let mut enrollment = fixture_enrollment(&format!("{}/", server.uri())); |
| 7908 | let mut runner_id = "runner_fixture".to_string(); |
| 7909 | let client = crate::tls::reqwest_client_builder() |
| 7910 | .redirect(reqwest::redirect::Policy::none()) |
| 7911 | .build() |
| 7912 | .expect("fixture client"); |
| 7913 | let (event_tx, _event_rx) = mpsc::unbounded_channel(); |
| 7914 | let mut outbox = RuntimeTransportOutbox::default(); |
| 7915 | outbox |
| 7916 | .enqueue( |
| 7917 | "run_fixture", |
| 7918 | runtime_envelope( |
| 7919 | 1, |
| 7920 | "turn.completed", |
| 7921 | Some("turn_fixture"), |
| 7922 | "2026-08-08T12:00:00Z".to_string(), |
| 7923 | json!({ "turn": { "status": "completed", "usage": {} } }), |
| 7924 | ), |
| 7925 | ) |
| 7926 | .expect("queue terminal envelope"); |
| 7927 | |
| 7928 | let error = drain_runtime_outbox_for_stop( |
| 7929 | &client, |
| 7930 | &mut enrollment, |
| 7931 | &mut runner_id, |
| 7932 | &fixture_start(), |
| 7933 | &event_tx, |
| 7934 | &mut outbox, |
| 7935 | Instant::now() + Duration::from_millis(700), |
| 7936 | ) |
| 7937 | .await |
| 7938 | .expect_err("an undrained outbox must fail the stop"); |
| 7939 | assert!(error.contains("not confirmed"), "{error}"); |
| 7940 | assert!( |
| 7941 | !outbox.events.is_empty(), |
| 7942 | "the exact unacknowledged envelope must be retained for the reconnect resend" |
| 7943 | ); |
| 7944 | } |
| 7945 | |
| 7946 | #[test] |
| 7947 | fn journal_roundtrip_restores_unacknowledged_envelopes_byte_identically() { |
| 7948 | let dir = tempfile::tempdir().expect("journal tempdir"); |
| 7949 | let journal = |
| 7950 | RuntimeEventJournal::open(dir.path(), "target:fixture@01", "session:fixture@01") |
| 7951 | .expect("journal setup"); |
| 7952 | assert!(journal.load().expect("missing file is empty").is_empty()); |
| 7953 | |
| 7954 | let delta = runtime_envelope( |
| 7955 | 1, |
| 7956 | "item.delta", |
| 7957 | Some("turn_fixture"), |
| 7958 | "2026-08-08T12:00:00Z".to_string(), |
| 7959 | json!({ "kind": "agent_message", "delta": "exact 🫧 body" }), |
| 7960 | ); |
| 7961 | let terminal = runtime_envelope( |
| 7962 | 2, |
| 7963 | "turn.completed", |
| 7964 | Some("turn_fixture"), |
| 7965 | "2026-08-08T12:00:01Z".to_string(), |
| 7966 | json!({ "turn": { "status": "completed", "usage": {} } }), |
| 7967 | ); |
| 7968 | let mut pending: HashMap<String, BTreeMap<u64, PendingRuntimeEnvelope>> = HashMap::new(); |
| 7969 | let mut events = BTreeMap::new(); |
| 7970 | for envelope in [delta.clone(), terminal.clone()] { |
| 7971 | let seq = runtime_envelope_seq(&envelope).unwrap(); |
| 7972 | let encoded_len = serde_json::to_vec(&envelope).unwrap().len(); |
| 7973 | let integrity = runtime_envelope_event(&envelope).is_some_and(integrity_critical_event); |
| 7974 | events.insert( |
| 7975 | seq, |
| 7976 | PendingRuntimeEnvelope { |
| 7977 | envelope, |
| 7978 | encoded_len, |
| 7979 | integrity, |
| 7980 | handed_off: true, |
| 7981 | }, |
| 7982 | ); |
| 7983 | } |
| 7984 | pending.insert("run_fixture".to_string(), events); |
| 7985 | journal.persist(&pending).expect("atomic persist"); |
| 7986 | drop(journal); |
| 7987 | |
| 7988 | let reopened = |
| 7989 | RuntimeEventJournal::open(dir.path(), "target:fixture@01", "session:fixture@01") |
| 7990 | .expect("journal reopen"); |
| 7991 | let restored = reopened.load().expect("verified load"); |
| 7992 | let events = restored.get("run_fixture").expect("restored run"); |
| 7993 | assert_eq!(events.len(), 2); |
| 7994 | assert_eq!( |
| 7995 | serde_json::to_vec(&events[&1]).unwrap(), |
| 7996 | serde_json::to_vec(&delta).unwrap(), |
| 7997 | "a restored envelope must re-serialize byte-identically for ambiguous retries" |
| 7998 | ); |
| 7999 | assert_eq!( |
| 8000 | serde_json::to_vec(&events[&2]).unwrap(), |
| 8001 | serde_json::to_vec(&terminal).unwrap() |
| 8002 | ); |
| 8003 | |
| 8004 | // Compaction: an empty pending set removes the file entirely. |
| 8005 | pending.get_mut("run_fixture").unwrap().clear(); |
| 8006 | reopened.persist(&pending).expect("compacting persist"); |
| 8007 | assert!(!reopened.path.exists(), "acknowledged journals are deleted"); |
| 8008 | } |
| 8009 | |
| 8010 | #[test] |
| 8011 | fn legacy_session_only_journal_is_preserved_and_repeated_starts_fail_closed() { |
| 8012 | let dir = tempfile::tempdir().expect("journal tempdir"); |
| 8013 | let session_id = "session_legacy_shared"; |
| 8014 | let journal_a = RuntimeEventJournal::open(dir.path(), "target_a", session_id).unwrap(); |
| 8015 | let mut hasher = Sha256::new(); |
| 8016 | hasher.update(b"cwc-remote-control-journal\0"); |
| 8017 | hasher.update(session_id.as_bytes()); |
