| 1 | package host |
| 2 | |
| 3 | // What is testable here without a live S3 is the GATING — who contends, who |
| 4 | // runs duties, and how the box behaves when it loses or never wins. The lock |
| 5 | // semantics themselves (a second claimant gets a *lease.HeldError; a dark |
| 6 | // holder cannot reclaim after failover) belong to internal/lease and are |
| 7 | // covered there against a real S3-compatible store — internal/lease's own |
| 8 | // fake (endpoint_test.go's newFakeS3) answers every GET with NoSuchKey, so it |
| 9 | // proves where the traffic goes and nothing about contention. This file |
| 10 | // deliberately invents no second fake S3: the election transport is injected |
| 11 | // as a seam instead. |
| 12 | |
| 13 | import ( |
| 14 | "context" |
| 15 | "errors" |
| 16 | "fmt" |
| 17 | "io" |
| 18 | "log" |
| 19 | "os" |
| 20 | "strings" |
| 21 | "sync" |
| 22 | "sync/atomic" |
| 23 | "testing" |
| 24 | "time" |
| 25 | |
| 26 | "github.com/carlosframework/platform/internal/lease" |
| 27 | ) |
| 28 | |
| 29 | // ---- helpers --------------------------------------------------------------- |
| 30 | |
| 31 | // lockedBuffer is an io.Writer safe for the election goroutine and the test |
| 32 | // body to touch at once — log output is the assertion in the "stays quiet" |
| 33 | // tests, so it has to be read while the loop is still running. |
| 34 | type lockedBuffer struct { |
| 35 | mu sync.Mutex |
| 36 | b strings.Builder |
| 37 | } |
| 38 | |
| 39 | func (l *lockedBuffer) Write(p []byte) (int, error) { |
| 40 | l.mu.Lock() |
| 41 | defer l.mu.Unlock() |
| 42 | return l.b.Write(p) |
| 43 | } |
| 44 | |
| 45 | func (l *lockedBuffer) String() string { |
| 46 | l.mu.Lock() |
| 47 | defer l.mu.Unlock() |
| 48 | return l.b.String() |
| 49 | } |
| 50 | |
| 51 | // captureRacyLog redirects the standard logger for one test, through a |
| 52 | // mutexed buffer: unlike this package's other captureLog, the writer here is |
| 53 | // shared with a live election goroutine. |
| 54 | func captureRacyLog(t *testing.T) *lockedBuffer { |
| 55 | t.Helper() |
| 56 | buf := &lockedBuffer{} |
| 57 | log.SetOutput(buf) |
| 58 | t.Cleanup(func() { log.SetOutput(os.Stderr) }) |
| 59 | return buf |
| 60 | } |
| 61 | |
| 62 | // quietLog silences the standard logger for a test that doesn't assert on it. |
| 63 | func quietLog(t *testing.T) { |
| 64 | t.Helper() |
| 65 | log.SetOutput(io.Discard) |
| 66 | t.Cleanup(func() { log.SetOutput(os.Stderr) }) |
| 67 | } |
| 68 | |
| 69 | // fakeLease stands in for *lease.Held: renewal health is a func the test |
| 70 | // flips, and releases are counted. |
| 71 | type fakeLease struct { |
| 72 | mu sync.Mutex |
| 73 | err error |
| 74 | released int |
| 75 | } |
| 76 | |
| 77 | func (f *fakeLease) fail(err error) { |
| 78 | f.mu.Lock() |
| 79 | defer f.mu.Unlock() |
| 80 | f.err = err |
| 81 | } |
| 82 | |
| 83 | func (f *fakeLease) Err() error { |
| 84 | f.mu.Lock() |
| 85 | defer f.mu.Unlock() |
| 86 | return f.err |
| 87 | } |
| 88 | |
| 89 | func (f *fakeLease) Release(context.Context) error { |
| 90 | f.mu.Lock() |
| 91 | defer f.mu.Unlock() |
| 92 | f.released++ |
| 93 | return nil |
| 94 | } |
| 95 | |
| 96 | func (f *fakeLease) releases() int { |
| 97 | f.mu.Lock() |
| 98 | defer f.mu.Unlock() |
| 99 | return f.released |
| 100 | } |
| 101 | |
