| 1 | package taskmonitor |
| 2 | |
| 3 | import ( |
| 4 | "context" |
| 5 | "encoding/json" |
| 6 | "errors" |
| 7 | "fmt" |
| 8 | "io" |
| 9 | "os" |
| 10 | "path/filepath" |
| 11 | "sort" |
| 12 | "strings" |
| 13 | "time" |
| 14 | |
| 15 | "reasonix/internal/fileutil" |
| 16 | ) |
| 17 | |
| 18 | // FileStore is a Store backed by a JSON file tree under a project-local |
| 19 | // directory. Tasks are stored as <dir>/<task-id>/snapshot.json and |
| 20 | // <dir>/<task-id>/events.jsonl. It is read-only in TM-02; write support |
| 21 | // is added in TM-04. |
| 22 | type FileStore struct { |
| 23 | baseDir string // projectDir → task data root (e.g. ".reasonix/tasks") |
| 24 | sink ProjectionSink |
| 25 | } |
| 26 | |
| 27 | // NewFileStore returns a FileStore rooted at baseDir. baseDir is typically |
| 28 | // ".reasonix/tasks" relative to the project root. |
| 29 | func NewFileStore(baseDir string) *FileStore { |
| 30 | return &FileStore{baseDir: baseDir} |
| 31 | } |
| 32 | |
| 33 | func NewObservedFileStore(baseDir string, sink ProjectionSink) *FileStore { |
| 34 | return &FileStore{baseDir: baseDir, sink: sink} |
| 35 | } |
| 36 | |
| 37 | // safeID validates a user-supplied identifier for use as a filesystem path |
| 38 | // component. It rejects empty strings, ".", "..", and values containing a |
| 39 | // path separator. Used for both taskID and idempotency keys. |
| 40 | func safeID(name string) (string, error) { |
| 41 | if name == "" { |
| 42 | return "", errors.New("identifier must not be empty") |
| 43 | } |
| 44 | cleaned := filepath.Base(name) |
| 45 | if cleaned == "." || cleaned == ".." { |
| 46 | return "", fmt.Errorf("invalid identifier %q", name) |
| 47 | } |
| 48 | // Windows accepts both slash styles as path separators. Check both so |
| 49 | // validation has the same traversal behavior on every platform. |
| 50 | if strings.ContainsAny(name, `/\\`) { |
| 51 | return "", fmt.Errorf("identifier %q contains path separator", name) |
| 52 | } |
| 53 | return cleaned, nil |
| 54 | } |
| 55 | |
| 56 | // taskRoot returns the cleaned directory holding task data for projectDir. |
| 57 | // projectDir is the caller-selected project scope, not a path relative to a |
| 58 | // separate containment root. Parent-relative paths such as ../project and |
| 59 | // directory names containing ".." are therefore valid inputs. |
| 60 | func (s *FileStore) taskRoot(projectDir string) (string, error) { |
| 61 | if projectDir == "" { |
| 62 | projectDir = "." |
| 63 | } |
| 64 | cleaned := filepath.Clean(projectDir) |
| 65 | root := filepath.Join(cleaned, s.baseDir) |
| 66 | if err := rejectStoreParents(cleaned, root); err != nil { |
| 67 | return "", err |
| 68 | } |
| 69 | return root, nil |
| 70 | } |
| 71 | |
| 72 | func rejectSymlink(path string) error { |
| 73 | info, err := os.Lstat(path) |
| 74 | if err != nil { |
| 75 | if os.IsNotExist(err) { |
| 76 | return nil |
| 77 | } |
| 78 | return err |
| 79 | } |
| 80 | if info.Mode()&os.ModeSymlink != 0 { |
| 81 | return fmt.Errorf("task store path %q is a symlink", path) |
| 82 | } |
| 83 | return nil |
| 84 | } |
| 85 | |
| 86 | // rejectSymlinkChain rejects symlinks in the store path itself and all of its |
| 87 | // descendants up to target. This keeps a project-local task id from redirecting |
| 88 | // reads or writes outside the project through an intermediate directory. |
| 89 | func rejectSymlinkChain(root, target string) error { |
| 90 | rel, err := filepath.Rel(root, target) |
| 91 | if err != nil { |
| 92 | return err |
