| 1 | package activator |
| 2 | |
| 3 | import ( |
| 4 | "context" |
| 5 | "errors" |
| 6 | "fmt" |
| 7 | "os" |
| 8 | "slices" |
| 9 | "strings" |
| 10 | "sync" |
| 11 | "testing" |
| 12 | "time" |
| 13 | |
| 14 | "github.com/carlosframework/platform/internal/hostclaim" |
| 15 | "github.com/carlosframework/platform/internal/hostwrite" |
| 16 | "github.com/carlosframework/platform/internal/registry" |
| 17 | "github.com/carlosframework/platform/internal/release" |
| 18 | ) |
| 19 | |
| 20 | // A hold is stop-then-start through the barrier (spec §7.8). A request that |
| 21 | // arrives during the stop is held, not refused, and is served by the new |
| 22 | // generation, which execs the attempt's staged build while the live link |
| 23 | // names something else. This is the witness for mutation M3. |
| 24 | func TestSwapWitnessHoldExecsTheStagedBuild(t *testing.T) { |
| 25 | t.Parallel() |
| 26 | f := newSwapFixture(t) |
| 27 | f.b.execVerFn = func() string { return "v3" } |
| 28 | gate := make(chan struct{}) |
| 29 | f.b.lastProc().setStopGate(gate) |
| 30 | results := make(chan SwapResult, 1) |
| 31 | go func() { results <- f.a.Swap(context.Background(), f.attempt(ModeHold), f.newest) }() |
| 32 | waitFor(t, "the hold's stop", func() bool { |
| 33 | f.a.mu.Lock() |
| 34 | defer f.a.mu.Unlock() |
| 35 | e := f.a.entries[f.host] |
| 36 | return e != nil && e.state == actStopping |
| 37 | }) |
| 38 | stopSeen := time.Now() // the old generation's stop began before this |
| 39 | served := make(chan error, 1) |
| 40 | waiting := &doneAskedCtx{Context: context.Background(), asked: make(chan struct{})} |
| 41 | go func() { |
| 42 | release, err := f.a.beginRequest(waiting, testRoute(f.host)) |
| 43 | if err == nil { |
| 44 | release() |
| 45 | } |
| 46 | served <- err |
| 47 | }() |
| 48 | // The request has reached its wait on the stopping entry: begin asks |
| 49 | // its context's Done only there (the first select it reaches). |
| 50 | <-waiting.asked |
| 51 | select { |
| 52 | case err := <-served: |
| 53 | t.Fatalf("the request was answered (%v) while the old generation was still stopping", err) |
| 54 | default: |
| 55 | } |
| 56 | gateOpen := time.Now() |
| 57 | close(gate) |
| 58 | var res SwapResult |
| 59 | select { |
| 60 | case res = <-results: |
| 61 | case <-time.After(10 * time.Second): |
| 62 | t.Fatal("the hold never finished") |
| 63 | } |
| 64 | if res.Outcome != SwapDone || res.Gen.Build != "v2" || res.Gen.StagedPath != f.staged { |
| 65 | t.Fatalf("Swap = %+v", res) |
| 66 | } |
| 67 | if held := gateOpen.Sub(stopSeen); res.Drain < held { |
| 68 | t.Fatalf("Drain = %v, want at least the old generation's gated stop, %v (spec §9.4)", res.Drain, held) |
| 69 | } |
| 70 | if err := <-served; err != nil { |
| 71 | t.Fatalf("the held request was refused: %v", err) |
| 72 | } |
| 73 | // #231: one live-link Start (the original wake). The hold's wake went |
| 74 | // through StartGen alone. |
| 75 | if got := f.b.startCount(); got != 1 { |
| 76 | t.Fatalf("Start (live link) calls = %d, want 1", got) |
| 77 | } |
| 78 | if gens, _, _ := f.b.calls(); gens != 1 { |
| 79 | t.Fatalf("StartGen calls = %d, want 1", gens) |
| 80 | } |
| 81 | if got := f.appSocket(t); got != res.Gen.Socket { |
| 82 | t.Fatalf("AppSocket = %q, want the new generation's %q", got, res.Gen.Socket) |
| 83 | } |
| 84 | } |
| 85 | |
| 86 | // doneAskedCtx is a context that signals the first time its Done channel |
| 87 | // is asked for: begin() asks only when it reaches a wait, so the signal is |
| 88 | // the request held on the entry. |
| 89 | type doneAskedCtx struct { |
| 90 | context.Context |
| 91 | once sync.Once |
| 92 | asked chan struct{} |
| 93 | } |
| 94 | |
| 95 | func (c *doneAskedCtx) Done() <-chan struct{} { |
| 96 | c.once.Do(func() { close(c.asked) }) |
| 97 | return c.Context.Done() |
| 98 | } |
| 99 | |
| 100 | // A hold canary is not done when it is healthy: it must survive the 10 s |
| 101 | // settle (spec §6.2, §8.5). Without the settle, swapHold answers SwapDone |
| 102 | // the moment the pinned wake is ready, and a crash a second later neither |
| 103 | // fails the attempt nor halts the rollout. |
| 104 | func TestSwapHoldCanarySettles(t *testing.T) { |
| 105 | old := swapSettle |
| 106 | t.Cleanup(func() { swapSettle = old }) |
| 107 | start := func(t *testing.T, f *swapFixture) (chan SwapResult, Gen, *fakeProc) { |
| 108 | t.Helper() |
| 109 | at := f.attempt(ModeHold) |
| 110 | at.Canary = true |
| 111 | results := make(chan SwapResult, 1) |
| 112 | go func() { results <- f.a.Swap(context.Background(), at, f.newest) }() |
| 113 | var g Gen |
| 114 | var p *fakeProc |
| 115 | waitFor(t, "the pinned generation awake", func() bool { |
| 116 | g, p = f.b.lastGen() |
| 117 | f.a.mu.Lock() |
| 118 | defer f.a.mu.Unlock() |
| 119 | e := f.a.entries[f.host] |
| 120 | return p != nil && e != nil && e.state == actAwake && e.gen.N == g.N |
| 121 | }) |
| 122 | return results, g, p |
| 123 | } |
