| 1 | // SPDX-License-Identifier: GPL-3.0-or-later |
| 2 | |
| 3 | package functions |
| 4 | |
| 5 | import ( |
| 6 | "bytes" |
| 7 | "context" |
| 8 | "fmt" |
| 9 | "strings" |
| 10 | "sync" |
| 11 | "sync/atomic" |
| 12 | "testing" |
| 13 | "time" |
| 14 | |
| 15 | "github.com/netdata/netdata/go/plugins/pkg/netdataapi" |
| 16 | "github.com/stretchr/testify/assert" |
| 17 | "github.com/stretchr/testify/require" |
| 18 | ) |
| 19 | |
| 20 | const ( |
| 21 | testPermissions = "0xFFFF" |
| 22 | testSource = "method=api,role=test" |
| 23 | ) |
| 24 | |
| 25 | type chanInput struct { |
| 26 | ch chan string |
| 27 | } |
| 28 | |
| 29 | func (m *chanInput) lines() <-chan string { |
| 30 | return m.ch |
| 31 | } |
| 32 | |
| 33 | type safeBuffer struct { |
| 34 | mu sync.Mutex |
| 35 | b bytes.Buffer |
| 36 | } |
| 37 | |
| 38 | func (s *safeBuffer) Write(p []byte) (int, error) { |
| 39 | s.mu.Lock() |
| 40 | defer s.mu.Unlock() |
| 41 | return s.b.Write(p) |
| 42 | } |
| 43 | |
| 44 | func (s *safeBuffer) String() string { |
| 45 | s.mu.Lock() |
| 46 | defer s.mu.Unlock() |
| 47 | return s.b.String() |
| 48 | } |
| 49 | |
| 50 | func newFlowManager() (*Manager, *safeBuffer) { |
| 51 | mgr := NewManager() |
| 52 | buf := &safeBuffer{} |
| 53 | mgr.api = netdataapi.New(buf) |
| 54 | return mgr, buf |
| 55 | } |
| 56 | |
| 57 | func functionLine(uid, name string) string { |
| 58 | return fmt.Sprintf(`FUNCTION %s 10 "%s" %s "%s"`, uid, name, testPermissions, testSource) |
| 59 | } |
| 60 | |
| 61 | func payloadStartCmd(uid, name string) string { |
| 62 | return fmt.Sprintf(`FUNCTION_PAYLOAD %s 10 "%s" %s "%s" application/json`, uid, name, testPermissions, testSource) |
| 63 | } |
| 64 | |
| 65 | func waitForSubstring(t *testing.T, f func() string, substr string, timeout time.Duration) { |
| 66 | t.Helper() |
| 67 | |
| 68 | deadline := time.Now().Add(timeout) |
| 69 | for time.Now().Before(deadline) { |
| 70 | if strings.Contains(f(), substr) { |
| 71 | return |
| 72 | } |
| 73 | time.Sleep(10 * time.Millisecond) |
| 74 | } |
| 75 | t.Fatalf("timeout waiting for substring %q in output: %s", substr, f()) |
| 76 | } |
| 77 | |
| 78 | func startFlowManager(t *testing.T, mgr *Manager) (context.CancelFunc, chan struct{}) { |
| 79 | t.Helper() |
| 80 | |
| 81 | ctx, cancel := context.WithTimeout(context.Background(), 3*time.Second) |
| 82 | done := make(chan struct{}) |
| 83 | go func() { |
| 84 | defer close(done) |
| 85 | mgr.Run(ctx, nil) |
| 86 | }() |
| 87 | return cancel, done |
| 88 | } |
| 89 | |
| 90 | func waitForDone(t *testing.T, done <-chan struct{}) { |
| 91 | t.Helper() |
| 92 | |
| 93 | select { |
| 94 | case <-done: |
| 95 | case <-time.After(3 * time.Second): |
| 96 | t.Fatal("timeout waiting for manager run to complete") |
| 97 | } |
| 98 | } |
| 99 | |
| 100 | func waitForCondition(t *testing.T, timeout time.Duration, fn func() bool, desc string) { |
| 101 | t.Helper() |
| 102 | |
| 103 | deadline := time.Now().Add(timeout) |
| 104 | for time.Now().Before(deadline) { |
| 105 | if fn() { |
| 106 | return |
| 107 | } |
| 108 | time.Sleep(10 * time.Millisecond) |
| 109 | } |
| 110 | t.Fatalf("timeout waiting for condition: %s", desc) |
| 111 | } |
| 112 | |
| 113 | func TestManager_FlowScenarios(t *testing.T) { |
| 114 | tests := map[string]struct { |