| 93 | } |
| 94 | cur := root |
| 95 | if err := rejectSymlink(cur); err != nil { |
| 96 | return err |
| 97 | } |
| 98 | if rel == "." { |
| 99 | return nil |
| 100 | } |
| 101 | for part := range strings.SplitSeq(rel, string(filepath.Separator)) { |
| 102 | cur = filepath.Join(cur, part) |
| 103 | if err := rejectSymlink(cur); err != nil { |
| 104 | return err |
| 105 | } |
| 106 | } |
| 107 | return nil |
| 108 | } |
| 109 | |
| 110 | func rejectStoreParents(projectDir, root string) error { |
| 111 | rel, err := filepath.Rel(projectDir, root) |
| 112 | if err != nil { |
| 113 | return err |
| 114 | } |
| 115 | cur := projectDir |
| 116 | for part := range strings.SplitSeq(rel, string(filepath.Separator)) { |
| 117 | if part == "." || part == "" { |
| 118 | continue |
| 119 | } |
| 120 | cur = filepath.Join(cur, part) |
| 121 | if err := rejectSymlink(cur); err != nil { |
| 122 | return err |
| 123 | } |
| 124 | } |
| 125 | return nil |
| 126 | } |
| 127 | |
| 128 | func prepareTaskDir(root, id string) (string, error) { |
| 129 | taskDir := filepath.Join(root, id) |
| 130 | if err := rejectSymlinkChain(root, taskDir); err != nil { |
| 131 | return "", err |
| 132 | } |
| 133 | if err := os.MkdirAll(root, 0o700); err != nil { |
| 134 | return "", err |
| 135 | } |
| 136 | if err := os.Chmod(root, 0o700); err != nil { |
| 137 | return "", err |
| 138 | } |
| 139 | if err := os.MkdirAll(taskDir, 0o700); err != nil { |
| 140 | return "", err |
| 141 | } |
| 142 | if err := os.Chmod(taskDir, 0o700); err != nil { |
| 143 | return "", err |
| 144 | } |
| 145 | return taskDir, nil |
| 146 | } |
| 147 | |
| 148 | // ListTasks implements Store. |
| 149 | func (s *FileStore) ListTasks(ctx context.Context, projectDir string) ([]TaskSnapshot, error) { |
| 150 | if err := ctx.Err(); err != nil { |
| 151 | return nil, err |
| 152 | } |
| 153 | root, err := s.taskRoot(projectDir) |
| 154 | if err != nil { |
| 155 | return nil, err |
| 156 | } |
| 157 | if err := rejectSymlink(root); err != nil { |
| 158 | return nil, err |
| 159 | } |
| 160 | entries, err := os.ReadDir(root) |
| 161 | if err != nil { |
| 162 | if os.IsNotExist(err) { |
| 163 | return []TaskSnapshot{}, nil |
| 164 | } |
| 165 | return nil, fmt.Errorf("read task dir %s: %w", root, err) |
| 166 | } |
| 167 | result := make([]TaskSnapshot, 0) |
| 168 | for _, e := range entries { |
| 169 | if !e.IsDir() { |
| 170 | continue |
| 171 | } |
| 172 | taskDir := filepath.Join(root, e.Name()) |
| 173 | if err := rejectSymlinkChain(root, taskDir); err != nil { |
| 174 | continue |
| 175 | } |
| 176 | snap, err := s.readSnapshot(taskDir) |
| 177 | if err != nil { |
| 178 | continue // skip corrupt entries |
| 179 | } |
| 180 | reconcileRuntime(&snap, timeNow()) |
| 181 | result = append(result, snap) |
| 182 | } |
| 183 | sort.Slice(result, func(i, j int) bool { |
| 184 | return result[i].UpdatedAt.After(result[j].UpdatedAt) |
| 185 | }) |
| 186 | return result, nil |
| 187 | } |
| 188 | |
| 189 | // GetTask implements Store. |
| 190 | func (s *FileStore) GetTask(ctx context.Context, projectDir string, taskID string) (*TaskSnapshot, error) { |
| 191 | snap, err := s.getTaskRaw(ctx, projectDir, taskID) |
| 192 | if snap != nil { |
| 193 | reconcileRuntime(snap, timeNow()) |
| 194 | } |
| 195 | return snap, err |
| 196 | } |
| 197 | |
| 198 | // getTaskRaw returns the persisted snapshot without applying observer-side |
| 199 | // runtime lease reconciliation. Runtime owners use this path when renewing a |
| 200 | // lease after process suspension or system sleep. |
