| 1 | package main |
| 2 | |
| 3 | import ( |
| 4 | "context" |
| 5 | "crypto/rand" |
| 6 | "encoding/hex" |
| 7 | "errors" |
| 8 | "fmt" |
| 9 | "os" |
| 10 | "strings" |
| 11 | "time" |
| 12 | |
| 13 | "reasonix/internal/agent" |
| 14 | ) |
| 15 | |
| 16 | type SessionRuntimePhase string |
| 17 | |
| 18 | const ( |
| 19 | sessionRuntimeStarting SessionRuntimePhase = "starting" |
| 20 | sessionRuntimeReady SessionRuntimePhase = "ready" |
| 21 | sessionRuntimeLeaseBlocked SessionRuntimePhase = "lease_blocked" |
| 22 | sessionRuntimeFailed SessionRuntimePhase = "failed" |
| 23 | sessionRuntimeClosing SessionRuntimePhase = "closing" |
| 24 | ) |
| 25 | |
| 26 | type SessionRuntimeIssue struct { |
| 27 | Code string `json:"code"` |
| 28 | Message string `json:"message"` |
| 29 | Retryable bool `json:"retryable"` |
| 30 | HolderPID int `json:"holderPid,omitempty"` |
| 31 | HolderHost string `json:"holderHost,omitempty"` |
| 32 | AcquiredAt string `json:"acquiredAt,omitempty"` |
| 33 | } |
| 34 | |
| 35 | type SessionRuntimeView struct { |
| 36 | Phase SessionRuntimePhase `json:"phase"` |
| 37 | Epoch string `json:"epoch"` |
| 38 | Issue *SessionRuntimeIssue `json:"issue,omitempty"` |
| 39 | } |
| 40 | |
| 41 | // desktopSessionRuntime is the process-local ownership record for one writable |
| 42 | // session. WorkspaceTab still carries compatibility projections of controller |
| 43 | // and lease fields while the desktop code migrates, but this registry is the |
| 44 | // authority that prevents a second local build from competing for the same |
| 45 | // session path. |
| 46 | // |
| 47 | // All fields are guarded by App.mu. |
| 48 | type desktopSessionRuntime struct { |
| 49 | ID string |
| 50 | Key string |
| 51 | Epoch string |
| 52 | Phase SessionRuntimePhase |
| 53 | Issue *SessionRuntimeIssue |
| 54 | Owner *WorkspaceTab |
| 55 | readyCh chan struct{} |
| 56 | } |
| 57 | |
| 58 | func newSessionRuntimeID(prefix string) string { |
| 59 | var b [12]byte |
| 60 | if _, err := rand.Read(b[:]); err == nil { |
| 61 | return prefix + "_" + hex.EncodeToString(b[:]) |
| 62 | } |
| 63 | return prefix + "_" + time.Now().UTC().Format("20060102150405.000000000") |
| 64 | } |
| 65 | |
| 66 | func cloneSessionRuntimeIssue(issue *SessionRuntimeIssue) *SessionRuntimeIssue { |
| 67 | if issue == nil { |
| 68 | return nil |
| 69 | } |
| 70 | copy := *issue |
| 71 | return © |
| 72 | } |
| 73 | |
| 74 | func sessionRuntimeIssueForError(err error) *SessionRuntimeIssue { |
| 75 | if err == nil { |
| 76 | return nil |
| 77 | } |
| 78 | issue := &SessionRuntimeIssue{ |
| 79 | Code: "startup_failed", |
| 80 | Message: userFacingSessionLeaseError("", err).Error(), |
| 81 | Retryable: false, |
| 82 | } |
| 83 | if !errors.Is(err, agent.ErrSessionLeaseHeld) { |
| 84 | return issue |
| 85 | } |
| 86 | issue.Code = "session_lease_held" |
| 87 | issue.Retryable = true |
| 88 | var leaseErr *agent.SessionLeaseError |
| 89 | if errors.As(err, &leaseErr) && leaseErr != nil && leaseErr.Info != nil { |
| 90 | issue.HolderPID = leaseErr.Info.PID |
| 91 | issue.HolderHost = strings.TrimSpace(leaseErr.Info.Hostname) |
| 92 | if !leaseErr.Info.AcquiredAt.IsZero() { |
| 93 | issue.AcquiredAt = leaseErr.Info.AcquiredAt.UTC().Format(time.RFC3339) |
