| 1 | package taskmonitor |
| 2 | |
| 3 | import ( |
| 4 | "context" |
| 5 | "strings" |
| 6 | "sync" |
| 7 | "testing" |
| 8 | "time" |
| 9 | ) |
| 10 | |
| 11 | func TestControlService_StopTask(t *testing.T) { |
| 12 | s := NewInMemoryStore() |
| 13 | cs := NewControlService(s) |
| 14 | ctx := context.Background() |
| 15 | |
| 16 | mustUpsertControl(t, s, "/p", TaskSnapshot{ |
| 17 | SchemaVersion: 1, TaskID: "t1", SessionID: "s1", |
| 18 | State: TaskStateRunning, Version: 1, |
| 19 | CreatedAt: time.Now(), UpdatedAt: time.Now(), |
| 20 | }) |
| 21 | |
| 22 | res, err := cs.StopTaskWithKiller(ctx, "/p", "t1", 1, "user request", "idem-1", &mockKiller{fn: func(string, string) bool { return true }}) |
| 23 | if err != nil { |
| 24 | t.Fatalf("StopTask: %v", err) |
| 25 | } |
| 26 | if !res.Accepted { |
| 27 | t.Errorf("expected accepted, got %+v", res) |
| 28 | } |
| 29 | if res.State != TaskStateCancelled { |
| 30 | t.Errorf("expected cancelled, got %q", res.State) |
| 31 | } |
| 32 | if res.Version != 2 { |
| 33 | t.Errorf("expected version 2, got %d", res.Version) |
| 34 | } |
| 35 | } |
| 36 | |
| 37 | func TestControlService_StopRoutesNamespacedTaskToRuntimeJobID(t *testing.T) { |
| 38 | s := NewInMemoryStore() |
| 39 | cs := NewControlService(s) |
| 40 | now := time.Now() |
| 41 | mustUpsertControl(t, s, "/p", TaskSnapshot{ |
| 42 | SchemaVersion: 1, TaskID: "session-1--task-1", JobID: "task-1", SessionID: "session-1", |
| 43 | State: TaskStateRunning, RuntimeState: RuntimeStateAlive, Version: 1, |
| 44 | CreatedAt: now, UpdatedAt: now, |
| 45 | }) |
| 46 | |
| 47 | killer := &mockKiller{fn: func(sessionID, jobID string) bool { |
| 48 | return sessionID == "session-1" && jobID == "task-1" |
| 49 | }} |
| 50 | res, err := cs.StopTaskWithKiller(context.Background(), "/p", "session-1--task-1", 1, "", "", killer) |
| 51 | if err != nil || !res.Accepted { |
| 52 | t.Fatalf("namespaced stop: result=%+v err=%v", res, err) |
| 53 | } |
| 54 | } |
| 55 | |
| 56 | func TestRuntimeJobIDSupportsSnapshotsBeforeJobIDField(t *testing.T) { |
| 57 | longSession := strings.Repeat("s", maxFieldLen) |
| 58 | for _, tc := range []struct { |
| 59 | name string |
| 60 | snap TaskSnapshot |
| 61 | want string |
| 62 | }{ |
| 63 | {name: "legacy raw id", snap: TaskSnapshot{TaskID: "task-1", SessionID: "session-1"}, want: "task-1"}, |
| 64 | {name: "namespaced id", snap: TaskSnapshot{TaskID: "session-1--task-1", SessionID: "session-1"}, want: "task-1"}, |
| 65 | {name: "hashed namespace", snap: TaskSnapshot{TaskID: monitorTaskID(longSession, "task-1"), SessionID: longSession}, want: "task-1"}, |
| 66 | {name: "explicit id", snap: TaskSnapshot{TaskID: "monitor-id", JobID: "bash-2", SessionID: "session-1"}, want: "bash-2"}, |
| 67 | } { |
| 68 | t.Run(tc.name, func(t *testing.T) { |
| 69 | if got := runtimeJobID(&tc.snap); got != tc.want { |
| 70 | t.Fatalf("runtimeJobID() = %q, want %q", got, tc.want) |
| 71 | } |
| 72 | }) |
| 73 | } |
| 74 | } |
| 75 | |
| 76 | func TestControlService_StopRequiresRuntimeOwner(t *testing.T) { |
| 77 | s := NewInMemoryStore() |
| 78 | cs := NewControlService(s) |
| 79 | mustUpsertControl(t, s, "/p", TaskSnapshot{ |
| 80 | SchemaVersion: 1, TaskID: "t1", SessionID: "s1", |
| 81 | State: TaskStateRunning, Version: 1, |
| 82 | CreatedAt: time.Now(), UpdatedAt: time.Now(), |
| 83 | }) |
| 84 | |
| 85 | res, err := cs.StopTask(context.Background(), "/p", "t1", 1, "", "") |
