| 1 | package main |
| 2 | |
| 3 | import ( |
| 4 | "context" |
| 5 | "errors" |
| 6 | "os" |
| 7 | "sync" |
| 8 | "testing" |
| 9 | "time" |
| 10 | |
| 11 | "github.com/jackc/pgx/v5/pgxpool" |
| 12 | ) |
| 13 | |
| 14 | func TestBackoff(t * |
| 15 | maxRand := func(n |
| 16 | |
| 17 | tests := []struct |
| 18 | attempt int |
| 19 | want time. |
| 20 | }{ |
| 21 | {0, 10 * time |
| 22 | {1, 20 * time |
| 23 | {2, 40 * time |
| 24 | {3, 80 * time |
| 25 | {8, 2560 * time |
| 26 | {9, time.Hour |
| 27 | {30, time.Hour |
| 28 | {100, time.Hour |
| 29 | } |
| 30 | for _, tt |
| 31 | got := backoff( |
| 32 | if got != tt |
| 33 | t.Errorf("backoff( |
| 34 | } |
| 35 | } |
| 36 | } |
| 37 | |
| 38 | // Tests only touch rows whose idempotency_key starts with this prefix, so they are safe on a shared dev database. |
| 39 | const testKeyPrefix = "test: |
| 40 | |
| 41 | // testDB refuses to run while the queue holds real pending jobs, because claim would lease them. |
| 42 | func testDB(t * |
| 43 | t.Helper() |
| 44 | url := os. |
| 45 | if url == "" |
| 46 | t.Skip("TEST_DATABASE_URL is not set |
| 47 | } |
| 48 | db, err := |
| 49 | if err != nil |
| 50 | t.Fatal(err |
| 51 | } |
| 52 | |
| 53 | var real int |
| 54 | err = db. |
| 55 | `SELECT count(*) FROM delivery_job WHERE status = 'pending' AND idempotency_key NOT LIKE $1 || '%'`, |
| 56 | testKeyPrefix).Scan(& |
| 57 | if err != nil |
| 58 | t.Fatal(err |
| 59 | } |
| 60 | if real > 0 |
| 61 | t.Fatalf("%d |
| 62 | } |
| 63 | |
| 64 | t.Cleanup(func() |
| 65 | // t.Context() is already canceled when cleanup runs. |
| 66 | _, err := |
| 67 | `DELETE FROM delivery_job WHERE idempotency_key LIKE $1 || '%'`, testKeyPrefix) |
| 68 | if err != nil |
| 69 | t.Errorf("cleanup: |
| 70 | } |
| 71 | db.Close() |
| 72 | }) |
| 73 | return db |
| 74 | } |
| 75 | |
| 76 | func insertJobs(t * |
| 77 | t.Helper() |
| 78 | ids := make( |
| 79 | for range n { |
| 80 | var id string |
| 81 | err := db. |
| 82 | `INSERT INTO delivery_job (kind, idempotency_key, payload) |
| 83 | VALUES ('email', $1 || gen_random_uuid()::text, '{}') |
| 84 | RETURNING id`, testKeyPrefix).Scan |
| 85 | if err != nil |
| 86 | t.Fatal(err |
| 87 | } |
| 88 | ids[id] = |
| 89 | } |
| 90 | return ids |
| 91 | } |
| 92 | |
| 93 | func TestClaimNeverHandsOutAJobTwice(t * |
| 94 | db := testDB( |
| 95 | inserted := insertJobs( |
| 96 | |
| 97 | var ( |
| 98 | mu sync. |
| 99 | claimed = map[string |
| 100 | wg sync. |
| 101 | ) |
| 102 | for range 4 { |
| 103 | wg.Go(func() |
| 104 | for { |
| 105 | jobs, err := |
| 106 | if err != nil |
| 107 | t.Error(err |
| 108 | return |
| 109 | } |
| 110 | if len(jobs) |
| 111 | return |
| 112 | } |
| 113 | mu.Lock() |
| 114 | for _, j |
| 115 | claimed[j.ID |
| 116 | } |
| 117 | mu.Unlock() |
| 118 | } |
| 119 | }) |
| 120 | } |
| 121 | wg.Wait() |
| 122 | |
| 123 | if len(claimed) |