| 201 | func (s *FileStore) getTaskRaw(ctx context.Context, projectDir string, taskID string) (*TaskSnapshot, error) { |
| 202 | if err := ctx.Err(); err != nil { |
| 203 | return nil, err |
| 204 | } |
| 205 | id, err := safeID(taskID) |
| 206 | if err != nil { |
| 207 | return nil, err |
| 208 | } |
| 209 | root, err := s.taskRoot(projectDir) |
| 210 | if err != nil { |
| 211 | return nil, err |
| 212 | } |
| 213 | if err := rejectSymlinkChain(root, filepath.Join(root, id)); err != nil { |
| 214 | return nil, err |
| 215 | } |
| 216 | snap, err := s.readSnapshot(filepath.Join(root, id)) |
| 217 | if err != nil { |
| 218 | if os.IsNotExist(err) { |
| 219 | return nil, nil |
| 220 | } |
| 221 | return nil, err |
| 222 | } |
| 223 | return &snap, nil |
| 224 | } |
| 225 | |
| 226 | // RenewRuntimeLease implements WriteStore. The raw read plus SaveTask CAS |
| 227 | // ensures a delayed owner cannot overwrite a concurrent control/completion |
| 228 | // update or renew a newer recorder generation. |
| 229 | func (s *FileStore) RenewRuntimeLease(ctx context.Context, projectDir, taskID, ownerID string, leaseUntil time.Time) (bool, error) { |
| 230 | if ownerID == "" || leaseUntil.IsZero() { |
| 231 | return false, nil |
| 232 | } |
| 233 | const maxAttempts = 4 |
| 234 | for range maxAttempts { |
| 235 | snap, err := s.getTaskRaw(ctx, projectDir, taskID) |
| 236 | if err != nil || snap == nil { |
| 237 | return false, err |
| 238 | } |
| 239 | if snap.RuntimeOwnerID != ownerID || snap.State.Terminal() || snap.RuntimeState.Effective() != RuntimeStateAlive { |
| 240 | return false, nil |
| 241 | } |
| 242 | snap.Version++ |
| 243 | snap.RuntimeLeaseUntil = leaseUntil |
| 244 | if err := s.SaveTask(ctx, projectDir, *snap); err == nil { |
| 245 | return true, nil |
| 246 | } else if !errors.Is(err, ErrStoreVersionConflict) { |
| 247 | return false, err |
| 248 | } |
| 249 | } |
| 250 | return false, ErrStoreVersionConflict |
| 251 | } |
| 252 | |
| 253 | // ListEvents implements Store. |
| 254 | func (s *FileStore) ListEvents(ctx context.Context, projectDir string, taskID string, afterSequence int) ([]TaskEvent, error) { |
| 255 | if err := ctx.Err(); err != nil { |
| 256 | return nil, err |
| 257 | } |
| 258 | id, err := safeID(taskID) |
| 259 | if err != nil { |
| 260 | return nil, err |
| 261 | } |
| 262 | root, err := s.taskRoot(projectDir) |
| 263 | if err != nil { |
| 264 | return nil, err |
| 265 | } |
| 266 | if err := rejectSymlinkChain(root, filepath.Join(root, id)); err != nil { |
| 267 | return nil, err |
| 268 | } |
| 269 | events, err := s.readEvents(filepath.Join(root, id)) |
| 270 | if err != nil { |
| 271 | if os.IsNotExist(err) { |
| 272 | return []TaskEvent{}, nil |
| 273 | } |
| 274 | return nil, err |
| 275 | } |
| 276 | result := make([]TaskEvent, 0) |
| 277 | for _, e := range events { |
| 278 | if e.Sequence > afterSequence { |
| 279 | result = append(result, e) |
| 280 | } |
| 281 | } |
| 282 | sort.Slice(result, func(i, j int) bool { |
| 283 | return result[i].Sequence < result[j].Sequence |
| 284 | }) |
| 285 | return result, nil |
| 286 | } |
| 287 | |
| 288 | func (s *FileStore) readSnapshot(taskDir string) (TaskSnapshot, error) { |
| 289 | if err := rejectSymlink(filepath.Join(taskDir, "snapshot.json")); err != nil { |
| 290 | return TaskSnapshot{}, err |
| 291 | } |
| 292 | data, err := os.ReadFile(filepath.Join(taskDir, "snapshot.json")) |
| 293 | if err != nil { |
| 294 | return TaskSnapshot{}, err |
| 295 | } |
| 296 | var snap TaskSnapshot |
