| 1 | package main |
| 2 | |
| 3 | import ( |
| 4 | "crypto/rand" |
| 5 | "encoding/hex" |
| 6 | "encoding/json" |
| 7 | "os" |
| 8 | "path/filepath" |
| 9 | "sort" |
| 10 | "strconv" |
| 11 | "strings" |
| 12 | "sync" |
| 13 | "time" |
| 14 | |
| 15 | "reasonix/internal/config" |
| 16 | "reasonix/internal/fileutil" |
| 17 | filelock "reasonix/internal/identitylock" |
| 18 | "reasonix/internal/repair" |
| 19 | ) |
| 20 | |
| 21 | const ( |
| 22 | desktopLifecycleSchemaVersion = 3 |
| 23 | desktopLifecycleRetention = 30 * 24 * time.Hour |
| 24 | maxDesktopLifecycleRecords = 20 |
| 25 | desktopLifecycleReadRetries = 12 |
| 26 | desktopLifecycleReadBackoff = 20 * time.Millisecond |
| 27 | ) |
| 28 | |
| 29 | type desktopLifecycleState struct { |
| 30 | SchemaVersion int `json:"schemaVersion"` |
| 31 | PID int `json:"pid"` |
| 32 | RunID string `json:"runId"` |
| 33 | IncidentID string `json:"incidentId,omitempty"` |
| 34 | Version string `json:"version,omitempty"` |
| 35 | BuildCommit string `json:"buildCommit,omitempty"` |
| 36 | Channel string `json:"channel,omitempty"` |
| 37 | Phase string `json:"phase"` |
| 38 | TerminationReason string `json:"terminationReason,omitempty"` |
| 39 | CleanupOutcome string `json:"cleanupOutcome,omitempty"` |
| 40 | ProcessRole string `json:"processRole,omitempty"` |
| 41 | StartedAt string `json:"startedAt"` |
| 42 | UpdatedAt string `json:"updatedAt"` |
| 43 | } |
| 44 | |
| 45 | type desktopLifecycleObservation struct { |
| 46 | RunID string |
| 47 | IncidentID string |
| 48 | Version string |
| 49 | BuildCommit string |
| 50 | Channel string |
| 51 | Phase string |
| 52 | TerminationReason string |
| 53 | CleanupOutcome string |
| 54 | ProcessRole string |
| 55 | StartedAt string |
| 56 | UpdatedAt string |
| 57 | claimedPath string |
| 58 | originalPath string |
| 59 | } |
| 60 | |
| 61 | type desktopLifecycleRuntime struct { |
| 62 | previousRun repair.PreviousRunObservation |
| 63 | previousRuns []desktopLifecycleObservation |
| 64 | tracker *desktopLifecycleTracker |
| 65 | } |
| 66 | |
| 67 | type desktopLifecycleTracker struct { |
| 68 | mu sync.Mutex |
| 69 | dir string |
| 70 | path string |
| 71 | state desktopLifecycleState |
| 72 | now func() time.Time |
| 73 | processAlive func(int) bool |
| 74 | updates chan string |
| 75 | writerStop chan struct{} |
| 76 | writerDone chan struct{} |
| 77 | writerOnce sync.Once |
| 78 | stopOnce sync.Once |
| 79 | writerActive bool |
| 80 | } |
| 81 | |
| 82 | func newDesktopLifecycleTracker(root, appVersion, appChannel string) *desktopLifecycleTracker { |
| 83 | now := time.Now().UTC() |
| 84 | runID := newDesktopLifecycleRunID() |
| 85 | dir := filepath.Join(root, "diagnostics", "lifecycle") |
| 86 | return &desktopLifecycleTracker{ |
| 87 | dir: dir, |
| 88 | path: filepath.Join(dir, strconv.Itoa(os.Getpid())+"-"+runID+".json"), |
| 89 | state: desktopLifecycleState{ |
| 90 | SchemaVersion: desktopLifecycleSchemaVersion, |
| 91 | PID: os.Getpid(), |
| 92 | RunID: runID, |
| 93 | IncidentID: newDesktopLifecycleRunID(), |
| 94 | Version: appVersion, |
| 95 | BuildCommit: buildCommit(), |
| 96 | ProcessRole: "service", |
| 97 | Channel: appChannel, |
| 98 | Phase: "starting", |
| 99 | StartedAt: now.Format(time.RFC3339Nano), |
| 100 | UpdatedAt: now.Format(time.RFC3339Nano), |