| 124 | t.Errorf("claimed |
| 125 | } |
| 126 | for id, n |
| 127 | if n > 1 |
| 128 | t.Errorf("job |
| 129 | } |
| 130 | if !inserted[id |
| 131 | t.Errorf("claimed job |
| 132 | } |
| 133 | } |
| 134 | } |
| 135 | |
| 136 | func TestClaimReclaimsAfterLeaseExpires(t * |
| 137 | db := testDB( |
| 138 | insertJobs(t, db |
| 139 | |
| 140 | first, err := |
| 141 | if err != nil |
| 142 | t.Fatal(err |
| 143 | } |
| 144 | if len(first) |
| 145 | t.Fatalf("first claim got |
| 146 | } |
| 147 | |
| 148 | again, err := |
| 149 | if err != nil |
| 150 | t.Fatal(err |
| 151 | } |
| 152 | if len(again) |
| 153 | t.Fatalf("claimed |
| 154 | } |
| 155 | |
| 156 | // the worker died: expire the lease instead of sleeping 60s |
| 157 | _, err = |
| 158 | `UPDATE delivery_job SET locked_until = now() - interval '1 second' WHERE id = $1`, first[ |
| 159 | if err != nil |
| 160 | t.Fatal(err |
| 161 | } |
| 162 | |
| 163 | second, err := |
| 164 | if err != nil |
| 165 | t.Fatal(err |
| 166 | } |
| 167 | if len(second) |
| 168 | t.Fatalf("reclaim got |
| 169 | } |
| 170 | if second[0]. |
| 171 | t.Error("reclaim kept the old claim_token; a stale worker could overwrite the result |
| 172 | } |
| 173 | if second[0]. |
| 174 | t.Errorf("attempts = |
| 175 | } |
| 176 | |
| 177 | // The worker that lost its lease must not overwrite the new owner's result. |
| 178 | if err := complete |
| 179 | t.Errorf("complete with stale token = |
| 180 | } |
| 181 | if err := complete |
| 182 | t.Fatalf("complete with current token: |
| 183 | } |
| 184 | if got := jobStatus |
| 185 | t.Errorf("status = |
| 186 | } |
| 187 | } |
| 188 | |
| 189 | func TestFailSchedulesRetry(t * |
| 190 | db := testDB( |
| 191 | insertJobs(t, db |
| 192 | jobs, err := |
| 193 | if err != nil |
| 194 | t.Fatalf("claim = |
| 195 | } |
| 196 | |
| 197 | if err := fail |
| 198 | t.Fatal(err |
| 199 | } |
| 200 | |
| 201 | var ( |
| 202 | status, lastError string |
| 203 | delay time. |
| 204 | ) |
| 205 | err = db. |
| 206 | `SELECT status, last_error, next_attempt_at - now() FROM delivery_job WHERE id = $1`, |
| 207 | jobs[0].ID |
| 208 | if err != nil |
| 209 | t.Fatal(err |
| 210 | } |
| 211 | if status != " |
| 212 | t.Errorf("status = |
| 213 | } |
| 214 | if lastError != " |
| 215 | t.Errorf("last_error = |
| 216 | } |
| 217 | if delay < 29 |
| 218 | t.Errorf("next attempt in |
| 219 | } |
| 220 | |
| 221 | // Not due yet, so it must not be claimable. |
| 222 | again, err := |
| 223 | if err != nil |
| 224 | t.Fatal(err |
| 225 | } |
| 226 | if len(again) |
| 227 | t.Errorf("claimed |
| 228 | } |
| 229 | } |
| 230 | |
| 231 | func TestFailPastMaxAgeIsDead(t * |
| 232 | db := testDB( |
| 233 | _, err := |
| 234 | `INSERT INTO delivery_job (kind, idempotency_key, payload, created_at) |
| 235 | VALUES ('email', $1 || gen_random_uuid()::text, '{}', now() - interval '4 days')`, |
| 236 | testKeyPrefix) |