| 102 | type acquireCall struct{ bucket, prefix, region string } |
| 103 | |
| 104 | // acquirer records every acquisition attempt and answers from a per-attempt |
| 105 | // func, so a test can script "contended, contended, won". |
| 106 | type acquirer struct { |
| 107 | mu sync.Mutex |
| 108 | calls []acquireCall |
| 109 | answer func(n int) (heldLease, error) |
| 110 | } |
| 111 | |
| 112 | func (a *acquirer) acquire(_ context.Context, bucket, prefix, region string) (heldLease, error) { |
| 113 | a.mu.Lock() |
| 114 | a.calls = append(a.calls, acquireCall{bucket, prefix, region}) |
| 115 | n := len(a.calls) |
| 116 | answer := a.answer |
| 117 | a.mu.Unlock() |
| 118 | return answer(n) |
| 119 | } |
| 120 | |
| 121 | func (a *acquirer) count() int { |
| 122 | a.mu.Lock() |
| 123 | defer a.mu.Unlock() |
| 124 | return len(a.calls) |
| 125 | } |
| 126 | |
| 127 | func (a *acquirer) first() acquireCall { |
| 128 | a.mu.Lock() |
| 129 | defer a.mu.Unlock() |
| 130 | return a.calls[0] |
| 131 | } |
| 132 | |
| 133 | // contended is what a losing contender sees, wrapped exactly as |
| 134 | // lease.Acquire wraps it. |
| 135 | func contended() error { |
| 136 | return fmt.Errorf("acquire lease: %w", &lease.HeldError{Owner: "other-box", ExpiresAt: time.Now().Add(time.Minute)}) |
| 137 | } |
| 138 | |
| 139 | func waitFor(t *testing.T, what string, cond func() bool) { |
| 140 | t.Helper() |
| 141 | deadline := time.Now().Add(3 * time.Second) |
| 142 | for time.Now().Before(deadline) { |
| 143 | if cond() { |
| 144 | return |
| 145 | } |
| 146 | time.Sleep(time.Millisecond) |
| 147 | } |
| 148 | t.Fatalf("timed out waiting for %s", what) |
| 149 | } |
| 150 | |
| 151 | // runElection arms hk the way RunAgent does and registers its stand-down, so |
| 152 | // every test exercises the same start/stop pair production uses. The returned |
| 153 | // func is safe to call twice (the cleanup calls it again). |
| 154 | func runElection(t *testing.T, hk *housekeeper) func() { |
| 155 | t.Helper() |
| 156 | stop := hk.start(context.Background()) |
| 157 | stopped := false |
| 158 | once := func() { |
| 159 | if stopped { |
| 160 | return |
| 161 | } |
| 162 | stopped = true |
| 163 | stop() |
| 164 | } |
| 165 | t.Cleanup(once) |
| 166 | return once |
| 167 | } |
| 168 | |
| 169 | // heldGate is the gate a duty sees on a holder, for the tests that are about |
| 170 | // duty dispatch rather than the election. |
| 171 | func heldGate() (context.Context, bool) { return context.Background(), true } |
| 172 | |
| 173 | // noGate is the gate every non-holding box sees. |
| 174 | func noGate() (context.Context, bool) { return nil, false } |
| 175 | |
| 176 | func fastHousekeeper(getenv func(string) string, a *acquirer) *housekeeper { |
| 177 | hk := newHousekeeper(getenv, a.acquire) |
| 178 | hk.retry = time.Millisecond |
| 179 | hk.poll = time.Millisecond |
| 180 | return hk |
| 181 | } |
| 182 | |
| 183 | // ---- duty gating ----------------------------------------------------------- |
| 184 | |
| 185 | func TestRunHousekeeperDutiesRunsEachDutyOncePerTickWhenHeld(t *testing.T) { |
| 186 | quietLog(t) |
| 187 | a, b := 0, 0 |
| 188 | runHousekeeperDuties(heldGate, |
| 189 | housekeeperDuty{"a", func(context.Context) error { a++; return nil }}, |
| 190 | housekeeperDuty{"b", func(context.Context) error { b++; return nil }}, |
| 191 | ) |