| 8018 | let session_tag = bytes_to_hex(&hasher.finalize())[..32].to_string(); |
| 8019 | assert_eq!( |
| 8020 | journal_a.legacy_path, |
| 8021 | dir.path().join(format!("journal_{session_tag}.json")) |
| 8022 | ); |
| 8023 | let envelope = runtime_envelope( |
| 8024 | 1, |
| 8025 | "turn.completed", |
| 8026 | Some("turn_legacy"), |
| 8027 | "2026-08-08T12:00:00Z".to_string(), |
| 8028 | json!({ "turn": { "status": "completed" } }), |
| 8029 | ); |
| 8030 | crate::utils::write_atomic( |
| 8031 | &journal_a.legacy_path, |
| 8032 | &serde_json::to_vec(&json!({ |
| 8033 | "schemaVersion": 1, |
| 8034 | "session": session_tag, |
| 8035 | "runs": { "run_legacy": [envelope] }, |
| 8036 | })) |
| 8037 | .unwrap(), |
| 8038 | ) |
| 8039 | .unwrap(); |
| 8040 | |
| 8041 | assert_eq!(journal_a.load().unwrap_err(), JOURNAL_LEGACY_SCOPE_ERROR); |
| 8042 | journal_a.quarantine_legacy(); |
| 8043 | let unscoped_path = journal_a.legacy_unscoped_path.clone(); |
| 8044 | assert!(unscoped_path.exists()); |
| 8045 | assert!(!journal_a.path.exists()); |
| 8046 | drop(journal_a); |
| 8047 | |
| 8048 | let retry_a = RuntimeEventJournal::open(dir.path(), "target_a", session_id).unwrap(); |
| 8049 | assert_eq!(retry_a.load().unwrap_err(), JOURNAL_LEGACY_SCOPE_ERROR); |
| 8050 | drop(retry_a); |
| 8051 | assert_eq!( |
| 8052 | RuntimeEventJournal::open(dir.path(), "target_b", session_id) |
| 8053 | .err() |
| 8054 | .expect("the legacy session remains permanently bound to target A"), |
| 8055 | CLASSIC_LEASE_SCOPE_ERROR |
| 8056 | ); |
| 8057 | assert!(unscoped_path.exists()); |
| 8058 | } |
| 8059 | |
| 8060 | #[test] |
| 8061 | fn corrupt_v2_quarantine_never_moves_an_unrelated_legacy_journal() { |
| 8062 | let dir = tempfile::tempdir().expect("journal tempdir"); |
| 8063 | let journal = RuntimeEventJournal::open(dir.path(), "target_b", "session_shared").unwrap(); |
| 8064 | crate::utils::write_atomic(&journal.legacy_path, b"legacy target A recovery material") |
| 8065 | .unwrap(); |
| 8066 | crate::utils::write_atomic(&journal.path, b"{not-json").unwrap(); |
| 8067 | assert_eq!(journal.load().unwrap_err(), JOURNAL_UNTRUSTED_ERROR); |
| 8068 | journal.quarantine_current(); |
| 8069 | assert!(journal.legacy_path.exists()); |
| 8070 | assert!(!journal.legacy_unscoped_path.exists()); |
| 8071 | assert!(journal.path.with_extension("corrupt").exists()); |
| 8072 | } |
| 8073 | |
| 8074 | #[test] |
| 8075 | fn runtime_chat_projection_is_not_handed_off_before_durable_journal_write() { |
| 8076 | let dir = tempfile::tempdir().expect("journal tempdir"); |
| 8077 | let journal = |
| 8078 | RuntimeEventJournal::open(dir.path(), "target:fixture@01", "session:fixture@01") |
| 8079 | .expect("journal setup"); |
| 8080 | let journal_path = journal.path.clone(); |
| 8081 | let mut controller = RemoteControlController::default(); |
| 8082 | let (worker_tx, mut worker_rx) = mpsc::unbounded_channel(); |
| 8083 | controller.worker_tx = Some(worker_tx); |
| 8084 | controller.journal = Some(journal); |
| 8085 | let envelope = runtime_chat_envelope( |
| 8086 | 1, |
| 8087 | "turn.completed", |
| 8088 | Some("local_thread_aaaaaaaaaaaaaaaaaaaaaaaa"), |
| 8089 | Some("local_turn_bbbbbbbbbbbbbbbbbbbbbbbb"), |
| 8090 | Some("native_event_fixture"), |
| 8091 | RUNTIME_CHAT_CATALOG_TIMESTAMP.to_string(), |
| 8092 | json!({ "turn": { "status": "completed" } }), |
| 8093 | ); |
| 8094 | |
| 8095 | inject_journal_persist_failures(&journal_path, 1); |
| 8096 | assert!( |
| 8097 | !controller.queue_runtime_chat_envelope("run_fixture", envelope.clone()), |
| 8098 | "a failed durable append must reject native cursor advancement" |
| 8099 | ); |
| 8100 | assert!(controller.pending_runtime_events.is_empty()); |
| 8101 | assert!(!controller.event_seq.contains_key("run_fixture")); |
| 8102 | assert!(worker_rx.try_recv().is_err()); |
| 8103 | |
| 8104 | assert!(controller.queue_runtime_chat_envelope("run_fixture", envelope)); |
| 8105 | assert!(journal_path.exists()); |
| 8106 | let WorkerCommand::Upload { envelopes, .. } = worker_rx.try_recv().unwrap() else { |
| 8107 | panic!("durable retry must hand off the exact projection"); |
| 8108 | }; |
| 8109 | assert_eq!(envelopes[0]["seq"], 1); |
| 8110 | } |
| 8111 | |
| 8112 | #[test] |
| 8113 | fn classic_terminal_journal_failure_stays_retained_and_fail_closed_until_cursor() { |
| 8114 | let (mut controller, mut worker_rx, _event_tx, _journal_root) = wired_controller(); |
| 8115 | controller.status = Status::Connected; |
| 8116 | controller |
| 8117 | .activate_prompt("run_fixture", "turn_fixture") |
| 8118 | .unwrap(); |
| 8119 | let journal_path = controller.journal.as_ref().unwrap().path.clone(); |
| 8120 | inject_journal_persist_failures(&journal_path, 2); |