| 101 | }, |
| 102 | now: func() time.Time { return time.Now().UTC() }, |
| 103 | processAlive: desktopProcessAlive, |
| 104 | updates: make(chan string, 1), |
| 105 | writerStop: make(chan struct{}), |
| 106 | writerDone: make(chan struct{}), |
| 107 | } |
| 108 | } |
| 109 | |
| 110 | func prepareDesktopDiagnostics(app *App) { |
| 111 | if app == nil || version == "dev" { |
| 112 | return |
| 113 | } |
| 114 | root := config.MemoryUserDir() |
| 115 | if root == "" { |
| 116 | return |
| 117 | } |
| 118 | diagnosticsDir := filepath.Join(root, "diagnostics") |
| 119 | if err := os.MkdirAll(diagnosticsDir, 0o700); err != nil { |
| 120 | return |
| 121 | } |
| 122 | release, err := filelock.TryAcquire(filepath.Join(diagnosticsDir, "primary.lock")) |
| 123 | if err != nil { |
| 124 | return |
| 125 | } |
| 126 | app.diagnosticsOwner = true |
| 127 | app.diagnosticsOwnerRelease = release |
| 128 | |
| 129 | cfg, err := config.Load() |
| 130 | if err != nil { |
| 131 | return |
| 132 | } |
| 133 | app.diagnosticsConfigLoaded = true |
| 134 | if !cfg.DesktopTelemetry() { |
| 135 | return |
| 136 | } |
| 137 | app.diagnosticsTelemetry = true |
| 138 | tracker := newDesktopLifecycleTracker(root, version, channel) |
| 139 | if tracker.start() == nil { |
| 140 | app.lifecycle.tracker = tracker |
| 141 | } |
| 142 | } |
| 143 | |
| 144 | func (a *App) releaseDesktopDiagnosticsOwnership() { |
| 145 | if a == nil || a.diagnosticsOwnerRelease == nil { |
| 146 | return |
| 147 | } |
| 148 | a.diagnosticsOwnerRelease() |
| 149 | a.diagnosticsOwnerRelease = nil |
| 150 | a.diagnosticsOwner = false |
| 151 | a.diagnosticsConfigLoaded = false |
| 152 | a.diagnosticsTelemetry = false |
| 153 | } |
| 154 | |
| 155 | func initializeLifecycleDiagnostics(app *App) { |
| 156 | if app == nil { |
| 157 | return |
| 158 | } |
| 159 | if !app.diagnosticsOwner { |
| 160 | return |
| 161 | } |
| 162 | if !app.diagnosticsConfigLoaded || version == "dev" { |
| 163 | return |
| 164 | } |
| 165 | tracker := app.lifecycle.tracker |
| 166 | if tracker == nil { |
| 167 | tracker = newDesktopLifecycleTracker(config.MemoryUserDir(), version, channel) |
| 168 | } |
| 169 | enabled := app.diagnosticsTelemetry |
| 170 | legacy := repair.NewStartupTracker("").ObservePreviousRun() |
| 171 | if enabled { |
| 172 | app.lifecycle.previousRun = legacy |
| 173 | } |
| 174 | app.lifecycle.previousRuns = tracker.consumePrevious(enabled) |
| 175 | } |
| 176 | |
| 177 | func (a *App) markDesktopHealthy() { |
| 178 | a.startupReady.Store(true) |
| 179 | a.lifecycle.tracker.markAsync("healthy") |
| 180 | } |
| 181 | |
| 182 | func newDesktopLifecycleRunID() string { |
| 183 | var raw [8]byte |
| 184 | if _, err := rand.Read(raw[:]); err == nil { |
| 185 | return hex.EncodeToString(raw[:]) |
| 186 | } |
| 187 | return strconv.FormatInt(time.Now().UTC().UnixNano(), 16) |
| 188 | } |
| 189 | |
| 190 | func (t *desktopLifecycleTracker) start() error { |
| 191 | if t == nil || t.path == "" { |
| 192 | return nil |
| 193 | } |
| 194 | if err := t.writeState(); err != nil { |
| 195 | return err |
| 196 | } |
| 197 | t.writerOnce.Do(func() { |
| 198 | t.mu.Lock() |
| 199 | t.writerActive = true |
| 200 | t.mu.Unlock() |
| 201 | go t.runWriter() |
| 202 | }) |
| 203 | return nil |
| 204 | } |
| 205 | |