| 115 | run func(t *testing.T, mgr *Manager, in *chanInput, out *safeBuffer) |
| 116 | }{ |
| 117 | "queued cancel is ignored; function still executes": { |
| 118 | run: func(t *testing.T, mgr *Manager, in *chanInput, out *safeBuffer) { |
| 119 | // New semantics: a CANCEL for a queued function is a no-op at |
| 120 | // the framework level. The function stays in the scheduler, |
| 121 | // runs to completion when its turn comes, and any terminal |
| 122 | // response goes through normally. No 499 is emitted by the |
| 123 | // framework (netdata already considers the transaction done |
| 124 | // via its own per-transaction timer, so an extra 499 would |
| 125 | // just be noise). |
| 126 | var ( |
| 127 | mu sync.Mutex |
| 128 | executed []string |
| 129 | ) |
| 130 | started := make(chan struct{}, 1) |
| 131 | release := make(chan struct{}) |
| 132 | tx2Done := make(chan struct{}) |
| 133 | |
| 134 | mgr.Register("fn", func(fn Function) { |
| 135 | mu.Lock() |
| 136 | executed = append(executed, fn.UID) |
| 137 | mu.Unlock() |
| 138 | if fn.UID == "tx1" { |
| 139 | started <- struct{}{} |
| 140 | <-release |
| 141 | // Emit terminal response so the per-key lane advances |
| 142 | // to tx2; otherwise the lane stays "owned" by tx1 and |
| 143 | // tx2 never gets picked up. |
| 144 | mgr.respUID(fn.UID, 200, "ok") |
| 145 | } |
| 146 | if fn.UID == "tx2" { |
| 147 | mgr.respUID(fn.UID, 200, "ok") |
| 148 | close(tx2Done) |
| 149 | } |
| 150 | }) |
| 151 | |
| 152 | cancel, done := startFlowManager(t, mgr) |
| 153 | defer cancel() |
| 154 | |
| 155 | in.ch <- functionLine("tx1", "fn") |
| 156 | <-started |
| 157 | in.ch <- functionLine("tx2", "fn") |
| 158 | in.ch <- "FUNCTION_CANCEL tx2" |
| 159 | |
| 160 | // tx2 must NOT receive a 499 from the framework. We can't |
| 161 | // assert "no 499 ever" without waiting forever, but we can at |
| 162 | // least verify it isn't there immediately after CANCEL. |
| 163 | time.Sleep(50 * time.Millisecond) |
| 164 | assert.NotContains(t, out.String(), "FUNCTION_RESULT_BEGIN tx2 499") |
| 165 | |
| 166 | close(release) |
| 167 | select { |
| 168 | case <-tx2Done: |
| 169 | case <-time.After(time.Second): |
| 170 | t.Fatal("tx2 did not execute after release") |
| 171 | } |
| 172 | |
| 173 | close(in.ch) |
| 174 | waitForDone(t, done) |
| 175 | |
| 176 | // Both tx1 and tx2 should have executed (in order). |
| 177 | mu.Lock() |
| 178 | defer mu.Unlock() |
| 179 | assert.Equal(t, []string{"tx1", "tx2"}, executed) |
| 180 | // tx2's normal 200 response went through (no tombstone). |
| 181 | assert.Contains(t, out.String(), "FUNCTION_RESULT_BEGIN tx2 200") |
| 182 | }, |
| 183 | }, |
| 184 | "running cancel fallback tombstones silently (no emit)": { |
| 185 | run: func(t *testing.T, mgr *Manager, in *chanInput, out *safeBuffer) { |
| 186 | // New semantics: the cancel fallback timer no longer emits a |
| 187 | // 499 to netdata. By the time CANCEL arrives, netdata has |
| 188 | // already 504'd and removed the inflight transaction, so any |
| 189 | // response would just produce a "transaction not found" log. |
| 190 | // We only do the bookkeeping (tombstone + lane advance). |
| 191 | mgr.cancelFallbackDelay = 50 * time.Millisecond |
| 192 | started := make(chan struct{}, 1) |
| 193 | release := make(chan struct{}) |