| 8121 | |
| 8122 | controller.observe_engine_event(&turn_complete_event()); |
| 8123 | assert_eq!(controller.status, Status::Failed); |
| 8124 | assert!(worker_rx.try_recv().is_err(), "failed append must not send"); |
| 8125 | assert!( |
| 8126 | controller.has_unacknowledged_integrity_events(), |
| 8127 | "classic terminal must remain retryable in memory" |
| 8128 | ); |
| 8129 | assert!(controller.stop_refusal().is_some()); |
| 8130 | controller.ownership_blocked_until = Some(Instant::now() - Duration::from_millis(1)); |
| 8131 | assert!(controller.try_next_event().is_none()); |
| 8132 | assert_eq!( |
| 8133 | controller.status, |
| 8134 | Status::Failed, |
| 8135 | "lease expiry cannot discard an unacknowledged classic terminal" |
| 8136 | ); |
| 8137 | |
| 8138 | controller |
| 8139 | .journal |
| 8140 | .as_ref() |
| 8141 | .unwrap() |
| 8142 | .persist(&controller.pending_runtime_events) |
| 8143 | .expect("repair durable journal"); |
| 8144 | let (worker_tx, mut repaired_worker_rx) = mpsc::unbounded_channel(); |
| 8145 | controller.worker_tx = Some(worker_tx); |
| 8146 | controller.flush_pending_runtime_events("run_fixture"); |
| 8147 | let WorkerCommand::Upload { envelopes, .. } = repaired_worker_rx.try_recv().unwrap() else { |
| 8148 | panic!("repaired terminal must be handed off"); |
| 8149 | }; |
| 8150 | let seq = runtime_envelope_seq(&envelopes[0]).unwrap(); |
| 8151 | controller.reconcile_runtime_cursor("run_fixture", seq); |
| 8152 | assert!(!controller.has_unacknowledged_integrity_events()); |
| 8153 | } |
| 8154 | |
| 8155 | #[test] |
| 8156 | fn classic_worker_failure_retains_live_turn_until_exact_terminal_is_durable() { |
| 8157 | let (mut controller, _worker_rx, event_tx, _journal_root) = wired_controller(); |
| 8158 | controller.status = Status::Connected; |
| 8159 | controller |
| 8160 | .activate_prompt("run_fixture", "turn_fixture") |
| 8161 | .unwrap(); |
| 8162 | event_tx |
| 8163 | .send(RemoteEvent::Failed("fixture transport loss".to_string())) |
| 8164 | .unwrap(); |
| 8165 | assert!(matches!( |
| 8166 | controller.try_next_event(), |
| 8167 | Some(RemoteEvent::Failed(_)) |
| 8168 | )); |
| 8169 | assert!(controller.has_active_run()); |
| 8170 | assert!(controller.has_durable_classic_lease()); |
| 8171 | |
| 8172 | controller.ownership_blocked_until = Some(Instant::now() - Duration::from_millis(1)); |
| 8173 | assert!(controller.try_next_event().is_none()); |
| 8174 | assert_eq!(controller.status, Status::Failed); |
| 8175 | assert!(controller.has_active_run()); |
| 8176 | assert!(controller.start(fixture_start()).is_err()); |
| 8177 | |
| 8178 | let usage = codewhale_models::Usage { |
| 8179 | input_tokens: 17, |
| 8180 | output_tokens: 5, |
| 8181 | ..codewhale_models::Usage::default() |
| 8182 | }; |
| 8183 | controller.observe_engine_event(&EngineEvent::TurnComplete { |
| 8184 | usage: usage.clone(), |
| 8185 | parent_route_usage: usage.clone(), |
| 8186 | routed_usage_dropped_records: 0, |
| 8187 | status: TurnOutcomeStatus::Completed, |
| 8188 | error: None, |
| 8189 | tool_catalog: None, |
| 8190 | base_url: None, |
| 8191 | }); |
| 8192 | assert!(!controller.has_active_run()); |
| 8193 | assert!(!controller.has_durable_classic_lease()); |
| 8194 | let terminals: Vec<&PendingRuntimeEnvelope> = |
| 8195 | controller.pending_runtime_events["run_fixture"] |
| 8196 | .values() |
| 8197 | .filter(|entry| runtime_envelope_event(&entry.envelope) == Some("turn.completed")) |
| 8198 | .collect(); |
| 8199 | assert_eq!(terminals.len(), 1); |
| 8200 | assert_eq!( |
| 8201 | terminals[0].envelope["payload"]["turn"]["usage"], |
| 8202 | serde_json::to_value(usage).unwrap() |
| 8203 | ); |
| 8204 | } |
| 8205 | |
| 8206 | #[test] |
| 8207 | fn classic_recovery_uses_persisted_seq_floor_and_ignores_older_terminal() { |
| 8208 | let root = tempfile::tempdir().unwrap(); |
| 8209 | let old_terminal = runtime_envelope( |
| 8210 | 40, |
| 8211 | "turn.completed", |
| 8212 | Some("turn_floor_fixture"), |
| 8213 | "2026-08-08T12:00:00Z".to_string(), |
| 8214 | json!({ "turn": { "status": "completed", "usage": {} } }), |
| 8215 | ); |
| 8216 | { |
| 8217 | let mut first = RemoteControlController::default(); |
| 8218 | first.journal = Some( |
| 8219 | RuntimeEventJournal::open(root.path(), "target_floor", "session_floor").unwrap(), |
| 8220 | ); |
| 8221 | first.reset_pending_from(HashMap::from([( |
| 8222 | "run_floor".to_string(), |
| 8223 | BTreeMap::from([(40, old_terminal.clone())]), |
| 8224 | )])); |
| 8225 | first.event_seq.insert("run_floor".to_string(), 41); |
| 8226 | first |
| 8227 | .activate_prompt("run_floor", "turn_floor_fixture") |
| 8228 | .unwrap(); |