| 206 | func (t *desktopLifecycleTracker) runWriter() { |
| 207 | defer close(t.writerDone) |
| 208 | for { |
| 209 | select { |
| 210 | case phase := <-t.updates: |
| 211 | t.mark(phase) |
| 212 | case <-t.writerStop: |
| 213 | select { |
| 214 | case phase := <-t.updates: |
| 215 | t.mark(phase) |
| 216 | default: |
| 217 | } |
| 218 | return |
| 219 | } |
| 220 | } |
| 221 | } |
| 222 | |
| 223 | // markAsync keeps normal startup and DOM-ready paths free from diagnostic I/O. |
| 224 | // The single-slot queue is last-state-wins because lifecycle phases are |
| 225 | // monotonic and only the newest pending phase is useful. |
| 226 | func (t *desktopLifecycleTracker) markAsync(phase string) { |
| 227 | if t == nil || strings.TrimSpace(phase) == "" { |
| 228 | return |
| 229 | } |
| 230 | select { |
| 231 | case <-t.writerStop: |
| 232 | return |
| 233 | default: |
| 234 | } |
| 235 | select { |
| 236 | case t.updates <- phase: |
| 237 | return |
| 238 | default: |
| 239 | } |
| 240 | select { |
| 241 | case <-t.updates: |
| 242 | default: |
| 243 | } |
| 244 | select { |
| 245 | case t.updates <- phase: |
| 246 | default: |
| 247 | } |
| 248 | } |
| 249 | |
| 250 | func (t *desktopLifecycleTracker) incidentID() string { |
| 251 | if t == nil { |
| 252 | return "" |
| 253 | } |
| 254 | t.mu.Lock() |
| 255 | defer t.mu.Unlock() |
| 256 | return t.state.IncidentID |
| 257 | } |
| 258 | |
| 259 | func (t *desktopLifecycleTracker) stopWriter() { |
| 260 | if t == nil { |
| 261 | return |
| 262 | } |
| 263 | t.mu.Lock() |
| 264 | active := t.writerActive |
| 265 | t.mu.Unlock() |
| 266 | if !active { |
| 267 | return |
| 268 | } |
| 269 | t.stopOnce.Do(func() { close(t.writerStop) }) |
| 270 | timer := time.NewTimer(250 * time.Millisecond) |
| 271 | defer timer.Stop() |
| 272 | select { |
| 273 | case <-t.writerDone: |
| 274 | case <-timer.C: |
| 275 | } |
| 276 | } |
| 277 | |
| 278 | func (t *desktopLifecycleTracker) mark(phase string) { |
| 279 | if t == nil || t.path == "" || strings.TrimSpace(phase) == "" { |
| 280 | return |
| 281 | } |
| 282 | t.mu.Lock() |
| 283 | defer t.mu.Unlock() |
| 284 | t.state.Phase = phase |
| 285 | t.state.UpdatedAt = t.now().Format(time.RFC3339Nano) |
| 286 | _ = t.writeStateLocked() |
| 287 | } |
| 288 | |
| 289 | func (t *desktopLifecycleTracker) markShutdown(reason, phase, outcome string) { |
| 290 | if t == nil || t.path == "" { |
| 291 | return |
| 292 | } |
| 293 | t.stopWriter() |
| 294 | t.mu.Lock() |
| 295 | defer t.mu.Unlock() |
| 296 | t.state.TerminationReason = normalizeShutdownReason(reason) |
| 297 | t.state.CleanupOutcome = strings.TrimSpace(outcome) |
| 298 | if strings.TrimSpace(phase) != "" { |
| 299 | t.state.Phase = strings.TrimSpace(phase) |
| 300 | } |
| 301 | t.state.UpdatedAt = t.now().Format(time.RFC3339Nano) |
| 302 | _ = t.writeStateLocked() |
| 303 | } |
| 304 | |
| 305 | func (t *desktopLifecycleTracker) clean() { |
| 306 | if t == nil || t.path == "" { |
| 307 | return |
| 308 | } |
| 309 | t.stopWriter() |
| 310 | t.mu.Lock() |
| 311 | defer t.mu.Unlock() |
| 312 | _ = os.Remove(t.path) |
| 313 | } |
| 314 | |
| 315 | func (t *desktopLifecycleTracker) writeState() error { |
| 316 | t.mu.Lock() |
| 317 | defer t.mu.Unlock() |
| 318 | return t.writeStateLocked() |
| 319 | } |