| 86 | if err != nil || res.Accepted || res.Error == nil || res.Error.Code != ErrTaskRuntimeUnavailable { |
| 87 | t.Fatalf("expected unavailable runtime, got result=%+v err=%v", res, err) |
| 88 | } |
| 89 | snap, _ := s.GetTask(context.Background(), "/p", "t1") |
| 90 | if snap.State != TaskStateRunning || snap.Version != 1 { |
| 91 | t.Fatalf("failed stop mutated task: %+v", snap) |
| 92 | } |
| 93 | } |
| 94 | |
| 95 | func TestControlService_CancelRejectsUnreachableRuntime(t *testing.T) { |
| 96 | s := NewInMemoryStore() |
| 97 | cs := NewControlService(s) |
| 98 | mustUpsertControl(t, s, "/p", TaskSnapshot{ |
| 99 | SchemaVersion: 1, TaskID: "t1", SessionID: "s1", |
| 100 | State: TaskStateRunning, Version: 1, |
| 101 | CreatedAt: time.Now(), UpdatedAt: time.Now(), |
| 102 | }) |
| 103 | |
| 104 | killer := &mockKiller{fn: func(string, string) bool { return false }} |
| 105 | res, err := cs.CancelTaskWithKiller(context.Background(), "/p", "t1", 1, "", "", killer) |
| 106 | if err != nil || res.Accepted || res.Error == nil || res.Error.Code != ErrTaskRuntimeUnavailable { |
| 107 | t.Fatalf("expected rejected runtime control, got result=%+v err=%v", res, err) |
| 108 | } |
| 109 | snap, _ := s.GetTask(context.Background(), "/p", "t1") |
| 110 | if snap.State != TaskStateRunning || snap.Version != 1 { |
| 111 | t.Fatalf("failed cancel mutated task: %+v", snap) |
| 112 | } |
| 113 | } |
| 114 | |
| 115 | func TestControlService_VersionConflict(t *testing.T) { |
| 116 | s := NewInMemoryStore() |
| 117 | cs := NewControlService(s) |
| 118 | |
| 119 | mustUpsertControl(t, s, "/p", TaskSnapshot{ |
| 120 | SchemaVersion: 1, TaskID: "t1", SessionID: "s1", |
| 121 | State: TaskStateRunning, Version: 3, |
| 122 | CreatedAt: time.Now(), UpdatedAt: time.Now(), |
| 123 | }) |
| 124 | |
| 125 | res, _ := cs.StopTask(context.Background(), "/p", "t1", 1, "", "") |
| 126 | if res.Accepted || res.Error == nil || res.Error.Code != ErrTaskVersionConflict { |
| 127 | t.Errorf("expected version conflict, got %+v", res) |
| 128 | } |
| 129 | } |
| 130 | |
| 131 | func TestControlService_NotFound(t *testing.T) { |
| 132 | cs := NewControlService(NewInMemoryStore()) |
| 133 | res, _ := cs.StopTask(context.Background(), "/p", "ghost", 1, "", "") |
| 134 | if res.Error == nil || res.Error.Code != ErrTaskNotFound { |
| 135 | t.Errorf("expected not_found, got %+v", res.Error) |
| 136 | } |
| 137 | } |
| 138 | |
| 139 | func TestControlService_TerminalGuard(t *testing.T) { |
| 140 | s := NewInMemoryStore() |
| 141 | cs := NewControlService(s) |
| 142 | mustUpsertControl(t, s, "/p", TaskSnapshot{ |
| 143 | SchemaVersion: 1, TaskID: "t1", SessionID: "s1", |
| 144 | State: TaskStateSucceeded, Version: 1, |
| 145 | CreatedAt: time.Now(), UpdatedAt: time.Now(), |
| 146 | }) |
| 147 | res, _ := cs.StopTask(context.Background(), "/p", "t1", 1, "", "") |
| 148 | if res.Error == nil || res.Error.Code != ErrTaskAlreadyTerminal { |
| 149 | t.Errorf("expected terminal guard, got %+v", res.Error) |
| 150 | } |
| 151 | } |
| 152 | |
| 153 | func TestControlService_RequeueFailedTaskDoesNotClaimLiveRuntime(t *testing.T) { |
| 154 | s := NewInMemoryStore() |
| 155 | cs := NewControlService(s) |
| 156 | mustUpsertControl(t, s, "/p", TaskSnapshot{ |
| 157 | SchemaVersion: 1, TaskID: "failed", SessionID: "s1", |
| 158 | State: TaskStateFailed, RuntimeState: RuntimeStateExited, Version: 3, |
| 159 | CreatedAt: time.Now(), UpdatedAt: time.Now(), |
| 160 | }) |
| 161 | res, err := cs.RequeueTask(context.Background(), "/p", "failed", 3, "requeue-1") |