| 192 | if a != 1 || b != 1 { |
| 193 | t.Fatalf("duty runs = (%d, %d), want (1, 1)", a, b) |
| 194 | } |
| 195 | } |
| 196 | |
| 197 | func TestRunHousekeeperDutiesSkipsQuietlyWhenNotHeld(t *testing.T) { |
| 198 | buf := captureRacyLog(t) |
| 199 | ran := 0 |
| 200 | runHousekeeperDuties(noGate, |
| 201 | housekeeperDuty{"retention pass", func(context.Context) error { ran++; return nil }}, |
| 202 | ) |
| 203 | if ran != 0 { |
| 204 | t.Fatalf("duty ran %d times on a non-holder, want 0", ran) |
| 205 | } |
| 206 | // Not holding the lease is the NORMAL state on every box but one, every |
| 207 | // tick, forever: a log line here would bury the log. |
| 208 | if out := buf.String(); out != "" { |
| 209 | t.Fatalf("non-holder logged %q, want silence", out) |
| 210 | } |
| 211 | } |
| 212 | |
| 213 | func TestRunHousekeeperDutiesLogsOneDutysErrorAndRunsTheRest(t *testing.T) { |
| 214 | buf := captureRacyLog(t) |
| 215 | later := 0 |
| 216 | runHousekeeperDuties(heldGate, |
| 217 | housekeeperDuty{"retention pass", func(context.Context) error { return errors.New("boom") }}, |
| 218 | housekeeperDuty{"later duty", func(context.Context) error { later++; return nil }}, |
| 219 | ) |
| 220 | if later != 1 { |
| 221 | t.Fatalf("later duty ran %d times, want 1 — one duty's error must not stop the next", later) |
| 222 | } |
| 223 | out := buf.String() |
| 224 | if !strings.Contains(out, "retention pass") || !strings.Contains(out, "boom") { |
| 225 | t.Fatalf("log = %q, want the failing duty's name and error", out) |
| 226 | } |
| 227 | } |
| 228 | |
| 229 | // ---- who contends ---------------------------------------------------------- |
| 230 | |
| 231 | func TestHousekeeperFleetBoxNeverAttemptsAcquisition(t *testing.T) { |
| 232 | quietLog(t) |
| 233 | a := &acquirer{answer: func(int) (heldLease, error) { return &fakeLease{}, nil }} |
| 234 | hk := fastHousekeeper(envFunc(map[string]string{ |
| 235 | "CARLOS_FLEET": "acme", |
| 236 | "CARLOS_DEPLOYMENT_BUCKET": "dep-bucket", |
| 237 | }), a) |
| 238 | |
| 239 | // run returns immediately on a fleet box — nothing to loop over. |
| 240 | hk.run(context.Background()) |
| 241 | |
| 242 | if a.count() != 0 { |
| 243 | t.Fatalf("fleet box attempted acquisition %d times, want 0", a.count()) |
| 244 | } |
| 245 | if hk.held() { |
| 246 | t.Fatal("fleet box reports holding the housekeeper lease") |
| 247 | } |
| 248 | } |
| 249 | |
| 250 | func TestHousekeeperWithNoDeploymentStoreNeverHolds(t *testing.T) { |
| 251 | quietLog(t) |
| 252 | a := &acquirer{answer: func(int) (heldLease, error) { return &fakeLease{}, nil }} |
| 253 | hk := fastHousekeeper(envFunc(map[string]string{}), a) |
| 254 | |
| 255 | hk.run(context.Background()) |
| 256 | |
| 257 | if a.count() != 0 { |
| 258 | t.Fatalf("storeless box attempted acquisition %d times, want 0", a.count()) |
| 259 | } |
| 260 | if hk.held() { |
| 261 | t.Fatal("storeless box reports holding the housekeeper lease") |
| 262 | } |
| 263 | } |
| 264 | |
| 265 | func TestHousekeeperDirDeploymentIsItsOwnHousekeeper(t *testing.T) { |
| 266 | quietLog(t) |
| 267 | a := &acquirer{answer: func(int) (heldLease, error) { return &fakeLease{}, nil }} |
| 268 | hk := fastHousekeeper(envFunc(map[string]string{"CARLOS_DEPLOYMENT_DIR": t.TempDir()}), a) |
| 269 | stop := runElection(t, hk) |
| 270 | |
| 271 | // No goroutine to wait on: a lone box holds from the instant it is armed, |