| 320 | |
| 321 | func (t *desktopLifecycleTracker) writeStateLocked() error { |
| 322 | body, err := json.Marshal(t.state) |
| 323 | if err != nil { |
| 324 | return err |
| 325 | } |
| 326 | return fileutil.AtomicWriteFile(t.path, body, 0o600) |
| 327 | } |
| 328 | |
| 329 | // consumePrevious atomically owns every dead per-process record before |
| 330 | // returning it. Emitted records remain claimed until their pending event is |
| 331 | // durably written; finalizeObservation then deletes or restores the evidence. |
| 332 | func (t *desktopLifecycleTracker) consumePrevious(emit bool) []desktopLifecycleObservation { |
| 333 | if t == nil || t.dir == "" { |
| 334 | return nil |
| 335 | } |
| 336 | entries, err := os.ReadDir(t.dir) |
| 337 | if err != nil { |
| 338 | return nil |
| 339 | } |
| 340 | now := t.now() |
| 341 | observations := make([]desktopLifecycleObservation, 0) |
| 342 | for _, entry := range entries { |
| 343 | if entry.IsDir() || filepath.Ext(entry.Name()) != ".json" { |
| 344 | continue |
| 345 | } |
| 346 | path := filepath.Join(t.dir, entry.Name()) |
| 347 | state, readErr := readDesktopLifecycleState(path) |
| 348 | if readErr != nil { |
| 349 | if info, statErr := entry.Info(); statErr == nil && now.Sub(info.ModTime()) > desktopLifecycleRetention { |
| 350 | _ = os.Remove(path) |
| 351 | } |
| 352 | continue |
| 353 | } |
| 354 | // A newer Desktop may own a lifecycle schema this version cannot safely |
| 355 | // interpret. Preserve it verbatim so a downgrade never consumes or prunes |
| 356 | // future-format evidence. |
| 357 | if state.SchemaVersion != 2 && state.SchemaVersion != desktopLifecycleSchemaVersion { |
| 358 | continue |
| 359 | } |
| 360 | if state.PID <= 0 || state.Phase == "" { |
| 361 | if info, statErr := entry.Info(); statErr == nil && now.Sub(info.ModTime()) > desktopLifecycleRetention { |
| 362 | _ = os.Remove(path) |
| 363 | } |
| 364 | continue |
| 365 | } |
| 366 | if t.processAlive(state.PID) { |
| 367 | continue |
| 368 | } |
| 369 | claimed := path + ".claimed-" + t.state.RunID |
| 370 | // Windows fails a bare rename with a sharing violation while any other |
| 371 | // instance still holds the record open, so without the retry every |
| 372 | // claimant can lose the same race and the evidence is dropped by all. |
| 373 | if err := fileutil.ClaimRename(path, claimed); err != nil { |
| 374 | continue |
| 375 | } |
| 376 | state, readErr = readClaimedLifecycleState(claimed) |
| 377 | if readErr != nil || (state.SchemaVersion != 2 && state.SchemaVersion != desktopLifecycleSchemaVersion) { |
| 378 | // The file changed between inspection and claim. Put it back when |
| 379 | // possible instead of deleting data that may belong to another schema. |
| 380 | _ = os.Rename(claimed, path) |
| 381 | continue |
| 382 | } |
| 383 | if !emit { |
| 384 | _ = os.Remove(claimed) |
| 385 | continue |
| 386 | } |
| 387 | observations = append(observations, desktopLifecycleObservation{ |
| 388 | RunID: state.RunID, IncidentID: state.IncidentID, |
| 389 | Version: state.Version, BuildCommit: state.BuildCommit, Channel: state.Channel, Phase: state.Phase, |
| 390 | TerminationReason: state.TerminationReason, CleanupOutcome: state.CleanupOutcome, |
| 391 | ProcessRole: state.ProcessRole, |