| 8229 | assert_eq!( |
| 8230 | first |
| 8231 | .journal |
| 8232 | .as_ref() |
| 8233 | .unwrap() |
| 8234 | .classic_lease() |
| 8235 | .unwrap() |
| 8236 | .seq_floor, |
| 8237 | 41 |
| 8238 | ); |
| 8239 | } |
| 8240 | |
| 8241 | let mut reopened = RemoteControlController::default(); |
| 8242 | let journal = |
| 8243 | RuntimeEventJournal::open(root.path(), "target_floor", "session_floor").unwrap(); |
| 8244 | reopened.reset_pending_from(journal.load().unwrap()); |
| 8245 | reopened.journal = Some(journal); |
| 8246 | reopened.recover_classic_lease_before_worker().unwrap(); |
| 8247 | |
| 8248 | let events = &reopened.pending_runtime_events["run_floor"]; |
| 8249 | assert!(events.contains_key(&40)); |
| 8250 | let recovered = events.get(&42).expect("recovery follows the durable floor"); |
| 8251 | assert_eq!( |
| 8252 | runtime_envelope_event(&recovered.envelope), |
| 8253 | Some("turn.completed") |
| 8254 | ); |
| 8255 | assert_eq!(recovered.envelope["turn_id"], "turn_floor_fixture"); |
| 8256 | assert_eq!(recovered.envelope["payload"]["turn"]["status"], "failed"); |
| 8257 | assert!(!reopened.has_durable_classic_lease()); |
| 8258 | } |
| 8259 | |
| 8260 | #[test] |
| 8261 | fn blank_predispatch_lease_id_always_starts_a_fresh_generation() { |
| 8262 | let root = tempfile::tempdir().unwrap(); |
| 8263 | let mut controller = RemoteControlController::default(); |
| 8264 | controller.journal = Some( |
| 8265 | RuntimeEventJournal::open(root.path(), "target_generation", "session_generation") |
| 8266 | .unwrap(), |
| 8267 | ); |
| 8268 | controller |
| 8269 | .begin_classic_lease(ClassicRunLease { |
| 8270 | run_id: "run_generation".to_string(), |
| 8271 | turn_id: None, |
| 8272 | lease_id: String::new(), |
| 8273 | seq_floor: 0, |
| 8274 | }) |
| 8275 | .unwrap(); |
| 8276 | let stale_lease = controller |
| 8277 | .journal |
| 8278 | .as_ref() |
| 8279 | .and_then(RuntimeEventJournal::classic_lease) |
| 8280 | .unwrap(); |
| 8281 | |
| 8282 | controller |
| 8283 | .begin_classic_lease(ClassicRunLease { |
| 8284 | run_id: "run_generation".to_string(), |
| 8285 | turn_id: None, |
| 8286 | lease_id: String::new(), |
| 8287 | seq_floor: 0, |
| 8288 | }) |
| 8289 | .unwrap(); |
| 8290 | let fresh_lease = controller |
| 8291 | .journal |
| 8292 | .as_ref() |
| 8293 | .and_then(RuntimeEventJournal::classic_lease) |
| 8294 | .unwrap(); |
| 8295 | |
| 8296 | assert_ne!( |
| 8297 | fresh_lease.lease_id, stale_lease.lease_id, |
| 8298 | "a stale empty-turn lease cannot define the next recovery identity" |
| 8299 | ); |
| 8300 | } |
| 8301 | |
| 8302 | #[test] |
| 8303 | fn separate_predispatch_crashes_on_one_run_get_distinct_recovery_turn_ids() { |
| 8304 | let root = tempfile::tempdir().unwrap(); |
| 8305 | let first_recovery_turn; |
| 8306 | { |
| 8307 | let mut first = RemoteControlController::default(); |
| 8308 | first.journal = Some( |
| 8309 | RuntimeEventJournal::open(root.path(), "target_generation", "session_generation") |
| 8310 | .unwrap(), |
| 8311 | ); |
| 8312 | first |
| 8313 | .begin_classic_lease(ClassicRunLease { |
| 8314 | run_id: "run_generation".to_string(), |
| 8315 | turn_id: None, |
| 8316 | lease_id: String::new(), |
| 8317 | seq_floor: 0, |
| 8318 | }) |
| 8319 | .unwrap(); |
| 8320 | } |
| 8321 | { |
| 8322 | let mut recovered = RemoteControlController::default(); |
| 8323 | let journal = |
| 8324 | RuntimeEventJournal::open(root.path(), "target_generation", "session_generation") |
| 8325 | .unwrap(); |
| 8326 | recovered.reset_pending_from(journal.load().unwrap()); |
| 8327 | recovered.journal = Some(journal); |
| 8328 | recovered.recover_classic_lease_before_worker().unwrap(); |
| 8329 | first_recovery_turn = |
| 8330 | recovered.pending_runtime_events["run_generation"][&1].envelope["turn_id"] |
| 8331 | .as_str() |
| 8332 | .unwrap() |
| 8333 | .to_string(); |
| 8334 | recovered.reconcile_runtime_cursor("run_generation", 1); |
| 8335 | recovered |
| 8336 | .begin_classic_lease(ClassicRunLease { |
| 8337 | run_id: "run_generation".to_string(), |
| 8338 | turn_id: None, |
| 8339 | lease_id: String::new(), |
| 8340 | seq_floor: recovered.runtime_seq_floor("run_generation"), |
| 8341 | }) |
| 8342 | .unwrap(); |
| 8343 | } |
| 8344 | let mut recovered_again = RemoteControlController::default(); |
| 8345 | let journal = |
| 8346 | RuntimeEventJournal::open(root.path(), "target_generation", "session_generation") |
| 8347 | .unwrap(); |
| 8348 | recovered_again.reset_pending_from(journal.load().unwrap()); |
| 8349 | recovered_again.journal = Some(journal); |
| 8350 | recovered_again |
| 8351 | .recover_classic_lease_before_worker() |
| 8352 | .unwrap(); |