| 94 | } |
| 95 | } |
| 96 | return issue |
| 97 | } |
| 98 | |
| 99 | func (a *App) newSessionRuntimeLocked(tab *WorkspaceTab, key string) *desktopSessionRuntime { |
| 100 | if existing := a.runtimeBySessionKey[key]; key != "" && existing != nil && existing.Owner != tab { |
| 101 | // A second tab may begin restoring the same persisted path before it |
| 102 | // reaches claimSessionRuntime. Do not overwrite the first starting |
| 103 | // placeholder; the later build will wait for and attach to it. |
| 104 | key = "" |
| 105 | } |
| 106 | rt := &desktopSessionRuntime{ |
| 107 | ID: newSessionRuntimeID("runtime"), |
| 108 | Key: key, |
| 109 | Epoch: newSessionRuntimeID("epoch"), |
| 110 | Phase: sessionRuntimeStarting, |
| 111 | Owner: tab, |
| 112 | readyCh: make(chan struct{}), |
| 113 | } |
| 114 | if tab != nil && tab.Ctrl != nil && tab.Ready { |
| 115 | rt.Phase = sessionRuntimeReady |
| 116 | closeRuntimeReadyChannelLocked(rt) |
| 117 | } |
| 118 | if a.runtimeByID == nil { |
| 119 | a.runtimeByID = map[string]*desktopSessionRuntime{} |
| 120 | } |
| 121 | if a.runtimeBySessionKey == nil { |
| 122 | a.runtimeBySessionKey = map[string]*desktopSessionRuntime{} |
| 123 | } |
| 124 | a.runtimeByID[rt.ID] = rt |
| 125 | if key != "" { |
| 126 | a.runtimeBySessionKey[key] = rt |
| 127 | } |
| 128 | if tab != nil { |
| 129 | tab.runtimeID = rt.ID |
| 130 | if tab.sink != nil { |
| 131 | tab.sink.setRuntimeEpoch(rt.Epoch) |
| 132 | } |
| 133 | } |
| 134 | return rt |
| 135 | } |
| 136 | |
| 137 | func (a *App) runtimeForTabLocked(tab *WorkspaceTab) *desktopSessionRuntime { |
| 138 | if tab == nil || strings.TrimSpace(tab.runtimeID) == "" { |
| 139 | return nil |
| 140 | } |
| 141 | rt := a.runtimeByID[tab.runtimeID] |
| 142 | if rt == nil || rt.Owner != tab { |
| 143 | return nil |
| 144 | } |
| 145 | return rt |
| 146 | } |
| 147 | |
| 148 | func (a *App) runtimeOwnerLiveLocked(rt *desktopSessionRuntime) bool { |
| 149 | if rt == nil || rt.Owner == nil { |
| 150 | return false |
| 151 | } |
| 152 | if a.tabs[rt.Owner.ID] == rt.Owner { |
| 153 | return true |
| 154 | } |
| 155 | for _, detached := range a.detachedSessions { |
| 156 | if detached == rt.Owner { |
| 157 | return true |
| 158 | } |
| 159 | } |
| 160 | return false |
| 161 | } |
| 162 | |
| 163 | func (a *App) removeSessionRuntimeMappingsLocked(rt *desktopSessionRuntime) { |
| 164 | if rt == nil { |
| 165 | return |
| 166 | } |
| 167 | for key, candidate := range a.runtimeBySessionKey { |
| 168 | if candidate == rt { |
| 169 | delete(a.runtimeBySessionKey, key) |
| 170 | } |
| 171 | } |
| 172 | delete(a.runtimeByID, rt.ID) |
| 173 | } |
| 174 | |
| 175 | func closeRuntimeReadyChannelLocked(rt *desktopSessionRuntime) { |
| 176 | if rt == nil || rt.readyCh == nil { |
| 177 | return |
| 178 | } |
| 179 | close(rt.readyCh) |
| 180 | rt.readyCh = nil |
| 181 | } |
| 182 | |
| 183 | func (a *App) setSessionRuntimePhaseLocked(tab *WorkspaceTab, phase SessionRuntimePhase, err error) { |
| 184 | if tab == nil { |
| 185 | return |
| 186 | } |
| 187 | rt := a.runtimeForTabLocked(tab) |
| 188 | if rt == nil { |
| 189 | key := sessionRuntimeKey(tab.SessionPath) |
