| 1 | package activator |
| 2 | |
| 3 | import ( |
| 4 | "context" |
| 5 | "errors" |
| 6 | "fmt" |
| 7 | "slices" |
| 8 | "strings" |
| 9 | "sync" |
| 10 | "testing" |
| 11 | "time" |
| 12 | |
| 13 | "github.com/carlosframework/platform/internal/registry" |
| 14 | ) |
| 15 | |
| 16 | // execLog is a BeforeExec hook that records host@build, the backend's spawn |
| 17 | // count when it was asked (so a test can tell the record came before the |
| 18 | // spawn), and the build-state location the context carried. fail, when |
| 19 | // set, says which builds it refuses. |
| 20 | type execLog struct { |
| 21 | mu sync.Mutex |
| 22 | calls []string |
| 23 | at []int |
| 24 | locs []BuildStateLocation |
| 25 | spawns func() int |
| 26 | fail func(build string) bool |
| 27 | } |
| 28 | |
| 29 | func (l *execLog) hook(ctx context.Context, host, build string) error { |
| 30 | n := -1 |
| 31 | if l.spawns != nil { |
| 32 | n = l.spawns() |
| 33 | } |
| 34 | loc, _ := BuildStateLocationFrom(ctx) |
| 35 | l.mu.Lock() |
| 36 | defer l.mu.Unlock() |
| 37 | l.calls = append(l.calls, host+"@"+build) |
| 38 | l.at = append(l.at, n) |
| 39 | l.locs = append(l.locs, loc) |
| 40 | if l.fail != nil && l.fail(build) { |
| 41 | return errors.New("bucket refused the build state") |
| 42 | } |
| 43 | return nil |
| 44 | } |
| 45 | |
| 46 | func (l *execLog) seen() ([]string, []int, []BuildStateLocation) { |
| 47 | l.mu.Lock() |
| 48 | defer l.mu.Unlock() |
| 49 | return append([]string(nil), l.calls...), append([]int(nil), l.at...), append([]BuildStateLocation(nil), l.locs...) |
| 50 | } |
| 51 | |
| 52 | func failAll(string) bool { return true } |
| 53 | |
| 54 | // --- the hook itself --- |
| 55 | |
| 56 | // Ruling R-EXEC-NOVERSION: an identity outside the version store has no |
| 57 | // build a build state could name. With the hook set it is refused; with no |
| 58 | // hook (a box with no replica bucket) it runs as today. |
| 59 | func TestAnUnrecordableBuildIsRefused(t *testing.T) { |
| 60 | a := &Service{cfg: Config{Bucket: "b"}} |
| 61 | if err := a.recordBuild(context.Background(), "one.example.com", ""); err != nil { |
| 62 | t.Fatalf("a nil hook refused: %v", err) |
| 63 | } |
| 64 | var l execLog |
| 65 | a.BeforeExec = l.hook |
| 66 | err := a.recordBuild(context.Background(), "one.example.com", "") |
| 67 | if !errors.Is(err, errBuildStateUnwritten) { |
| 68 | t.Fatalf("err = %v, want errBuildStateUnwritten", err) |
| 69 | } |
| 70 | if calls, _, _ := l.seen(); len(calls) != 0 { |
| 71 | t.Fatalf("the hook was asked to record no build: %v", calls) |
| 72 | } |
| 73 | } |
| 74 | |
| 75 | // The unclaimed lease's location rides to the hook, so the agent writes |
| 76 | // beside this lease's lock.json; a location the caller already attached |
| 77 | // (the claimed path, Task 5c) is kept. A hook error is wrapped so every |
| 78 | // classifier reads it as a refusal. |
| 79 | func TestTheHookIsToldWhereTheLeaseIs(t *testing.T) { |
| 80 | a := &Service{cfg: Config{Bucket: "hib", Prefix: "inst"}} |
| 81 | var l execLog |
| 82 | a.BeforeExec = l.hook |