| 392 | StartedAt: state.StartedAt, UpdatedAt: state.UpdatedAt, |
| 393 | claimedPath: claimed, originalPath: path, |
| 394 | }) |
| 395 | } |
| 396 | t.pruneRecords() |
| 397 | return observations |
| 398 | } |
| 399 | |
| 400 | func (t *desktopLifecycleTracker) finalizeObservation(observation desktopLifecycleObservation, persisted bool) { |
| 401 | if observation.claimedPath == "" { |
| 402 | return |
| 403 | } |
| 404 | if persisted { |
| 405 | _ = os.Remove(observation.claimedPath) |
| 406 | return |
| 407 | } |
| 408 | if observation.originalPath != "" { |
| 409 | _ = os.Rename(observation.claimedPath, observation.originalPath) |
| 410 | } |
| 411 | } |
| 412 | |
| 413 | // readClaimedLifecycleState re-reads a just-claimed record. Winning the rename |
| 414 | // does not make the file readable yet: an instance that was inspecting it still |
| 415 | // holds a handle, and Windows answers with a sharing violation until it drops. |
| 416 | // Only the open is retried — content this build cannot parse is a definite |
| 417 | // answer about the record, not a lock waiting to clear. |
| 418 | func readClaimedLifecycleState(path string) (desktopLifecycleState, error) { |
| 419 | var body []byte |
| 420 | var err error |
| 421 | for attempt := range desktopLifecycleReadRetries { |
| 422 | if body, err = os.ReadFile(path); err == nil || os.IsNotExist(err) { |
| 423 | break |
| 424 | } |
| 425 | time.Sleep(time.Duration(attempt+1) * desktopLifecycleReadBackoff) |
| 426 | } |
| 427 | if err != nil { |
| 428 | return desktopLifecycleState{}, err |
| 429 | } |
| 430 | var state desktopLifecycleState |
| 431 | err = json.Unmarshal(body, &state) |
| 432 | return state, err |
| 433 | } |
| 434 | |
| 435 | func readDesktopLifecycleState(path string) (desktopLifecycleState, error) { |
| 436 | body, err := os.ReadFile(path) |
| 437 | if err != nil { |
| 438 | return desktopLifecycleState{}, err |
| 439 | } |
| 440 | var state desktopLifecycleState |
| 441 | if err := json.Unmarshal(body, &state); err != nil { |
| 442 | return desktopLifecycleState{}, err |
| 443 | } |
| 444 | return state, nil |
| 445 | } |
| 446 | |
| 447 | func (t *desktopLifecycleTracker) pruneRecords() { |
| 448 | entries, err := os.ReadDir(t.dir) |
| 449 | if err != nil || len(entries) <= maxDesktopLifecycleRecords { |
| 450 | return |
| 451 | } |
| 452 | type candidate struct { |
| 453 | path string |
| 454 | at time.Time |
| 455 | } |
| 456 | candidates := make([]candidate, 0, len(entries)) |
| 457 | for _, entry := range entries { |
| 458 | if entry.IsDir() || filepath.Ext(entry.Name()) != ".json" { |
| 459 | continue |
| 460 | } |
| 461 | path := filepath.Join(t.dir, entry.Name()) |
| 462 | state, err := readDesktopLifecycleState(path) |
| 463 | if err != nil || state.SchemaVersion != desktopLifecycleSchemaVersion || t.processAlive(state.PID) { |
| 464 | continue |
| 465 | } |
| 466 | if info, err := entry.Info(); err == nil { |
| 467 | candidates = append(candidates, candidate{path: path, at: info.ModTime()}) |
| 468 | } |
| 469 | } |
| 470 | sort.Slice(candidates, func(i, j int) bool { return candidates[i].at.Before(candidates[j].at) }) |
| 471 | for len(candidates) > maxDesktopLifecycleRecords { |
| 472 | _ = os.Remove(candidates[0].path) |
| 473 | candidates = candidates[1:] |
| 474 | } |
| 475 | } |
| 476 |