| 1 | package jobs |
| 2 | |
| 3 | import ( |
| 4 | "context" |
| 5 | "encoding/json" |
| 6 | "errors" |
| 7 | "io" |
| 8 | "os" |
| 9 | "path/filepath" |
| 10 | "runtime" |
| 11 | "strings" |
| 12 | "sync" |
| 13 | "testing" |
| 14 | "time" |
| 15 | |
| 16 | "reasonix/internal/event" |
| 17 | "reasonix/internal/evidence" |
| 18 | ) |
| 19 | |
| 20 | func TestCompletedJobPersistsOutputAndReleasesMemory(t *testing.T) { |
| 21 | sessionPath := filepath.Join(t.TempDir(), "session.jsonl") |
| 22 | m := NewManager(event.Discard) |
| 23 | defer m.Close() |
| 24 | m.SetActiveSessionPath("session", sessionPath) |
| 25 | |
| 26 | j := m.StartForSession("session", "bash", "persist", func(_ context.Context, out io.Writer) (string, error) { |
| 27 | _, _ = io.WriteString(out, strings.Repeat("x", defaultTailBytes+1024)) |
| 28 | return "", nil |
| 29 | }) |
| 30 | <-j.done |
| 31 | |
| 32 | j.mu.Lock() |
| 33 | tailLen := len(j.tail) |
| 34 | result := j.result |
| 35 | artifactPath := j.artifactPath |
| 36 | j.mu.Unlock() |
| 37 | |
| 38 | if tailLen != 0 { |
| 39 | t.Fatalf("completed artifact-backed job kept %d tail bytes, want 0", tailLen) |
| 40 | } |
| 41 | if result != "" { |
| 42 | t.Fatalf("completed artifact-backed job kept result %q, want empty", result) |
| 43 | } |
| 44 | if artifactPath == "" { |
| 45 | t.Fatal("artifact path should be set") |
| 46 | } |
| 47 | |
| 48 | res := m.WaitForSession(context.Background(), "session", []string{j.ID}, 1) |
| 49 | if len(res) != 1 || len(res[0].Output) != defaultTailBytes+1024 { |
| 50 | t.Fatalf("wait output len = %d, want %d", len(res[0].Output), defaultTailBytes+1024) |
| 51 | } |
| 52 | } |
| 53 | |
| 54 | func TestJobArtifactPreservesOutput(t *testing.T) { |
| 55 | sessionPath := filepath.Join(t.TempDir(), "session.jsonl") |
| 56 | m := NewManager(event.Discard) |
| 57 | defer m.Close() |
| 58 | m.SetActiveSessionPath("session", sessionPath) |
| 59 | secret := "sk-real-secret-value-123456" |
| 60 | |
| 61 | j := m.StartForSession("session", "bash", "persist secret", func(_ context.Context, out io.Writer) (string, error) { |
| 62 | _, _ = io.WriteString(out, "DEEPSEEK_API_KEY="+secret+"\n") |
| 63 | return "Authorization: Bearer ghp_abcdefghijklmnopqrstuvwxyz", nil |
| 64 | }) |
| 65 | <-j.done |
| 66 | |
| 67 | res := m.WaitForSession(context.Background(), "session", []string{j.ID}, 1) |
| 68 | if len(res) != 1 { |
| 69 | t.Fatalf("wait result = %+v", res) |
| 70 | } |
| 71 | if !strings.Contains(res[0].Output, secret) || !strings.Contains(res[0].Output, "ghp_abcdefghijklmnopqrstuvwxyz") { |
| 72 | t.Fatalf("wait output did not preserve job output:\n%s", res[0].Output) |
| 73 | } |
| 74 | |
| 75 | data, err := os.ReadFile(filepath.Join(ArtifactDir(sessionPath), j.ID+jobLogExt)) |
| 76 | if err != nil { |
| 77 | t.Fatalf("read artifact: %v", err) |
| 78 | } |
| 79 | if !strings.Contains(string(data), secret) || !strings.Contains(string(data), "ghp_abcdefghijklmnopqrstuvwxyz") { |
| 80 | t.Fatalf("artifact did not preserve job output:\n%s", data) |
| 81 | } |
| 82 | } |
| 83 | |
| 84 | func TestJobArtifactMetadataPreservesLabel(t *testing.T) { |
| 85 | sessionPath := filepath.Join(t.TempDir(), "session.jsonl") |
| 86 | m := NewManager(event.Discard) |
| 87 | defer m.Close() |
| 88 | m.SetActiveSessionPath("session", sessionPath) |
| 89 | const secret = "sk-real-secret-value-123456" |
| 90 | |
| 91 | j := m.StartForSession("session", "bash", "echo DEEPSEEK_API_KEY="+secret, func(context.Context, io.Writer) (string, error) { |
| 92 | return "", nil |
| 93 | }) |
| 94 | <-j.done |
| 95 | |
| 96 | data, err := os.ReadFile(filepath.Join(ArtifactDir(sessionPath), j.ID+jobMetaExt)) |
| 97 | if err != nil { |
| 98 | t.Fatal(err) |
| 99 | } |
| 100 | if !strings.Contains(string(data), secret) { |
| 101 | t.Fatalf("job metadata did not preserve label:\n%s", data) |
| 102 | } |
| 103 | } |
| 104 | |
| 105 | func TestListArtifactViewsVerifiesTerminalArtifactPresence(t *testing.T) { |
| 106 | sessionPath := filepath.Join(t.TempDir(), "session.jsonl") |
| 107 | dir := ArtifactDir(sessionPath) |
| 108 | if err := os.MkdirAll(dir, 0o700); err != nil { |
| 109 | t.Fatal(err) |
| 110 | } |
| 111 | type artifactCase struct { |
| 112 | id string |
| 113 | status Status |
| 114 | metaOK bool |
| 115 | metaErr string |
| 116 | artifact string |
| 117 | legacy bool |
| 118 | want bool |
| 119 | } |
| 120 | cases := []artifactCase{ |
| 121 | {id: "task-running", status: Running, metaOK: true, artifact: "file", want: false}, |
| 122 | {id: "task-unknown", status: Status("future"), metaOK: true, artifact: "file", want: false}, |
| 123 | {id: "task-missing", status: Done, metaOK: true, want: false}, |
| 124 | {id: "task-error", status: Done, metaOK: true, metaErr: "write failed", artifact: "file", want: false}, |
| 125 | {id: "task-directory", status: Done, metaOK: true, artifact: "directory", want: false}, |
| 126 | {id: "task-complete", status: Done, metaOK: true, artifact: "file", want: true}, |
| 127 | {id: "task-legacy", status: Done, metaOK: true, artifact: "file", legacy: true, want: true}, |
| 128 | } |
| 129 | for _, tc := range cases { |
| 130 | logName := tc.id + jobLogExt |
| 131 | metaLogPath := logName |
| 132 | if tc.legacy { |
| 133 | metaLogPath = "" |
| 134 | } |
| 135 | if err := writeMeta(filepath.Join(dir, tc.id+jobMetaExt), artifactMeta{ |
| 136 | ID: tc.id, |
| 137 | Kind: "task", |
| 138 | Status: tc.status, |
| 139 | StartedAt: time.Now().Add(-time.Minute).UnixMilli(), |
| 140 | FinishedAt: time.Now().UnixMilli(), |
| 141 | ArtifactComplete: tc.metaOK, |