| 272 | // because there is nothing to acquire. |
| 273 | if !hk.held() { |
| 274 | t.Fatal("dir deployment does not hold: a directory-backed deployment is one box by construction and must keep pruning") |
| 275 | } |
| 276 | if a.count() != 0 { |
| 277 | t.Fatalf("dir deployment attempted acquisition %d times, want 0 — there is no bucket to elect in", a.count()) |
| 278 | } |
| 279 | |
| 280 | // And it stands down with the process, so nothing believes it is the |
| 281 | // housekeeper after the agent has gone. |
| 282 | stop() |
| 283 | if hk.held() { |
| 284 | t.Fatal("dir deployment still reports holding after stand-down") |
| 285 | } |
| 286 | } |
| 287 | |
| 288 | // ---- the election ---------------------------------------------------------- |
| 289 | |
| 290 | func TestHousekeeperAcquiresTheDeploymentLeaseAndHolds(t *testing.T) { |
| 291 | quietLog(t) |
| 292 | a := &acquirer{answer: func(int) (heldLease, error) { return &fakeLease{}, nil }} |
| 293 | hk := fastHousekeeper(envFunc(map[string]string{"CARLOS_DEPLOYMENT_BUCKET": "dep-bucket"}), a) |
| 294 | runElection(t, hk) |
| 295 | |
| 296 | waitFor(t, "the lease to be held", hk.held) |
| 297 | |
| 298 | got := a.first() |
| 299 | want := acquireCall{bucket: "dep-bucket", prefix: housekeeperLeasePrefix, region: ""} |
| 300 | if got != want { |
| 301 | t.Fatalf("acquire call = %+v, want %+v", got, want) |
| 302 | } |
| 303 | if housekeeperLeasePrefix != "control/housekeeper-lease" { |
| 304 | t.Fatalf("lease prefix = %q, want control/housekeeper-lease (lock.json lands under it)", housekeeperLeasePrefix) |
| 305 | } |
| 306 | } |
| 307 | |
| 308 | func TestHousekeeperLosingTheElectionIsQuietAndKeepsTrying(t *testing.T) { |
| 309 | buf := captureRacyLog(t) |
| 310 | a := &acquirer{answer: func(int) (heldLease, error) { return nil, contended() }} |
| 311 | hk := fastHousekeeper(envFunc(map[string]string{"CARLOS_DEPLOYMENT_BUCKET": "dep-bucket"}), a) |
| 312 | // The arming line is not what this test is about; drop everything logged |
| 313 | // before the loop starts contending. |
| 314 | runElection(t, hk) |
| 315 | |
| 316 | waitFor(t, "several acquisition attempts", func() bool { return a.count() >= 3 }) |
| 317 | |
| 318 | if hk.held() { |
| 319 | t.Fatal("a losing contender reports holding the lease") |
| 320 | } |
| 321 | // Contention is normal on every tick a box doesn't win: an error loop |
| 322 | // here would be a line every retry, forever, on N-1 boxes. |
| 323 | for _, line := range strings.Split(strings.TrimSpace(buf.String()), "\n") { |
| 324 | if strings.Contains(line, "lease acquire") { |
| 325 | t.Fatalf("contention logged as a failure: %q", line) |
| 326 | } |
| 327 | } |
| 328 | } |
| 329 | |
| 330 | func TestHousekeeperLogsARealAcquireFailure(t *testing.T) { |
| 331 | buf := captureRacyLog(t) |
| 332 | a := &acquirer{answer: func(int) (heldLease, error) { return nil, errors.New("no credentials") }} |
| 333 | hk := fastHousekeeper(envFunc(map[string]string{"CARLOS_DEPLOYMENT_BUCKET": "dep-bucket"}), a) |
| 334 | runElection(t, hk) |
| 335 | |
| 336 | waitFor(t, "the failure to be logged", func() bool { return strings.Contains(buf.String(), "no credentials") }) |
| 337 | |
| 338 | if hk.held() { |
| 339 | t.Fatal("a box that could not acquire reports holding the lease") |
| 340 | } |
| 341 | } |