| 162 | if err != nil || !res.Accepted || res.State != TaskStateQueued || res.RuntimeState != RuntimeStateExited || res.Version != 4 { |
| 163 | t.Fatalf("expected failed task to be requeued without a live runtime, got result=%+v err=%v", res, err) |
| 164 | } |
| 165 | snap, _ := s.GetTask(context.Background(), "/p", "failed") |
| 166 | if snap.RuntimeState != RuntimeStateExited { |
| 167 | t.Fatalf("requeue changed runtime state to %q, want exited", snap.RuntimeState) |
| 168 | } |
| 169 | } |
| 170 | |
| 171 | func TestControlService_RequeueRejectsLiveRuntime(t *testing.T) { |
| 172 | s := NewInMemoryStore() |
| 173 | cs := NewControlService(s) |
| 174 | mustUpsertControl(t, s, "/p", TaskSnapshot{ |
| 175 | SchemaVersion: 1, TaskID: "failed", SessionID: "s1", |
| 176 | State: TaskStateFailed, RuntimeState: RuntimeStateAlive, Version: 3, |
| 177 | CreatedAt: time.Now(), UpdatedAt: time.Now(), |
| 178 | }) |
| 179 | res, err := cs.RequeueTask(context.Background(), "/p", "failed", 3, "") |
| 180 | if err != nil || res.Error == nil || res.Error.Code != ErrTaskInProgress { |
| 181 | t.Fatalf("expected live-runtime guard, got result=%+v err=%v", res, err) |
| 182 | } |
| 183 | } |
| 184 | |
| 185 | func TestControlService_RequeueAllowsExpiredRuntimeLease(t *testing.T) { |
| 186 | now := time.Now().UTC() |
| 187 | s := NewInMemoryStore() |
| 188 | cs := NewControlService(s) |
| 189 | mustUpsertControl(t, s, "/p", TaskSnapshot{ |
| 190 | SchemaVersion: 1, TaskID: "failed", SessionID: "s1", |
| 191 | State: TaskStateFailed, RuntimeState: RuntimeStateAlive, RuntimeLeaseUntil: now.Add(-time.Minute), Version: 3, |
| 192 | CreatedAt: now.Add(-time.Hour), UpdatedAt: now.Add(-time.Minute), |
| 193 | }) |
| 194 | res, err := cs.RequeueTask(context.Background(), "/p", "failed", 3, "") |
| 195 | if err != nil || !res.Accepted || res.State != TaskStateQueued || res.RuntimeState != RuntimeStateExited { |
| 196 | t.Fatalf("expected expired lease to requeue, got result=%+v err=%v", res, err) |
| 197 | } |
| 198 | } |
| 199 | |
| 200 | func TestControlService_RequeueRejectsNonFailedState(t *testing.T) { |
| 201 | s := NewInMemoryStore() |
| 202 | cs := NewControlService(s) |
| 203 | mustUpsertControl(t, s, "/p", TaskSnapshot{ |
| 204 | SchemaVersion: 1, TaskID: "done", SessionID: "s1", |
| 205 | State: TaskStateSucceeded, RuntimeState: RuntimeStateExited, Version: 3, |
| 206 | CreatedAt: time.Now(), UpdatedAt: time.Now(), |
| 207 | }) |
| 208 | res, err := cs.RequeueTask(context.Background(), "/p", "done", 3, "") |
| 209 | if err != nil || res.Error == nil || res.Error.Code != ErrTaskNotRequeueable { |
| 210 | t.Fatalf("expected not-requeueable guard, got result=%+v err=%v", res, err) |
| 211 | } |
| 212 | } |
| 213 | |
| 214 | func TestControlService_Idempotency(t *testing.T) { |
| 215 | s := NewInMemoryStore() |
| 216 | cs := NewControlService(s) |
| 217 | |
| 218 | mustUpsertControl(t, s, "/p", TaskSnapshot{ |
| 219 | SchemaVersion: 1, TaskID: "t1", SessionID: "s1", |
| 220 | State: TaskStateRunning, Version: 1, |
| 221 | CreatedAt: time.Now(), UpdatedAt: time.Now(), |
| 222 | }) |
| 223 | |
| 224 | // First call |
| 225 | killer := &mockKiller{fn: func(string, string) bool { return true }} |
| 226 | res1, err := cs.StopTaskWithKiller(context.Background(), "/p", "t1", 1, "", "key-1", killer) |
| 227 | if err != nil || !res1.Accepted { |
| 228 | t.Fatalf("first call failed: %v, %+v", err, res1) |