| 124 | t.Run("an exit in the window is down, with no fallback", func(t *testing.T) { |
| 125 | swapSettle = 5 * time.Second |
| 126 | f := newSwapFixture(t) |
| 127 | results, _, p := start(t, f) |
| 128 | select { |
| 129 | case res := <-results: |
| 130 | t.Fatalf("Swap answered %+v before the settle ended", res) |
| 131 | case <-time.After(50 * time.Millisecond): |
| 132 | } |
| 133 | p.exit() |
| 134 | res := <-results |
| 135 | if res.Outcome != SwapDown || !strings.Contains(res.Reason, errSettleExit.Error()) { |
| 136 | t.Fatalf("Swap = %+v, want down: %v", res, errSettleExit) |
| 137 | } |
| 138 | if gens, _, _ := f.b.calls(); gens != 1 || f.b.startCount() != 1 { |
| 139 | t.Fatalf("a failed hold canary fell back: StartGen %d, Start %d", gens, f.b.startCount()) |
| 140 | } |
| 141 | }) |
| 142 | t.Run("surviving the window is done", func(t *testing.T) { |
| 143 | swapSettle = 100 * time.Millisecond |
| 144 | f := newSwapFixture(t) |
| 145 | began := time.Now() |
| 146 | results, g, _ := start(t, f) |
| 147 | res := <-results |
| 148 | if res.Outcome != SwapDone || res.Gen.N != g.N { |
| 149 | t.Fatalf("Swap = %+v", res) |
| 150 | } |
| 151 | if took := time.Since(began); took < swapSettle { |
| 152 | t.Fatalf("done after %s, inside the %s settle", took, swapSettle) |
| 153 | } |
| 154 | }) |
| 155 | t.Run("an intentional stop in the window is not a failure", func(t *testing.T) { |
| 156 | swapSettle = 5 * time.Second |
| 157 | f := newSwapFixture(t) |
| 158 | results, _, _ := start(t, f) |
| 159 | f.a.Bounce(f.host) |
| 160 | res := <-results |
| 161 | if res.Outcome != SwapAbandoned || !strings.Contains(res.Reason, errSwapStopped.Error()) { |
| 162 | t.Fatalf("Swap = %+v, want abandoned: %v", res, errSwapStopped) |
| 163 | } |
| 164 | }) |
| 165 | } |
| 166 | |
| 167 | // A hold build that fails its health check leaves nothing serving and does |
| 168 | // not fall back (spec §8.3). |
| 169 | func TestSwapHoldFailureIsDown(t *testing.T) { |
| 170 | f := newSwapFixture(t) |
| 171 | f.a.genHealth = func(context.Context, Gen) error { return errors.New("unhealthy: HTTP 500") } |
| 172 | res := f.a.Swap(context.Background(), f.attempt(ModeHold), f.newest) |
| 173 | if res.Outcome != SwapDown { |
| 174 | t.Fatalf("Swap = %+v, want down", res) |
| 175 | } |
| 176 | if f.a.entryCount() != 0 { |
| 177 | t.Fatal("a failed hold left an entry behind") |
| 178 | } |
| 179 | if _, p := f.b.lastGen(); p == nil || !p.isStopped() { |
| 180 | t.Fatal("the failed hold generation was left running") |
| 181 | } |
| 182 | if got := f.b.startCount(); got != 1 { |
| 183 | t.Fatalf("Start (live link) calls = %d: a failed hold fell back", got) |
| 184 | } |
| 185 | } |
| 186 | |
| 187 | // A pinned wake that fails before it reaches the exec ran nothing of the |
| 188 | // build: the hold is abandoned, never down (index P5/P6, spec §6.4: only a |
| 189 | // hold build that executed is held down). Reported down, plan 5 records a |
| 190 | // failure of a build that never ran and halts its cohort. |
| 191 | func TestSwapHoldRefusedBeforeTheExecIsAbandoned(t *testing.T) { |
| 192 | t.Parallel() |
| 193 | for _, tc := range []struct { |
| 194 | name string |
| 195 | arrange func(f *swapFixture) |
| 196 | want string |
| 197 | }{ |
| 198 | {"the lease is held elsewhere after the stop", func(f *swapFixture) { |
| 199 | f.b.mu.Lock() |
| 200 | f.b.acquireErr = &errLeaseHeld{holder: "box-b", retryAfter: time.Minute} |
| 201 | f.b.mu.Unlock() |
| 202 | }, "instance lease held"}, |
| 203 | {"the restore fails", func(f *swapFixture) { |
| 204 | f.b.mu.Lock() |
| 205 | f.b.restoreErr = errors.New("replica unreadable") |
| 206 | f.b.mu.Unlock() |
| 207 | }, "replica unreadable"}, |
| 208 | {"the staged build changed after the stop", func(f *swapFixture) { |
| 209 | f.b.mu.Lock() |
| 210 | f.b.restoreErr = nil |
| 211 | f.b.mu.Unlock() |
| 212 | f.a.backend = &acquireHookBackend{fakeSwapBackend: f.b, hook: func() { |
| 213 | if err := os.WriteFile(f.staged, []byte("not build v2"), 0o755); err != nil { |
| 214 | panic(err) |
| 215 | } |
| 216 | }} |
| 217 | }, "staging"}, |
| 218 | } { |
| 219 | t.Run(tc.name, func(t *testing.T) { |
| 220 | f := newSwapFixture(t) |
| 221 | tc.arrange(f) |
| 222 | res := f.a.Swap(context.Background(), f.attempt(ModeHold), f.newest) |
| 223 | if res.Outcome != SwapAbandoned || !strings.Contains(res.Reason, tc.want) { |
| 224 | t.Fatalf("Swap = %+v, want abandoned: %s", res, tc.want) |
| 225 | } |
| 226 | if gens, _, _ := f.b.calls(); gens != 0 { |
| 227 | t.Fatalf("StartGen calls = %d, want 0", gens) |
| 228 | } |
| 229 | }) |
| 230 | } |
| 231 | } |
| 232 | |
| 233 | // lockedSink is an EventSink safe for the wake goroutine and the test. |
| 234 | type lockedSink struct { |
| 235 | mu sync.Mutex |
| 236 | events []eventRec |
| 237 | } |
| 238 | |
| 239 | func (s *lockedSink) App(string, string, string, string) {} |
| 240 | func (s *lockedSink) Event(acct, app, host, level, msg string) { |
| 241 | s.mu.Lock() |