| 237 | if err != nil |
| 238 | t.Fatal(err |
| 239 | } |
| 240 | jobs, err := |
| 241 | if err != nil |
| 242 | t.Fatalf("claim = |
| 243 | } |
| 244 | |
| 245 | if err := fail |
| 246 | t.Fatal(err |
| 247 | } |
| 248 | if got := jobStatus |
| 249 | t.Errorf("status = |
| 250 | } |
| 251 | } |
| 252 | |
| 253 | func jobStatus(t * |
| 254 | t.Helper() |
| 255 | var s string |
| 256 | if err := db |
| 257 | t.Fatal(err |
| 258 | } |
| 259 | return s |
| 260 | } |
| 261 | |
| 262 | func TestListenWakesOnNotify(t * |
| 263 | db := testDB( |
| 264 | wake := make( |
| 265 | go listen(t. |
| 266 | |
| 267 | waitWake := func(what |
| 268 | t.Helper() |
| 269 | select { |
| 270 | case <-wake: |
| 271 | case <-time.After |
| 272 | t.Fatalf("no wake-up for |
| 273 | } |
| 274 | } |
| 275 | |
| 276 | waitWake("initial LISTEN") |
| 277 | if _, err |
| 278 | t.Fatal(err |
| 279 | } |
| 280 | waitWake("NOTIFY delivery") |
| 281 | } |
| 282 | |
| 283 | // The returned stop cancels run and waits for it to drain. |
| 284 | func startRun(t * |
| 285 | t.Helper() |
| 286 | ctx, cancel := |
| 287 | done := make( |
| 288 | go func() { |
| 289 | run(ctx, db |
| 290 | close(done) |
| 291 | }() |
| 292 | return func() { |
| 293 | cancel() |
| 294 | select { |
| 295 | case <-done: |
| 296 | case <-time.After |
| 297 | t.Fatal("run did not return after cancel |
| 298 | } |
| 299 | } |
| 300 | } |
| 301 | |
| 302 | func TestRunDeliversEveryJob(t * |
| 303 | db := testDB( |
| 304 | inserted := insertJobs( |
| 305 | |
| 306 | var ( |
| 307 | mu sync. |
| 308 | sent = map[string |
| 309 | all = make( |
| 310 | ) |
| 311 | send := func(ctx |
| 312 | time.Sleep(10 |
| 313 | mu.Lock() |
| 314 | defer mu.Unlock() |
| 315 | sent[j.ID |
| 316 | if len(sent) |
| 317 | close(all) |
| 318 | } |
| 319 | return nil |
| 320 | } |
| 321 | |
| 322 | stop := startRun( |
| 323 | select { |
| 324 | case <-all: |
| 325 | case <-time.After |
| 326 | t.Fatal("not every job was sent within 5s |
| 327 | } |
| 328 | stop() |
| 329 | |
| 330 | for id := range |
| 331 | if sent[id] |
| 332 | t.Errorf("job |
| 333 | } |
| 334 | if got := jobStatus |
| 335 | t.Errorf("job |
| 336 | } |
| 337 | } |
| 338 | } |
| 339 | |
| 340 | func TestRunDrainsInFlightSendOnShutdown(t * |
| 341 | db := testDB( |
| 342 | inserted := insertJobs( |
| 343 | |
| 344 | started := make( |
| 345 | release := make( |
| 346 | send := func(ctx |
| 347 | close(started) |
| 348 | <-release |
| 349 | return ctx.Err() |
| 350 | } |
| 351 | |
| 352 | stop := startRun( |
| 353 | <-started |
| 354 | go func() { |
| 355 | time.Sleep(100 |
| 356 | close(release) |
| 357 | }() |
| 358 | stop() |
| 359 | |
| 360 | for id := range |
| 361 | if got := jobStatus |
| 362 | t.Errorf("status = |
| 363 | } |
| 364 | } |
| 365 | } |
| 366 | |
| 367 | // Runs in a transaction that is rolled back, so it prunes nothing for real. |