| 229 | } |
| 230 | |
| 231 | // Second call with same key, op, task, version — idempotent |
| 232 | res2, err := cs.StopTaskWithKiller(context.Background(), "/p", "t1", 1, "", "key-1", killer) |
| 233 | if err != nil { |
| 234 | t.Fatalf("second call: %v", err) |
| 235 | } |
| 236 | if !res2.Idempotent || !res2.Accepted { |
| 237 | t.Errorf("expected idempotent accepted, got %+v", res2) |
| 238 | } |
| 239 | } |
| 240 | |
| 241 | func TestControlService_IdempotencyConflict_DifferentOp(t *testing.T) { |
| 242 | s := NewInMemoryStore() |
| 243 | cs := NewControlService(s) |
| 244 | killer := &mockKiller{fn: func(string, string) bool { return true }} |
| 245 | |
| 246 | mustUpsertControl(t, s, "/p", TaskSnapshot{ |
| 247 | SchemaVersion: 1, TaskID: "t1", SessionID: "s1", |
| 248 | State: TaskStateRunning, Version: 1, |
| 249 | CreatedAt: time.Now(), UpdatedAt: time.Now(), |
| 250 | }) |
| 251 | |
| 252 | cs.StopTaskWithKiller(context.Background(), "/p", "t1", 1, "", "key-1", killer) |
| 253 | // Same key but different command |
| 254 | res, _ := cs.CancelTask(context.Background(), "/p", "t1", 1, "", "key-1") |
| 255 | if !strings.Contains(res.Error.Code, "idempotency") { |
| 256 | t.Errorf("expected idempotency conflict, got %+v", res.Error) |
| 257 | } |
| 258 | } |
| 259 | |
| 260 | func TestControlService_IdempotencyConflict_DifferentVersion(t *testing.T) { |
| 261 | s := NewInMemoryStore() |
| 262 | cs := NewControlService(s) |
| 263 | killer := &mockKiller{fn: func(string, string) bool { return true }} |
| 264 | |
| 265 | mustUpsertControl(t, s, "/p", TaskSnapshot{ |
| 266 | SchemaVersion: 1, TaskID: "t1", SessionID: "s1", |
| 267 | State: TaskStateRunning, Version: 1, |
| 268 | CreatedAt: time.Now(), UpdatedAt: time.Now(), |
| 269 | }) |
| 270 | |
| 271 | cs.StopTaskWithKiller(context.Background(), "/p", "t1", 1, "", "key-1", killer) |
| 272 | res, _ := cs.StopTaskWithKiller(context.Background(), "/p", "t1", 2, "", "key-1", killer) |
| 273 | if !strings.Contains(res.Error.Code, "idempotency") { |
| 274 | t.Errorf("expected idempotency conflict for different version, got %+v", res.Error) |
| 275 | } |
| 276 | } |
| 277 | |
| 278 | func TestControlService_AuditEvent(t *testing.T) { |
| 279 | s := NewInMemoryStore() |
| 280 | cs := NewControlService(s) |
| 281 | killer := &mockKiller{fn: func(string, string) bool { return true }} |
| 282 | |
| 283 | mustUpsertControl(t, s, "/p", TaskSnapshot{ |
| 284 | SchemaVersion: 1, TaskID: "t1", SessionID: "s1", |
| 285 | State: TaskStateRunning, Version: 1, |
| 286 | CreatedAt: time.Now(), UpdatedAt: time.Now(), |
| 287 | }) |
| 288 | |
| 289 | cs.StopTaskWithKiller(context.Background(), "/p", "t1", 1, `stop command "rm -rf ./private" in /Users/alice/project`, "", killer) |
| 290 | |
| 291 | events, _ := s.ListEvents(context.Background(), "/p", "t1", 0) |
| 292 | found := false |
| 293 | for _, ev := range events { |
| 294 | if ev.EventType == "control_stop" { |
| 295 | found = true |
| 296 | if ev.Sequence < 1 { |
| 297 | t.Errorf("expected positive sequence, got %d", ev.Sequence) |
| 298 | } |
| 299 | if ev.ErrorSummary != "" { |
| 300 | t.Errorf("control reason leaked into event: %q", ev.ErrorSummary) |
| 301 | } |
| 302 | if ev.SessionID != "s1" { |
| 303 | t.Errorf("expected session s1, got %q", ev.SessionID) |
| 304 | } |
| 305 | if ev.TaskID != "t1" { |
| 306 | t.Errorf("expected task t1, got %q", ev.TaskID) |
| 307 | } |
| 308 | } |
| 309 | } |
| 310 | if !found { |