| 242 | s.events = append(s.events, eventRec{acct, app, host, level, msg}) |
| 243 | s.mu.Unlock() |
| 244 | } |
| 245 | |
| 246 | func (s *lockedSink) errors() []string { |
| 247 | s.mu.Lock() |
| 248 | defer s.mu.Unlock() |
| 249 | var out []string |
| 250 | for _, e := range s.events { |
| 251 | if e.level == "error" { |
| 252 | out = append(out, e.msg) |
| 253 | } |
| 254 | } |
| 255 | return out |
| 256 | } |
| 257 | |
| 258 | // A pinned wake's refusal is the attempt's business, not a member's: an |
| 259 | // abandoned hold emits no "wake failed" event, and no swap text reaches the |
| 260 | // member's timeline (plan 5's EvSwapDown carries a failed hold's line). A |
| 261 | // pinned build that ran and failed is reported in the wording any failed |
| 262 | // wake already uses. |
| 263 | func TestSwapHoldWakeEventsCarryNoSwapText(t *testing.T) { |
| 264 | t.Parallel() |
| 265 | t.Run("superseded before the exec emits nothing", func(t *testing.T) { |
| 266 | f := newSwapFixture(t) |
| 267 | sink := &lockedSink{} |
| 268 | f.a.Logs = sink |
| 269 | var once sync.Once |
| 270 | f.a.backend = &acquireHookBackend{fakeSwapBackend: f.b, hook: func() { once.Do(func() { f.rev.Add(1) }) }} |
| 271 | if res := f.a.Swap(context.Background(), f.attempt(ModeHold), f.newest); res.Outcome != SwapAbandoned { |
| 272 | t.Fatalf("Swap = %+v, want abandoned", res) |
| 273 | } |
| 274 | if got := sink.errors(); len(got) != 0 { |
| 275 | t.Fatalf("member error events %q, want none for an abandoned hold", got) |
| 276 | } |
| 277 | }) |
| 278 | t.Run("a mismatched spawn's text stays in the log", func(t *testing.T) { |
| 279 | f := newSwapFixture(t) |
| 280 | sink := &lockedSink{} |
| 281 | f.a.Logs = sink |
| 282 | f.a.backend = &mismatchSwapBackend{fakeSwapBackend: f.b} |
| 283 | if res := f.a.Swap(context.Background(), f.attempt(ModeHold), f.newest); res.Outcome != SwapDown { |
| 284 | t.Fatalf("Swap = %+v, want down: the mismatched process ran", res) |
| 285 | } |
| 286 | for _, msg := range sink.errors() { |
| 287 | if strings.Contains(msg, "the hold chose") || strings.Contains(msg, "swap:") { |
| 288 | t.Fatalf("member event %q carries swap text", msg) |
| 289 | } |
| 290 | } |
| 291 | }) |
| 292 | t.Run("a pinned build that failed its health uses the existing wording", func(t *testing.T) { |
| 293 | f := newSwapFixture(t) |
| 294 | sink := &lockedSink{} |
| 295 | f.a.Logs = sink |
| 296 | f.a.genHealth = func(context.Context, Gen) error { return errors.New("unhealthy: HTTP 500") } |
| 297 | if res := f.a.Swap(context.Background(), f.attempt(ModeHold), f.newest); res.Outcome != SwapDown { |
| 298 | t.Fatalf("Swap = %+v, want down", res) |
| 299 | } |
| 300 | got := sink.errors() |
| 301 | if len(got) != 1 || !strings.HasPrefix(got[0], "wake failed: ") || strings.Contains(got[0], "swap:") { |
| 302 | t.Fatalf("member error events %q, want one existing \"wake failed: …\" line", got) |
| 303 | } |
| 304 | }) |
| 305 | } |
| 306 | |
| 307 | // mismatchSwapBackend runs a different build from the one it was asked for. |
| 308 | type mismatchSwapBackend struct { |
| 309 | *fakeSwapBackend |
| 310 | } |
| 311 | |
| 312 | func (b *mismatchSwapBackend) StartStaged(ctx context.Context, rt registry.Route, g Gen) (Gen, Proc, error) { |
| 313 | ran, p, err := b.fakeSwapBackend.StartStaged(ctx, rt, g) |
| 314 | ran.Build = "v9" |
| 315 | return ran, p, err |
| 316 | } |
| 317 | |
| 318 | // acquireHookBackend runs hook inside Acquire: the window between a hold's |
| 319 | // re-check at its stop and its pinned wake's exec. |
| 320 | type acquireHookBackend struct { |
| 321 | *fakeSwapBackend |
| 322 | hook func() |
| 323 | } |
| 324 | |
| 325 | func (b *acquireHookBackend) Acquire(ctx context.Context, rt registry.Route) (Lease, error) { |
| 326 | b.hook() |
| 327 | return b.fakeSwapBackend.Acquire(ctx, rt) |
| 328 | } |
| 329 | |
| 330 | // Spec §6.3, §7.4, §8.1: a hold's attempt checks run again at the pinned |
| 331 | // wake's exec and once more before its result counts. Checked only when |
| 332 | // the stop finished, a newer attempt (or a halted cohort, which bumps the |
| 333 | // revision) landing during the wake still exec'd the superseded build and |
| 334 | // answered SwapDone, which records it good. |
| 335 | func TestSwapHoldReChecksAtTheExecAndBeforeTheResult(t *testing.T) { |
| 336 | t.Run("superseded before the exec", func(t *testing.T) { |
| 337 | f := newSwapFixture(t) |
| 338 | var once sync.Once |
| 339 | f.a.backend = &acquireHookBackend{fakeSwapBackend: f.b, hook: func() { once.Do(func() { f.rev.Add(1) }) }} |
| 340 | res := f.a.Swap(context.Background(), f.attempt(ModeHold), f.newest) |
| 341 | if res.Outcome != SwapAbandoned || !strings.Contains(res.Reason, errSwapSuperseded.Error()) { |
| 342 | t.Fatalf("Swap = %+v, want abandoned: superseded", res) |
| 343 | } |
| 344 | if gens, _, _ := f.b.calls(); gens != 0 { |
| 345 | t.Fatalf("the superseded hold build was exec'd (%d)", gens) |
| 346 | } |
| 347 | }) |
| 348 | t.Run("superseded before the result", func(t *testing.T) { |