| 83 | if err := a.recordBuild(context.Background(), "one.example.com", "v1"); err != nil { |
| 84 | t.Fatal(err) |
| 85 | } |
| 86 | claimed := BuildStateLocation{Bucket: "data", Prefix: "instances/one.example.com/inc1"} |
| 87 | if err := a.recordBuild(WithBuildStateLocation(context.Background(), claimed), "one.example.com", "v1"); err != nil { |
| 88 | t.Fatal(err) |
| 89 | } |
| 90 | calls, _, locs := l.seen() |
| 91 | if len(calls) != 2 || calls[0] != "one.example.com@v1" { |
| 92 | t.Fatalf("calls = %v", calls) |
| 93 | } |
| 94 | if want := (BuildStateLocation{Bucket: "hib", Prefix: HibernatePrefix("inst", "one.example.com")}); locs[0] != want { |
| 95 | t.Fatalf("location = %+v, want the hibernate lease's %+v", locs[0], want) |
| 96 | } |
| 97 | if locs[1] != claimed { |
| 98 | t.Fatalf("location = %+v, want the caller's %+v kept", locs[1], claimed) |
| 99 | } |
| 100 | l.fail = failAll |
| 101 | err := a.recordBuild(context.Background(), "one.example.com", "v1") |
| 102 | if !errors.Is(err, errBuildStateUnwritten) || !strings.Contains(err.Error(), "bucket refused") { |
| 103 | t.Fatalf("err = %v, want errBuildStateUnwritten carrying the cause", err) |
| 104 | } |
| 105 | } |
| 106 | |
| 107 | // A wake's own record (Task 5b) is pending until its first exec (amadan |
| 108 | // #272): a call for that build writes nothing, the exec seam writes it |
| 109 | // once, and another build is written at once. A record of another build |
| 110 | // supersedes the pending one, which the exec seam then never writes. |
| 111 | func TestARecordedBuildIsNotWrittenTwice(t *testing.T) { |
| 112 | rt := registry.Route{Host: "one.example.com"} |
| 113 | a := &Service{cfg: Config{Bucket: "b"}} |
| 114 | var l execLog |
| 115 | a.BeforeExec = l.hook |
| 116 | ctx := wakeRecord(a, rt.Host, "v1") |
| 117 | if err := a.recordBuild(ctx, rt.Host, "v1"); err != nil { |
| 118 | t.Fatal(err) |
| 119 | } |
| 120 | if calls, _, _ := l.seen(); len(calls) != 0 { |
| 121 | t.Fatalf("calls = %v before any exec, want none", calls) |
| 122 | } |
| 123 | for range 2 { |
| 124 | if _, err := runPendingRecord(ctx); err != nil { |
| 125 | t.Fatal(err) |
| 126 | } |
| 127 | } |
| 128 | if err := a.recordBuild(ctx, rt.Host, "v2"); err != nil { |
| 129 | t.Fatal(err) |
| 130 | } |
| 131 | if calls, _, _ := l.seen(); !slices.Equal(calls, []string{"one.example.com@v1", "one.example.com@v2"}) { |
| 132 | t.Fatalf("calls = %v, want v1 once at the exec, then the fallback's v2", calls) |
| 133 | } |
| 134 | |
| 135 | var s execLog |
| 136 | a.BeforeExec = s.hook |
| 137 | ctx = wakeRecord(a, rt.Host, "v1") |
| 138 | if err := a.recordBuild(ctx, rt.Host, "v2"); err != nil { |
| 139 | t.Fatal(err) |
| 140 | } |
| 141 | if _, err := runPendingRecord(ctx); err != nil { |
| 142 | t.Fatal(err) |
| 143 | } |
| 144 | if calls, _, _ := s.seen(); !slices.Equal(calls, []string{"one.example.com@v2"}) { |
| 145 | t.Fatalf("calls = %v, want only v2: the superseded v1 is never written", calls) |
| 146 | } |
| 147 | |
| 148 | // A write that fails refuses every exec of the wake, and is not retried. |