| 342 | |
| 343 | func TestHousekeeperStandsDownWhenRenewalFails(t *testing.T) { |
| 344 | quietLog(t) |
| 345 | first := &fakeLease{} |
| 346 | third := &fakeLease{} |
| 347 | // The takeover, scripted: this box holds, loses renewal, and finds the |
| 348 | // lock already taken by whoever won it after expiry — for as long as the |
| 349 | // test says so. That middle state is what makes "stood down" observable |
| 350 | // at all: if a lease were waiting on the very next attempt, the box would |
| 351 | // be holding again within a retry and a housekeeper that never cleared |
| 352 | // its flag would look identical to one that did. |
| 353 | var freed atomic.Bool |
| 354 | a := &acquirer{answer: func(n int) (heldLease, error) { |
| 355 | switch { |
| 356 | case n == 1: |
| 357 | return first, nil |
| 358 | case !freed.Load(): |
| 359 | return nil, contended() |
| 360 | default: |
| 361 | return third, nil |
| 362 | } |
| 363 | }} |
| 364 | hk := fastHousekeeper(envFunc(map[string]string{"CARLOS_DEPLOYMENT_BUCKET": "dep-bucket"}), a) |
| 365 | runElection(t, hk) |
| 366 | |
| 367 | waitFor(t, "the first lease to be held", hk.held) |
| 368 | |
| 369 | // Renewal fails: leadership is lost, so this box must stop reporting |
| 370 | // itself as the housekeeper — a stale holder still running duties (or, |
| 371 | // from Part 2, still issuing certs) against a lease another box now owns |
| 372 | // is the whole reason this check exists. |
| 373 | first.fail(errors.New("renew: precondition failed")) |
| 374 | waitFor(t, "the stand-down", func() bool { return !hk.held() }) |
| 375 | waitFor(t, "a fresh acquisition attempt", func() bool { return a.count() > 1 }) |
| 376 | |
| 377 | // A lost lease is NOT released: our etag is stale, and the TTL is what |
| 378 | // hands it over. |
| 379 | if first.releases() != 0 { |
| 380 | t.Fatalf("released a lost lease %d times, want 0", first.releases()) |
| 381 | } |
| 382 | |
| 383 | // And it recovers on its own once the lock frees up again. |
| 384 | freed.Store(true) |
| 385 | waitFor(t, "the lease to be won back", hk.held) |
| 386 | if third.releases() != 0 { |
| 387 | t.Fatalf("released the recovered lease %d times, want 0 while still holding it", third.releases()) |
| 388 | } |
| 389 | } |
| 390 | |
| 391 | // The stand-down is the path an ordinary `systemctl restart` takes, so it has |
| 392 | // to release the lock AND come back — a stop that hung would wedge the |
| 393 | // restart, and one that returned before the loop had released would strand |
| 394 | // control/housekeeper-lease/lock.json for its 15-minute TTL. |
| 395 | func TestHousekeeperStopReleasesTheLeaseAndReturns(t *testing.T) { |
| 396 | quietLog(t) |
| 397 | held := &fakeLease{} |
| 398 | a := &acquirer{answer: func(int) (heldLease, error) { return held, nil }} |
| 399 | hk := fastHousekeeper(envFunc(map[string]string{"CARLOS_DEPLOYMENT_BUCKET": "dep-bucket"}), a) |
| 400 | stop := runElection(t, hk) |
| 401 | |
| 402 | waitFor(t, "the lease to be held", hk.held) |
| 403 | |
| 404 | returned := make(chan struct{}) |
| 405 | go func() { |
| 406 | stop() |
| 407 | close(returned) |
| 408 | }() |
| 409 | select { |
| 410 | case <-returned: |
| 411 | case <-time.After(3 * time.Second): |
| 412 | t.Fatal("stop() did not return — a shutdown would hang here") |
| 413 | } |
| 414 | |
| 415 | if hk.held() { |