| 190 | if existing := a.runtimeBySessionKey[key]; key != "" && existing != nil && existing.Owner != tab { |
| 191 | return |
| 192 | } |
| 193 | rt = a.newSessionRuntimeLocked(tab, key) |
| 194 | } |
| 195 | rt.Phase = phase |
| 196 | rt.Issue = sessionRuntimeIssueForError(err) |
| 197 | if phase == sessionRuntimeStarting { |
| 198 | rt.Issue = nil |
| 199 | if rt.readyCh == nil { |
| 200 | rt.readyCh = make(chan struct{}) |
| 201 | } |
| 202 | } else { |
| 203 | closeRuntimeReadyChannelLocked(rt) |
| 204 | } |
| 205 | if tab.sink != nil { |
| 206 | tab.sink.setRuntimeEpoch(rt.Epoch) |
| 207 | } |
| 208 | } |
| 209 | |
| 210 | func (a *App) advanceSessionRuntimeEpochLocked(tab *WorkspaceTab) string { |
| 211 | if tab == nil { |
| 212 | return "" |
| 213 | } |
| 214 | rt := a.runtimeForTabLocked(tab) |
| 215 | if rt == nil { |
| 216 | rt = a.newSessionRuntimeLocked(tab, sessionRuntimeKey(tab.SessionPath)) |
| 217 | } |
| 218 | rt.Epoch = newSessionRuntimeID("epoch") |
| 219 | rt.Phase = sessionRuntimeReady |
| 220 | rt.Issue = nil |
| 221 | closeRuntimeReadyChannelLocked(rt) |
| 222 | if tab.sink != nil { |
| 223 | tab.sink.setRuntimeEpoch(rt.Epoch) |
| 224 | } |
| 225 | return rt.Epoch |
| 226 | } |
| 227 | |
| 228 | func (a *App) sessionRuntimeViewLocked(tab *WorkspaceTab) SessionRuntimeView { |
| 229 | if tab == nil { |
| 230 | return SessionRuntimeView{Phase: sessionRuntimeStarting} |
| 231 | } |
| 232 | if rt := a.runtimeForTabLocked(tab); rt != nil { |
| 233 | return SessionRuntimeView{ |
| 234 | Phase: rt.Phase, |
| 235 | Epoch: rt.Epoch, |
| 236 | Issue: cloneSessionRuntimeIssue(rt.Issue), |
| 237 | } |
| 238 | } |
| 239 | view := SessionRuntimeView{Phase: sessionRuntimeStarting} |
| 240 | switch { |
| 241 | case tab.Ctrl != nil && tab.Ready: |
| 242 | view.Phase = sessionRuntimeReady |
| 243 | case tab.StartupErrLeaseHeld: |
| 244 | view.Phase = sessionRuntimeLeaseBlocked |
| 245 | view.Issue = sessionRuntimeIssueForError(&agent.SessionLeaseError{}) |
| 246 | case strings.TrimSpace(tab.StartupErr) != "": |
| 247 | view.Phase = sessionRuntimeFailed |
| 248 | view.Issue = &SessionRuntimeIssue{Code: "startup_failed", Message: tab.StartupErr} |
| 249 | } |
| 250 | return view |
| 251 | } |
| 252 | |
| 253 | func (a *App) bindSessionRuntimeKeyLocked(tab *WorkspaceTab, path string) bool { |
| 254 | if tab == nil { |
| 255 | return false |
| 256 | } |
| 257 | key := sessionRuntimeKey(path) |
| 258 | if key == "" { |
| 259 | return true |
| 260 | } |
| 261 | if existing := a.runtimeBySessionKey[key]; existing != nil && existing.Owner != tab { |
| 262 | return false |
| 263 | } |
| 264 | rt := a.runtimeForTabLocked(tab) |
| 265 | if rt == nil { |
| 266 | a.newSessionRuntimeLocked(tab, key) |
| 267 | return true |
| 268 | } |
| 269 | if rt.Key != "" && rt.Key != key && a.runtimeBySessionKey[rt.Key] == rt { |
| 270 | delete(a.runtimeBySessionKey, rt.Key) |
| 271 | } |
| 272 | rt.Key = key |
| 273 | a.runtimeBySessionKey[key] = rt |
| 274 | return true |
| 275 | } |
| 276 | |
| 277 | type sessionRuntimePathTransition struct { |
| 278 | runtime *desktopSessionRuntime |
| 279 | owner *WorkspaceTab |
| 280 | oldKey string |
| 281 | targetKey string |
| 282 | expectedEpoch string |
| 283 | } |
| 284 | |