| 194 | |
| 195 | mgr.Register("fn", func(fn Function) { |
| 196 | if fn.UID == "tx1" { |
| 197 | started <- struct{}{} |
| 198 | <-release |
| 199 | } |
| 200 | }) |
| 201 | |
| 202 | cancel, done := startFlowManager(t, mgr) |
| 203 | defer cancel() |
| 204 | |
| 205 | in.ch <- functionLine("tx1", "fn") |
| 206 | <-started |
| 207 | in.ch <- "FUNCTION_CANCEL tx1" |
| 208 | // Wait long enough for the fallback timer to fire. |
| 209 | time.Sleep(150 * time.Millisecond) |
| 210 | |
| 211 | close(release) |
| 212 | close(in.ch) |
| 213 | waitForDone(t, done) |
| 214 | |
| 215 | // No 499 should be emitted, ever. |
| 216 | assert.Equal(t, 0, strings.Count(out.String(), "FUNCTION_RESULT_BEGIN tx1 499")) |
| 217 | }, |
| 218 | }, |
| 219 | "repeated cancel for same uid is idempotent and silent": { |
| 220 | run: func(t *testing.T, mgr *Manager, in *chanInput, out *safeBuffer) { |
| 221 | mgr.cancelFallbackDelay = 50 * time.Millisecond |
| 222 | started := make(chan struct{}, 1) |
| 223 | release := make(chan struct{}) |
| 224 | |
| 225 | mgr.Register("fn", func(fn Function) { |
| 226 | if fn.UID == "tx1" { |
| 227 | started <- struct{}{} |
| 228 | <-release |
| 229 | } |
| 230 | }) |
| 231 | |
| 232 | cancel, done := startFlowManager(t, mgr) |
| 233 | defer cancel() |
| 234 | |
| 235 | in.ch <- functionLine("tx1", "fn") |
| 236 | <-started |
| 237 | in.ch <- "FUNCTION_CANCEL tx1" |
| 238 | in.ch <- "FUNCTION_CANCEL tx1" |
| 239 | time.Sleep(150 * time.Millisecond) |
| 240 | |
| 241 | close(release) |
| 242 | close(in.ch) |
| 243 | waitForDone(t, done) |
| 244 | |
| 245 | assert.Equal(t, 0, strings.Count(out.String(), "FUNCTION_RESULT_BEGIN tx1 499")) |
| 246 | }, |
| 247 | }, |
| 248 | "running cancel drops late terminal response": { |
| 249 | run: func(t *testing.T, mgr *Manager, in *chanInput, out *safeBuffer) { |
| 250 | mgr.cancelFallbackDelay = 50 * time.Millisecond |
| 251 | started := make(chan struct{}, 1) |
| 252 | release := make(chan struct{}) |
| 253 | |
| 254 | mgr.Register("fn", func(fn Function) { |
| 255 | if fn.UID == "tx1" { |
| 256 | started <- struct{}{} |
| 257 | <-release |
| 258 | mgr.respUID(fn.UID, 200, "late response") |
| 259 | } |
| 260 | }) |
| 261 | |
| 262 | cancel, done := startFlowManager(t, mgr) |
| 263 | defer cancel() |
| 264 | |
| 265 | in.ch <- functionLine("tx1", "fn") |
| 266 | <-started |
| 267 | in.ch <- "FUNCTION_CANCEL tx1" |
| 268 | // Let the fallback timer fire (tombstones the UID). |
| 269 | time.Sleep(150 * time.Millisecond) |
| 270 | |
| 271 | close(release) |
| 272 | close(in.ch) |
| 273 | waitForDone(t, done) |
| 274 | |
| 275 | got := out.String() |
| 276 | // No 499 emitted by the fallback path. |
| 277 | assert.Equal(t, 0, strings.Count(got, "FUNCTION_RESULT_BEGIN tx1 499")) |
| 278 | // Late 200 from the handler is dropped by the tombstone. |
| 279 | assert.Equal(t, 0, strings.Count(got, "FUNCTION_RESULT_BEGIN tx1 200")) |
| 280 | }, |
| 281 | }, |
| 282 | "payload pre-admission cancel emits 499 and skips handler": { |
| 283 | run: func(t *testing.T, mgr *Manager, in *chanInput, out *safeBuffer) { |
| 284 | var calls atomic.Int32 |
| 285 | mgr.Register("fn", func(Function) { calls.Add(1) }) |
| 286 | |
| 287 | cancel, done := startFlowManager(t, mgr) |
| 288 | defer cancel() |
| 289 | |