| 349 | f := newSwapFixture(t) |
| 350 | f.a.genHealth = func(context.Context, Gen) error { f.rev.Add(1); return nil } |
| 351 | res := f.a.Swap(context.Background(), f.attempt(ModeHold), f.newest) |
| 352 | if res.Outcome != SwapAbandoned || !strings.Contains(res.Reason, errSwapSuperseded.Error()) { |
| 353 | t.Fatalf("Swap = %+v, want abandoned: a superseded result is never done", res) |
| 354 | } |
| 355 | }) |
| 356 | } |
| 357 | |
| 358 | // killingSwapBackend is fakeSwapBackend plus plan 2's KillGen. fail decides, |
| 359 | // per generation, whether the child is still occupied. |
| 360 | type killingSwapBackend struct { |
| 361 | *fakeSwapBackend |
| 362 | kmu sync.Mutex |
| 363 | fail func(n uint64) bool |
| 364 | } |
| 365 | |
| 366 | func (k *killingSwapBackend) KillGen(_ context.Context, _ registry.Route, n uint64) error { |
| 367 | k.kmu.Lock() |
| 368 | defer k.kmu.Unlock() |
| 369 | if k.fail != nil && k.fail(n) { |
| 370 | return errors.New("the child still holds a process") |
| 371 | } |
| 372 | return nil |
| 373 | } |
| 374 | |
| 375 | func (k *killingSwapBackend) setFail(f func(uint64) bool) { |
| 376 | k.kmu.Lock() |
| 377 | k.fail = f |
| 378 | k.kmu.Unlock() |
| 379 | } |
| 380 | |
| 381 | // A failed wake gives its lease back only once its generation's child is |
| 382 | // proven empty (spec §7.4). Until then the host takes no new wake, and the |
| 383 | // sweep retries the proof. cleanupFailedWake that accepts the leader's exit |
| 384 | // releases the lease over a surviving descendant and lets a new wake start |
| 385 | // beside it. |
| 386 | func TestSwapHoldFailedWakeKeepsTheLeaseUntilItsChildIsEmpty(t *testing.T) { |
| 387 | t.Parallel() |
| 388 | f := newSwapFixture(t) |
| 389 | kb := &killingSwapBackend{fakeSwapBackend: f.b} |
| 390 | kb.setFail(func(n uint64) bool { return n != f.incumbent.N }) // the hold's new generation only |
| 391 | f.a.backend = kb |
| 392 | f.a.genHealth = func(context.Context, Gen) error { return errors.New("unhealthy: HTTP 500") } |
| 393 | before := f.b.releaseCount() |
| 394 | if res := f.a.Swap(context.Background(), f.attempt(ModeHold), f.newest); res.Outcome != SwapDown { |
| 395 | t.Fatalf("Swap = %+v, want down", res) |
| 396 | } |
| 397 | if got := f.b.releaseCount(); got != before+1 { |
| 398 | t.Fatalf("lease releases %d -> %d: want the old generation's only, the failed wake's kept", before, got) |
| 399 | } |
| 400 | if _, err := f.a.beginRequest(context.Background(), testRoute(f.host)); err == nil || !strings.Contains(err.Error(), errWakeUnproven.Error()) { |
| 401 | t.Fatalf("a new wake = %v, want refused while the failed generation is unproven", err) |
| 402 | } |
| 403 | kb.setFail(nil) |
| 404 | f.a.sweepOnce() |
| 405 | waitFor(t, "the failed wake's lease given back", func() bool { return f.b.releaseCount() == before+2 }) |
| 406 | } |
| 407 | |
| 408 | // The same when the failed wake's leader itself never stops: an unproven |
| 409 | // stop (ErrStillRunning) is no termination, so the lease is kept until the |
| 410 | // sweep's retried stop and proof hold. Retaining only when the leader had |
| 411 | // exited released the lease over a live process. |
| 412 | func TestSwapHoldFailedWakeWhoseLeaderNeverStopsKeepsTheLease(t *testing.T) { |
| 413 | t.Parallel() |
| 414 | f := newSwapFixture(t) |
| 415 | f.b.gmu.Lock() |
| 416 | f.b.genStuck = true |
| 417 | f.b.gmu.Unlock() |
| 418 | f.a.genHealth = func(context.Context, Gen) error { return errors.New("unhealthy: HTTP 500") } |
| 419 | before := f.b.releaseCount() |
| 420 | if res := f.a.Swap(context.Background(), f.attempt(ModeHold), f.newest); res.Outcome != SwapDown { |
| 421 | t.Fatalf("Swap = %+v, want down", res) |
| 422 | } |
| 423 | if got := f.b.releaseCount(); got != before+1 { |
| 424 | t.Fatalf("lease releases %d -> %d: the failed wake's lease was given back over a live leader", before, got) |
| 425 | } |
| 426 | _, p := f.b.lastGen() |
| 427 | p.exit() |
| 428 | f.a.sweepOnce() |
| 429 | waitFor(t, "the failed wake's lease given back", func() bool { return f.b.releaseCount() == before+2 }) |
| 430 | } |
| 431 | |
| 432 | // The re-check before the spawn happens when the stop finishes. A fence, a |
| 433 | // newer attempt or the agent's shutdown abandons the hold, and nothing is |
| 434 | // spawned. |
| 435 | func TestSwapHoldRefusedAtTheStop(t *testing.T) { |
| 436 | t.Parallel() |
| 437 | for _, tc := range []struct { |
| 438 | name string |
| 439 | act func(t *testing.T, f *swapFixture) |
| 440 | want error |
| 441 | }{ |
| 442 | {"fenced", func(_ *testing.T, f *swapFixture) { |
| 443 | f.a.mu.Lock() |
| 444 | if f.a.fences == nil { |
| 445 | f.a.fences = map[string]uint64{} |
| 446 | } |
| 447 | f.a.fences[f.host] = 5 |
| 448 | f.a.mu.Unlock() |
| 449 | }, errSwapFenced}, |
| 450 | {"superseded", func(_ *testing.T, f *swapFixture) { f.rev.Add(1) }, errSwapSuperseded}, |
| 451 | {"shutdown", func(t *testing.T, f *swapFixture) { |