| 416 | t.Fatal("still reports holding the lease after shutdown") |
| 417 | } |
| 418 | if held.releases() != 1 { |
| 419 | t.Fatalf("releases = %d, want 1 — a graceful stand-down hands the lease over immediately", held.releases()) |
| 420 | } |
| 421 | // The election goroutine is gone, not merely told to go: stop() waits on |
| 422 | // it, so a leaked loop would still be acquiring after this point. |
| 423 | before := a.count() |
| 424 | time.Sleep(20 * time.Millisecond) // many retries at the 1ms test cadence |
| 425 | if after := a.count(); after != before { |
| 426 | t.Fatalf("acquisitions after stop: %d → %d, want no further attempts (the loop leaked)", before, after) |
| 427 | } |
| 428 | } |
| 429 | |
| 430 | // ---- duties are bound to the holding --------------------------------------- |
| 431 | |
| 432 | // A duty must not outlive the authority it started under: when renewal fails |
| 433 | // mid-prune, the sweep's context is cancelled rather than left to finish its |
| 434 | // deletes against a lock another box now owns. |
| 435 | func TestHousekeeperCancelsDutiesWhenTheHoldEnds(t *testing.T) { |
| 436 | quietLog(t) |
| 437 | held := &fakeLease{} |
| 438 | a := &acquirer{answer: func(n int) (heldLease, error) { |
| 439 | if n == 1 { |
| 440 | return held, nil |
| 441 | } |
| 442 | return nil, contended() |
| 443 | }} |
| 444 | hk := fastHousekeeper(envFunc(map[string]string{"CARLOS_DEPLOYMENT_BUCKET": "dep-bucket"}), a) |
| 445 | runElection(t, hk) |
| 446 | waitFor(t, "the lease to be held", hk.held) |
| 447 | |
| 448 | started := make(chan struct{}) |
| 449 | observed := make(chan error, 1) |
| 450 | go func() { |
| 451 | runHousekeeperDuties(hk.gate, housekeeperDuty{"slow duty", func(ctx context.Context) error { |
| 452 | close(started) |
| 453 | select { |
| 454 | case <-ctx.Done(): |
| 455 | observed <- ctx.Err() |
| 456 | case <-time.After(3 * time.Second): |
| 457 | observed <- nil |
| 458 | } |
| 459 | return ctx.Err() |
| 460 | }}) |
| 461 | }() |
| 462 | |
| 463 | <-started |
| 464 | held.fail(errors.New("renew: precondition failed")) |
| 465 | |
| 466 | select { |
| 467 | case err := <-observed: |
| 468 | if err == nil { |
| 469 | t.Fatal("the duty ran to completion after the hold ended — it must be cancelled") |
| 470 | } |
| 471 | case <-time.After(3 * time.Second): |
| 472 | t.Fatal("timed out waiting for the duty to observe the stand-down") |
| 473 | } |
| 474 | |
| 475 | // And the gate is shut for the next tick, so nothing new starts either. |
| 476 | if _, ok := hk.gate(); ok { |
| 477 | t.Fatal("the gate is still open after the hold ended") |
| 478 | } |
| 479 | } |
| 480 | |
| 481 | // A duty that starts on a holder and is cancelled mid-flight stood down; it |
| 482 | // did not fail, so it must not be logged as a failure — and the duties behind |
| 483 | // it in the same tick are no longer this box's to run. |
| 484 | func TestRunHousekeeperDutiesStopQuietlyWhenTheHoldEndsMidTick(t *testing.T) { |
| 485 | buf := captureRacyLog(t) |
| 486 | ctx, cancel := context.WithCancel(context.Background()) |
| 487 | later := 0 |
| 488 | runHousekeeperDuties( |
| 489 | func() (context.Context, bool) { return ctx, true }, |
| 490 | housekeeperDuty{"retention pass", func(ctx context.Context) error { |
| 491 | cancel() // the hold ends while this duty is running |