| 290 | in.ch <- payloadStartCmd("tx1", "fn") |
| 291 | in.ch <- "payload line" |
| 292 | in.ch <- "FUNCTION_CANCEL tx1" |
| 293 | close(in.ch) |
| 294 | waitForDone(t, done) |
| 295 | |
| 296 | waitForSubstring(t, out.String, "FUNCTION_RESULT_BEGIN tx1 499", time.Second) |
| 297 | assert.EqualValues(t, 0, calls.Load()) |
| 298 | }, |
| 299 | }, |
| 300 | "duplicate active uid is ignored without corrupting lane progression": { |
| 301 | run: func(t *testing.T, mgr *Manager, in *chanInput, out *safeBuffer) { |
| 302 | mgr.workerCount = 2 |
| 303 | started := make(chan struct{}, 1) |
| 304 | tx2Started := make(chan struct{}, 1) |
| 305 | release := make(chan struct{}) |
| 306 | var calls atomic.Int32 |
| 307 | |
| 308 | mgr.Register("fn", func(fn Function) { |
| 309 | calls.Add(1) |
| 310 | |
| 311 | if fn.UID == "tx1" { |
| 312 | started <- struct{}{} |
| 313 | <-release |
| 314 | mgr.respUID(fn.UID, 200, "ok") |
| 315 | return |
| 316 | } |
| 317 | |
| 318 | if fn.UID == "tx2" { |
| 319 | select { |
| 320 | case tx2Started <- struct{}{}: |
| 321 | default: |
| 322 | } |
| 323 | } |
| 324 | |
| 325 | mgr.respUID(fn.UID, 200, "ok") |
| 326 | }) |
| 327 | |
| 328 | cancel, done := startFlowManager(t, mgr) |
| 329 | defer cancel() |
| 330 | |
| 331 | in.ch <- functionLine("tx1", "fn") |
| 332 | <-started |
| 333 | // Same-key request should stay queued until tx1 completes. |
| 334 | in.ch <- functionLine("tx2", "fn") |
| 335 | // Duplicate active UID must not advance lanes or finalize tx1. |
| 336 | in.ch <- functionLine("tx1", "fn") |
| 337 | |
| 338 | select { |
| 339 | case <-tx2Started: |
| 340 | t.Fatal("tx2 started before tx1 was released") |
| 341 | default: |
| 342 | } |
| 343 | |
| 344 | close(release) |
| 345 | select { |
| 346 | case <-tx2Started: |
| 347 | case <-time.After(time.Second): |
| 348 | t.Fatal("tx2 did not start after tx1 was released") |
| 349 | } |
| 350 | close(in.ch) |
| 351 | waitForDone(t, done) |
| 352 | |
| 353 | got := out.String() |
| 354 | assert.Equal(t, 1, strings.Count(got, "FUNCTION_RESULT_BEGIN tx1 200")) |
| 355 | assert.Equal(t, 1, strings.Count(got, "FUNCTION_RESULT_BEGIN tx2 200")) |
| 356 | assert.Equal(t, 0, strings.Count(got, "FUNCTION_RESULT_BEGIN tx1 409")) |
| 357 | assert.EqualValues(t, 2, calls.Load()) |
| 358 | }, |
| 359 | }, |
| 360 | "duplicate tombstoned uid is ignored without extra terminal output": { |
| 361 | run: func(t *testing.T, mgr *Manager, in *chanInput, out *safeBuffer) { |
| 362 | var calls atomic.Int32 |
| 363 | |
| 364 | mgr.Register("fn", func(fn Function) { |
| 365 | calls.Add(1) |
| 366 | mgr.respUID(fn.UID, 200, "ok") |
| 367 | }) |
| 368 | |
| 369 | cancel, done := startFlowManager(t, mgr) |
| 370 | defer cancel() |
| 371 | |
| 372 | in.ch <- functionLine("tx1", "fn") |
| 373 | waitForSubstring(t, out.String, "FUNCTION_RESULT_BEGIN tx1 200", time.Second) |
| 374 | |
| 375 | // Re-send same UID while tombstone is still active. |
| 376 | in.ch <- functionLine("tx1", "fn") |
| 377 | close(in.ch) |
| 378 | waitForDone(t, done) |
| 379 | |
| 380 | got := out.String() |
| 381 | assert.Equal(t, 1, strings.Count(got, "FUNCTION_RESULT_BEGIN tx1 200")) |
| 382 | assert.Equal(t, 0, strings.Count(got, "FUNCTION_RESULT_BEGIN tx1 409")) |
| 383 | assert.EqualValues(t, 1, calls.Load()) |
| 384 | }, |
| 385 | }, |