| 142 | ArtifactError: tc.metaErr, |
| 143 | LogPath: metaLogPath, |
| 144 | }); err != nil { |
| 145 | t.Fatalf("write %s metadata: %v", tc.id, err) |
| 146 | } |
| 147 | switch tc.artifact { |
| 148 | case "file": |
| 149 | if err := os.WriteFile(filepath.Join(dir, logName), []byte("persisted output"), 0o600); err != nil { |
| 150 | t.Fatalf("write %s artifact: %v", tc.id, err) |
| 151 | } |
| 152 | case "directory": |
| 153 | if err := os.Mkdir(filepath.Join(dir, logName), 0o700); err != nil { |
| 154 | t.Fatalf("create %s artifact directory: %v", tc.id, err) |
| 155 | } |
| 156 | } |
| 157 | } |
| 158 | |
| 159 | views, err := ListArtifactViews(sessionPath) |
| 160 | if err != nil { |
| 161 | t.Fatal(err) |
| 162 | } |
| 163 | got := make(map[string]bool, len(views)) |
| 164 | for _, view := range views { |
| 165 | got[view.ID] = view.ArtifactComplete |
| 166 | } |
| 167 | if len(got) != len(cases) { |
| 168 | t.Fatalf("artifact views = %+v, want %d entries", views, len(cases)) |
| 169 | } |
| 170 | for _, tc := range cases { |
| 171 | complete, ok := got[tc.id] |
| 172 | if !ok { |
| 173 | t.Errorf("%s artifact view is missing", tc.id) |
| 174 | continue |
| 175 | } |
| 176 | if complete != tc.want { |
| 177 | t.Errorf("%s artifact complete = %v, want %v", tc.id, complete, tc.want) |
| 178 | } |
| 179 | } |
| 180 | } |
| 181 | |
| 182 | func TestJobArtifactUsesPrivatePermissions(t *testing.T) { |
| 183 | if runtime.GOOS == "windows" { |
| 184 | t.Skip("Windows file ACLs are not represented by Unix permission bits") |
| 185 | } |
| 186 | sessionPath := filepath.Join(t.TempDir(), "session.jsonl") |
| 187 | m := NewManager(event.Discard) |
| 188 | defer m.Close() |
| 189 | m.SetActiveSessionPath("session", sessionPath) |
| 190 | |
| 191 | // Simulate a legacy artifact at the path the first job will reuse. |
| 192 | dir := ArtifactDir(sessionPath) |
| 193 | if err := os.MkdirAll(dir, 0o755); err != nil { |
| 194 | t.Fatal(err) |
| 195 | } |
| 196 | logPath := filepath.Join(dir, "bash-1"+jobLogExt) |
| 197 | if err := os.WriteFile(logPath, []byte("old redacted output"), 0o644); err != nil { |
| 198 | t.Fatal(err) |
| 199 | } |
| 200 | |
| 201 | j := m.StartForSession("session", "bash", "echo API_KEY=raw-secret", func(_ context.Context, out io.Writer) (string, error) { |
| 202 | _, _ = io.WriteString(out, "API_KEY=raw-secret\n") |
| 203 | return "", nil |
| 204 | }) |
| 205 | <-j.done |
| 206 | if j.ID != "bash-1" { |
| 207 | t.Fatalf("job id = %q, want bash-1", j.ID) |
| 208 | } |
| 209 | assertPrivateArtifactMode(t, logPath) |
| 210 | assertPrivateArtifactMode(t, filepath.Join(dir, j.ID+jobMetaExt)) |
| 211 | info, err := os.Stat(dir) |
| 212 | if err != nil { |
| 213 | t.Fatal(err) |
| 214 | } |
| 215 | if got := info.Mode().Perm(); got != 0o700 { |
| 216 | t.Fatalf("artifact dir mode = %04o, want 0700", got) |
| 217 | } |
| 218 | } |
| 219 | |
| 220 | func assertPrivateArtifactMode(t *testing.T, path string) { |
| 221 | t.Helper() |
| 222 | info, err := os.Stat(path) |
| 223 | if err != nil { |
| 224 | t.Fatalf("stat %s: %v", path, err) |
| 225 | } |
| 226 | if got := info.Mode().Perm(); got != 0o600 { |
| 227 | t.Fatalf("%s mode = %04o, want 0600", path, got) |
| 228 | } |
| 229 | } |
| 230 | |
| 231 | func TestRestoreSessionArtifactsAndAdvanceSequence(t *testing.T) { |
| 232 | sessionPath := filepath.Join(t.TempDir(), "session.jsonl") |
| 233 | first := NewManager(event.Discard) |
| 234 | first.SetActiveSessionPath("session", sessionPath) |
| 235 | j := first.StartForSession("session", "task", "answer", func(context.Context, io.Writer) (string, error) { |
| 236 | return "persisted answer", nil |
| 237 | }) |
| 238 | <-j.done |
| 239 | first.Close() |
| 240 | |
| 241 | second := NewManager(event.Discard) |
| 242 | defer second.Close() |
| 243 | second.SetActiveSessionPath("session", sessionPath) |
| 244 | |
| 245 | res := second.WaitForSession(context.Background(), "session", []string{j.ID}, 1) |
| 246 | if len(res) != 1 || !strings.Contains(res[0].Output, "persisted answer") { |
| 247 | t.Fatalf("restored wait = %+v, want persisted answer", res) |
| 248 | } |
| 249 | if got := second.WaitForSession(context.Background(), "session", nil, 1); len(got) != 0 { |
| 250 | t.Fatalf("wait without ids should ignore restored completed artifacts, got %+v", got) |
| 251 | } |
| 252 | if got := second.LeaseEvidenceForSession("session", j.ID); len(got.Receipts) != 0 { |
| 253 | t.Fatalf("mutation-free task restored mutation evidence: %+v", got) |
| 254 | } |
| 255 | |
| 256 | next := second.StartForSession("session", "bash", "next", func(context.Context, io.Writer) (string, error) { |
| 257 | return "", nil |
| 258 | }) |
| 259 | <-next.done |
| 260 | if next.ID == j.ID { |
| 261 | t.Fatalf("new job reused restored id %q", next.ID) |
| 262 | } |
| 263 | } |
| 264 | |
| 265 | func TestRestoreRunningArtifactAsInterrupted(t *testing.T) { |
| 266 | sessionPath := filepath.Join(t.TempDir(), "session.jsonl") |
| 267 | dir := ArtifactDir(sessionPath) |
| 268 | metaPath := filepath.Join(dir, "task-1"+jobMetaExt) |
| 269 | if err := writeMeta(metaPath, artifactMeta{ |
| 270 | ID: "task-1", |
| 271 | Kind: "task", |
| 272 | Status: Running, |
| 273 | StartedAt: time.Now().Add(-time.Minute).UnixMilli(), |
| 274 | ArtifactComplete: true, |
| 275 | }); err != nil { |
| 276 | t.Fatal(err) |
| 277 | } |
| 278 | |
| 279 | m := NewManager(event.Discard, WithSessionOwnershipProbe(func(path string) bool { |
| 280 | return path == sessionPath |
| 281 | })) |
| 282 | defer m.Close() |