| 311 | t.Error("expected audit event for stop") |
| 312 | } |
| 313 | } |
| 314 | |
| 315 | func TestControlService_StopPreservesRuntimeLeaseUntilExit(t *testing.T) { |
| 316 | s := NewInMemoryStore() |
| 317 | cs := NewControlService(s) |
| 318 | now := time.Now() |
| 319 | leaseUntil := now.Add(time.Minute) |
| 320 | mustUpsertControl(t, s, "/p", TaskSnapshot{ |
| 321 | SchemaVersion: 1, TaskID: "t1", SessionID: "s1", |
| 322 | State: TaskStateRunning, RuntimeState: RuntimeStateAlive, |
| 323 | RuntimeLeaseUntil: leaseUntil, RuntimeOwnerID: "owner-1", Version: 1, |
| 324 | CreatedAt: now, UpdatedAt: now, |
| 325 | }) |
| 326 | |
| 327 | res, err := cs.StopTaskWithKiller(context.Background(), "/p", "t1", 1, "", "", &mockKiller{fn: func(string, string) bool { return true }}) |
| 328 | if err != nil || !res.Accepted { |
| 329 | t.Fatalf("stop: result=%+v err=%v", res, err) |
| 330 | } |
| 331 | snap, err := s.GetTask(context.Background(), "/p", "t1") |
| 332 | if err != nil || snap == nil { |
| 333 | t.Fatalf("snapshot: %+v err=%v", snap, err) |
| 334 | } |
| 335 | if snap.RuntimeState != RuntimeStateAlive || snap.RuntimeOwnerID != "owner-1" || !snap.RuntimeLeaseUntil.Equal(leaseUntil) { |
| 336 | t.Fatalf("stop discarded live runtime ownership: %+v", snap) |
| 337 | } |
| 338 | reconciled := *snap |
| 339 | reconcileRuntime(&reconciled, leaseUntil.Add(time.Second)) |
| 340 | if reconciled.State != TaskStateCancelled || reconciled.RuntimeState != RuntimeStateExited { |
| 341 | t.Fatalf("expired cancelled runtime did not reconcile: %+v", reconciled) |
| 342 | } |
| 343 | } |
| 344 | |
| 345 | func TestControlService_StopBoundsLegacyLeaseLessRuntime(t *testing.T) { |
| 346 | s := NewInMemoryStore() |
| 347 | cs := NewControlService(s) |
| 348 | now := time.Now() |
| 349 | mustUpsertControl(t, s, "/p", TaskSnapshot{ |
| 350 | SchemaVersion: 1, TaskID: "t1", SessionID: "s1", |
| 351 | State: TaskStateRunning, RuntimeState: RuntimeStateAlive, |
| 352 | RuntimeOwnerID: "owner-1", Version: 1, CreatedAt: now, UpdatedAt: now, |
| 353 | }) |
| 354 | |
| 355 | res, err := cs.StopTaskWithKiller(context.Background(), "/p", "t1", 1, "", "", &mockKiller{fn: func(string, string) bool { return true }}) |
| 356 | if err != nil || !res.Accepted { |
| 357 | t.Fatalf("stop: result=%+v err=%v", res, err) |
| 358 | } |
| 359 | snap, err := s.GetTask(context.Background(), "/p", "t1") |
| 360 | if err != nil || snap == nil { |
| 361 | t.Fatalf("snapshot: %+v err=%v", snap, err) |
| 362 | } |
| 363 | if snap.RuntimeState != RuntimeStateAlive || snap.RuntimeLeaseUntil.IsZero() || snap.RuntimeOwnerID != "owner-1" { |
| 364 | t.Fatalf("legacy runtime did not receive bounded lease: %+v", snap) |
| 365 | } |
| 366 | if got := snap.RuntimeLeaseUntil.Sub(snap.UpdatedAt); got != runtimeLeaseTTL { |
| 367 | t.Fatalf("lease duration = %v, want %v", got, runtimeLeaseTTL) |
| 368 | } |
| 369 | reconciled := *snap |
| 370 | reconcileRuntime(&reconciled, snap.RuntimeLeaseUntil.Add(time.Second)) |
| 371 | if reconciled.State != TaskStateCancelled || reconciled.RuntimeState != RuntimeStateExited { |
| 372 | t.Fatalf("expired legacy runtime did not reconcile: %+v", reconciled) |
| 373 | } |
| 374 | } |
| 375 | |
| 376 | func TestControlService_FileStoreClaimsIdempotencyBeforeSideEffects(t *testing.T) { |
| 377 | project := t.TempDir() |
| 378 | store := NewFileStore(".reasonix/tasks") |
| 379 | now := time.Now() |
| 380 | if err := store.SaveTask(context.Background(), project, TaskSnapshot{ |