| 386 | "panic in handler emits 500": { |
| 387 | run: func(t *testing.T, mgr *Manager, in *chanInput, out *safeBuffer) { |
| 388 | mgr.Register("fn", func(Function) { panic("boom") }) |
| 389 | |
| 390 | cancel, done := startFlowManager(t, mgr) |
| 391 | defer cancel() |
| 392 | |
| 393 | in.ch <- functionLine("tx1", "fn") |
| 394 | close(in.ch) |
| 395 | waitForDone(t, done) |
| 396 | |
| 397 | waitForSubstring(t, out.String, "FUNCTION_RESULT_BEGIN tx1 500", time.Second) |
| 398 | }, |
| 399 | }, |
| 400 | "cancel for unknown uid is no-op": { |
| 401 | run: func(t *testing.T, mgr *Manager, in *chanInput, out *safeBuffer) { |
| 402 | cancel, done := startFlowManager(t, mgr) |
| 403 | defer cancel() |
| 404 | |
| 405 | in.ch <- "FUNCTION_CANCEL unknown" |
| 406 | close(in.ch) |
| 407 | waitForDone(t, done) |
| 408 | |
| 409 | assert.Equal(t, 0, strings.Count(out.String(), "FUNCTION_RESULT_BEGIN")) |
| 410 | }, |
| 411 | }, |
| 412 | "cancel after completion is no-op": { |
| 413 | run: func(t *testing.T, mgr *Manager, in *chanInput, out *safeBuffer) { |
| 414 | mgr.Register("fn", func(fn Function) { |
| 415 | mgr.respUID(fn.UID, 200, "done") |
| 416 | }) |
| 417 | |
| 418 | cancel, done := startFlowManager(t, mgr) |
| 419 | defer cancel() |
| 420 | |
| 421 | in.ch <- functionLine("tx1", "fn") |
| 422 | waitForSubstring(t, out.String, "FUNCTION_RESULT_BEGIN tx1 200", time.Second) |
| 423 | in.ch <- "FUNCTION_CANCEL tx1" |
| 424 | close(in.ch) |
| 425 | waitForDone(t, done) |
| 426 | |
| 427 | got := out.String() |
| 428 | assert.Equal(t, 1, strings.Count(got, "FUNCTION_RESULT_BEGIN tx1 200")) |
| 429 | assert.Equal(t, 0, strings.Count(got, "FUNCTION_RESULT_BEGIN tx1 499")) |
| 430 | }, |
| 431 | }, |
| 432 | "stdin close uses canceling shutdown and force-finalizes unresolved requests": { |
| 433 | run: func(t *testing.T, mgr *Manager, in *chanInput, out *safeBuffer) { |
| 434 | mgr.shutdownDrainTimeout = 50 * time.Millisecond |
| 435 | started := make(chan struct{}, 1) |
| 436 | block := make(chan struct{}) |
| 437 | |
| 438 | mgr.Register("fn", func(fn Function) { |
| 439 | started <- struct{}{} |
| 440 | <-block |
| 441 | mgr.respUID(fn.UID, 200, "late") |
| 442 | }) |
| 443 | |
| 444 | cancel, done := startFlowManager(t, mgr) |
| 445 | defer cancel() |
| 446 | |
| 447 | in.ch <- functionLine("tx1", "fn") |
| 448 | <-started |
| 449 | close(in.ch) |
| 450 | waitForDone(t, done) |
| 451 | |
| 452 | got := out.String() |
| 453 | assert.Equal(t, 1, strings.Count(got, "FUNCTION_RESULT_BEGIN tx1 499")) |
| 454 | assert.Equal(t, 0, strings.Count(got, "FUNCTION_RESULT_BEGIN tx1 200")) |
| 455 | }, |
| 456 | }, |
| 457 | "late terminal output after shutdown is dropped by tombstone guard": { |
| 458 | run: func(t *testing.T, mgr *Manager, in *chanInput, out *safeBuffer) { |
| 459 | mgr.shutdownDrainTimeout = 50 * time.Millisecond |
| 460 | started := make(chan struct{}, 1) |
| 461 | block := make(chan struct{}) |
| 462 | doneResp := make(chan struct{}) |
| 463 | |
| 464 | mgr.Register("fn", func(fn Function) { |
| 465 | started <- struct{}{} |
| 466 | <-block |
| 467 | mgr.respUID(fn.UID, 200, "late") |
| 468 | close(doneResp) |
| 469 | }) |
| 470 | |
| 471 | cancel, done := startFlowManager(t, mgr) |
| 472 | defer cancel() |
| 473 | |
| 474 | in.ch <- functionLine("tx1", "fn") |