| 149 | f := execLog{fail: failAll} |
| 150 | a.BeforeExec = f.hook |
| 151 | ctx = wakeRecord(a, rt.Host, "v1") |
| 152 | for range 2 { |
| 153 | if _, err := runPendingRecord(ctx); !errors.Is(err, errBuildStateUnwritten) { |
| 154 | t.Fatalf("err = %v, want errBuildStateUnwritten", err) |
| 155 | } |
| 156 | } |
| 157 | if calls, _, _ := f.seen(); len(calls) != 1 { |
| 158 | t.Fatalf("calls = %v, want one attempt", calls) |
| 159 | } |
| 160 | } |
| 161 | |
| 162 | // wakeRecord is the context settleWakeIdentity returns for a wake of host |
| 163 | // that settled build: its record pending. |
| 164 | func wakeRecord(a *Service, host, build string) context.Context { |
| 165 | return withPendingRecord(context.Background(), build, func() error { return a.recordBuild(context.Background(), host, build) }, nil) |
| 166 | } |
| 167 | |
| 168 | // Ruling R-WAKEWORDS: a member never sees the internal text. |
| 169 | func TestMemberSeesFencedForAnUnwrittenBuildState(t *testing.T) { |
| 170 | err := fmt.Errorf("start one.example.com: %w", fmt.Errorf("%w: s3: access denied", errBuildStateUnwritten)) |
| 171 | if got, want := memberSafeWakeReason(err), memberSafeWakeReason(ErrFenced); got != want { |
| 172 | t.Fatalf("member reads %q, want the fence's %q", got, want) |
| 173 | } |
| 174 | } |
| 175 | |
| 176 | // A refused record is never a failure of the build, whichever classifier |
| 177 | // reads it. |
| 178 | func TestAnUnwrittenBuildStateIsARefusal(t *testing.T) { |
| 179 | err := fmt.Errorf("spawn: %w", fmt.Errorf("%w: s3: access denied", errBuildStateUnwritten)) |
| 180 | if got := classifyStep(nil, err); got != stepStale { |
| 181 | t.Fatalf("classifyStep = %v, want stepStale", got) |
| 182 | } |
| 183 | if got := verdictOf(err); got != gateNoOwnership { |
| 184 | t.Fatalf("verdictOf = %v, want a refusal", got) |
| 185 | } |
| 186 | } |
| 187 | |
| 188 | // --- every chokepoint that execs a captured identity (anchors-exec §1) --- |
| 189 | |
| 190 | // #11: a hold swap's pinned wake records the attempt's build before its |
| 191 | // exec; a refused record abandons the attempt with nothing exec'd. |
| 192 | func TestBeforeExecRunsBeforeAHoldSwap(t *testing.T) { |
| 193 | t.Parallel() |
| 194 | f := newSwapFixture(t) |
| 195 | l := &execLog{spawns: func() int { n, _, _ := f.b.calls(); return n }} |
| 196 | f.a.BeforeExec = l.hook |
| 197 | res := f.a.Swap(context.Background(), f.attempt(ModeHold), f.newest) |
| 198 | if res.Outcome != SwapDone { |
| 199 | t.Fatalf("Swap = %+v", res) |
| 200 | } |
| 201 | calls, at, locs := l.seen() |
| 202 | if len(calls) != 1 || calls[0] != f.host+"@v2" || at[0] != 0 { |
| 203 | t.Fatalf("records %v at spawn counts %v, want %s@v2 before the exec", calls, at, f.host) |
| 204 | } |
| 205 | if locs[0] != (BuildStateLocation{Bucket: "test-bucket", Prefix: HibernatePrefix("", f.host)}) { |
| 206 | t.Fatalf("location = %+v", locs[0]) |
| 207 | } |
| 208 | if gens, _, _ := f.b.calls(); gens != 1 { |
| 209 | t.Fatalf("StartStaged calls = %d, want 1", gens) |
| 210 | } |
| 211 | |
| 212 | f = newSwapFixture(t) |
| 213 | f.a.BeforeExec = (&execLog{fail: failAll}).hook |