| 381 | SchemaVersion: 1, TaskID: "t1", SessionID: "s1", State: TaskStateRunning, |
| 382 | RuntimeState: RuntimeStateAlive, Version: 1, CreatedAt: now, UpdatedAt: now, |
| 383 | }); err != nil { |
| 384 | t.Fatal(err) |
| 385 | } |
| 386 | claimed := make(chan struct{}) |
| 387 | release := make(chan struct{}) |
| 388 | killer := &mockKiller{fn: func(string, string) bool { |
| 389 | close(claimed) |
| 390 | <-release |
| 391 | return true |
| 392 | }} |
| 393 | firstDone := make(chan ControlResult, 1) |
| 394 | go func() { |
| 395 | res, _ := NewControlService(store).StopTaskWithKiller(context.Background(), project, "t1", 1, "", "same-key", killer) |
| 396 | firstDone <- res |
| 397 | }() |
| 398 | <-claimed |
| 399 | second, err := NewControlService(store).StopTaskWithKiller(context.Background(), project, "t1", 1, "", "same-key", &mockKiller{fn: func(string, string) bool { t.Fatal("second request reached runtime"); return true }}) |
| 400 | if err != nil || second.Error == nil || second.Error.Code != ErrTaskInProgress { |
| 401 | t.Fatalf("expected pending idempotency claim, got result=%+v err=%v", second, err) |
| 402 | } |
| 403 | close(release) |
| 404 | first := <-firstDone |
| 405 | if !first.Accepted || first.State != TaskStateCancelled { |
| 406 | t.Fatalf("first operation not accepted: %+v", first) |
| 407 | } |
| 408 | } |
| 409 | |
| 410 | func TestInMemoryStore_IdempotencyClaimIsPendingUntilFinalized(t *testing.T) { |
| 411 | store := NewInMemoryStore() |
| 412 | r := IdempotencyRecord{Key: "same-key", Op: "stop", TaskID: "t1", Version: 1} |
| 413 | first, err := store.ClaimIdempotency(context.Background(), "/p", r) |
| 414 | if err != nil || first != nil { |
| 415 | t.Fatalf("first claim = %+v, err=%v", first, err) |
| 416 | } |
| 417 | second, err := store.ClaimIdempotency(context.Background(), "/p", r) |
| 418 | if err != nil || second == nil || !second.Pending { |
| 419 | t.Fatalf("second claim = %+v, err=%v; want pending record", second, err) |
| 420 | } |
| 421 | if err := store.FinalizeIdempotency(context.Background(), "/p", r); err != nil { |
| 422 | t.Fatal(err) |
| 423 | } |
| 424 | final, err := store.ClaimIdempotency(context.Background(), "/p", r) |
| 425 | if err != nil || final == nil || final.Pending { |
| 426 | t.Fatalf("final claim = %+v, err=%v; want finalized record", final, err) |
| 427 | } |
| 428 | } |
| 429 | |
| 430 | func TestControlService_AuditSequenceMonotonic(t *testing.T) { |
| 431 | s := NewInMemoryStore() |
| 432 | cs := NewControlService(s) |
| 433 | killer := &mockKiller{fn: func(string, string) bool { return true }} |
| 434 | |
| 435 | mustUpsertControl(t, s, "/p", TaskSnapshot{ |
| 436 | SchemaVersion: 1, TaskID: "t1", SessionID: "s1", |
| 437 | State: TaskStateRunning, Version: 1, |
| 438 | CreatedAt: time.Now(), UpdatedAt: time.Now(), |
| 439 | }) |
| 440 | |
| 441 | // Stop creates audit event sequence 1 |
| 442 | cs.StopTaskWithKiller(context.Background(), "/p", "t1", 1, "", "", killer) |
| 443 | |
| 444 | // Reset task to running (simulate a new execution lifecycle) |
| 445 | s.UpsertTask("/p", TaskSnapshot{ |
| 446 | SchemaVersion: 1, TaskID: "t1", SessionID: "s1", |
| 447 | State: TaskStateRunning, Version: 2, |
| 448 | CreatedAt: time.Now(), UpdatedAt: time.Now(), |
| 449 | }) |
| 450 | |
| 451 | // Cancel should get sequence 2 from NextSequence |
| 452 | res, _ := cs.CancelTaskWithKiller(context.Background(), "/p", "t1", 2, "", "", killer) |