| 475 | <-started |
| 476 | close(in.ch) |
| 477 | waitForDone(t, done) |
| 478 | |
| 479 | // Release handler after manager has already force-finalized and returned. |
| 480 | close(block) |
| 481 | select { |
| 482 | case <-doneResp: |
| 483 | case <-time.After(time.Second): |
| 484 | t.Fatal("timed out waiting for late handler response") |
| 485 | } |
| 486 | |
| 487 | got := out.String() |
| 488 | assert.Equal(t, 1, strings.Count(got, "FUNCTION_RESULT_BEGIN tx1 499")) |
| 489 | assert.Equal(t, 0, strings.Count(got, "FUNCTION_RESULT_BEGIN tx1 200")) |
| 490 | }, |
| 491 | }, |
| 492 | "shutdown finalizes unresolved awaiting_result even when workers are drained": { |
| 493 | run: func(t *testing.T, mgr *Manager, in *chanInput, out *safeBuffer) { |
| 494 | mgr.shutdownDrainTimeout = 250 * time.Millisecond |
| 495 | mgr.awaitingWarnDelay = time.Second |
| 496 | returned := make(chan struct{}, 1) |
| 497 | |
| 498 | mgr.Register("fn", func(Function) { |
| 499 | // Return without terminal response. |
| 500 | returned <- struct{}{} |
| 501 | }) |
| 502 | |
| 503 | cancel, done := startFlowManager(t, mgr) |
| 504 | defer cancel() |
| 505 | |
| 506 | in.ch <- functionLine("tx1", "fn") |
| 507 | <-returned |
| 508 | waitForCondition(t, time.Second, func() bool { |
| 509 | mgr.invStateMux.Lock() |
| 510 | defer mgr.invStateMux.Unlock() |
| 511 | rec, ok := mgr.invState["tx1"] |
| 512 | return ok && rec != nil && rec.state == stateAwaitingResult |
| 513 | }, "tx1 reaches awaiting_result before shutdown") |
| 514 | |
| 515 | close(in.ch) |
| 516 | waitForDone(t, done) |
| 517 | |
| 518 | got := out.String() |
| 519 | assert.Equal(t, 1, strings.Count(got, "FUNCTION_RESULT_BEGIN tx1 499")) |
| 520 | assert.Equal(t, 0, strings.Count(got, "FUNCTION_RESULT_BEGIN tx1 200")) |
| 521 | }, |
| 522 | }, |
| 523 | } |
| 524 | |
| 525 | for name, tc := range tests { |
| 526 | t.Run(name, func(t *testing.T) { |
| 527 | mgr, out := newFlowManager() |
| 528 | in := &chanInput{ch: make(chan string, 16)} |
| 529 | mgr.input = in |
| 530 | tc.run(t, mgr, in, out) |
| 531 | }) |
| 532 | } |
| 533 | } |
| 534 | |
| 535 | func TestManager_InvocationStateScenarios(t *testing.T) { |
| 536 | tests := map[string]struct { |
| 537 | run func(t *testing.T, mgr *Manager, in *chanInput, out *safeBuffer) |
| 538 | }{ |
| 539 | "worker return transitions to awaiting_result until terminal output": { |
| 540 | run: func(t *testing.T, mgr *Manager, in *chanInput, out *safeBuffer) { |
| 541 | mgr.workerCount = 1 |
| 542 | returned := make(chan struct{}, 1) |
| 543 | releaseResponse := make(chan struct{}) |
| 544 | |
| 545 | mgr.Register("fn", func(fn Function) { |
| 546 | go func() { |
| 547 | <-releaseResponse |
| 548 | mgr.respUID(fn.UID, 200, "ok") |
| 549 | }() |
| 550 | returned <- struct{}{} |
| 551 | }) |
| 552 | |
| 553 | cancel, done := startFlowManager(t, mgr) |
| 554 | defer cancel() |
| 555 | |
| 556 | in.ch <- functionLine("tx1", "fn") |
| 557 | <-returned |
| 558 | |
| 559 | waitForCondition(t, time.Second, func() bool { |
| 560 | mgr.invStateMux.Lock() |
| 561 | defer mgr.invStateMux.Unlock() |
| 562 | rec, ok := mgr.invState["tx1"] |
| 563 | return ok && rec != nil && rec.state == stateAwaitingResult |
| 564 | }, "transaction tx1 reaches awaiting_result state") |
| 565 | |