| 452 | go f.a.Shutdown(context.Background()) |
| 453 | // The shutdown is announced before the old generation's stop is |
| 454 | // let go, so the outcome does not hang on goroutine scheduling. |
| 455 | waitFor(t, "the shutdown announced", func() bool { |
| 456 | f.a.mu.Lock() |
| 457 | defer f.a.mu.Unlock() |
| 458 | return f.a.shuttingDown |
| 459 | }) |
| 460 | }, errSwapShutdown}, |
| 461 | } { |
| 462 | t.Run(tc.name, func(t *testing.T) { |
| 463 | f := newSwapFixture(t) |
| 464 | gate := make(chan struct{}) |
| 465 | f.b.lastProc().setStopGate(gate) |
| 466 | results := make(chan SwapResult, 1) |
| 467 | go func() { results <- f.a.Swap(context.Background(), f.attempt(ModeHold), f.newest) }() |
| 468 | waitFor(t, "the hold's stop", func() bool { |
| 469 | f.a.mu.Lock() |
| 470 | defer f.a.mu.Unlock() |
| 471 | e := f.a.entries[f.host] |
| 472 | return e != nil && e.state == actStopping |
| 473 | }) |
| 474 | tc.act(t, f) |
| 475 | close(gate) |
| 476 | res := <-results |
| 477 | if res.Outcome != SwapAbandoned || !strings.Contains(res.Reason, tc.want.Error()) { |
| 478 | t.Fatalf("Swap = %+v, want abandoned: %v", res, tc.want) |
| 479 | } |
| 480 | if gens, _, _ := f.b.calls(); gens != 0 { |
| 481 | t.Fatal("a refused hold spawned a generation") |
| 482 | } |
| 483 | }) |
| 484 | } |
| 485 | } |
| 486 | |
| 487 | // Claims on: a hold needs a claim runtime that can start a pinned |
| 488 | // generation of its own. Without one it is refused BEFORE anything stops. |
| 489 | func TestSwapHoldClaimsOnNeedsPins(t *testing.T) { |
| 490 | f := newSwapFixture(t) |
| 491 | f.a.cfg.HostWrites = hostwrite.Config{Enabled: true, Selector: &swapSelector{work: &swapWork{}}} |
| 492 | res := f.a.Swap(context.Background(), f.attempt(ModeHold), f.newest) |
| 493 | if res.Outcome != SwapAbandoned || !strings.Contains(res.Reason, errClaimsNoGenerations.Error()) { |
| 494 | t.Fatalf("Swap = %+v", res) |
| 495 | } |
| 496 | if f.b.lastProc().isStopped() { |
| 497 | t.Fatal("a refused hold stopped the incumbent") |
| 498 | } |
| 499 | } |
| 500 | |
| 501 | // A row that cannot run a staged generation is refused BEFORE anything |
| 502 | // stops (spec §7.8 is stop-then-start: a start that cannot happen must not |
| 503 | // cost the instance its running process). Refused after the stop, the |
| 504 | // pinned wake failed and the hold answered SwapDown for a build that never |
| 505 | // ran. Swaps off refuses a hold as it refuses an overlap. |
| 506 | func TestSwapHoldRefusesARowThatCannotRunAStagedBuild(t *testing.T) { |
| 507 | t.Parallel() |
| 508 | for _, tc := range []struct { |
| 509 | name string |
| 510 | arrange func(f *swapFixture) |
| 511 | want error |
| 512 | }{ |
| 513 | {"swap off", func(f *swapFixture) { f.a.gens = false }, errSwapOff}, |
| 514 | {"unit-backed", func(f *swapFixture) { |
| 515 | f.a.mu.Lock() |
| 516 | f.a.entries[f.host].route.Backing = registry.BackingUnit |
| 517 | f.a.mu.Unlock() |
| 518 | }, errSwapKind}, |
| 519 | {"fenced layout", func(f *swapFixture) { |
| 520 | f.b.layoutFn = func(registry.Route) (Layout, string) { return LayoutFenced, "a migration journal" } |
| 521 | }, errSwapLayout}, |
| 522 | {"worker", func(f *swapFixture) { |
| 523 | f.a.mu.Lock() |
| 524 | f.a.entries[f.host].worker = true |
| 525 | f.a.mu.Unlock() |
| 526 | }, errSwapKind}, |
| 527 | } { |
| 528 | t.Run(tc.name, func(t *testing.T) { |
| 529 | f := newSwapFixture(t) |
| 530 | tc.arrange(f) |
| 531 | res := f.a.Swap(context.Background(), f.attempt(ModeHold), f.newest) |
| 532 | if res.Outcome != SwapAbandoned || !strings.Contains(res.Reason, tc.want.Error()) { |
| 533 | t.Fatalf("Swap = %+v, want abandoned: %v", res, tc.want) |
| 534 | } |
| 535 | if f.b.lastProc().isStopped() { |
| 536 | t.Fatal("a refused hold stopped the incumbent") |
| 537 | } |
| 538 | if gens, _, _ := f.b.calls(); gens != 0 || f.b.startCount() != 1 { |
| 539 | t.Fatalf("a refused hold spawned: StartGen %d, Start %d", gens, f.b.startCount()) |
| 540 | } |
| 541 | f.a.mu.Lock() |
| 542 | e := f.a.entries[f.host] |
| 543 | held := e != nil && e.holdNext != nil |
| 544 | f.a.mu.Unlock() |
| 545 | if held { |
| 546 | t.Fatal("a refused hold left its spec on the entry") |
| 547 | } |
| 548 | }) |
| 549 | } |
| 550 | } |
| 551 | |
| 552 | // The caller is released when the old generation's stop outlasts |
| 553 | // wakeWaitBound, the bound §7.8 holds requests for, or when the caller's own |
| 554 | // context ends. The hold is withdrawn under the entry lock, so the stop's |
| 555 | // eventual finish installs no pinned wake and the next request wakes the |
| 556 | // ordinary way. Waiting on the stop alone held the caller for as long as an |
| 557 | // unproven stop stayed pending. |
| 558 | func TestSwapHoldReleasesItsCallerWhileTheStopIsPending(t *testing.T) { |
| 559 | for _, tc := range []struct { |
| 560 | name string |
| 561 | run func(f *swapFixture) SwapResult |
| 562 | want error |
| 563 | }{ |
| 564 | {"the stop outlasts wakeWaitBound", func(f *swapFixture) SwapResult { |