| 283 | m.SetActiveSessionPath("session", sessionPath) |
| 284 | |
| 285 | if got := m.RunningForSession("session"); len(got) != 0 { |
| 286 | t.Fatalf("restored stale job remained live: %+v", got) |
| 287 | } |
| 288 | if m.KillForSession("session", "task-1") { |
| 289 | t.Fatal("restored interrupted job must not be killable") |
| 290 | } |
| 291 | result := m.WaitForSession(context.Background(), "session", []string{"task-1"}, 1) |
| 292 | if len(result) != 1 || result[0].Status != Interrupted { |
| 293 | t.Fatalf("restored result = %+v, want interrupted", result) |
| 294 | } |
| 295 | |
| 296 | persisted, err := readMeta(metaPath) |
| 297 | if err != nil { |
| 298 | t.Fatal(err) |
| 299 | } |
| 300 | if persisted.Status != Interrupted || persisted.FinishedAt == 0 || persisted.ArtifactComplete { |
| 301 | t.Fatalf("persisted restored metadata = %+v", persisted) |
| 302 | } |
| 303 | } |
| 304 | |
| 305 | func TestRestoreRunningArtifactDefersTombstoneWhenRepairWriteFails(t *testing.T) { |
| 306 | sessionPath := filepath.Join(t.TempDir(), "session.jsonl") |
| 307 | dir := ArtifactDir(sessionPath) |
| 308 | metaPath := filepath.Join(dir, "task-1"+jobMetaExt) |
| 309 | if err := writeMeta(metaPath, artifactMeta{ |
| 310 | ID: "task-1", |
| 311 | Kind: "task", |
| 312 | Status: Running, |
| 313 | StartedAt: time.Now().Add(-time.Minute).UnixMilli(), |
| 314 | }); err != nil { |
| 315 | t.Fatal(err) |
| 316 | } |
| 317 | |
| 318 | originalRepair := repairArtifactMeta |
| 319 | repairArtifactMeta = func(string, artifactMeta) error { |
| 320 | return errors.New("simulated repair failure") |
| 321 | } |
| 322 | t.Cleanup(func() { repairArtifactMeta = originalRepair }) |
| 323 | |
| 324 | sink := &recordingSink{} |
| 325 | m := NewManager(sink, WithSessionOwnershipProbe(func(path string) bool { |
| 326 | return path == sessionPath |
| 327 | })) |
| 328 | defer m.Close() |
| 329 | m.SetActiveSessionPath("session", sessionPath) |
| 330 | |
| 331 | if result := m.WaitForSession(context.Background(), "session", []string{"task-1"}, 1); len(result) != 0 { |
| 332 | t.Fatalf("failed repair published an in-memory tombstone: %+v", result) |
| 333 | } |
| 334 | persisted, err := readMeta(metaPath) |
| 335 | if err != nil { |
| 336 | t.Fatal(err) |
| 337 | } |
| 338 | if persisted.Status != Running || persisted.FinishedAt != 0 { |
| 339 | t.Fatalf("failed repair changed durable metadata: %+v", persisted) |
| 340 | } |
| 341 | m.mu.Lock() |
| 342 | loaded := m.loaded["session"] |
| 343 | m.mu.Unlock() |
| 344 | if loaded { |
| 345 | t.Fatal("failed repair marked the session loaded and prevented retry") |
| 346 | } |
| 347 | sink.mu.Lock() |
| 348 | events := append([]event.Event(nil), sink.events...) |
| 349 | sink.mu.Unlock() |
| 350 | if len(events) != 1 || events[0].Kind != event.Notice || events[0].Level != event.LevelWarn || |
| 351 | events[0].Text != "Background job recovery did not complete." || |
| 352 | !strings.Contains(events[0].Detail, "task-1") || !strings.Contains(events[0].Detail, "simulated repair failure") { |
| 353 | t.Fatalf("repair failure notice = %+v", events) |
| 354 | } |
| 355 | |
| 356 | repairArtifactMeta = originalRepair |
| 357 | m.SetActiveSessionPath("session", sessionPath) |
| 358 | result := m.WaitForSession(context.Background(), "session", []string{"task-1"}, 1) |
| 359 | if len(result) != 1 || result[0].Status != Interrupted { |
| 360 | t.Fatalf("retried repair result = %+v, want interrupted", result) |
| 361 | } |
| 362 | m.mu.Lock() |
| 363 | loaded = m.loaded["session"] |
| 364 | m.mu.Unlock() |
| 365 | if !loaded { |
| 366 | t.Fatal("successful retry did not mark the session loaded") |
| 367 | } |
| 368 | persisted, err = readMeta(metaPath) |
| 369 | if err != nil { |
| 370 | t.Fatal(err) |
| 371 | } |
| 372 | if persisted.Status != Interrupted || persisted.FinishedAt == 0 || persisted.ArtifactComplete { |
| 373 | t.Fatalf("retried repair metadata = %+v, want durable interrupted tombstone", persisted) |
| 374 | } |
| 375 | } |
| 376 | |
| 377 | func TestRestoreRunningArtifactFromClosedOwnerAsInterrupted(t *testing.T) { |
| 378 | sessionPath := filepath.Join(t.TempDir(), "session.jsonl") |
| 379 | dir := ArtifactDir(sessionPath) |
| 380 | metaPath := filepath.Join(dir, "task-1"+jobMetaExt) |
| 381 | owner := NewManager(event.Discard) |
| 382 | ownerID := owner.ownerID |
| 383 | owner.Close() |
| 384 | if err := writeMeta(metaPath, artifactMeta{ |
| 385 | ID: "task-1", |
| 386 | Kind: "task", |
| 387 | OwnerID: ownerID, |
| 388 | Status: Running, |
| 389 | StartedAt: time.Now().Add(-time.Minute).UnixMilli(), |
| 390 | }); err != nil { |
| 391 | t.Fatal(err) |
| 392 | } |
| 393 | |
| 394 | restored := NewManager(event.Discard, WithSessionOwnershipProbe(func(path string) bool { |
| 395 | return path == sessionPath |
| 396 | })) |
| 397 | defer restored.Close() |
| 398 | restored.SetActiveSessionPath("session", sessionPath) |
| 399 | result := restored.WaitForSession(context.Background(), "session", []string{"task-1"}, 1) |
| 400 | if len(result) != 1 || result[0].Status != Interrupted { |
| 401 | t.Fatalf("restored result = %+v, want interrupted after the original owner closed", result) |
| 402 | } |
| 403 | } |
| 404 | |
| 405 | func TestRestoreRunningArtifactWithoutSessionOwnershipDefersRepair(t *testing.T) { |
| 406 | sessionPath := filepath.Join(t.TempDir(), "session.jsonl") |
| 407 | metaPath := filepath.Join(ArtifactDir(sessionPath), "task-1"+jobMetaExt) |
| 408 | if err := writeMeta(metaPath, artifactMeta{ |
| 409 | ID: "task-1", |
| 410 | Kind: "task", |
| 411 | Status: Running, |