| 214 | res = f.a.Swap(context.Background(), f.attempt(ModeHold), f.newest) |
| 215 | if res.Outcome != SwapAbandoned || !strings.Contains(res.Reason, errBuildStateUnwritten.Error()) { |
| 216 | t.Fatalf("Swap = %+v, want abandoned for the unwritten build state", res) |
| 217 | } |
| 218 | if gens, _, _ := f.b.calls(); gens != 0 { |
| 219 | t.Fatalf("StartStaged calls = %d: a build ran unrecorded", gens) |
| 220 | } |
| 221 | } |
| 222 | |
| 223 | // #12: an overlap successor records its build before its exec; a refused |
| 224 | // record is SwapAbandoned, never SwapFailed (which the scheduler writes as |
| 225 | // the build's failure), and the incumbent keeps serving. |
| 226 | func TestBeforeExecRunsBeforeAnOverlapSuccessor(t *testing.T) { |
| 227 | t.Parallel() |
| 228 | f := newSwapFixture(t) |
| 229 | l := &execLog{spawns: func() int { n, _, _ := f.b.calls(); return n }} |
| 230 | f.a.BeforeExec = l.hook |
| 231 | res := f.a.Swap(context.Background(), f.attempt(ModeOverlap), f.newest) |
| 232 | if res.Outcome != SwapDone { |
| 233 | t.Fatalf("Swap = %+v", res) |
| 234 | } |
| 235 | if calls, at, _ := l.seen(); len(calls) != 1 || calls[0] != f.host+"@v2" || at[0] != 0 { |
| 236 | t.Fatalf("records %v at spawn counts %v, want %s@v2 before the exec", calls, at, f.host) |
| 237 | } |
| 238 | |
| 239 | f = newSwapFixture(t) |
| 240 | f.a.BeforeExec = (&execLog{fail: failAll}).hook |
| 241 | res = f.a.Swap(context.Background(), f.attempt(ModeOverlap), f.newest) |
| 242 | if res.Outcome != SwapAbandoned || !strings.Contains(res.Reason, errBuildStateUnwritten.Error()) { |
| 243 | t.Fatalf("Swap = %+v, want abandoned (never failed) for the unwritten build state", res) |
| 244 | } |
| 245 | if gens, _, _ := f.b.calls(); gens != 0 { |
| 246 | t.Fatalf("StartStaged calls = %d: a build ran unrecorded", gens) |
| 247 | } |
| 248 | if got := f.appSocket(t); got != f.incumbent.Socket { |
| 249 | t.Fatalf("AppSocket = %q, want the incumbent's", got) |
| 250 | } |
| 251 | if _, raises, restores := f.b.calls(); raises != restores { |
| 252 | t.Fatalf("raises %d restores %d: the parent limit was not restored", raises, restores) |
| 253 | } |
| 254 | } |
| 255 | |
| 256 | // #13: a planned boot records B before its exec; a refused record execs |
| 257 | // nothing and records nothing against B. |
| 258 | func TestBeforeExecRunsBeforeAPlannedBoot(t *testing.T) { |
| 259 | t.Parallel() |
| 260 | dir := t.TempDir() |
| 261 | next := stagedID(t, dir, "v2") |
| 262 | a, b, _ := planActivator(t, WakePlan{Boot: next, Mode: ModeOverlap}) |
| 263 | l := &execLog{spawns: func() int { return len(b.started()) }} |
| 264 | a.BeforeExec = l.hook |
| 265 | done, err := a.beginRequest(context.Background(), testRoute("one.example.com")) |
| 266 | if err != nil { |
| 267 | t.Fatalf("the wake: %v", err) |
| 268 | } |
| 269 | done() |
| 270 | if calls, at, _ := l.seen(); len(calls) != 1 || calls[0] != "one.example.com@v2" || at[0] != 0 { |
| 271 | t.Fatalf("records %v at spawn counts %v, want one.example.com@v2 before the exec", calls, at) |
| 272 | } |