| 566 | close(releaseResponse) |
| 567 | waitForSubstring(t, out.String, "FUNCTION_RESULT_BEGIN tx1 200", time.Second) |
| 568 | |
| 569 | waitForCondition(t, time.Second, func() bool { |
| 570 | mgr.invStateMux.Lock() |
| 571 | defer mgr.invStateMux.Unlock() |
| 572 | _, ok := mgr.invState["tx1"] |
| 573 | return !ok |
| 574 | }, "transaction tx1 is removed from active state after finalization") |
| 575 | |
| 576 | close(in.ch) |
| 577 | waitForDone(t, done) |
| 578 | }, |
| 579 | }, |
| 580 | } |
| 581 | |
| 582 | for name, tc := range tests { |
| 583 | t.Run(name, func(t *testing.T) { |
| 584 | mgr, out := newFlowManager() |
| 585 | in := &chanInput{ch: make(chan string, 16)} |
| 586 | mgr.input = in |
| 587 | tc.run(t, mgr, in, out) |
| 588 | }) |
| 589 | } |
| 590 | } |
| 591 | |
| 592 | func TestManager_tryFinalize(t *testing.T) { |
| 593 | tests := map[string]struct { |
| 594 | run func(t *testing.T, mgr *Manager) |
| 595 | }{ |
| 596 | "first finalization wins and tombstone blocks immediate reuse": { |
| 597 | run: func(t *testing.T, mgr *Manager) { |
| 598 | calls := 0 |
| 599 | ok := mgr.tryFinalize("uid1", "test.first", func() { calls++ }) |
| 600 | require.True(t, ok) |
| 601 | assert.Equal(t, 1, calls) |
| 602 | |
| 603 | ok = mgr.tryFinalize("uid1", "test.late", func() { calls++ }) |
| 604 | require.False(t, ok) |
| 605 | assert.Equal(t, 1, calls) |
| 606 | |
| 607 | admitted := mgr.trySetInvocationState("uid1", stateQueued, func() {}, "fn") |
| 608 | assert.Equal(t, invocationAdmissionDuplicateTombstone, admitted) |
| 609 | |
| 610 | mgr.invStateMux.Lock() |
| 611 | mgr.tombstones["uid1"] = time.Now().Add(-time.Second) |
| 612 | mgr.invStateMux.Unlock() |
| 613 | |
| 614 | admitted = mgr.trySetInvocationState("uid1", stateQueued, func() {}, "fn") |
| 615 | assert.Equal(t, invocationAdmissionAccepted, admitted) |
| 616 | }, |
| 617 | }, |
| 618 | "empty uid or nil emitter is rejected": { |
| 619 | run: func(t *testing.T, mgr *Manager) { |
| 620 | assert.False(t, mgr.tryFinalize("", "source", func() {})) |
| 621 | assert.False(t, mgr.tryFinalize("uid1", "source", nil)) |
| 622 | }, |
| 623 | }, |
| 624 | } |
| 625 | |
| 626 | for name, tc := range tests { |
| 627 | t.Run(name, func(t *testing.T) { |
| 628 | tc.run(t, NewManager()) |
| 629 | }) |
| 630 | } |
| 631 | } |
| 632 | |
| 633 | func TestManager_WorkerPoolConcurrencyBound(t *testing.T) { |
| 634 | tests := map[string]struct { |
| 635 | workerCount int |
| 636 | input []string |
| 637 | register func(t *testing.T, mgr *Manager, current, maxSeen *atomic.Int32) |
| 638 | assertions func(t *testing.T, maxSeen int32) |
| 639 | }{ |
| 640 | "same key is serialized": { |
| 641 | workerCount: 4, |
| 642 | input: []string{ |
| 643 | functionLine("tx-1", "fn"), |
| 644 | functionLine("tx-2", "fn"), |
| 645 | functionLine("tx-3", "fn"), |
| 646 | functionLine("tx-4", "fn"), |
| 647 | }, |
| 648 | register: func(t *testing.T, mgr *Manager, current, maxSeen *atomic.Int32) { |
| 649 | t.Helper() |
| 650 | mgr.Register("fn", func(fn Function) { |
| 651 | c := current.Add(1) |
| 652 | for { |
| 653 | prev := maxSeen.Load() |
| 654 | if c <= prev || maxSeen.CompareAndSwap(prev, c) { |
| 655 | break |
| 656 | } |
| 657 | } |
| 658 | |
| 659 | time.Sleep(30 * time.Millisecond) |
| 660 | current.Add(-1) |