| 412 | StartedAt: time.Now().Add(-time.Minute).UnixMilli(), |
| 413 | }); err != nil { |
| 414 | t.Fatal(err) |
| 415 | } |
| 416 | |
| 417 | observer := NewManager(event.Discard) |
| 418 | observer.SetActiveSessionPath("session", sessionPath) |
| 419 | if result := observer.WaitForSession(context.Background(), "session", []string{"task-1"}, 1); len(result) != 0 { |
| 420 | observer.Close() |
| 421 | t.Fatalf("unowned observer published a running artifact: %+v", result) |
| 422 | } |
| 423 | meta, err := readMeta(metaPath) |
| 424 | if err != nil { |
| 425 | observer.Close() |
| 426 | t.Fatal(err) |
| 427 | } |
| 428 | if meta.Status != Running || meta.FinishedAt != 0 { |
| 429 | observer.Close() |
| 430 | t.Fatalf("unowned observer rewrote running metadata: %+v", meta) |
| 431 | } |
| 432 | observer.Close() |
| 433 | |
| 434 | owner := NewManager(event.Discard, WithSessionOwnershipProbe(func(path string) bool { |
| 435 | return path == sessionPath |
| 436 | })) |
| 437 | defer owner.Close() |
| 438 | owner.SetActiveSessionPath("session", sessionPath) |
| 439 | result := owner.WaitForSession(context.Background(), "session", []string{"task-1"}, 1) |
| 440 | if len(result) != 1 || result[0].Status != Interrupted { |
| 441 | t.Fatalf("owned reload = %+v, want interrupted", result) |
| 442 | } |
| 443 | } |
| 444 | |
| 445 | func TestRestoreDoesNotInterruptJobOwnedByLiveManager(t *testing.T) { |
| 446 | sessionPath := filepath.Join(t.TempDir(), "session.jsonl") |
| 447 | first := NewManager(event.Discard) |
| 448 | defer first.Close() |
| 449 | first.SetActiveSessionPath("session", sessionPath) |
| 450 | |
| 451 | release := make(chan struct{}) |
| 452 | job := first.StartForSession("session", "task", "running", func(context.Context, io.Writer) (string, error) { |
| 453 | <-release |
| 454 | return "done", nil |
| 455 | }) |
| 456 | metaPath := filepath.Join(ArtifactDir(sessionPath), job.ID+jobMetaExt) |
| 457 | |
| 458 | second := NewManager(event.Discard) |
| 459 | defer second.Close() |
| 460 | second.SetActiveSessionPath("session", sessionPath) |
| 461 | |
| 462 | meta, err := readMeta(metaPath) |
| 463 | if err != nil { |
| 464 | close(release) |
| 465 | t.Fatal(err) |
| 466 | } |
| 467 | if meta.Status != Running { |
| 468 | close(release) |
| 469 | t.Fatalf("a replacement manager interrupted a still-live job: status=%q", meta.Status) |
| 470 | } |
| 471 | if got := second.RunningForSession("session"); len(got) != 0 { |
| 472 | close(release) |
| 473 | t.Fatalf("replacement manager published an unowned live job: %+v", got) |
| 474 | } |
| 475 | |
| 476 | close(release) |
| 477 | <-job.done |
| 478 | second.SetActiveSessionPath("session", sessionPath) |
| 479 | result := second.WaitForSession(context.Background(), "session", []string{job.ID}, 1) |
| 480 | if len(result) != 1 || result[0].Status != Done { |
| 481 | t.Fatalf("replacement manager did not load the terminal artifact after owner completion: %+v", result) |
| 482 | } |
| 483 | } |
| 484 | |
| 485 | func TestRunningJobArtifactMetadataIsIncomplete(t *testing.T) { |
| 486 | sessionPath := filepath.Join(t.TempDir(), "session.jsonl") |
| 487 | m := NewManager(event.Discard) |
| 488 | defer m.Close() |
| 489 | m.SetActiveSessionPath("session", sessionPath) |
| 490 | |
| 491 | release := make(chan struct{}) |
| 492 | job := m.StartForSession("session", "task", "running", func(context.Context, io.Writer) (string, error) { |
| 493 | <-release |
| 494 | return "done", nil |
| 495 | }) |
| 496 | metaPath := filepath.Join(ArtifactDir(sessionPath), job.ID+jobMetaExt) |
| 497 | meta, err := readMeta(metaPath) |
| 498 | if err != nil { |
| 499 | close(release) |
| 500 | t.Fatal(err) |
| 501 | } |
| 502 | if meta.Status != Running || meta.FinishedAt != 0 || meta.ArtifactComplete { |
| 503 | close(release) |
| 504 | t.Fatalf("running metadata = %+v", meta) |
| 505 | } |
| 506 | |
| 507 | close(release) |
| 508 | <-job.done |
| 509 | meta, err = readMeta(metaPath) |
| 510 | if err != nil { |
| 511 | t.Fatal(err) |
| 512 | } |
| 513 | if meta.Status != Done || meta.FinishedAt == 0 || !meta.ArtifactComplete { |
| 514 | t.Fatalf("terminal metadata = %+v", meta) |
| 515 | } |
| 516 | } |
| 517 | |
| 518 | func TestTaskMutationEvidencePersistsWithoutSensitiveReceiptData(t *testing.T) { |
| 519 | sessionPath := filepath.Join(t.TempDir(), "session.jsonl") |
| 520 | first := NewManager(event.Discard) |
| 521 | first.SetActiveSessionPath("session", sessionPath) |
| 522 | const secret = "private-receipt-value-123456" |
| 523 | j := first.StartForSession("session", "task", "writer", func(ctx context.Context, _ io.Writer) (string, error) { |
| 524 | PublishEvidence(ctx, evidence.ChildEvidenceSummary{Receipts: []evidence.Receipt{ |
| 525 | { |
| 526 | ToolName: "write_file", |
| 527 | Args: json.RawMessage(`{"path":"internal/agent/task.go","content":"` + secret + `"}`), |
| 528 | Success: true, |
| 529 | Write: true, |
| 530 | Mutation: true, |
| 531 | Paths: []string{"internal/agent/task.go"}, |
| 532 | }, |
| 533 | { |
| 534 | ToolName: "bash", |
| 535 | Success: true, |
| 536 | Command: "go test ./... --token=" + secret, |
| 537 | }, |
| 538 | }}) |
| 539 | return "persisted answer", nil |
| 540 | }) |
| 541 | <-j.done |
| 542 | first.Close() |
| 543 | |
| 544 | metaPath := filepath.Join(ArtifactDir(sessionPath), j.ID+jobMetaExt) |
| 545 | data, err := os.ReadFile(metaPath) |
| 546 | if err != nil { |
| 547 | t.Fatal(err) |
| 548 | } |
| 549 | text := string(data) |
| 550 | for _, leaked := range []string{secret, "content", "go test ./..."} { |