| 8353 | let second_recovery_turn = recovered_again.pending_runtime_events["run_generation"][&2] |
| 8354 | .envelope["turn_id"] |
| 8355 | .as_str() |
| 8356 | .unwrap(); |
| 8357 | assert_ne!(first_recovery_turn, second_recovery_turn); |
| 8358 | } |
| 8359 | |
| 8360 | #[test] |
| 8361 | fn classic_terminal_cursor_repairs_failed_canonical_clear_before_compaction() { |
| 8362 | let (mut controller, _worker_rx, _event_tx, journal_root) = wired_controller(); |
| 8363 | controller.status = Status::Connected; |
| 8364 | controller |
| 8365 | .activate_prompt("run_fixture", "turn_fixture") |
| 8366 | .unwrap(); |
| 8367 | let state_path = controller |
| 8368 | .journal |
| 8369 | .as_ref() |
| 8370 | .unwrap() |
| 8371 | .active_index_path |
| 8372 | .clone(); |
| 8373 | inject_journal_persist_failures(&state_path, 1); |
| 8374 | |
| 8375 | controller.observe_engine_event(&turn_complete_event()); |
| 8376 | assert!(controller.has_durable_classic_lease()); |
| 8377 | assert!(controller.has_active_run()); |
| 8378 | let terminal_seq = *controller.pending_runtime_events["run_fixture"] |
| 8379 | .last_key_value() |
| 8380 | .unwrap() |
| 8381 | .0; |
| 8382 | controller.reconcile_runtime_cursor("run_fixture", terminal_seq); |
| 8383 | assert!(!controller.has_durable_classic_lease()); |
| 8384 | assert!(!controller.has_active_run()); |
| 8385 | assert!(controller.pending_runtime_events.is_empty()); |
| 8386 | drop(controller); |
| 8387 | |
| 8388 | let journal = RuntimeEventJournal::open( |
| 8389 | journal_root.path(), |
| 8390 | "target_wired_fixture", |
| 8391 | "session_wired_fixture", |
| 8392 | ) |
| 8393 | .unwrap(); |
| 8394 | assert!(journal.classic_lease().is_none()); |
| 8395 | assert!(journal.load().unwrap().is_empty()); |
| 8396 | } |
| 8397 | |
| 8398 | #[test] |
| 8399 | fn saved_session_is_permanently_bound_to_its_first_remote_target() { |
| 8400 | let dir = tempfile::tempdir().expect("journal tempdir"); |
| 8401 | let first = RuntimeEventJournal::open(dir.path(), "target_a", "session_shared") |
| 8402 | .expect("first target journal"); |
| 8403 | assert_eq!( |
| 8404 | RuntimeEventJournal::open(dir.path(), "target_b", "session_shared") |
| 8405 | .err() |
| 8406 | .expect("a second target must fail closed"), |
| 8407 | CLASSIC_LEASE_SCOPE_ERROR |
| 8408 | ); |
| 8409 | |
| 8410 | let pending_for = |run_id: &str, marker: &str| { |
| 8411 | let envelope = runtime_envelope( |
| 8412 | 1, |
| 8413 | "turn.completed", |
| 8414 | None, |
| 8415 | "2026-08-08T12:00:00Z".to_string(), |
| 8416 | json!({ "turn": { "status": "completed" }, "marker": marker }), |
| 8417 | ); |
| 8418 | let encoded_len = serde_json::to_vec(&envelope).unwrap().len(); |
| 8419 | HashMap::from([( |
| 8420 | run_id.to_string(), |
| 8421 | BTreeMap::from([( |
| 8422 | 1, |
| 8423 | PendingRuntimeEnvelope { |
| 8424 | envelope, |
| 8425 | encoded_len, |
| 8426 | integrity: true, |
| 8427 | handed_off: true, |
| 8428 | }, |
| 8429 | )]), |
| 8430 | )]) |
| 8431 | }; |
| 8432 | first.persist(&pending_for("run_a", "a")).unwrap(); |
| 8433 | assert_eq!( |
| 8434 | first.load().unwrap().keys().collect::<Vec<_>>(), |
| 8435 | vec!["run_a"] |
| 8436 | ); |
| 8437 | drop(first); |
| 8438 | assert_eq!( |
| 8439 | RuntimeEventJournal::open(dir.path(), "target_b", "session_shared") |
| 8440 | .err() |
| 8441 | .expect("the permanent target binding must fail closed"), |
| 8442 | CLASSIC_LEASE_SCOPE_ERROR, |
| 8443 | "the target binding survives terminal settlement and process restart" |
| 8444 | ); |
| 8445 | RuntimeEventJournal::open(dir.path(), "target_a", "session_shared") |
| 8446 | .expect("the original target can reopen its saved session"); |
| 8447 | } |
| 8448 | |
| 8449 | #[cfg(unix)] |
| 8450 | #[test] |
| 8451 | fn journal_directory_and_file_are_owner_only() { |
| 8452 | use std::os::unix::fs::PermissionsExt; |
| 8453 | let base = tempfile::tempdir().expect("journal tempdir"); |
| 8454 | let dir = base.path().join("journal"); |
| 8455 | let journal = RuntimeEventJournal::open(&dir, "target:fixture@01", "session:fixture@01") |
| 8456 | .expect("journal setup"); |
| 8457 | let mut pending: HashMap<String, BTreeMap<u64, PendingRuntimeEnvelope>> = HashMap::new(); |
| 8458 | let envelope = runtime_envelope( |
| 8459 | 1, |
| 8460 | "turn.completed", |
| 8461 | None, |
| 8462 | "2026-08-08T12:00:00Z".to_string(), |
| 8463 | json!({ "turn": { "status": "completed", "usage": {} } }), |
| 8464 | ); |
| 8465 | let encoded_len = serde_json::to_vec(&envelope).unwrap().len(); |
| 8466 | pending.insert( |
| 8467 | "run_fixture".to_string(), |
| 8468 | BTreeMap::from([( |
| 8469 | 1, |
| 8470 | PendingRuntimeEnvelope { |
| 8471 | envelope, |
| 8472 | encoded_len, |