| 492 | return ctx.Err() |
| 493 | }}, |
| 494 | housekeeperDuty{"later duty", func(context.Context) error { later++; return nil }}, |
| 495 | ) |
| 496 | if later != 0 { |
| 497 | t.Fatalf("later duty ran %d times after the hold ended, want 0", later) |
| 498 | } |
| 499 | if out := buf.String(); out != "" { |
| 500 | t.Fatalf("stand-down logged as a failure: %q", out) |
| 501 | } |
| 502 | } |
| 503 | |
| 504 | // ---- the process-wide gate ------------------------------------------------- |
| 505 | |
| 506 | func TestHousekeeperHeldIsFalseBeforeAnyElectionIsArmed(t *testing.T) { |
| 507 | if housekeeperHeld() { |
| 508 | t.Fatal("housekeeperHeld() is true with no housekeeper armed") |
| 509 | } |
| 510 | if _, ok := housekeeperGate(); ok { |
| 511 | t.Fatal("housekeeperGate() reports held with no housekeeper armed") |
| 512 | } |
| 513 | } |
| 514 | |
| 515 | // ---- mintMeshKeyLoop (multi-region backhaul spec, Part 3 Task 2's review |
| 516 | // fix): the mesh key's own arming loop, independent of retentionLoop ------- |
| 517 | |
| 518 | // Mirrors TestRetentionLoopIsAPlatformBoxJob's shape (agent_test.go): a |
| 519 | // fleet box must skip before ever opening a store. |
| 520 | func TestMintMeshKeyLoopIsAPlatformBoxJob(t *testing.T) { |
| 521 | const skipLine = "agent: mesh key mint loop is a platform-box job — skipped on a fleet box" |
| 522 | |
| 523 | logs := captureLog(t) |
| 524 | door := newDoorFake(t) |
| 525 | for k, v := range fleetDoorEnv(door.url) { |
| 526 | t.Setenv(k, v) |
| 527 | } |
| 528 | t.Setenv("CARLOS_DEPLOYMENT_BUCKET", "") |
| 529 | t.Setenv("CARLOS_DEPLOYMENT_DIR", "") |
| 530 | |
| 531 | mintMeshKeyLoop() // must return; a fleet box never reaches the loop |
| 532 | |
| 533 | if got := logs(); !strings.Contains(got, skipLine) { |
| 534 | t.Fatalf("log = %q, want the skip line verbatim", got) |
| 535 | } |
| 536 | if hits := door.seen(); len(hits) != 0 { |
| 537 | t.Fatalf("door saw %+v, want zero — the skip must land BEFORE the store opens", hits) |
| 538 | } |
| 539 | } |
| 540 | |
| 541 | // The regression guard for the Critical review finding: the mint must NOT be |
| 542 | // gated on CARLOS_RELEASE_PUBKEY (retentionLoop's own gate, unrelated to the |
| 543 | // mesh key) — a box missing that var still wins the housekeeper election, so |
| 544 | // a mint duty tied to retentionLoop's arming would silently never run |
| 545 | // fleet-wide. Proven by explicitly leaving CARLOS_RELEASE_PUBKEY unset and |
| 546 | // confirming the loop still arms. |
| 547 | func TestMintMeshKeyLoopArmsWithoutReleasePubkey(t *testing.T) { |
| 548 | t.Setenv("CARLOS_FLEET", "") |
| 549 | t.Setenv("CARLOS_DEPLOYMENT_BUCKET", "") |
| 550 | t.Setenv("CARLOS_DEPLOYMENT_DIR", t.TempDir()) |
| 551 | t.Setenv("CARLOS_RELEASE_PUBKEY", "") // deliberately unset — the whole point |
| 552 | |
| 553 | buf := captureRacyLog(t) |
| 554 | // mintMeshKeyLoop never returns once armed: run it in its own goroutine |
| 555 | // (it parks in a time.Sleep for a minute, long past this test's end) and |
| 556 | // wait for the arming log line against a synchronized sink. |
| 557 | go mintMeshKeyLoop() |
| 558 | |
| 559 | waitFor(t, "the mint loop to arm despite no CARLOS_RELEASE_PUBKEY", func() bool { |
| 560 | return strings.Contains(buf.String(), "agent: mesh key mint loop armed") |
| 561 | }) |
| 562 | } |
| 563 | |
amadan