| 551 | if strings.Contains(text, leaked) { |
| 552 | t.Fatalf("job metadata persisted sensitive receipt data %q:\n%s", leaked, text) |
| 553 | } |
| 554 | } |
| 555 | for _, want := range []string{`"mutationEvidenceVersion": 1`, `"risk": "medium"`, `"internal/agent/task.go"`} { |
| 556 | if !strings.Contains(text, want) { |
| 557 | t.Fatalf("job metadata missing %q:\n%s", want, text) |
| 558 | } |
| 559 | } |
| 560 | |
| 561 | second := NewManager(event.Discard) |
| 562 | defer second.Close() |
| 563 | second.SetActiveSessionPath("session", sessionPath) |
| 564 | summary := second.LeaseEvidenceForSession("session", j.ID) |
| 565 | if len(summary.Receipts) != 1 { |
| 566 | t.Fatalf("restored evidence = %+v, want one synthetic mutation", summary) |
| 567 | } |
| 568 | receipt := summary.Receipts[0] |
| 569 | if !receipt.Success || !receipt.Mutation || !receipt.Write || receipt.ToolName != recoveredBackgroundTaskToolName { |
| 570 | t.Fatalf("restored receipt = %+v, want successful recovered mutation", receipt) |
| 571 | } |
| 572 | if len(receipt.Paths) != 1 || filepath.ToSlash(receipt.Paths[0]) != "internal/agent/task.go" { |
| 573 | t.Fatalf("restored paths = %v, want internal/agent/task.go", receipt.Paths) |
| 574 | } |
| 575 | if len(receipt.Args) != 0 || receipt.Command != "" || receipt.Read { |
| 576 | t.Fatalf("restored receipt retained stale sign-off data: %+v", receipt) |
| 577 | } |
| 578 | |
| 579 | ledger := evidence.NewLedger() |
| 580 | ledger.MergeChild(summary) |
| 581 | mutation, ok := ledger.LatestSuccessfulMutationIndex() |
| 582 | if !ok || ledger.HasSuccessfulReviewAfter(mutation) || ledger.HasSuccessfulVerificationCommand() { |
| 583 | t.Fatalf("restored evidence bypassed fresh review/verification: %+v", ledger.Summary()) |
| 584 | } |
| 585 | if got := ledger.MutationRiskAfter(mutation); got != evidence.RiskMedium { |
| 586 | t.Fatalf("restored mutation risk = %s, want medium", got) |
| 587 | } |
| 588 | // Lease does not consume: the receipts stay available until the collecting |
| 589 | // turn commits. Only then is the persisted summary drained. |
| 590 | if again := second.LeaseEvidenceForSession("session", j.ID); len(again.Receipts) != 1 { |
| 591 | t.Fatalf("restored evidence not re-leasable before commit: %+v", again) |
| 592 | } |
| 593 | second.CommitEvidenceForSession("session", j.ID) |
| 594 | if afterCommit := second.LeaseEvidenceForSession("session", j.ID); len(afterCommit.Receipts) != 0 { |
| 595 | t.Fatalf("committed evidence still leasable: %+v", afterCommit) |
| 596 | } |
| 597 | |
| 598 | // The commit drained the persisted copy too — a further restart must not |
| 599 | // offer the same mutation again. |
| 600 | third := NewManager(event.Discard) |
| 601 | defer third.Close() |
| 602 | third.SetActiveSessionPath("session", sessionPath) |
| 603 | if thirdLease := third.LeaseEvidenceForSession("session", j.ID); len(thirdLease.Receipts) != 0 { |
| 604 | t.Fatalf("committed evidence resurrected after restart: %+v", thirdLease) |
| 605 | } |
| 606 | } |
| 607 | |
| 608 | func TestHighRiskTaskMutationEvidenceRestoresAsOpaque(t *testing.T) { |
| 609 | meta := artifactMeta{ |
| 610 | Kind: "task", |
| 611 | MutationEvidenceVersion: mutationEvidenceVersion, |
| 612 | MutationEvidence: &artifactMutationEvidence{ |
| 613 | Risk: string(evidence.RiskHigh), |
| 614 | Paths: []string{"ordinary-looking.go"}, |
| 615 | }, |
| 616 | } |
| 617 | summary := mutationEvidenceFromArtifact(meta) |
| 618 | if len(summary.Receipts) != 1 || len(summary.Receipts[0].Paths) != 0 { |
| 619 | t.Fatalf("high-risk restored evidence = %+v, want opaque mutation", summary) |
| 620 | } |
| 621 | ledger := evidence.NewLedger() |
| 622 | ledger.MergeChild(summary) |
| 623 | mutation, ok := ledger.LatestSuccessfulMutationIndex() |
| 624 | if !ok || ledger.MutationRiskAfter(mutation) != evidence.RiskHigh { |
| 625 | t.Fatalf("high-risk mutation was downgraded during recovery: %+v", ledger.Summary()) |
| 626 | } |
| 627 | } |
| 628 | |
| 629 | func TestLegacyTaskArtifactRecoversAsOpaqueHighRiskMutation(t *testing.T) { |
| 630 | // A pre-feature artifact (no mutationEvidenceVersion) proves only that the |
| 631 | // mutation state was never recorded — not that the task made no changes. A |
| 632 | // legacy background writer task collected after upgrade could carry real, |
| 633 | // unreviewed edits, so recovery must be conservative: an opaque RiskHigh |
| 634 | // mutation that forces fresh inspection and review rather than silently |
| 635 | // skipping it. |
| 636 | sessionPath := filepath.Join(t.TempDir(), "session.jsonl") |
| 637 | dir := ArtifactDir(sessionPath) |
| 638 | if err := os.MkdirAll(dir, 0o755); err != nil { |
| 639 | t.Fatal(err) |
| 640 | } |
| 641 | if err := os.WriteFile(filepath.Join(dir, "task-1"+jobLogExt), []byte("legacy answer"), 0o600); err != nil { |
| 642 | t.Fatal(err) |
| 643 | } |
| 644 | if err := writeMeta(filepath.Join(dir, "task-1"+jobMetaExt), artifactMeta{ |
| 645 | ID: "task-1", |
| 646 | Kind: "task", |
| 647 | Status: Done, |
| 648 | ArtifactComplete: true, |
| 649 | LogPath: "task-1" + jobLogExt, |
| 650 | }); err != nil { |
| 651 | t.Fatal(err) |
| 652 | } |
| 653 | |
| 654 | m := NewManager(event.Discard) |
| 655 | defer m.Close() |
| 656 | m.SetActiveSessionPath("session", sessionPath) |
| 657 | summary := m.LeaseEvidenceForSession("session", "task-1") |
| 658 | if len(summary.Receipts) != 1 || !summary.HasMutation() || len(summary.MutationPaths()) != 0 { |
| 659 | t.Fatalf("legacy task evidence = %+v, want one opaque mutation", summary) |
| 660 | } |