| 8473 | integrity: true, |
| 8474 | handed_off: true, |
| 8475 | }, |
| 8476 | )]), |
| 8477 | ); |
| 8478 | journal.persist(&pending).expect("atomic persist"); |
| 8479 | let dir_mode = std::fs::metadata(&dir).unwrap().permissions().mode() & 0o777; |
| 8480 | assert_eq!(dir_mode, 0o700, "journal directory must be private"); |
| 8481 | let file_mode = std::fs::metadata(&journal.path) |
| 8482 | .unwrap() |
| 8483 | .permissions() |
| 8484 | .mode() |
| 8485 | & 0o777; |
| 8486 | assert_eq!(file_mode, 0o600, "journal file must be owner-only"); |
| 8487 | } |
| 8488 | |
| 8489 | #[test] |
| 8490 | fn corrupt_journal_fails_closed_and_start_quarantines_it() { |
| 8491 | let dir = tempfile::tempdir().expect("journal tempdir"); |
| 8492 | let probe = RuntimeEventJournal::open(dir.path(), "target_fixture", "session:fixture@01") |
| 8493 | .expect("journal setup"); |
| 8494 | std::fs::write(&probe.path, b"{ not json").expect("plant corrupt journal"); |
| 8495 | let corrupt_path = probe.path.clone(); |
| 8496 | drop(probe); |
| 8497 | |
| 8498 | let mut controller = RemoteControlController::default(); |
| 8499 | let error = controller |
| 8500 | .start(RemoteStart { |
| 8501 | journal_dir: Some(dir.path().to_path_buf()), |
| 8502 | ..fixture_start() |
| 8503 | }) |
| 8504 | .expect_err("a corrupt journal must fail closed"); |
| 8505 | assert_eq!(error, JOURNAL_UNTRUSTED_ERROR); |
| 8506 | assert_eq!(controller.status, Status::Off, "no relay may start"); |
| 8507 | assert!( |
| 8508 | !corrupt_path.exists(), |
| 8509 | "the untrusted journal must not stay in place" |
| 8510 | ); |
| 8511 | assert!( |
| 8512 | corrupt_path.with_extension("corrupt").exists(), |
| 8513 | "the untrusted journal is quarantined, not silently discarded" |
| 8514 | ); |
| 8515 | |
| 8516 | // A mismatched session tag is equally untrusted. |
| 8517 | let other = |
| 8518 | RuntimeEventJournal::open(dir.path(), "target:fixture@01", "session:fixture@02") |
| 8519 | .expect("journal setup"); |
| 8520 | std::fs::write( |
| 8521 | &other.path, |
| 8522 | serde_json::to_vec(&json!({ |
| 8523 | "schemaVersion": JOURNAL_SCHEMA_VERSION, |
| 8524 | "scope": "00000000000000000000000000000000", |
| 8525 | "runs": {}, |
| 8526 | })) |
| 8527 | .unwrap(), |
| 8528 | ) |
| 8529 | .expect("plant mismatched journal"); |
| 8530 | assert_eq!(other.load().unwrap_err(), JOURNAL_UNTRUSTED_ERROR); |
| 8531 | } |
| 8532 | |
| 8533 | #[test] |
| 8534 | fn start_recovers_journaled_envelopes_and_resends_on_connect() { |
| 8535 | let dir = tempfile::tempdir().expect("journal tempdir"); |
| 8536 | { |
| 8537 | let journal = |
| 8538 | RuntimeEventJournal::open(dir.path(), "target:fixture@01", "session:fixture@01") |
| 8539 | .expect("journal setup"); |
| 8540 | let envelope = runtime_envelope( |
| 8541 | 3, |
| 8542 | "turn.completed", |
| 8543 | Some("turn_fixture"), |
| 8544 | "2026-08-08T12:00:00Z".to_string(), |
| 8545 | json!({ "turn": { "status": "completed", "usage": {} } }), |
| 8546 | ); |
| 8547 | let encoded_len = serde_json::to_vec(&envelope).unwrap().len(); |
| 8548 | let pending = HashMap::from([( |
| 8549 | "run_fixture".to_string(), |
| 8550 | BTreeMap::from([( |
| 8551 | 3, |
| 8552 | PendingRuntimeEnvelope { |
| 8553 | envelope, |
| 8554 | encoded_len, |
| 8555 | integrity: true, |
| 8556 | handed_off: true, |
| 8557 | }, |
| 8558 | )]), |
| 8559 | )]); |
| 8560 | journal |
| 8561 | .persist(&pending) |
| 8562 | .expect("previous process persisted"); |
| 8563 | } |
| 8564 | |
| 8565 | let mut controller = RemoteControlController::default(); |
| 8566 | let journal = |
| 8567 | RuntimeEventJournal::open(dir.path(), "target:fixture@01", "session:fixture@01") |
| 8568 | .expect("journal setup"); |
| 8569 | controller.reset_pending_from(journal.load().expect("clean recovery")); |
| 8570 | controller.journal = Some(journal); |
| 8571 | assert!( |
| 8572 | controller.has_unacknowledged_integrity_events(), |
| 8573 | "recovered terminal state must gate /rc stop until acknowledged" |
| 8574 | ); |
| 8575 | assert!(controller.stop_refusal().is_some()); |
| 8576 | |
| 8577 | let (worker_tx, mut worker_rx) = mpsc::unbounded_channel(); |
| 8578 | let (event_tx, event_rx) = mpsc::unbounded_channel(); |
| 8579 | controller.worker_tx = Some(worker_tx); |
| 8580 | controller.event_rx = Some(event_rx); |
| 8581 | event_tx |
| 8582 | .send(RemoteEvent::Connected { |
| 8583 | account_ref: "account_fixture".to_string(), |
| 8584 | runner_id: "runner_fixture".to_string(), |
| 8585 | target_ref: "target_fixture".to_string(), |
| 8586 | attachment: RemoteAttachment { |
| 8587 | run_id: "run_other".to_string(), |
| 8588 | workspace_id: "workspace_fixture".to_string(), |