| 273 | |
| 274 | dir = t.TempDir() |
| 275 | good, next := stagedID(t, dir, "v1"), stagedID(t, dir, "v2") |
| 276 | a, b, evs := planActivator(t, WakePlan{Boot: next, Mode: ModeOverlap, FirstWake: true, Fallback: &good}) |
| 277 | a.BeforeExec = (&execLog{fail: failAll}).hook |
| 278 | _, err = a.beginRequest(context.Background(), testRoute("one.example.com")) |
| 279 | if !errors.Is(err, errBuildStateUnwritten) { |
| 280 | t.Fatalf("err = %v, want errBuildStateUnwritten", err) |
| 281 | } |
| 282 | if gens := b.started(); len(gens) != 0 { |
| 283 | t.Fatalf("spawned %+v: a build ran unrecorded", gens) |
| 284 | } |
| 285 | time.Sleep(50 * time.Millisecond) |
| 286 | if k := evs.kinds(); len(k) != 0 { |
| 287 | t.Fatalf("events %v: a build that never ran was recorded", k) |
| 288 | } |
| 289 | } |
| 290 | |
| 291 | // #15: the fallback is a different build touching the data, so it records |
| 292 | // again before its own exec. A refused record starts no good build; B's own |
| 293 | // failure (it did fail) stays its own, and nothing falls back. |
| 294 | func TestBeforeExecRunsBeforeAPlannedFallback(t *testing.T) { |
| 295 | t.Parallel() |
| 296 | dir := t.TempDir() |
| 297 | good, next := stagedID(t, dir, "v1"), stagedID(t, dir, "v2") |
| 298 | a, b, evs := planActivator(t, WakePlan{Boot: next, Mode: ModeOverlap, FirstWake: true, Fallback: &good}) |
| 299 | a.genHealth = func(_ context.Context, g Gen) error { |
| 300 | if g.Build == "v2" { |
| 301 | return errors.New("GET / answered 500") |
| 302 | } |
| 303 | return nil |
| 304 | } |
| 305 | l := &execLog{spawns: func() int { return len(b.started()) }} |
| 306 | a.BeforeExec = l.hook |
| 307 | done, err := a.beginRequest(context.Background(), testRoute("one.example.com")) |
| 308 | if err != nil { |
| 309 | t.Fatalf("the wake: %v", err) |
| 310 | } |
| 311 | done() |
| 312 | calls, at, _ := l.seen() |
| 313 | if len(calls) != 2 || calls[0] != "one.example.com@v2" || calls[1] != "one.example.com@v1" || at[0] != 0 || at[1] != 1 { |
| 314 | t.Fatalf("records %v at spawn counts %v, want v2 before B's exec, then v1 before the fallback's", calls, at) |
| 315 | } |
| 316 | evs.wait(t, GenFellBack) |
| 317 | |
| 318 | a, b, evs = planActivator(t, WakePlan{Boot: next, Mode: ModeOverlap, FirstWake: true, Fallback: &good}) |
| 319 | a.genHealth = func(context.Context, Gen) error { return errors.New("GET / answered 500") } |
| 320 | a.BeforeExec = (&execLog{fail: func(build string) bool { return build == "v1" }}).hook |
| 321 | // The wake's error is B's failure (spawnPlanned wraps it); the |
| 322 | // fallback's refusal is its detail. |
| 323 | if _, err := a.beginRequest(context.Background(), testRoute("one.example.com")); err == nil || !strings.Contains(err.Error(), errBuildStateUnwritten.Error()) { |
| 324 | t.Fatalf("err = %v, want B's failure with the fallback's unwritten build state", err) |
| 325 | } |
| 326 | if gens := b.started(); len(gens) != 1 || gens[0].Build != "v2" { |
| 327 | t.Fatalf("spawned %+v, want B alone", gens) |
| 328 | } |
| 329 | if ev := evs.wait(t, GenFailed); ev.Failed.Version != "v2" { |