| 297 | if err := json.Unmarshal(data, &snap); err != nil { |
| 298 | return TaskSnapshot{}, fmt.Errorf("parse snapshot: %w", err) |
| 299 | } |
| 300 | return snap, nil |
| 301 | } |
| 302 | |
| 303 | func (s *FileStore) readEvents(taskDir string) ([]TaskEvent, error) { |
| 304 | if err := rejectSymlink(filepath.Join(taskDir, "events.jsonl")); err != nil { |
| 305 | return nil, err |
| 306 | } |
| 307 | data, err := os.ReadFile(filepath.Join(taskDir, "events.jsonl")) |
| 308 | if err != nil { |
| 309 | return nil, err |
| 310 | } |
| 311 | // JSONL: one JSON object per line |
| 312 | var events []TaskEvent |
| 313 | raw := string(data) |
| 314 | for raw != "" { |
| 315 | idx := 0 |
| 316 | // find newline |
| 317 | for idx < len(raw) && raw[idx] != '\n' { |
| 318 | idx++ |
| 319 | } |
| 320 | line := raw[:idx] |
| 321 | raw = raw[idx:] |
| 322 | if len(raw) > 0 { |
| 323 | raw = raw[1:] // skip newline |
| 324 | } |
| 325 | if line == "" { |
| 326 | continue |
| 327 | } |
| 328 | var ev TaskEvent |
| 329 | if err := json.Unmarshal([]byte(line), &ev); err != nil { |
| 330 | continue // skip corrupt lines |
| 331 | } |
| 332 | events = append(events, ev) |
| 333 | } |
| 334 | return events, nil |
| 335 | } |
| 336 | |
| 337 | // SaveTask implements WriteStore. It atomically writes the snapshot, |
| 338 | // failing if a concurrent write has changed the version. |
| 339 | func (s *FileStore) SaveTask(ctx context.Context, projectDir string, snap TaskSnapshot) (retErr error) { |
| 340 | committed := false |
| 341 | defer func() { |
| 342 | if committed && s.sink != nil { |
| 343 | s.sink.SnapshotChanged(projectDir, snap.TaskID) |
| 344 | } |
| 345 | }() |
| 346 | if err := ctx.Err(); err != nil { |
| 347 | return err |
| 348 | } |
| 349 | id, err := safeID(snap.TaskID) |
| 350 | if err != nil { |
| 351 | return err |
| 352 | } |
| 353 | root, err := s.taskRoot(projectDir) |
| 354 | if err != nil { |
| 355 | return err |
| 356 | } |
| 357 | taskDir, err := prepareTaskDir(root, id) |
| 358 | if err != nil { |
| 359 | return fmt.Errorf("save task: %w", err) |
| 360 | } |
| 361 | |
| 362 | // Cross-process CAS: hold the per-task lock while reading the current |
| 363 | // version and replacing snapshot.json, so two writers (CLI + Desktop, |
| 364 | // or two control operations) cannot both pass the version check and |
| 365 | // clobber each other. A dedicated lock file is used — never snapshot.json |
| 366 | // itself, since rename swaps the inode and would orphan the lock. |
| 367 | lockPath := filepath.Join(taskDir, "task.lock") |
| 368 | if err := rejectSymlink(lockPath); err != nil { |
| 369 | return fmt.Errorf("save task: %w", err) |
| 370 | } |
| 371 | lf, err := os.OpenFile(lockPath, os.O_CREATE|os.O_RDWR, 0o600) |
| 372 | if err != nil { |
| 373 | return fmt.Errorf("save task: open lock: %w", err) |
| 374 | } |
| 375 | defer lf.Close() |
| 376 | if err := lockTaskFile(lf); err != nil { |
| 377 | return fmt.Errorf("save task: lock: %w", err) |
| 378 | } |
| 379 | _ = lf.Chmod(0o600) |
| 380 | defer func() { |
| 381 | if unlockErr := unlockTaskFile(lf); unlockErr != nil && retErr == nil { |
| 382 | retErr = fmt.Errorf("save task: unlock: %w", unlockErr) |
| 383 | } |
| 384 | }() |
| 385 | |
| 386 | target := filepath.Join(taskDir, "snapshot.json") |
| 387 | // Read current version for CAS check (inside the lock). |
| 388 | current, err := s.readSnapshot(taskDir) |
| 389 | switch { |
| 390 | case err == nil && snap.Version <= current.Version: |