| 565 | return f.a.Swap(context.Background(), f.attempt(ModeHold), f.newest) |
| 566 | }, errHoldStopSlow}, |
| 567 | {"the caller's context ends", func(f *swapFixture) SwapResult { |
| 568 | ctx, cancel := context.WithCancel(context.Background()) |
| 569 | defer cancel() |
| 570 | go func() { |
| 571 | // Cancel once the hold's stop has begun (bounded: the test's |
| 572 | // own assertion reports a hold that never got there). |
| 573 | for deadline := time.Now().Add(5 * time.Second); time.Now().Before(deadline); time.Sleep(time.Millisecond) { |
| 574 | f.a.mu.Lock() |
| 575 | e := f.a.entries[f.host] |
| 576 | stopping := e != nil && e.state == actStopping |
| 577 | f.a.mu.Unlock() |
| 578 | if stopping { |
| 579 | break |
| 580 | } |
| 581 | } |
| 582 | cancel() |
| 583 | }() |
| 584 | return f.a.Swap(ctx, f.attempt(ModeHold), f.newest) |
| 585 | }, errSwapStopped}, |
| 586 | } { |
| 587 | t.Run(tc.name, func(t *testing.T) { |
| 588 | old := wakeWaitBound |
| 589 | t.Cleanup(func() { wakeWaitBound = old }) |
| 590 | f := newSwapFixture(t) |
| 591 | if errors.Is(tc.want, errHoldStopSlow) { |
| 592 | wakeWaitBound = 100 * time.Millisecond |
| 593 | } |
| 594 | gate := make(chan struct{}) |
| 595 | incumbent := f.b.lastProc() |
| 596 | incumbent.setStopGate(gate) |
| 597 | res := tc.run(f) |
| 598 | if res.Outcome != SwapAbandoned || !strings.Contains(res.Reason, tc.want.Error()) { |
| 599 | close(gate) |
| 600 | t.Fatalf("Swap = %+v, want abandoned: %v", res, tc.want) |
| 601 | } |
| 602 | f.a.mu.Lock() |
| 603 | prev := f.a.entries[f.host] |
| 604 | if prev == nil || prev.state != actStopping || prev.holdNext != nil { |
| 605 | f.a.mu.Unlock() |
| 606 | close(gate) |
| 607 | t.Fatal("want the old entry still stopping, with its hold withdrawn") |
| 608 | } |
| 609 | stopped := prev.stopped |
| 610 | f.a.mu.Unlock() |
| 611 | close(gate) |
| 612 | <-stopped |
| 613 | // finishStop would install the pinned wake in the critical section |
| 614 | // that closes stopped, so its absence now is final. |
| 615 | f.a.mu.Lock() |
| 616 | next := f.a.entries[f.host] |
| 617 | f.a.mu.Unlock() |
| 618 | if next != nil { |
| 619 | t.Fatalf("a withdrawn hold installed an entry (state %v)", next.state) |
| 620 | } |
| 621 | if gens, _, _ := f.b.calls(); gens != 0 { |
| 622 | t.Fatalf("a withdrawn hold exec'd its build (%d)", gens) |
| 623 | } |
| 624 | }) |
| 625 | } |
| 626 | } |
| 627 | |
| 628 | // A claims-on pinned wake runs the hold's attempt gate before anything is |
| 629 | // admitted: a superseded hold reaches neither the prerequisites nor the |
| 630 | // admission. |
| 631 | func TestDoClaimWakeRunsThePinGateFirst(t *testing.T) { |
| 632 | obs := hostclaim.Observed{Host: "x.example", ETag: "one", Claim: &hostclaim.Claim{ |
| 633 | Holder: &hostclaim.Holder{Location: "instances/x.example/inc", Session: 1, Generation: "one", WriterBox: "box"}, |
| 634 | Op: hostclaim.Op{Kind: hostclaim.KindServing, Exec: "box"}, |
| 635 | }} |
| 636 | admitted := false |
| 637 | a := &Service{cfg: Config{HostWrites: hostwrite.Config{Enabled: true}, Claims: &ClaimRuntime{Box: "box", Pins: true, |
| 638 | Read: func(context.Context, string) (hostclaim.Observed, error) { return obs, nil }, |
| 639 | StopForeign: func(context.Context, string) error { return nil }, |
| 640 | FailedAdmission: func(context.Context, hostclaim.Observed, error) error { return nil }, |
| 641 | Prerequisites: func(context.Context, registry.Route, hostclaim.Observed, startPolicy) error { |
| 642 | admitted = true |
| 643 | return nil |
| 644 | }, |
| 645 | Admit: func(context.Context, registry.Route, hostclaim.Observed, startPolicy) (hostclaim.Observed, error) { |
| 646 | admitted = true |
| 647 | return obs, nil |
| 648 | }, |
| 649 | Prepare: func(context.Context, registry.Route, hostclaim.Observed, startPolicy) (ClaimStart, error) { |
| 650 | admitted = true |
| 651 | return ClaimStart{}, nil |
| 652 | }}}} |
| 653 | e := &actEntry{host: obs.Host, route: registry.Route{Host: obs.Host, Kind: "instance"}, pin: &Gen{Build: "v2"}, |
| 654 | pinCheck: func() error { return errSwapSuperseded }} |
| 655 | if _, err := a.doClaimWake(context.Background(), e); !errors.Is(err, errSwapSuperseded) || admitted { |
| 656 | t.Fatalf("doClaimWake = %v (admitted %v), want the pin gate's refusal before admission", err, admitted) |
| 657 | } |
| 658 | } |
| 659 | |
| 660 | // A claim runtime that prepared a different build from the pinned one is |
| 661 | // refused before Restore touches the database. |
| 662 | func TestStartAdmittedRefusesAPinMismatch(t *testing.T) { |
| 663 | obs := hostclaim.Observed{Host: "x.example", ETag: "one", Claim: &hostclaim.Claim{ |
| 664 | Holder: &hostclaim.Holder{Location: "instances/x.example/inc", Session: 1, Generation: "one"}, |
| 665 | Op: hostclaim.Op{Kind: hostclaim.KindServing, Exec: "box"}, |
| 666 | }} |
| 667 | work, _ := (hostwrite.Config{}).Select(context.Background(), obs.Host, "spawn") |
| 668 | stop := errors.New("stop at restore") |