| 453 | if !res.Accepted { |
| 454 | t.Fatalf("cancel failed: %+v", res) |
| 455 | } |
| 456 | |
| 457 | events, _ := s.ListEvents(context.Background(), "/p", "t1", 0) |
| 458 | if len(events) != 2 { |
| 459 | t.Fatalf("expected 2 events, got %d", len(events)) |
| 460 | } |
| 461 | if events[1].Sequence != 2 { |
| 462 | t.Errorf("expected sequence 2, got %d", events[1].Sequence) |
| 463 | } |
| 464 | } |
| 465 | |
| 466 | func TestControlService_KillJob(t *testing.T) { |
| 467 | s := NewInMemoryStore() |
| 468 | cs := NewControlService(s) |
| 469 | |
| 470 | killed := false |
| 471 | mk := &mockKiller{fn: func(sessionID, id string) bool { |
| 472 | killed = true |
| 473 | return sessionID == "s1" && id == "t1" |
| 474 | }} |
| 475 | |
| 476 | mustUpsertControl(t, s, "/p", TaskSnapshot{ |
| 477 | SchemaVersion: 1, TaskID: "t1", SessionID: "s1", |
| 478 | State: TaskStateRunning, Version: 1, |
| 479 | CreatedAt: time.Now(), UpdatedAt: time.Now(), |
| 480 | }) |
| 481 | |
| 482 | res, _ := cs.StopTaskWithKiller(context.Background(), "/p", "t1", 1, "", "", mk) |
| 483 | if !res.Accepted { |
| 484 | t.Fatalf("stop failed: %+v", res) |
| 485 | } |
| 486 | if !killed { |
| 487 | t.Error("expected Kill to be called for stop") |
| 488 | } |
| 489 | } |
| 490 | |
| 491 | func TestControlService_KillNotCalledForTerminalTask(t *testing.T) { |
| 492 | s := NewInMemoryStore() |
| 493 | cs := NewControlService(s) |
| 494 | |
| 495 | killed := false |
| 496 | mk := &mockKiller{fn: func(_, _ string) bool { killed = true; return true }} |
| 497 | |
| 498 | mustUpsertControl(t, s, "/p", TaskSnapshot{ |
| 499 | SchemaVersion: 1, TaskID: "t1", SessionID: "s1", |
| 500 | State: TaskStateSucceeded, Version: 1, |
| 501 | CreatedAt: time.Now(), UpdatedAt: time.Now(), |
| 502 | }) |
| 503 | |
| 504 | cs.StopTaskWithKiller(context.Background(), "/p", "t1", 1, "", "", mk) |
| 505 | if killed { |
| 506 | t.Error("Kill should not be called for terminal tasks") |
| 507 | } |
| 508 | } |
| 509 | |
| 510 | func TestControlService_ConcurrentKillersRemainCallScoped(t *testing.T) { |
| 511 | s := NewInMemoryStore() |
| 512 | cs := NewControlService(s) |
| 513 | now := time.Now() |
| 514 | for _, snap := range []TaskSnapshot{ |
| 515 | {SchemaVersion: 1, TaskID: "task-a", SessionID: "session-a", State: TaskStateRunning, RuntimeState: RuntimeStateAlive, Version: 1, CreatedAt: now, UpdatedAt: now}, |
| 516 | {SchemaVersion: 1, TaskID: "task-b", SessionID: "session-b", State: TaskStateRunning, RuntimeState: RuntimeStateAlive, Version: 1, CreatedAt: now, UpdatedAt: now}, |
| 517 | } { |
| 518 | mustUpsertControl(t, s, "/p", snap) |
| 519 | } |
| 520 | |
| 521 | started := make(chan struct{}) |
| 522 | killed := make(chan string, 2) |
| 523 | var wg sync.WaitGroup |
| 524 | for _, tc := range []struct { |
| 525 | taskID, sessionID string |
| 526 | }{ |
| 527 | {taskID: "task-a", sessionID: "session-a"}, |
| 528 | {taskID: "task-b", sessionID: "session-b"}, |
| 529 | } { |
| 530 | tc := tc |
| 531 | wg.Add(1) |
| 532 | go func() { |
| 533 | defer wg.Done() |
| 534 | <-started |
| 535 | killer := &mockKiller{fn: func(sessionID, taskID string) bool { |
| 536 | killed <- sessionID + "/" + taskID |
| 537 | return sessionID == tc.sessionID && taskID == tc.taskID |
| 538 | }} |
| 539 | res, err := cs.StopTaskWithKiller(context.Background(), "/p", tc.taskID, 1, "", "", killer) |
| 540 | if err != nil || !res.Accepted { |
| 541 | t.Errorf("StopTaskWithKiller(%s): result=%+v err=%v", tc.taskID, res, err) |