| 391 | return fmt.Errorf("save task: %w: stored=%d, given=%d", ErrStoreVersionConflict, current.Version, snap.Version) |
| 392 | case err != nil && !os.IsNotExist(err): |
| 393 | // A corrupt snapshot must fail loudly, never bypass the CAS check. |
| 394 | return fmt.Errorf("save task: read current snapshot: %w", err) |
| 395 | } |
| 396 | |
| 397 | data, err := json.Marshal(snap) |
| 398 | if err != nil { |
| 399 | return fmt.Errorf("save task: marshal: %w", err) |
| 400 | } |
| 401 | |
| 402 | // Atomic write via temp file + rename |
| 403 | tmp, err := os.CreateTemp(taskDir, ".snapshot-*.tmp") |
| 404 | if err != nil { |
| 405 | return fmt.Errorf("save task: %w", err) |
| 406 | } |
| 407 | tmpName := tmp.Name() |
| 408 | if _, err := tmp.Write(data); err != nil { |
| 409 | tmp.Close() |
| 410 | os.Remove(tmpName) |
| 411 | return fmt.Errorf("save task: %w", err) |
| 412 | } |
| 413 | if err := tmp.Sync(); err != nil { |
| 414 | tmp.Close() |
| 415 | os.Remove(tmpName) |
| 416 | return fmt.Errorf("save task: %w", err) |
| 417 | } |
| 418 | if err := tmp.Close(); err != nil { |
| 419 | os.Remove(tmpName) |
| 420 | return fmt.Errorf("save task: %w", err) |
| 421 | } |
| 422 | // Keep the CAS lock across bounded retries for Windows readers or filter |
| 423 | // drivers. Publication must remain an atomic rename, never a copy fallback. |
| 424 | if err := fileutil.ClaimRename(tmpName, target); err != nil { |
| 425 | os.Remove(tmpName) |
| 426 | return fmt.Errorf("save task: %w", err) |
| 427 | } |
| 428 | _ = os.Chmod(target, 0o600) |
| 429 | committed = true |
| 430 | return nil |
| 431 | } |
| 432 | |
| 433 | // SaveEvent implements WriteStore. |
| 434 | // AppendAuditEvent implements WriteStore. It atomically assigns the next |
| 435 | // monotonic sequence number and appends the event to the JSONL file. |
| 436 | func (s *FileStore) AppendAuditEvent(ctx context.Context, projectDir string, ev TaskEvent) (retErr error) { |
| 437 | committed := false |
| 438 | defer func() { |
| 439 | if committed && s.sink != nil { |
| 440 | s.sink.EventsChanged(projectDir, ev.TaskID) |
| 441 | } |
| 442 | }() |
| 443 | if err := ctx.Err(); err != nil { |
| 444 | return err |
| 445 | } |
| 446 | id, err := safeID(ev.TaskID) |
| 447 | if err != nil { |
| 448 | return err |
| 449 | } |
| 450 | root, err := s.taskRoot(projectDir) |
| 451 | if err != nil { |
| 452 | return err |
| 453 | } |
| 454 | taskDir, err := prepareTaskDir(root, id) |
| 455 | if err != nil { |
| 456 | return fmt.Errorf("append audit event: %w", err) |
| 457 | } |
| 458 | |
| 459 | // Cross-process atomicity: take the per-task lock (shared with SaveTask) |
| 460 | // so sequence assignment and snapshot writes never interleave. The |
| 461 | // events file itself is never renamed, so a dedicated task.lock is |
| 462 | // sufficient and keeps exactly one lock per task directory. |
| 463 | lockPath := filepath.Join(taskDir, "task.lock") |
| 464 | if err := rejectSymlink(lockPath); err != nil { |
| 465 | return fmt.Errorf("append audit event: %w", err) |
| 466 | } |
| 467 | lf, err := os.OpenFile(lockPath, os.O_CREATE|os.O_RDWR, 0o600) |
| 468 | if err != nil { |
| 469 | return fmt.Errorf("append audit event: open lock: %w", err) |
| 470 | } |
| 471 | defer lf.Close() |
| 472 | if err := lockTaskFile(lf); err != nil { |
| 473 | return fmt.Errorf("append audit event: lock: %w", err) |
| 474 | } |
| 475 | defer func() { |
| 476 | if unlockErr := unlockTaskFile(lf); unlockErr != nil && retErr == nil { |
| 477 | retErr = fmt.Errorf("append audit event: unlock: %w", unlockErr) |
| 478 | } |