| 8589 | runtime_cursor: 0, |
| 8590 | snapshot_present: false, |
| 8591 | runtime_chat_relay_protocol: RUNTIME_CHAT_RELAY_PROTOCOL.to_string(), |
| 8592 | runtime_chat_relay_challenge: "a".repeat(32), |
| 8593 | }, |
| 8594 | links: RemoteLinks::default(), |
| 8595 | }) |
| 8596 | .unwrap(); |
| 8597 | controller.try_next_event().unwrap(); |
| 8598 | let WorkerCommand::Upload { envelopes, .. } = worker_rx.try_recv().unwrap() else { |
| 8599 | panic!("recovered envelopes must resend on connect"); |
| 8600 | }; |
| 8601 | assert_eq!(envelopes[0]["seq"], 3); |
| 8602 | assert_eq!(envelopes[0]["event"], "turn.completed"); |
| 8603 | } |
| 8604 | |
| 8605 | #[test] |
| 8606 | fn delta_pressure_sheds_to_resync_and_preserves_integrity_capacity() { |
| 8607 | let (mut controller, _worker_rx, _event_tx, _journal_root) = wired_controller(); |
| 8608 | let delta_budget = MAX_JOURNAL_EVENTS - JOURNAL_RESERVED_INTEGRITY_EVENTS; |
| 8609 | for index in 0..delta_budget { |
| 8610 | assert!( |
| 8611 | controller.queue_runtime_envelope( |
| 8612 | "run_fixture", |
| 8613 | runtime_envelope( |
| 8614 | (index + 1) as u64, |
| 8615 | "item.delta", |
| 8616 | Some("turn_fixture"), |
| 8617 | "2026-08-08T12:00:00Z".to_string(), |
| 8618 | json!({ "kind": "agent_message", "delta": index.to_string() }), |
| 8619 | ), |
| 8620 | ), |
| 8621 | "delta {index} fits the unreserved budget" |
| 8622 | ); |
| 8623 | } |
| 8624 | assert_eq!(controller.pending_event_count, delta_budget); |
| 8625 | |
| 8626 | let shed_seq = (delta_budget + 1) as u64; |
| 8627 | assert!( |
| 8628 | !controller.queue_runtime_envelope( |
| 8629 | "run_fixture", |
| 8630 | runtime_envelope( |
| 8631 | shed_seq, |
| 8632 | "item.delta", |
| 8633 | Some("turn_fixture"), |
| 8634 | "2026-08-08T12:00:00Z".to_string(), |
| 8635 | json!({ "kind": "agent_message", "delta": "over budget" }), |
| 8636 | ), |
| 8637 | ), |
| 8638 | "a delta beyond the unreserved budget is shed" |
| 8639 | ); |
| 8640 | assert_eq!(controller.pending_event_count, delta_budget); |
| 8641 | assert!(controller.resync_required.contains("run_fixture")); |
| 8642 | assert_ne!( |
| 8643 | controller.status, |
| 8644 | Status::Failed, |
| 8645 | "delta pressure is ordinary and must not fail the relay" |
| 8646 | ); |
| 8647 | |
| 8648 | // Reserved capacity keeps the terminal boundary deliverable, and the |
| 8649 | // terminal boundary schedules the resynchronization snapshot. |
| 8650 | controller |
| 8651 | .activate_prompt("run_fixture", "turn_fixture") |
| 8652 | .unwrap(); |
| 8653 | controller.observe_engine_event(&turn_complete_event()); |
| 8654 | assert!( |
| 8655 | controller.has_unacknowledged_integrity_events(), |
| 8656 | "the terminal envelope must use the reserved capacity" |
| 8657 | ); |
| 8658 | assert_eq!( |
| 8659 | controller.take_pending_resync().as_deref(), |
| 8660 | Some("run_fixture"), |
| 8661 | "the shed run resynchronizes at its terminal boundary" |
| 8662 | ); |
| 8663 | controller.upload_resync_snapshot("run_fixture", &[]); |
| 8664 | assert!( |
| 8665 | controller |
| 8666 | .pending_runtime_events |
| 8667 | .get("run_fixture") |
| 8668 | .is_some_and(|events| events.values().any(|entry| runtime_envelope_event( |
| 8669 | &entry.envelope |
| 8670 | ) == Some("session.snapshot"))), |
| 8671 | "the bounded snapshot restores account truth" |
| 8672 | ); |
| 8673 | } |
| 8674 | |
| 8675 | #[test] |
| 8676 | fn integrity_overflow_fails_closed_without_restoring_input() { |
| 8677 | let (mut controller, _worker_rx, _event_tx, _journal_root) = wired_controller(); |
| 8678 | controller.status = Status::Connected; |
| 8679 | for index in 0..MAX_JOURNAL_EVENTS { |
| 8680 | assert!(controller.queue_runtime_envelope( |
| 8681 | "run_fixture", |
| 8682 | runtime_envelope( |
| 8683 | (index + 1) as u64, |
| 8684 | "item.failed", |
| 8685 | Some("turn_fixture"), |
| 8686 | "2026-08-08T12:00:00Z".to_string(), |
| 8687 | json!({ "item": { "id": index.to_string(), "kind": "error" } }), |
| 8688 | ), |
| 8689 | )); |
| 8690 | } |
| 8691 | |
| 8692 | assert!(!controller.queue_runtime_envelope( |
| 8693 | "run_fixture", |
| 8694 | runtime_envelope( |
| 8695 | (MAX_JOURNAL_EVENTS + 1) as u64, |
| 8696 | "turn.completed", |
| 8697 | Some("turn_fixture"), |
| 8698 | "2026-08-08T12:00:00Z".to_string(), |
| 8699 | json!({ "turn": { "status": "failed", "usage": {} } }), |
| 8700 | ), |
| 8701 | )); |
| 8702 | assert_eq!( |
| 8703 | controller.status, |
| 8704 | Status::Failed, |
| 8705 | "losing integrity state can never be silent" |
| 8706 | ); |
| 8707 | assert!( |
| 8708 | !controller.can_share_approval_with_web(), |
| 8709 | "a failed-closed relay keeps the shared-decision channel closed" |