| 330 | t.Fatalf("failure event = %+v, want B's own", ev) |
| 331 | } |
| 332 | time.Sleep(50 * time.Millisecond) |
| 333 | for _, k := range evs.kinds() { |
| 334 | if k == GenFellBack { |
| 335 | t.Fatalf("events %v: a fallback that never ran was reported", evs.kinds()) |
| 336 | } |
| 337 | } |
| 338 | } |
| 339 | |
| 340 | // #5: a worker wake records the identity it is about to exec, after its |
| 341 | // publication; a refused record execs nothing and fails the wake. |
| 342 | func TestBeforeExecRunsBeforeAWorkerWake(t *testing.T) { |
| 343 | t.Parallel() |
| 344 | a, b, _, _, _ := stagedActivator(t) |
| 345 | host := "bot.x.internal" |
| 346 | l := &execLog{spawns: b.workerStarts} |
| 347 | a.BeforeExec = l.hook |
| 348 | if err := a.WakeWorker(alwaysOnWorkerRoute(host)); err != nil { |
| 349 | t.Fatalf("WakeWorker: %v", err) |
| 350 | } |
| 351 | calls, at, locs := l.seen() |
| 352 | if len(calls) != 1 || calls[0] != host+"@v1" || at[0] != 0 { |
| 353 | t.Fatalf("records %v at spawn counts %v, want %s@v1 before the exec", calls, at, host) |
| 354 | } |
| 355 | if locs[0] != (BuildStateLocation{Bucket: "test-bucket", Prefix: HibernatePrefix("", host)}) { |
| 356 | t.Fatalf("location = %+v, want the worker's own lease", locs[0]) |
| 357 | } |
| 358 | |
| 359 | a, b, _, _, _ = stagedActivator(t) |
| 360 | a.BeforeExec = (&execLog{fail: failAll}).hook |
| 361 | if err := a.WakeWorker(alwaysOnWorkerRoute(host)); !errors.Is(err, errBuildStateUnwritten) { |
| 362 | t.Fatalf("WakeWorker = %v, want errBuildStateUnwritten", err) |
| 363 | } |
| 364 | if n := b.workerStarts(); n != 0 { |
| 365 | t.Fatalf("%d worker children: a build ran unrecorded", n) |
| 366 | } |
| 367 | } |
| 368 | |
| 369 | // #7: a crash restart records the captured build before it respawns it. A |
| 370 | // refused record is a refusal: nothing exec'd, no "failed to respawn" |
| 371 | // member error event, and the sweep asks again. |
| 372 | func TestBeforeExecRunsBeforeAWorkerRecycle(t *testing.T) { |
| 373 | t.Parallel() |
| 374 | a, b, clk, _, _ := stagedActivator(t) |
| 375 | sink := &lockedSink{} |
| 376 | a.Logs = sink |
| 377 | host := "bot.x.internal" |
| 378 | l := &execLog{spawns: b.workerStarts} |
| 379 | a.BeforeExec = l.hook |
| 380 | if err := a.WakeWorker(alwaysOnWorkerRoute(host)); err != nil { |
| 381 | t.Fatalf("WakeWorker: %v", err) |
| 382 | } |
| 383 | b.lastWorker().kill() |
| 384 | a.sweepOnce() |
| 385 | waitForRestart(t, a, b.fakeBackend, host, 2) |
| 386 | calls, at, _ := l.seen() |
| 387 | if len(calls) != 2 || calls[1] != host+"@v1" || at[1] != 1 { |
| 388 | t.Fatalf("records %v at spawn counts %v, want the recycle's %s@v1 before its exec", calls, at, host) |
| 389 | } |
| 390 | |
| 391 | l.mu.Lock() |
| 392 | l.fail = failAll |
| 393 | l.mu.Unlock() |
| 394 | b.lastWorker().kill() |
| 395 | clk.advance(sidecarRestartMax + time.Second) |
| 396 | a.sweepOnce() |
| 397 | waitForSpawnCalls(t, a, b, host, 3) |
| 398 | if n := b.workerStarts(); n != 2 { |
| 399 | t.Fatalf("%d worker children: a build ran unrecorded", n) |
| 400 | } |