| 479 | }() |
| 480 | |
| 481 | eventsPath := filepath.Join(taskDir, "events.jsonl") |
| 482 | if err := rejectSymlink(eventsPath); err != nil { |
| 483 | return fmt.Errorf("append audit event: %w", err) |
| 484 | } |
| 485 | f, err := os.OpenFile(eventsPath, os.O_RDWR|os.O_CREATE, 0o600) |
| 486 | if err != nil { |
| 487 | return err |
| 488 | } |
| 489 | defer f.Close() |
| 490 | _ = f.Chmod(0o600) |
| 491 | |
| 492 | // Read current events to compute next sequence (safe under lock) |
| 493 | if _, err := f.Seek(0, 0); err != nil { |
| 494 | return err |
| 495 | } |
| 496 | raw, err := io.ReadAll(f) |
| 497 | if err != nil { |
| 498 | return err |
| 499 | } |
| 500 | max := 0 |
| 501 | for line := range strings.SplitSeq(string(raw), "\n") { |
| 502 | line = strings.TrimSpace(line) |
| 503 | if line == "" { |
| 504 | continue |
| 505 | } |
| 506 | var existing TaskEvent |
| 507 | if err := json.Unmarshal([]byte(line), &existing); err != nil { |
| 508 | continue |
| 509 | } |
| 510 | if existing.Sequence > max { |
| 511 | max = existing.Sequence |
| 512 | } |
| 513 | } |
| 514 | ev.Sequence = max + 1 |
| 515 | if err := ev.Validate(); err != nil { |
| 516 | return fmt.Errorf("append audit event: %w", err) |
| 517 | } |
| 518 | data, err := json.Marshal(ev) |
| 519 | if err != nil { |
| 520 | return err |
| 521 | } |
| 522 | // Append at end of locked file |
| 523 | if _, err := f.Seek(0, 2); err != nil { |
| 524 | return err |
| 525 | } |
| 526 | if _, err := f.WriteString(string(data) + "\n"); err != nil { |
| 527 | return err |
| 528 | } |
| 529 | committed = true |
| 530 | return nil |
| 531 | } |
| 532 | |
| 533 | // ── deprecated: removed NextSequence, SaveEvent — use AppendAuditEvent ── |
| 534 | |
| 535 | // CheckIdempotency implements WriteStore. |
| 536 | func (s *FileStore) CheckIdempotency(ctx context.Context, projectDir string, key string) (*IdempotencyRecord, error) { |
| 537 | root, err := s.taskRoot(projectDir) |
| 538 | if err != nil { |
| 539 | return nil, err |
| 540 | } |
| 541 | id, err := safeID(key) |
| 542 | if err != nil { |
| 543 | return nil, err |
| 544 | } |
| 545 | idemDir := filepath.Join(root, ".idempotency") |
| 546 | if err := rejectSymlink(idemDir); err != nil { |
| 547 | return nil, err |
| 548 | } |
| 549 | if err := rejectSymlink(filepath.Join(idemDir, id+".json")); err != nil { |
| 550 | return nil, err |
| 551 | } |
| 552 | data, err := os.ReadFile(filepath.Join(idemDir, id+".json")) |
| 553 | if err != nil { |
| 554 | if os.IsNotExist(err) { |
| 555 | return nil, nil |
| 556 | } |
| 557 | return nil, err |
| 558 | } |
| 559 | var rec IdempotencyRecord |
| 560 | if err := json.Unmarshal(data, &rec); err != nil { |
| 561 | return nil, nil |
| 562 | } |
| 563 | return &rec, nil |
| 564 | } |
| 565 | |
| 566 | func (s *FileStore) idempotencyPaths(projectDir, key string) (string, string, string, error) { |
| 567 | root, err := s.taskRoot(projectDir) |
| 568 | if err != nil { |
| 569 | return "", "", "", err |
| 570 | } |
| 571 | id, err := safeID(key) |
| 572 | if err != nil { |
| 573 | return "", "", "", err |
| 574 | } |
| 575 | dir := filepath.Join(root, ".idempotency") |
| 576 | if err := rejectSymlink(dir); err != nil { |
| 577 | return "", "", "", err |
| 578 | } |
| 579 | if err := os.MkdirAll(dir, 0o700); err != nil { |
| 580 | return "", "", "", err |
| 581 | } |
| 582 | _ = os.Chmod(dir, 0o700) |
| 583 | target := filepath.Join(dir, id+".json") |
| 584 | lock := filepath.Join(dir, id+".lock") |
| 585 | if err := rejectSymlink(target); err != nil { |