| 661 | mgr.respUID(fn.UID, 200, "ok") |
| 662 | }) |
| 663 | }, |
| 664 | assertions: func(t *testing.T, maxSeen int32) { |
| 665 | t.Helper() |
| 666 | assert.Equal(t, int32(1), maxSeen) |
| 667 | }, |
| 668 | }, |
| 669 | "different keys run concurrently": { |
| 670 | workerCount: 4, |
| 671 | input: []string{ |
| 672 | functionLine("tx-a1", "fnA"), |
| 673 | functionLine("tx-b1", "fnB"), |
| 674 | functionLine("tx-a2", "fnA"), |
| 675 | functionLine("tx-b2", "fnB"), |
| 676 | }, |
| 677 | register: func(t *testing.T, mgr *Manager, current, maxSeen *atomic.Int32) { |
| 678 | t.Helper() |
| 679 | registerFn := func(name string) { |
| 680 | mgr.Register(name, func(fn Function) { |
| 681 | c := current.Add(1) |
| 682 | for { |
| 683 | prev := maxSeen.Load() |
| 684 | if c <= prev || maxSeen.CompareAndSwap(prev, c) { |
| 685 | break |
| 686 | } |
| 687 | } |
| 688 | |
| 689 | time.Sleep(40 * time.Millisecond) |
| 690 | current.Add(-1) |
| 691 | mgr.respUID(fn.UID, 200, "ok") |
| 692 | }) |
| 693 | } |
| 694 | registerFn("fnA") |
| 695 | registerFn("fnB") |
| 696 | }, |
| 697 | assertions: func(t *testing.T, maxSeen int32) { |
| 698 | t.Helper() |
| 699 | assert.GreaterOrEqual(t, maxSeen, int32(2)) |
| 700 | assert.LessOrEqual(t, maxSeen, int32(4)) |
| 701 | }, |
| 702 | }, |
| 703 | } |
| 704 | |
| 705 | for name, tc := range tests { |
| 706 | t.Run(name, func(t *testing.T) { |
| 707 | mgr, out := newFlowManager() |
| 708 | mgr.workerCount = tc.workerCount |
| 709 | mgr.queueSize = len(tc.input) + tc.workerCount |
| 710 | in := &chanInput{ch: make(chan string, len(tc.input)+tc.workerCount)} |
| 711 | mgr.input = in |
| 712 | |
| 713 | var current atomic.Int32 |
| 714 | var maxSeen atomic.Int32 |
| 715 | |
| 716 | tc.register(t, mgr, ¤t, &maxSeen) |
| 717 | |
| 718 | cancel, done := startFlowManager(t, mgr) |
| 719 | defer cancel() |
| 720 | |
| 721 | for _, line := range tc.input { |
| 722 | in.ch <- line |
| 723 | } |
| 724 | close(in.ch) |
| 725 | waitForDone(t, done) |
| 726 | |
| 727 | got := out.String() |
| 728 | assert.Equal(t, len(tc.input), strings.Count(got, "FUNCTION_RESULT_BEGIN tx-")) |
| 729 | tc.assertions(t, maxSeen.Load()) |
| 730 | }) |
| 731 | } |
| 732 | } |
| 733 | |
| 734 | func TestManager_DispatchInvocationStopping(t *testing.T) { |
| 735 | tests := map[string]struct { |
| 736 | run func(t *testing.T, mgr *Manager, out *safeBuffer) |
| 737 | }{ |
| 738 | "stopping manager rejects dispatch without tracking state": { |
| 739 | run: func(t *testing.T, mgr *Manager, out *safeBuffer) { |
| 740 | mgr.Register("fn", func(Function) { t.Fatal("handler should not execute when manager is stopping") }) |
| 741 | mgr.setStopping(true) |
| 742 | |
| 743 | fn := &Function{UID: "tx-stop", Name: "fn"} |
| 744 | mgr.dispatchInvocation(context.Background(), fn) |
| 745 | |
| 746 | waitForSubstring(t, out.String, "FUNCTION_RESULT_BEGIN tx-stop 503", time.Second) |
| 747 | |
| 748 | mgr.invStateMux.Lock() |
| 749 | _, ok := mgr.invState["tx-stop"] |
| 750 | mgr.invStateMux.Unlock() |
| 751 | assert.False(t, ok) |
| 752 | }, |
| 753 | }, |
| 754 | } |
| 755 | |
| 756 | for name, tc := range tests { |
| 757 | t.Run(name, func(t *testing.T) { |
| 758 | mgr, out := newFlowManager() |
| 759 | tc.run(t, mgr, out) |
| 760 | }) |
| 761 | } |
| 762 | } |