| 661 | ledger := evidence.NewLedger() |
| 662 | ledger.MergeChild(summary) |
| 663 | mutation, ok := ledger.LatestSuccessfulMutationIndex() |
| 664 | if !ok || ledger.MutationRiskAfter(mutation) != evidence.RiskHigh { |
| 665 | t.Fatalf("legacy task mutation was not recovered conservatively: %+v", ledger.Summary()) |
| 666 | } |
| 667 | } |
| 668 | |
| 669 | func TestFutureVersionTaskArtifactRecoversAsOpaqueHighRiskMutation(t *testing.T) { |
| 670 | // A meta written by a newer build (unknown non-zero version) may contain |
| 671 | // real evidence in a shape this build cannot parse: recover it as an opaque |
| 672 | // mutation so downgrade coexistence cannot skip review. |
| 673 | meta := artifactMeta{ |
| 674 | Kind: "task", |
| 675 | MutationEvidenceVersion: mutationEvidenceVersion + 1, |
| 676 | } |
| 677 | summary := mutationEvidenceFromArtifact(meta) |
| 678 | if len(summary.Receipts) != 1 || !summary.HasMutation() || len(summary.MutationPaths()) != 0 { |
| 679 | t.Fatalf("future-version evidence = %+v, want one opaque mutation", summary) |
| 680 | } |
| 681 | ledger := evidence.NewLedger() |
| 682 | ledger.MergeChild(summary) |
| 683 | mutation, ok := ledger.LatestSuccessfulMutationIndex() |
| 684 | if !ok || ledger.MutationRiskAfter(mutation) != evidence.RiskHigh { |
| 685 | t.Fatalf("future-version mutation was not recovered conservatively: %+v", ledger.Summary()) |
| 686 | } |
| 687 | } |
| 688 | |
| 689 | func TestLeasedEvidenceResurrectsUntilCommitted(t *testing.T) { |
| 690 | // Collection is provisional. A lease that is never committed — the |
| 691 | // collecting turn was cancelled, errored, or the process exited before |
| 692 | // delivery — must leave the mutation recoverable after a restart, so a |
| 693 | // background change can never ship unreviewed. Only a commit drains the |
| 694 | // persisted copy. |
| 695 | sessionPath := filepath.Join(t.TempDir(), "session.jsonl") |
| 696 | first := NewManager(event.Discard) |
| 697 | first.SetActiveSessionPath("session", sessionPath) |
| 698 | j := first.StartForSession("session", "task", "writer", func(ctx context.Context, _ io.Writer) (string, error) { |
| 699 | PublishEvidence(ctx, evidence.ChildEvidenceSummary{Receipts: []evidence.Receipt{{ |
| 700 | ToolName: "write_file", Success: true, Write: true, Mutation: true, Paths: []string{"changed.go"}, |
| 701 | }}}) |
| 702 | return "done", nil |
| 703 | }) |
| 704 | <-j.done |
| 705 | // Lease without committing (the turn never delivered), then restart. |
| 706 | if leased := first.LeaseEvidenceForSession("session", j.ID); !leased.HasMutation() { |
| 707 | t.Fatalf("live lease = %+v, want the published mutation", leased) |
| 708 | } |
| 709 | first.Close() |
| 710 | |
| 711 | data, err := os.ReadFile(filepath.Join(ArtifactDir(sessionPath), j.ID+jobMetaExt)) |
| 712 | if err != nil { |
| 713 | t.Fatal(err) |
| 714 | } |
| 715 | if !strings.Contains(string(data), `"mutationEvidence"`) { |
| 716 | t.Fatalf("uncommitted lease drained the persisted mutation summary:\n%s", data) |
| 717 | } |
| 718 | |
| 719 | second := NewManager(event.Discard) |
| 720 | defer second.Close() |
| 721 | second.SetActiveSessionPath("session", sessionPath) |
| 722 | if summary := second.LeaseEvidenceForSession("session", j.ID); !summary.HasMutation() { |
| 723 | t.Fatalf("uncommitted evidence lost after restart: %+v", summary) |
| 724 | } |
| 725 | // Committing after the restart drains it; a further restart offers nothing. |
| 726 | second.CommitEvidenceForSession("session", j.ID) |
| 727 | second.Close() |
| 728 | third := NewManager(event.Discard) |
| 729 | defer third.Close() |
| 730 | third.SetActiveSessionPath("session", sessionPath) |
| 731 | if summary := third.LeaseEvidenceForSession("session", j.ID); len(summary.Receipts) != 0 { |
| 732 | t.Fatalf("committed evidence resurrected after restart: %+v", summary) |
| 733 | } |
| 734 | } |
| 735 | |
| 736 | func TestFinishDestroySessionPurgesOwnedJobs(t *testing.T) { |
| 737 | m := NewManager(event.Discard) |
| 738 | defer m.Close() |
| 739 | |
| 740 | j := m.StartForSession("session", "task", "done", func(context.Context, io.Writer) (string, error) { |
| 741 | return "answer", nil |
| 742 | }) |
| 743 | <-j.done |
| 744 | |
| 745 | done := m.DestroySession("session") |
| 746 | if len(done) != 0 { |
| 747 | t.Fatalf("finished job should not need destroy wait, got %d handles", len(done)) |
| 748 | } |
| 749 | m.FinishDestroySession("session") |
| 750 | |
| 751 | if _, _, ok := m.OutputForSession("session", j.ID); ok { |
| 752 | t.Fatalf("destroyed session job %s should be purged", j.ID) |
| 753 | } |
| 754 | } |
| 755 | |
| 756 | func TestSetActiveSessionPathMigratesRunningJobArtifacts(t *testing.T) { |
| 757 | sessionPath := filepath.Join(t.TempDir(), "session.jsonl") |
| 758 | m := NewManager(event.Discard) |
| 759 | defer m.Close() |
| 760 | |
| 761 | wroteBefore := make(chan struct{}) |
| 762 | release := make(chan struct{}) |
| 763 | var releaseOnce sync.Once |
| 764 | defer releaseOnce.Do(func() { close(release) }) |
| 765 | j := m.StartForSession("session", "bash", "migrate", func(_ context.Context, out io.Writer) (string, error) { |
| 766 | _, _ = io.WriteString(out, "before\n") |
| 767 | close(wroteBefore) |
| 768 | <-release |
| 769 | _, _ = io.WriteString(out, "after\n") |
| 770 | return "", nil |
| 771 | }) |
| 772 | <-wroteBefore |
| 773 | j.mu.Lock() |
| 774 | oldPath := j.artifactPath |
| 775 | j.mu.Unlock() |
| 776 | |
| 777 | m.SetActiveSessionPath("session", sessionPath) |
| 778 | j.mu.Lock() |
| 779 | gotPath := j.artifactPath |