| 586 | return "", "", "", err |
| 587 | } |
| 588 | if err := rejectSymlink(lock); err != nil { |
| 589 | return "", "", "", err |
| 590 | } |
| 591 | return dir, target, lock, nil |
| 592 | } |
| 593 | |
| 594 | // quarantineCorruptIdempotency moves an unreadable record out of the active |
| 595 | // key path without deleting it. A corrupt record cannot safely describe either |
| 596 | // a pending or finalized operation; keeping it as evidence prevents it from |
| 597 | // permanently blocking future claims while preserving forensic data. |
| 598 | func quarantineCorruptIdempotency(target string) error { |
| 599 | backup := fmt.Sprintf("%s.corrupt-%d", target, timeNow().UnixNano()) |
| 600 | if err := os.Rename(target, backup); err != nil { |
| 601 | if os.IsNotExist(err) { |
| 602 | return nil |
| 603 | } |
| 604 | return err |
| 605 | } |
| 606 | return nil |
| 607 | } |
| 608 | |
| 609 | func (s *FileStore) ClaimIdempotency(ctx context.Context, projectDir string, r IdempotencyRecord) (*IdempotencyRecord, error) { |
| 610 | if err := ctx.Err(); err != nil { |
| 611 | return nil, err |
| 612 | } |
| 613 | _, target, lockPath, err := s.idempotencyPaths(projectDir, r.Key) |
| 614 | if err != nil { |
| 615 | return nil, err |
| 616 | } |
| 617 | lf, err := os.OpenFile(lockPath, os.O_CREATE|os.O_RDWR, 0o600) |
| 618 | if err != nil { |
| 619 | return nil, err |
| 620 | } |
| 621 | defer lf.Close() |
| 622 | _ = lf.Chmod(0o600) |
| 623 | if err := lockTaskFile(lf); err != nil { |
| 624 | return nil, err |
| 625 | } |
| 626 | defer func() { _ = unlockTaskFile(lf) }() |
| 627 | data, err := os.ReadFile(target) |
| 628 | if err == nil { |
| 629 | var existing IdempotencyRecord |
| 630 | if jsonErr := json.Unmarshal(data, &existing); jsonErr != nil { |
| 631 | if quarantineErr := quarantineCorruptIdempotency(target); quarantineErr != nil { |
| 632 | return nil, fmt.Errorf("idempotency claim: parse existing record: %w (quarantine: %w)", jsonErr, quarantineErr) |
| 633 | } |
| 634 | // Continue with a fresh claim after preserving the corrupt record. |
| 635 | } else if existing.Pending && timeNow().Sub(existing.ClaimedAt) > 5*time.Minute { |
| 636 | _ = os.Remove(target) |
| 637 | } else { |
| 638 | return &existing, nil |
| 639 | } |
| 640 | } else if !os.IsNotExist(err) { |
| 641 | return nil, err |
| 642 | } |
| 643 | if r.ClaimedAt.IsZero() { |
| 644 | r.ClaimedAt = timeNow() |
| 645 | } |
| 646 | r.Pending = true |
| 647 | data, err = json.Marshal(r) |
| 648 | if err != nil { |
| 649 | return nil, err |
| 650 | } |
| 651 | if err := fileutil.AtomicWriteFile(target, data, 0o600); err != nil { |
| 652 | return nil, err |
| 653 | } |
| 654 | _ = os.Chmod(target, 0o600) |
| 655 | return nil, nil |
| 656 | } |
| 657 | |
| 658 | func (s *FileStore) FinalizeIdempotency(ctx context.Context, projectDir string, r IdempotencyRecord) error { |
| 659 | if err := ctx.Err(); err != nil { |
| 660 | return err |
| 661 | } |
| 662 | _, target, lockPath, err := s.idempotencyPaths(projectDir, r.Key) |
| 663 | if err != nil { |
| 664 | return err |
| 665 | } |
| 666 | lf, err := os.OpenFile(lockPath, os.O_CREATE|os.O_RDWR, 0o600) |
| 667 | if err != nil { |
| 668 | return err |
| 669 | } |
| 670 | defer lf.Close() |
| 671 | if err := lockTaskFile(lf); err != nil { |
| 672 | return err |
| 673 | } |
| 674 | defer func() { _ = unlockTaskFile(lf) }() |
| 675 | data, err := os.ReadFile(target) |
| 676 | if err != nil { |
| 677 | return err |
| 678 | } |
| 679 | var existing IdempotencyRecord |