| 285 | func (a *App) reserveSessionRuntimePath(tab *WorkspaceTab, path string) (sessionRuntimePathTransition, error) { |
| 286 | targetKey := sessionRuntimeKey(path) |
| 287 | if tab == nil || targetKey == "" { |
| 288 | return sessionRuntimePathTransition{}, nil |
| 289 | } |
| 290 | a.mu.Lock() |
| 291 | defer a.mu.Unlock() |
| 292 | if existing := a.runtimeBySessionKey[targetKey]; existing != nil && existing.Owner != tab { |
| 293 | return sessionRuntimePathTransition{}, fmt.Errorf("%w: local runtime already owns session", agent.ErrSessionLeaseHeld) |
| 294 | } |
| 295 | rt := a.runtimeForTabLocked(tab) |
| 296 | if rt == nil { |
| 297 | // A path transition must retain the source identity until commit. Using |
| 298 | // targetKey here would make a failed first rebind forget the still-live |
| 299 | // source controller and its lease. |
| 300 | rt = a.newSessionRuntimeLocked(tab, sessionRuntimeKey(tab.currentSessionPath())) |
| 301 | } |
| 302 | transition := sessionRuntimePathTransition{ |
| 303 | runtime: rt, |
| 304 | owner: tab, |
| 305 | oldKey: rt.Key, |
| 306 | targetKey: targetKey, |
| 307 | expectedEpoch: rt.Epoch, |
| 308 | } |
| 309 | // Keep the old key mapped until the lease rebind succeeds. The target alias |
| 310 | // prevents another local startup from claiming it during the off-lock file |
| 311 | // operation. |
| 312 | a.runtimeBySessionKey[targetKey] = rt |
| 313 | return transition, nil |
| 314 | } |
| 315 | |
| 316 | func (a *App) commitSessionRuntimePath(transition sessionRuntimePathTransition) { |
| 317 | if transition.runtime == nil || transition.targetKey == "" { |
| 318 | return |
| 319 | } |
| 320 | a.mu.Lock() |
| 321 | defer a.mu.Unlock() |
| 322 | rt := transition.runtime |
| 323 | if transition.oldKey != "" && transition.oldKey != transition.targetKey && a.runtimeBySessionKey[transition.oldKey] == rt { |
| 324 | delete(a.runtimeBySessionKey, transition.oldKey) |
| 325 | } |
| 326 | rt.Key = transition.targetKey |
| 327 | a.runtimeBySessionKey[transition.targetKey] = rt |
| 328 | } |
| 329 | |
| 330 | // commitSessionRuntimePathLocked commits a previously reserved path only when |
| 331 | // the same runtime generation still owns both aliases. It lets controller |
| 332 | // swaps make the registry update part of their single App.mu commit. |
| 333 | func (a *App) commitSessionRuntimePathLocked(transition sessionRuntimePathTransition) bool { |
| 334 | if transition.runtime == nil || transition.targetKey == "" { |
| 335 | return false |
| 336 | } |
| 337 | if !a.sessionRuntimePathTransitionValidLocked(transition) { |
| 338 | return false |
| 339 | } |
| 340 | rt := transition.runtime |
| 341 | if transition.oldKey != "" && transition.oldKey != transition.targetKey && a.runtimeBySessionKey[transition.oldKey] == rt { |
| 342 | delete(a.runtimeBySessionKey, transition.oldKey) |
| 343 | } |
| 344 | rt.Key = transition.targetKey |
| 345 | a.runtimeBySessionKey[transition.targetKey] = rt |
| 346 | return true |
| 347 | } |
| 348 | |
| 349 | func (a *App) sessionRuntimePathTransitionValidLocked(transition sessionRuntimePathTransition) bool { |
| 350 | if transition.runtime == nil || transition.targetKey == "" { |
| 351 | return false |