| 780 | j.mu.Unlock() |
| 781 | if gotPath != oldPath { |
| 782 | t.Fatalf("running artifact path = %q, want unchanged %q before completion", gotPath, oldPath) |
| 783 | } |
| 784 | |
| 785 | releaseOnce.Do(func() { close(release) }) |
| 786 | <-j.done |
| 787 | j.mu.Lock() |
| 788 | donePath := j.artifactPath |
| 789 | j.mu.Unlock() |
| 790 | if !strings.HasPrefix(donePath, ArtifactDir(sessionPath)+string(filepath.Separator)) { |
| 791 | t.Fatalf("completed artifact path = %q, want under %q", donePath, ArtifactDir(sessionPath)) |
| 792 | } |
| 793 | res := m.WaitForSession(context.Background(), "session", []string{j.ID}, 1) |
| 794 | if len(res) != 1 || !strings.Contains(res[0].Output, "before\n") || !strings.Contains(res[0].Output, "after\n") { |
| 795 | t.Fatalf("wait after migration = %+v, want before and after output", res) |
| 796 | } |
| 797 | } |
| 798 | |
| 799 | func TestArtifactFailureDoesNotFailSuccessfulJob(t *testing.T) { |
| 800 | dir := t.TempDir() |
| 801 | sessionPath := filepath.Join(dir, "session.jsonl") |
| 802 | if err := os.WriteFile(ArtifactDir(sessionPath), []byte("not a dir"), 0o644); err != nil { |
| 803 | t.Fatal(err) |
| 804 | } |
| 805 | m := NewManager(event.Discard) |
| 806 | defer m.Close() |
| 807 | m.SetActiveSessionPath("session", sessionPath) |
| 808 | |
| 809 | j := m.StartForSession("session", "task", "artifact fail", func(context.Context, io.Writer) (string, error) { |
| 810 | return "successful result", nil |
| 811 | }) |
| 812 | <-j.done |
| 813 | |
| 814 | res := m.WaitForSession(context.Background(), "session", []string{j.ID}, 1) |
| 815 | if len(res) != 1 || res[0].Status != Done { |
| 816 | t.Fatalf("wait = %+v, want one done result", res) |
| 817 | } |
| 818 | if !strings.Contains(res[0].Output, "successful result") || !strings.Contains(res[0].Output, "job artifact incomplete:") { |
| 819 | t.Fatalf("output = %q, want result and artifact warning", res[0].Output) |
| 820 | } |
| 821 | } |
| 822 | |
| 823 | func TestMigrateArtifactDirFallsBackToCopyWhenRenameFails(t *testing.T) { |
| 824 | root := t.TempDir() |
| 825 | src := filepath.Join(root, "src") |
| 826 | dst := filepath.Join(root, "dst") |
| 827 | if err := os.MkdirAll(src, 0o755); err != nil { |
| 828 | t.Fatal(err) |
| 829 | } |
| 830 | if err := os.WriteFile(filepath.Join(src, "bash-1.log"), []byte("persisted output"), 0o644); err != nil { |
| 831 | t.Fatal(err) |
| 832 | } |
| 833 | |
| 834 | oldRename := renamePath |
| 835 | renamePath = func(_, _ string) error { |
| 836 | return errors.New("forced rename failure") |
| 837 | } |
| 838 | t.Cleanup(func() { renamePath = oldRename }) |
| 839 | |
| 840 | if err := migrateArtifactDir(src, dst); err != nil { |
| 841 | t.Fatalf("migrateArtifactDir: %v", err) |
| 842 | } |
| 843 | got, err := os.ReadFile(filepath.Join(dst, "bash-1.log")) |
| 844 | if err != nil { |
| 845 | t.Fatalf("read migrated artifact: %v", err) |
| 846 | } |
| 847 | if string(got) != "persisted output" { |
| 848 | t.Fatalf("migrated artifact = %q, want persisted output", got) |
| 849 | } |
| 850 | if runtime.GOOS != "windows" { |
| 851 | assertPrivateArtifactMode(t, filepath.Join(dst, "bash-1.log")) |
| 852 | } |
| 853 | if _, err := os.Stat(filepath.Join(src, "bash-1.log")); !os.IsNotExist(err) { |
| 854 | t.Fatalf("source artifact should be removed after copy fallback, stat err = %v", err) |
| 855 | } |
| 856 | } |
| 857 | |
| 858 | func TestSetActiveSessionPathAdoptsUnscopedTemporaryJobs(t *testing.T) { |
| 859 | sessionPath := filepath.Join(t.TempDir(), "session.jsonl") |
| 860 | m := NewManager(event.Discard) |
| 861 | defer m.Close() |
| 862 | |
| 863 | j := m.StartForSession("", "task", "temporary", func(context.Context, io.Writer) (string, error) { |
| 864 | return "temporary answer", nil |
| 865 | }) |
| 866 | <-j.done |
| 867 | |
| 868 | m.SetActiveSessionPath("session", sessionPath) |
| 869 | res := m.WaitForSession(context.Background(), "session", []string{j.ID}, 1) |
| 870 | if len(res) != 1 || !strings.Contains(res[0].Output, "temporary answer") { |
| 871 | t.Fatalf("adopted wait = %+v, want temporary answer", res) |
| 872 | } |
| 873 | if _, _, ok := m.OutputForSession("", j.ID); !ok { |
| 874 | t.Fatalf("legacy unscoped lookup should still find adopted job %s", j.ID) |
| 875 | } |
| 876 | if _, err := os.Stat(filepath.Join(ArtifactDir(sessionPath), j.ID+jobLogExt)); err != nil { |
| 877 | t.Fatalf("adopted artifact should be under persistent sidecar: %v", err) |
| 878 | } |
| 879 | } |
| 880 | |
| 881 | func TestSetActiveSessionPathAdoptsUnscopedJobsOnMigrationFailure(t *testing.T) { |
| 882 | dir := t.TempDir() |
| 883 | sessionPath := filepath.Join(dir, "session.jsonl") |
| 884 | if err := os.WriteFile(ArtifactDir(sessionPath), []byte("not a dir"), 0o644); err != nil { |
| 885 | t.Fatal(err) |
| 886 | } |
| 887 | m := NewManager(event.Discard) |
| 888 | defer m.Close() |
| 889 | |
| 890 | j := m.StartForSession("", "task", "temporary", func(context.Context, io.Writer) (string, error) { |
| 891 | return "temporary answer", nil |
| 892 | }) |
| 893 | <-j.done |
| 894 | |
| 895 | m.SetActiveSessionPath("session", sessionPath) |
| 896 | |
| 897 | out, status, ok := m.OutputForSession("session", j.ID) |
| 898 | if !ok || status != Done { |
| 899 | t.Fatalf("scoped output ok/status = %v/%s, want true/done", ok, status) |
| 900 | } |
| 901 | if !strings.Contains(out, "temporary answer") || !strings.Contains(out, "job artifact incomplete: migration:") { |
| 902 | t.Fatalf("scoped output = %q, want answer and migration error", out) |
| 903 | } |