| 669 | for _, tc := range []struct { |
| 670 | version string |
| 671 | restored bool |
| 672 | }{{"v1", false}, {"v2", true}} { |
| 673 | restored := false |
| 674 | a := &Service{cfg: Config{HostWrites: hostwrite.Config{Enabled: true}, Claims: &ClaimRuntime{Box: "box", |
| 675 | Read: func(context.Context, string) (hostclaim.Observed, error) { return obs, nil }, |
| 676 | Prepare: func(context.Context, registry.Route, hostclaim.Observed, startPolicy) (ClaimStart, error) { |
| 677 | s := claimedStartFor(obs, work, tc.version) |
| 678 | s.Restore = func(context.Context) error { restored = true; return stop } |
| 679 | return s, nil |
| 680 | }}}} |
| 681 | policy := startPolicy{MaxStarts: 1, Window: time.Second, Pin: &Gen{Build: "v2"}} |
| 682 | err := a.startAdmitted(context.Background(), &actEntry{host: obs.Host}, obs, policy) |
| 683 | if restored != tc.restored { |
| 684 | t.Fatalf("prepared %s: restored = %v, want %v (err %v)", tc.version, restored, tc.restored, err) |
| 685 | } |
| 686 | if !tc.restored && !errors.Is(err, hostwrite.ErrUnavailable) { |
| 687 | t.Fatalf("prepared %s: %v, want hostwrite.ErrUnavailable", tc.version, err) |
| 688 | } |
| 689 | if tc.restored && !errors.Is(err, stop) { |
| 690 | t.Fatalf("prepared %s: %v, want the restore's own error", tc.version, err) |
| 691 | } |
| 692 | } |
| 693 | } |
| 694 | |
| 695 | // A foreign row keeps its canonical socket, so an overlap attempt on it |
| 696 | // runs as a hold (plan 1's reading): the old process stops, then the staged |
| 697 | // build starts on the same socket. Refusing it (errSwapLayout) left a |
| 698 | // foreign row unable to take any release with the flag on. |
| 699 | func TestSwapOverlapOnAForeignRowRunsAsAHold(t *testing.T) { |
| 700 | f := newSwapFixture(t) |
| 701 | f.b.layoutFn = func(registry.Route) (Layout, string) { return LayoutForeign, "hand-set socket" } |
| 702 | incumbent := f.b.lastProc() |
| 703 | res := f.a.Swap(context.Background(), f.attempt(ModeOverlap), f.newest) |
| 704 | if res.Outcome != SwapDone || res.Gen.StagedPath != f.staged || res.Mode != ModeHold { |
| 705 | t.Fatalf("Swap = %+v, want done through the hold, reported as a hold (Mode)", res) |
| 706 | } |
| 707 | ev := f.b.events() |
| 708 | if i, j := slices.Index(ev, "stop:instance"), indexPrefix(ev, "startGen:"+f.staged); i < 0 || j < 0 || i > j { |
| 709 | t.Fatalf("order %v: a foreign row stops before its replacement starts", ev) |
| 710 | } |
| 711 | if indexPrefix(ev, "raise") >= 0 { |
| 712 | t.Fatalf("events %v: a foreign row never raises an overlap parent", ev) |
| 713 | } |
| 714 | if !incumbent.isStopped() { |
| 715 | t.Fatal("the old process outlived the hold") |
| 716 | } |
| 717 | } |
| 718 | |
| 719 | // Task 5, plan review #2: the scheduler writes an attempt's intent in the |
| 720 | // mode Swap will run it in, which EffectiveMode answers with Swap's own |
| 721 | // override: an overlap on a foreign row is a hold, anything else is as given. |
| 722 | func TestEffectiveModeIsTheModeSwapRuns(t *testing.T) { |
| 723 | f := newSwapFixture(t) |
| 724 | host := f.attempt(ModeOverlap).Host |
| 725 | if m := f.a.EffectiveMode(host, ModeOverlap); m != ModeOverlap { |
| 726 | t.Fatalf("flat row: %s, want overlap", m) |
| 727 | } |
| 728 | f.b.layoutFn = func(registry.Route) (Layout, string) { return LayoutForeign, "hand-set socket" } |
| 729 | if m := f.a.EffectiveMode(host, ModeOverlap); m != ModeHold { |
| 730 | t.Fatalf("foreign row: %s, want hold", m) |
| 731 | } |
| 732 | if m := f.a.EffectiveMode(host, ModeHold); m != ModeHold { |
| 733 | t.Fatalf("foreign row, hold: %s", m) |
| 734 | } |
| 735 | } |
| 736 | |
| 737 | // A nested row can hold-swap now that root mounts the staged build (Task 9): |
| 738 | // the refusal Task 6 made for it is gone, and the hold runs to SwapDone. |
| 739 | func TestSwapHoldAcceptsANestedRow(t *testing.T) { |
| 740 | f := newSwapFixture(t) |
| 741 | f.b.layoutFn = func(registry.Route) (Layout, string) { return LayoutNested, "" } |
| 742 | res := f.a.Swap(context.Background(), f.attempt(ModeHold), f.newest) |
| 743 | if res.Outcome != SwapDone || res.Gen.StagedPath != f.staged { |
| 744 | t.Fatalf("Swap = %+v, want a done hold on the staged build", res) |
| 745 | } |
| 746 | } |
| 747 | |
| 748 | // zdd-readiness Task 13 (Astra R1 T12, R3 §5 item 9): a hold stops its |
| 749 | // incumbent before its own build execs, so the build has to be in the |
| 750 | // generations file before that stop begins. Published only at the pinned |
| 751 | // wake's exec, root could sweep or restage the staged build in the window |
| 752 | // where nothing serves at all. |
| 753 | |
| 754 | // The generations file names the hold's build when the incumbent's stop |
| 755 | // begins. The fake records what the file said at that moment, before the |
| 756 | // stop's own gate. |
| 757 | func TestSwapHoldPublishesItsBuildBeforeTheIncumbentStops(t *testing.T) { |
| 758 | f := newSwapFixture(t) |