| 401 | sink.mu.Lock() |
| 402 | defer sink.mu.Unlock() |
| 403 | for _, ev := range sink.events { |
| 404 | if ev.level == "error" { |
| 405 | t.Fatalf("member event %+v: a refused record read as a failed respawn", ev) |
| 406 | } |
| 407 | } |
| 408 | } |
| 409 | |
| 410 | // #10: ReplaceWorker's attempt records its build before the exec; a refused |
| 411 | // record is abandonment, never a failure of the build. |
| 412 | func TestBeforeExecRunsBeforeReplaceWorker(t *testing.T) { |
| 413 | shrinkFirstWake(t, 55*time.Second, 25*time.Second, 100*time.Millisecond) |
| 414 | a, b, _, _, _ := stagedActivator(t) |
| 415 | host := "bot.x.internal" |
| 416 | l := &execLog{spawns: b.workerStarts} |
| 417 | a.BeforeExec = l.hook |
| 418 | if err := a.WakeWorker(alwaysOnWorkerRoute(host)); err != nil { |
| 419 | t.Fatal(err) |
| 420 | } |
| 421 | res := a.ReplaceWorker(context.Background(), workerAttempt(CauseRelease, ModeOverlap, fakeIdentity("v2")), nil, one) |
| 422 | if res.Outcome != SwapDone { |
| 423 | t.Fatalf("ReplaceWorker = %+v", res) |
| 424 | } |
| 425 | if calls, at, _ := l.seen(); len(calls) != 2 || calls[1] != host+"@v2" || at[1] != 1 { |
| 426 | t.Fatalf("records %v at spawn counts %v, want %s@v2 before the attempt's exec", calls, at, host) |
| 427 | } |
| 428 | |
| 429 | a, b, _, _, _ = stagedActivator(t) |
| 430 | l = &execLog{fail: func(build string) bool { return build == "v2" }} |
| 431 | a.BeforeExec = l.hook |
| 432 | if err := a.WakeWorker(alwaysOnWorkerRoute(host)); err != nil { |
| 433 | t.Fatal(err) |
| 434 | } |
| 435 | res = a.ReplaceWorker(context.Background(), workerAttempt(CauseRelease, ModeOverlap, fakeIdentity("v2")), nil, one) |
| 436 | if res.Outcome != SwapAbandoned || !strings.Contains(res.Reason, errBuildStateUnwritten.Error()) { |
| 437 | t.Fatalf("ReplaceWorker = %+v, want abandoned: an unrecorded build never ran", res) |
| 438 | } |
| 439 | waitFor(t, "the respawns to finish", func() bool { return !a.entryRespawning(host) }) |
| 440 | // Every child had a recorded build before it, and v2 never served. |
| 441 | calls, _, _ := l.seen() |
| 442 | recorded := 0 |
| 443 | for _, c := range calls { |
| 444 | if c != host+"@v2" { |
| 445 | recorded++ |
| 446 | } |
| 447 | } |
| 448 | if !slices.Contains(calls, host+"@v2") || b.workerStarts() != recorded { |
| 449 | t.Fatalf("records %v, %d worker children: a child ran without its build recorded", calls, b.workerStarts()) |
| 450 | } |
| 451 | if got := a.entryServing(host); got.Version == "v2" { |
| 452 | t.Fatalf("serving %+v: v2 ran unrecorded", got) |
| 453 | } |
| 454 | // Fix round 1: nothing of v2 ran, so the replaced v1 is restored at |
| 455 | // once, with no sweep: the worker is not left down until the next |
| 456 | // crash restart. |
| 457 | if got, w := a.entryServing(host), b.lastWorker(); got != fakeIdentity("v1") || b.workerStarts() != 2 || w == nil || w.Exited() { |
| 458 | t.Fatalf("serving %+v with %d children (last exited: %v); want v1 restored and running without a sweep", got, b.workerStarts(), w != nil && w.Exited()) |
| 459 | } |
| 460 | } |
| 461 | |
amadan