| 904 | res := m.WaitForSession(context.Background(), "session", []string{j.ID}, 1) |
| 905 | if len(res) != 1 || !strings.Contains(res[0].Output, "job artifact incomplete: migration:") { |
| 906 | t.Fatalf("scoped wait = %+v, want migration error", res) |
| 907 | } |
| 908 | if note := m.DrainCompletedNoteForSession("session"); !strings.Contains(note, j.ID) { |
| 909 | t.Fatalf("adopted completion note = %q, want job id %s", note, j.ID) |
| 910 | } |
| 911 | } |
| 912 | |
| 913 | func TestSetActiveSessionPathReportsMigrationFailure(t *testing.T) { |
| 914 | dir := t.TempDir() |
| 915 | sessionPath := filepath.Join(dir, "session.jsonl") |
| 916 | if err := os.WriteFile(ArtifactDir(sessionPath), []byte("not a dir"), 0o644); err != nil { |
| 917 | t.Fatal(err) |
| 918 | } |
| 919 | var noticesMu sync.Mutex |
| 920 | var notices []event.Event |
| 921 | m := NewManager(event.FuncSink(func(e event.Event) { |
| 922 | if e.Kind == event.Notice { |
| 923 | noticesMu.Lock() |
| 924 | notices = append(notices, e) |
| 925 | noticesMu.Unlock() |
| 926 | } |
| 927 | })) |
| 928 | defer m.Close() |
| 929 | |
| 930 | j := m.StartForSession("session", "task", "migrate fail", func(context.Context, io.Writer) (string, error) { |
| 931 | return "answer", nil |
| 932 | }) |
| 933 | m.SetActiveSessionPath("session", sessionPath) |
| 934 | <-j.done |
| 935 | |
| 936 | res := m.WaitForSession(context.Background(), "session", []string{j.ID}, 1) |
| 937 | if len(res) != 1 || !strings.Contains(res[0].Output, "job artifact incomplete: migration:") { |
| 938 | t.Fatalf("wait after migration failure = %+v, want artifact error", res) |
| 939 | } |
| 940 | noticesMu.Lock() |
| 941 | capturedNotices := append([]event.Event(nil), notices...) |
| 942 | noticesMu.Unlock() |
| 943 | found := false |
| 944 | for _, notice := range capturedNotices { |
| 945 | if notice.Text == "Job artifact migration failed." && strings.Contains(notice.Detail, "job artifact migration failed") { |
| 946 | found = true |
| 947 | break |
| 948 | } |
| 949 | } |
| 950 | if !found { |
| 951 | t.Fatalf("migration failure notice not emitted, got %+v", capturedNotices) |
| 952 | } |
| 953 | } |
| 954 | |
| 955 | func TestOutputReadsArtifactFromOffset(t *testing.T) { |
| 956 | dir := t.TempDir() |
| 957 | path := filepath.Join(dir, "bash-1.log") |
| 958 | prefix := strings.Repeat("x", 2*defaultTailBytes) |
| 959 | suffix := "new output\n" |
| 960 | if err := os.WriteFile(path, []byte(prefix+suffix), 0o644); err != nil { |
| 961 | t.Fatal(err) |
| 962 | } |
| 963 | |
| 964 | m := NewManager(event.Discard) |
| 965 | defer m.Close() |
| 966 | j := &Job{ |
| 967 | ID: "bash-1", |
| 968 | Kind: "bash", |
| 969 | SessionID: "session", |
| 970 | status: Running, |
| 971 | readOffset: int64(len(prefix)), |
| 972 | artifactPath: path, |
| 973 | done: make(chan struct{}), |
| 974 | } |
| 975 | m.jobs[jobKey("session", j.ID)] = j |
| 976 | m.order = append(m.order, jobKey("session", j.ID)) |
| 977 | |
| 978 | text, status, ok := m.OutputForSession("session", j.ID) |
| 979 | if !ok || status != Running { |
| 980 | t.Fatalf("OutputForSession ok/status = %v/%s, want true/running", ok, status) |
| 981 | } |
| 982 | if text != suffix { |
| 983 | t.Fatalf("OutputForSession text = %q, want %q", text, suffix) |
| 984 | } |
| 985 | if j.readOffset != int64(len(prefix)+len(suffix)) { |
| 986 | t.Fatalf("readOffset = %d, want %d", j.readOffset, len(prefix)+len(suffix)) |
| 987 | } |
| 988 | } |
| 989 | |
| 990 | func TestRestoredArtifactsAreScopedBySession(t *testing.T) { |
| 991 | root := t.TempDir() |
| 992 | pathA := filepath.Join(root, "a.jsonl") |
| 993 | pathB := filepath.Join(root, "b.jsonl") |
| 994 | |
| 995 | first := NewManager(event.Discard) |
| 996 | first.SetActiveSessionPath("session-a", pathA) |
| 997 | jobA := first.StartForSession("session-a", "bash", "a", func(_ context.Context, out io.Writer) (string, error) { |
| 998 | _, _ = io.WriteString(out, "from-a") |
| 999 | return "", nil |
| 1000 | }) |
| 1001 | <-jobA.done |
| 1002 | first.Close() |
| 1003 | |
| 1004 | second := NewManager(event.Discard) |
| 1005 | second.SetActiveSessionPath("session-b", pathB) |
| 1006 | jobB := second.StartForSession("session-b", "bash", "b", func(_ context.Context, out io.Writer) (string, error) { |
| 1007 | _, _ = io.WriteString(out, "from-b") |
| 1008 | return "", nil |
| 1009 | }) |
| 1010 | <-jobB.done |
| 1011 | second.Close() |
| 1012 | |
| 1013 | if jobA.ID != "bash-1" || jobB.ID != "bash-1" { |
| 1014 | t.Fatalf("test setup expected duplicate ids, got %s and %s", jobA.ID, jobB.ID) |
| 1015 | } |
| 1016 | |
| 1017 | m := NewManager(event.Discard) |
| 1018 | defer m.Close() |
| 1019 | m.SetActiveSessionPath("session-a", pathA) |
| 1020 | m.SetActiveSessionPath("session-b", pathB) |
| 1021 | |
| 1022 | resA := m.WaitForSession(context.Background(), "session-a", []string{"bash-1"}, 1) |
| 1023 | if len(resA) != 1 || resA[0].Output != "from-a" { |
| 1024 | t.Fatalf("session-a wait = %+v, want from-a", resA) |
| 1025 | } |
| 1026 | resB := m.WaitForSession(context.Background(), "session-b", []string{"bash-1"}, 1) |
| 1027 | if len(resB) != 1 || resB[0].Output != "from-b" { |
| 1028 | t.Fatalf("session-b wait = %+v, want from-b", resB) |
| 1029 | } |
| 1030 | |
| 1031 | next := m.StartForSession("session-b", "bash", "next", func(context.Context, io.Writer) (string, error) { |
| 1032 | return "", nil |
| 1033 | }) |
| 1034 | <-next.done |
| 1035 | if next.ID == "bash-1" { |
| 1036 | t.Fatal("new job should not reuse restored bash-1") |
| 1037 | } |
| 1038 | } |
| 1039 |