| 759 | f.a.GenerationsFile = release.GenerationsPath(t.TempDir()) |
| 760 | incumbent := f.b.lastProc() |
| 761 | incumbent.mu.Lock() |
| 762 | incumbent.onStopEnter = func() { |
| 763 | f.b.record(fmt.Sprintf("stop-begins:file=%s", strings.Join(hostsOf(f.a.GenerationsFile)[f.host], ","))) |
| 764 | } |
| 765 | incumbent.mu.Unlock() |
| 766 | res := f.a.Swap(context.Background(), f.attempt(ModeHold), f.newest) |
| 767 | if res.Outcome != SwapDone || res.Gen.StagedPath != f.staged { |
| 768 | t.Fatalf("Swap = %+v, want a done hold on the staged build", res) |
| 769 | } |
| 770 | evs := f.b.events() |
| 771 | began := slices.IndexFunc(evs, func(ev string) bool { return strings.HasPrefix(ev, "stop-begins:") }) |
| 772 | if began < 0 { |
| 773 | t.Fatalf("events %q: the incumbent's stop never began", evs) |
| 774 | } |
| 775 | if evs[began] != "stop-begins:file=v1,v2" { |
| 776 | t.Fatalf("when the incumbent's stop began the file named %q, want the hold's build v2 beside v1", evs[began]) |
| 777 | } |
| 778 | if i := slices.Index(evs, "startGen:"+f.staged); i < began { |
| 779 | t.Fatalf("events %q: the hold's exec at %d, before its incumbent's stop at %d", evs, i, began) |
| 780 | } |
| 781 | // The pending registration went with the swap: the file names what runs. |
| 782 | waitForGenerations(t, f.a.GenerationsFile, map[string][]string{f.host: {"v2"}}) |
| 783 | f.a.mu.Lock() |
| 784 | pending := len(f.a.genPending) |
| 785 | f.a.mu.Unlock() |
| 786 | if pending != 0 { |
| 787 | t.Fatalf("%d pending generations left after the hold finished", pending) |
| 788 | } |
| 789 | } |
| 790 | |
| 791 | // A hold whose build cannot be published stops nothing: the incumbent keeps |
| 792 | // serving and the attempt is abandoned with nothing executed. |
| 793 | func TestSwapHoldUnpublishedLeavesTheIncumbentServing(t *testing.T) { |
| 794 | f := newSwapFixture(t) |
| 795 | f.a.GenerationsFile = unwritableGenerationsFile(t) |
| 796 | incumbent := f.b.lastProc() |
| 797 | res := f.a.Swap(context.Background(), f.attempt(ModeHold), f.newest) |
| 798 | if res.Outcome != SwapAbandoned || res.Executed || !strings.Contains(res.Reason, errStagedBuildNotPublished.Error()) { |
| 799 | t.Fatalf("Swap = %+v, want abandoned, not executed: unpublished", res) |
| 800 | } |
| 801 | if n := incumbent.stopEntered(); n != 0 || incumbent.isStopped() { |
| 802 | t.Fatalf("the incumbent's stop was entered %d times for a hold whose build was never published", n) |
| 803 | } |
| 804 | f.a.mu.Lock() |
| 805 | e := f.a.entries[f.host] |
| 806 | awake := e != nil && e.state == actAwake && e.gen.N == f.incumbent.N && e.holdNext == nil |
| 807 | pending := len(f.a.genPending) |
| 808 | f.a.mu.Unlock() |
| 809 | if !awake { |
| 810 | t.Fatal("want the incumbent's entry still awake on its generation, with no hold on it") |
| 811 | } |
| 812 | if pending != 0 { |
| 813 | t.Fatalf("%d pending generations left by an abandoned hold", pending) |
| 814 | } |
| 815 | if gens, _, _ := f.b.calls(); gens != 0 { |
| 816 | t.Fatalf("StartGen calls = %d, want 0", gens) |
| 817 | } |
| 818 | } |
| 819 | |
| 820 | // The publication runs off a.mu, so the admission is asked again once it is |
| 821 | // written. An attempt superseded while the file was written withdraws its |
| 822 | // pending build and stops nothing. |
| 823 | func TestSwapHoldReadmitsAfterThePublication(t *testing.T) { |
| 824 | f := newSwapFixture(t) |
| 825 | f.a.GenerationsFile = release.GenerationsPath(t.TempDir()) |
| 826 | incumbent := f.b.lastProc() |
| 827 | at := f.attempt(ModeHold) |
| 828 | // newest is asked under a.mu at each admission. The newer attempt |
| 829 | // arrives exactly when the hold's build is in the file: after the |
| 830 | // first admission, before the second. |
| 831 | newest := func(host string) uint64 { |
| 832 | if slices.Contains(hostsOf(f.a.GenerationsFile)[host], "v2") { |
| 833 | return at.SchedRev + 1 |
| 834 | } |
| 835 | return at.SchedRev |
| 836 | } |
| 837 | res := f.a.Swap(context.Background(), at, newest) |
| 838 | if res.Outcome != SwapAbandoned || res.Executed || !strings.Contains(res.Reason, errSwapSuperseded.Error()) { |
| 839 | t.Fatalf("Swap = %+v, want abandoned, not executed: superseded", res) |
| 840 | } |
| 841 | if n := incumbent.stopEntered(); n != 0 { |
| 842 | t.Fatalf("the incumbent's stop was entered %d times for a superseded hold", n) |
| 843 | } |
| 844 | f.a.mu.Lock() |
| 845 | e := f.a.entries[f.host] |
| 846 | awake := e != nil && e.state == actAwake && e.holdNext == nil |
| 847 | pending := len(f.a.genPending) |
| 848 | f.a.mu.Unlock() |
| 849 | if !awake || pending != 0 { |
| 850 | t.Fatalf("awake with no hold = %v, pending generations = %d", awake, pending) |
| 851 | } |
| 852 | // The withdrawn build is no longer published. |
| 853 | waitForGenerations(t, f.a.GenerationsFile, map[string][]string{f.host: {"v1"}}) |
| 854 | if gens, _, _ := f.b.calls(); gens != 0 { |
| 855 | t.Fatalf("StartGen calls = %d, want 0", gens) |
| 856 | } |
| 857 | } |
| 858 | |
amadan