| 352 | } |
| 353 | rt := transition.runtime |
| 354 | if a.runtimeByID[rt.ID] != rt || |
| 355 | rt.Owner != transition.owner || |
| 356 | rt.Epoch != transition.expectedEpoch || |
| 357 | (rt.Key != transition.oldKey && rt.Key != transition.targetKey) || |
| 358 | a.runtimeBySessionKey[transition.targetKey] != rt { |
| 359 | return false |
| 360 | } |
| 361 | return true |
| 362 | } |
| 363 | |
| 364 | func (a *App) rollbackSessionRuntimePath(transition sessionRuntimePathTransition) { |
| 365 | if transition.runtime == nil || transition.targetKey == "" || transition.targetKey == transition.oldKey { |
| 366 | return |
| 367 | } |
| 368 | a.mu.Lock() |
| 369 | if a.runtimeBySessionKey[transition.targetKey] == transition.runtime { |
| 370 | delete(a.runtimeBySessionKey, transition.targetKey) |
| 371 | } |
| 372 | a.mu.Unlock() |
| 373 | } |
| 374 | |
| 375 | // claimSessionRuntime reserves path for tab, or waits for/attaches the existing |
| 376 | // local runtime. The caller owns its not-yet-published candidate controller; a |
| 377 | // true return means that candidate must be closed because tab now uses the |
| 378 | // already registered runtime. |
| 379 | func (a *App) claimSessionRuntime(tab *WorkspaceTab, path string, ctx context.Context) bool { |
| 380 | key := sessionRuntimeKey(path) |
| 381 | if tab == nil || key == "" { |
| 382 | return false |
| 383 | } |
| 384 | for { |
| 385 | a.mu.Lock() |
| 386 | if tab.removed || a.tabs[tab.ID] != tab { |
| 387 | a.mu.Unlock() |
| 388 | return false |
| 389 | } |
| 390 | rt := a.runtimeBySessionKey[key] |
| 391 | if rt != nil && !a.runtimeOwnerLiveLocked(rt) { |
| 392 | a.removeSessionRuntimeMappingsLocked(rt) |
| 393 | rt = nil |
| 394 | } |
| 395 | switch { |
| 396 | case rt == nil: |
| 397 | // Detached runtimes created before the admission registry was |
| 398 | // populated are a compatibility edge. Attach them before claiming |
| 399 | // the key so applyRuntimeTab can publish one authoritative runtime |
| 400 | // instead of leaving behind an unused placeholder. |
| 401 | if detached := a.detachedSessions[key]; detached != nil && detached.Ctrl != nil { |
| 402 | a.mu.Unlock() |
| 403 | return a.attachExistingSessionRuntime(tab, path, a.ctx) |
| 404 | } |
| 405 | a.bindSessionRuntimeKeyLocked(tab, path) |
| 406 | a.mu.Unlock() |
| 407 | return false |
| 408 | case rt.Owner == tab: |
| 409 | a.mu.Unlock() |
| 410 | // The target may own the starting placeholder while a legacy |
| 411 | // visible/detached runtime for the same session predates the |
| 412 | // registry. Let attachExistingSessionRuntime adopt that usable |
| 413 | // controller; otherwise this remains the owner build. |
| 414 | return a.attachExistingSessionRuntime(tab, path, a.ctx) |
| 415 | case rt.Phase == sessionRuntimeStarting && rt.readyCh != nil: |
| 416 | wait := rt.readyCh |
| 417 | a.mu.Unlock() |
| 418 | select { |
| 419 | case <-wait: |
| 420 | continue |
| 421 | case <-ctx.Done(): |
| 422 | return false |
| 423 | case <-time.After(250 * time.Millisecond): |
| 424 | // Re-check owner liveness even when a superseded build exited |
| 425 | // before publishing a terminal phase. |
| 426 | continue |
| 427 | } |
| 428 | default: |
| 429 | a.mu.Unlock() |