| 680 | if err := json.Unmarshal(data, &existing); err != nil { |
| 681 | return err |
| 682 | } |
| 683 | if existing.Op != r.Op || existing.TaskID != r.TaskID || existing.Version != r.Version { |
| 684 | return fmt.Errorf("idempotency key conflict: different params") |
| 685 | } |
| 686 | existing.Pending = false |
| 687 | data, err = json.Marshal(existing) |
| 688 | if err != nil { |
| 689 | return err |
| 690 | } |
| 691 | if err := fileutil.AtomicWriteFile(target, data, 0o600); err != nil { |
| 692 | return err |
| 693 | } |
| 694 | _ = os.Chmod(target, 0o600) |
| 695 | return nil |
| 696 | } |
| 697 | |
| 698 | func (s *FileStore) ReleaseIdempotency(ctx context.Context, projectDir, key string) error { |
| 699 | if err := ctx.Err(); err != nil { |
| 700 | return err |
| 701 | } |
| 702 | _, target, lockPath, err := s.idempotencyPaths(projectDir, key) |
| 703 | if err != nil { |
| 704 | return err |
| 705 | } |
| 706 | lf, err := os.OpenFile(lockPath, os.O_CREATE|os.O_RDWR, 0o600) |
| 707 | if err != nil { |
| 708 | return err |
| 709 | } |
| 710 | defer lf.Close() |
| 711 | if err := lockTaskFile(lf); err != nil { |
| 712 | return err |
| 713 | } |
| 714 | defer func() { _ = unlockTaskFile(lf) }() |
| 715 | data, err := os.ReadFile(target) |
| 716 | if os.IsNotExist(err) { |
| 717 | return nil |
| 718 | } |
| 719 | if err != nil { |
| 720 | return err |
| 721 | } |
| 722 | var existing IdempotencyRecord |
| 723 | if err := json.Unmarshal(data, &existing); err != nil { |
| 724 | return err |
| 725 | } |
| 726 | if existing.Pending { |
| 727 | return os.Remove(target) |
| 728 | } |
| 729 | return nil |
| 730 | } |
| 731 | |
| 732 | // RecordIdempotency implements WriteStore. |
| 733 | func (s *FileStore) RecordIdempotency(ctx context.Context, projectDir string, r IdempotencyRecord) error { |
| 734 | root, err := s.taskRoot(projectDir) |
| 735 | if err != nil { |
| 736 | return err |
| 737 | } |
| 738 | id, err := safeID(r.Key) |
| 739 | if err != nil { |
| 740 | return err |
| 741 | } |
| 742 | idemDir := filepath.Join(root, ".idempotency") |
| 743 | if err := rejectSymlink(idemDir); err != nil { |
| 744 | return err |
| 745 | } |
| 746 | if err := os.MkdirAll(idemDir, 0o700); err != nil { |
| 747 | return err |
| 748 | } |
| 749 | if err := os.Chmod(idemDir, 0o700); err != nil { |
| 750 | return err |
| 751 | } |
| 752 | data, err := json.Marshal(r) |
| 753 | if err != nil { |
| 754 | return err |
| 755 | } |
| 756 | target := filepath.Join(idemDir, id+".json") |
| 757 | if err := rejectSymlink(target); err != nil { |
| 758 | return err |
| 759 | } |
| 760 | // Publish the complete record only when the key is still absent. A crash |
| 761 | // during the old direct write could leave a permanently unparsable record. |
| 762 | if err := fileutil.AtomicCreateFile(target, data, 0o600); err == nil { |
| 763 | return nil |
| 764 | } else if !os.IsExist(err) { |
| 765 | return err |
| 766 | } |
| 767 | existing, rdErr := os.ReadFile(target) |
| 768 | if rdErr != nil { |
| 769 | return fmt.Errorf("idempotency conflict: cannot read existing record: %w", rdErr) |
| 770 | } |
| 771 | var prev IdempotencyRecord |
| 772 | if err := json.Unmarshal(existing, &prev); err != nil { |
| 773 | return fmt.Errorf("idempotency conflict: cannot parse existing record: %w", err) |
| 774 | } |
| 775 | if prev.Op != r.Op || prev.TaskID != r.TaskID || prev.Version != r.Version { |
| 776 | return fmt.Errorf("idempotency key conflict: different params") |
| 777 | } |
| 778 | return nil // idempotent |
| 779 | } |
| 780 |