| 8710 | ); |
| 8711 | } |
| 8712 | |
| 8713 | #[test] |
| 8714 | fn message_deltas_coalesce_until_a_handoff_boundary() { |
| 8715 | let (mut controller, mut worker_rx, _event_tx, _journal_root) = wired_controller(); |
| 8716 | controller |
| 8717 | .activate_prompt("run_fixture", "turn_fixture") |
| 8718 | .unwrap(); |
| 8719 | controller.observe_engine_event(&EngineEvent::MessageDelta { |
| 8720 | index: 0, |
| 8721 | content: "Hello ".to_string(), |
| 8722 | }); |
| 8723 | controller.observe_engine_event(&EngineEvent::MessageDelta { |
| 8724 | index: 0, |
| 8725 | content: "world".to_string(), |
| 8726 | }); |
| 8727 | assert!( |
| 8728 | worker_rx.try_recv().is_err(), |
| 8729 | "deferred deltas coalesce before any transport handoff" |
| 8730 | ); |
| 8731 | let events = controller |
| 8732 | .pending_runtime_events |
| 8733 | .get("run_fixture") |
| 8734 | .unwrap(); |
| 8735 | assert_eq!(events.len(), 1, "both deltas share one envelope"); |
| 8736 | assert_eq!( |
| 8737 | events.values().next().unwrap().envelope["payload"]["delta"], |
| 8738 | "Hello world" |
| 8739 | ); |
| 8740 | |
| 8741 | controller.observe_engine_event(&EngineEvent::ToolCallStarted { |
| 8742 | id: "tool_fixture".to_string(), |
| 8743 | name: "shell".to_string(), |
| 8744 | input: json!({}), |
| 8745 | }); |
| 8746 | let WorkerCommand::Upload { envelopes, .. } = worker_rx.try_recv().unwrap() else { |
| 8747 | panic!("the coalesced delta must hand off before a later event"); |
| 8748 | }; |
| 8749 | assert_eq!(envelopes[0]["event"], "item.delta"); |
| 8750 | assert_eq!(envelopes[0]["payload"]["delta"], "Hello world"); |
| 8751 | let WorkerCommand::Upload { envelopes, .. } = worker_rx.try_recv().unwrap() else { |
| 8752 | panic!("the tool event follows the coalesced delta"); |
| 8753 | }; |
| 8754 | assert_eq!(envelopes[0]["event"], "item.started"); |
| 8755 | |
| 8756 | // Once handed off an envelope is immutable; new deltas open a fresh |
| 8757 | // envelope that flushes on the next UI poll. |
| 8758 | controller.observe_engine_event(&EngineEvent::MessageDelta { |
| 8759 | index: 0, |
| 8760 | content: "again".to_string(), |
| 8761 | }); |
| 8762 | assert!(worker_rx.try_recv().is_err()); |
| 8763 | assert!(controller.try_next_event().is_none()); |
| 8764 | let WorkerCommand::Upload { envelopes, .. } = worker_rx.try_recv().unwrap() else { |
| 8765 | panic!("the UI poll hands the deferred delta to the transport"); |
| 8766 | }; |
| 8767 | assert_eq!(envelopes[0]["payload"]["delta"], "again"); |
| 8768 | } |
| 8769 | #[test] |
| 8770 | fn runtime_image_legacy_work_never_downgrades_to_text() { |
| 8771 | let image = crate::image_attach::tests::runtime_image_fixture(1); |
| 8772 | let command = json!({"type":"prompt.request","runId":"run_fixture","turnId":"turn_fixture","prompt":"look","images":[image]}); |
| 8773 | assert!( |
| 8774 | parse_remote_command(&command, "run_fixture") |
| 8775 | .unwrap_err() |
| 8776 | .contains("legacy remote Work") |
| 8777 | ); |
| 8778 | let mut text = command; |
| 8779 | text.as_object_mut().unwrap().remove("images"); |
| 8780 | assert!(matches!( |
| 8781 | parse_remote_command(&text, "run_fixture").unwrap(), |
| 8782 | RemoteCommand::Prompt { .. } |
| 8783 | )); |
| 8784 | } |
| 8785 | |
| 8786 | #[tokio::test] |
| 8787 | async fn runtime_image_command_response_has_bounded_larger_budget() { |
| 8788 | use axum::{Router, routing::get}; |
| 8789 | let payload = serde_json::to_string( |
| 8790 | &json!({"commands":[],"fixture": "x".repeat(MAX_RESPONSE_BYTES + 1)}), |
| 8791 | ) |
| 8792 | .unwrap(); |
| 8793 | let app = Router::new().route( |
| 8794 | "/", |
| 8795 | get(move || { |
| 8796 | let payload = payload.clone(); |
| 8797 | async move { payload } |
| 8798 | }), |
| 8799 | ); |
| 8800 | let listener = tokio::net::TcpListener::bind("127.0.0.1:0").await.unwrap(); |
| 8801 | let addr = listener.local_addr().unwrap(); |
| 8802 | let server = tokio::spawn(async move { |
| 8803 | axum::serve(listener, app).await.unwrap(); |
| 8804 | }); |
| 8805 | let client = crate::tls::reqwest_client(); |
| 8806 | let url = format!("http://{addr}/"); |
| 8807 | assert!( |
| 8808 | read_bounded_json(client.get(&url).send().await.unwrap()) |
| 8809 | .await |
| 8810 | .is_err() |
| 8811 | ); |
| 8812 | let parsed = read_bounded_json_with_limit( |
| 8813 | client.get(&url).send().await.unwrap(), |
| 8814 | codewhale_protocol::runtime::MAX_RUNTIME_IMAGE_BODY_BYTES, |
| 8815 | ) |
| 8816 | .await |
| 8817 | .unwrap(); |
| 8818 | assert_eq!( |
| 8819 | parsed["fixture"].as_str().unwrap().len(), |
| 8820 | MAX_RESPONSE_BYTES + 1 |
| 8821 | ); |
| 8822 | server.abort(); |
| 8823 | } |
| 8824 | } |
| 8825 |