| 368 | func TestPruneDeletesOnlyOldFinishedJobs(t * |
| 369 | db := testDB( |
| 370 | tx, err := |
| 371 | if err != nil |
| 372 | t.Fatal(err |
| 373 | } |
| 374 | defer tx.Rollback( |
| 375 | |
| 376 | insert := func(status |
| 377 | var id string |
| 378 | err := tx. |
| 379 | `INSERT INTO delivery_job (kind, idempotency_key, payload, status, updated_at) |
| 380 | VALUES ('email', $1 || gen_random_uuid()::text, '{}', $2, now() - $3::interval) |
| 381 | RETURNING id`, testKeyPrefix, status |
| 382 | if err != nil |
| 383 | t.Fatal(err |
| 384 | } |
| 385 | return id |
| 386 | } |
| 387 | oldDone := insert( |
| 388 | oldDead := insert( |
| 389 | recent := insert( |
| 390 | oldPending := insert( |
| 391 | |
| 392 | if _, err |
| 393 | t.Fatal(err |
| 394 | } |
| 395 | |
| 396 | exists := func(id |
| 397 | var n int |
| 398 | if err := tx |
| 399 | t.Fatal(err |
| 400 | } |
| 401 | return n == 1 |
| 402 | } |
| 403 | for id, want |
| 404 | if got := exists |
| 405 | t.Errorf("job |
| 406 | } |
| 407 | } |
| 408 | } |
| 409 | |
| 410 | func TestRunRoutesSendOutcomes(t * |
| 411 | db := testDB( |
| 412 | tests := []struct |
| 413 | name string |
| 414 | err error |
| 415 | wantStatus string |
| 416 | wantDelay time. |
| 417 | }{ |
| 418 | {"permanent failure gives up at once", permanentError |
| 419 | {"Retry-After sets the next try", retryAfterError |
| 420 | {"Retry-After is capped", retryAfterError |
| 421 | } |
| 422 | for _, tt |
| 423 | t.Run(tt |
| 424 | ids := insertJobs( |
| 425 | done := make( |
| 426 | stop := startRun( |
| 427 | defer close(done) |
| 428 | return tt.err |
| 429 | }) |
| 430 | <-done |
| 431 | stop() |
| 432 | |
| 433 | for id := range |
| 434 | var ( |
| 435 | status string |
| 436 | delay time. |
| 437 | ) |
| 438 | err := db. |
| 439 | if err != nil |
| 440 | t.Fatal(err |
| 441 | } |
| 442 | if status != tt |
| 443 | t.Errorf("status = |
| 444 | } |
| 445 | if tt.wantDelay |
| 446 | t.Errorf("next try in |
| 447 | } |
| 448 | } |
| 449 | }) |
| 450 | } |
| 451 | } |
| 452 | |
| 453 | func TestQueueCallsReturnDatabaseErrors(t * |
| 454 | url := os. |
| 455 | if url == "" |
| 456 | t.Skip("TEST_DATABASE_URL is not set |
| 457 | } |
| 458 | closed, err := |
| 459 | if err != nil |
| 460 | t.Fatal(err |
| 461 | } |
| 462 | closed.Close() |
| 463 | j := Job |
| 464 | |
| 465 | if _, err |
| 466 | t.Error("claim on a closed pool succeeded |
| 467 | } |
| 468 | if err := complete |
| 469 | t.Errorf("complete = |
| 470 | } |
| 471 | if err := fail |
| 472 | t.Errorf("fail = |
| 473 | } |
| 474 | if err := giveUp |
| 475 | t.Errorf("giveUp = |
| 476 | } |
| 477 | if _, err |
| 478 | t.Error("prune on a closed pool succeeded |
| 479 | } |
| 480 | |
| 481 | // a canceled pruneEvery returns after its first pass instead of waiting for the tick |
| 482 | ctx, cancel := |
| 483 | cancel() |
| 484 | pruneEvery(ctx, closed |
| 485 | } |