| 430 | return a.attachExistingSessionRuntime(tab, path, a.ctx) |
| 431 | } |
| 432 | } |
| 433 | } |
| 434 | |
| 435 | func (a *App) releaseSessionRuntimeLocked(tab *WorkspaceTab) { |
| 436 | rt := a.runtimeForTabLocked(tab) |
| 437 | if rt == nil { |
| 438 | if tab != nil { |
| 439 | tab.runtimeID = "" |
| 440 | } |
| 441 | return |
| 442 | } |
| 443 | rt.Phase = sessionRuntimeClosing |
| 444 | closeRuntimeReadyChannelLocked(rt) |
| 445 | a.removeSessionRuntimeMappingsLocked(rt) |
| 446 | if tab != nil { |
| 447 | tab.runtimeID = "" |
| 448 | } |
| 449 | } |
| 450 | |
| 451 | func sameCurrentProcessLease(err error) bool { |
| 452 | var leaseErr *agent.SessionLeaseError |
| 453 | if !errors.As(err, &leaseErr) || leaseErr == nil || leaseErr.Info == nil { |
| 454 | return false |
| 455 | } |
| 456 | if leaseErr.Info.PID != os.Getpid() || leaseErr.Info.WriterID != agent.SessionWriterID() { |
| 457 | return false |
| 458 | } |
| 459 | host, _ := os.Hostname() |
| 460 | return strings.TrimSpace(leaseErr.Info.Hostname) == strings.TrimSpace(host) |
| 461 | } |
| 462 | |
| 463 | // sessionParentLive reports whether a desktop tab or detached runtime in this |
| 464 | // process currently owns, or is still building, the requested session. It is |
| 465 | // intentionally checked before stale-subagent cleanup probes the durable lease: |
| 466 | // a starting tab has published SessionPath but may not have bound that lease yet. |
| 467 | func (a *App) sessionParentLive(sessionPath string) bool { |
| 468 | return a.sessionParentLiveForBuild(sessionPath, nil) |
| 469 | } |
| 470 | |
| 471 | // subagentParentProbeForBuild excludes an initial build's own unbound tab: that |
| 472 | // build can safely repair its crash leftovers before it binds the session lease. |
| 473 | // Other live tabs remain protected from the sweep. |
| 474 | func (a *App) subagentParentProbeForBuild(building *WorkspaceTab) func(string) bool { |
| 475 | return func(sessionPath string) bool { |
| 476 | return a.sessionParentLiveForBuild(sessionPath, building) |
| 477 | } |
| 478 | } |
| 479 | |
| 480 | func (a *App) sessionParentLiveForBuild(sessionPath string, building *WorkspaceTab) bool { |
| 481 | key := sessionRuntimeKey(sessionPath) |
| 482 | if a == nil || key == "" { |
| 483 | return false |
| 484 | } |
| 485 | |
| 486 | a.mu.RLock() |
| 487 | defer a.mu.RUnlock() |
| 488 | if rt := a.runtimeBySessionKey[key]; rt != nil && a.runtimeOwnerLiveLocked(rt) && |
| 489 | !(rt.Owner == building && building.Ctrl == nil) { |
| 490 | return true |
| 491 | } |
| 492 | liveTab := func(tab *WorkspaceTab) bool { |
| 493 | if tab == nil { |
| 494 | return false |
| 495 | } |
| 496 | if tab == building && tab.Ctrl == nil { |
| 497 | return false |
| 498 | } |
| 499 | if sessionRuntimeKey(tab.SessionPath) == key { |
| 500 | return true |
| 501 | } |
| 502 | return tab.Ctrl != nil && sessionRuntimeKey(tab.Ctrl.SessionPath()) == key |
| 503 | } |
| 504 | for _, tab := range a.tabs { |
| 505 | if liveTab(tab) { |
| 506 | return true |
| 507 | } |
| 508 | } |
| 509 | for _, tab := range a.detachedSessions { |
| 510 | if liveTab(tab) { |
| 511 | return true |
| 512 | } |
| 513 | } |
| 514 | return false |
| 515 | } |
| 516 |