| 542 | } |
| 543 | }() |
| 544 | } |
| 545 | close(started) |
| 546 | wg.Wait() |
| 547 | close(killed) |
| 548 | |
| 549 | got := map[string]bool{} |
| 550 | for target := range killed { |
| 551 | got[target] = true |
| 552 | } |
| 553 | for _, want := range []string{"session-a/task-a", "session-b/task-b"} { |
| 554 | if !got[want] { |
| 555 | t.Fatalf("missing call-scoped kill %q; got %v", want, got) |
| 556 | } |
| 557 | } |
| 558 | } |
| 559 | |
| 560 | func TestControlService_ConcurrentAccess(t *testing.T) { |
| 561 | s := NewInMemoryStore() |
| 562 | cs := NewControlService(s) |
| 563 | killer := &mockKiller{fn: func(string, string) bool { return true }} |
| 564 | |
| 565 | mustUpsertControl(t, s, "/p", TaskSnapshot{ |
| 566 | SchemaVersion: 1, TaskID: "t1", SessionID: "s1", |
| 567 | State: TaskStateRunning, Version: 1, |
| 568 | CreatedAt: time.Now(), UpdatedAt: time.Now(), |
| 569 | }) |
| 570 | |
| 571 | var wg sync.WaitGroup |
| 572 | success := 0 |
| 573 | var mu sync.Mutex |
| 574 | |
| 575 | for i := 0; i < 10; i++ { |
| 576 | wg.Add(1) |
| 577 | go func() { |
| 578 | defer wg.Done() |
| 579 | res, _ := cs.StopTaskWithKiller(context.Background(), "/p", "t1", 1, "", "", killer) |
| 580 | if res.Accepted { |
| 581 | mu.Lock() |
| 582 | success++ |
| 583 | mu.Unlock() |
| 584 | } |
| 585 | }() |
| 586 | } |
| 587 | wg.Wait() |
| 588 | // Exactly one caller should succeed due to mutex + version CAS |
| 589 | if success != 1 { |
| 590 | t.Errorf("expected exactly 1 success, got %d", success) |
| 591 | } |
| 592 | } |
| 593 | |
| 594 | func TestControlService_CancelTask(t *testing.T) { |
| 595 | s := NewInMemoryStore() |
| 596 | cs := NewControlService(s) |
| 597 | killer := &mockKiller{fn: func(string, string) bool { return true }} |
| 598 | |
| 599 | mustUpsertControl(t, s, "/p", TaskSnapshot{ |
| 600 | SchemaVersion: 1, TaskID: "t1", SessionID: "s1", |
| 601 | State: TaskStateWaiting, Version: 1, |
| 602 | CreatedAt: time.Now(), UpdatedAt: time.Now(), |
| 603 | }) |
| 604 | |
| 605 | res, _ := cs.CancelTaskWithKiller(context.Background(), "/p", "t1", 1, "timeout", "", killer) |
| 606 | if !res.Accepted || res.State != TaskStateCancelled { |
| 607 | t.Errorf("expected cancelled, got %+v", res) |
| 608 | } |
| 609 | } |
| 610 | |
| 611 | func TestControlService_OpenSession(t *testing.T) { |
| 612 | s := NewInMemoryStore() |
| 613 | cs := NewControlService(s) |
| 614 | |
| 615 | mustUpsertControl(t, s, "/p", TaskSnapshot{ |
| 616 | SchemaVersion: 1, TaskID: "t1", SessionID: "sess-abc", |
| 617 | State: TaskStateRunning, Version: 1, |
| 618 | CreatedAt: time.Now(), UpdatedAt: time.Now(), |
| 619 | }) |
| 620 | |
| 621 | res, _ := cs.OpenTaskSession(context.Background(), "/p", "t1") |
| 622 | if res.SessionID != "sess-abc" || !res.Accepted { |
| 623 | t.Errorf("expected sess-abc, got %+v", res) |
| 624 | } |
| 625 | } |
| 626 | |
| 627 | // mockKiller implements JobKiller for tests. |
| 628 | type mockKiller struct { |
| 629 | fn func(string, string) bool |
| 630 | } |
| 631 | |
| 632 | func (m *mockKiller) Kill(sessionID, id string) bool { |
| 633 | if m.fn != nil { |
| 634 | return m.fn(sessionID, id) |
| 635 | } |
| 636 | return false |
| 637 | } |
| 638 | |
| 639 | func mustUpsertControl(t *testing.T, s *InMemoryStore, proj string, snap TaskSnapshot) { |
| 640 | t.Helper() |
| 641 | if err := s.UpsertTask(proj, snap); err != nil { |
| 642 | t.Fatal(err) |
| 643 | } |
| 644 | } |
| 645 |