| 1 | package rollout |
| 2 | |
| 3 | import ( |
| 4 | "context" |
| 5 | "fmt" |
| 6 | "sort" |
| 7 | "strings" |
| 8 | "sync" |
| 9 | "sync/atomic" |
| 10 | "testing" |
| 11 | "time" |
| 12 | |
| 13 | "github.com/carlosframework/platform/internal/activator" |
| 14 | "github.com/carlosframework/platform/internal/appconfig" |
| 15 | "github.com/carlosframework/platform/internal/registry" |
| 16 | "github.com/carlosframework/platform/internal/release" |
| 17 | ) |
| 18 | |
| 19 | type swapCall struct { |
| 20 | a activator.Attempt |
| 21 | outcome activator.SwapOutcome |
| 22 | } |
| 23 | |
| 24 | type replaceCall struct { |
| 25 | a activator.Attempt |
| 26 | fallback *release.StagedIdentity |
| 27 | } |
| 28 | |
| 29 | // fakeAct models plan 3's Swap faithfully on the one point this plan relies |
| 30 | // on: the re-check before the switch. A superseded attempt never switches. |
| 31 | type fakeAct struct { |
| 32 | mu sync.Mutex |
| 33 | gens map[string]activator.Gen |
| 34 | owned map[string]bool |
| 35 | n uint64 |
| 36 | gates map[string]chan struct{} |
| 37 | outcome func(a activator.Attempt) activator.SwapOutcome |
| 38 | started []string |
| 39 | calls []swapCall |
| 40 | inFlight int |
| 41 | maxFlight int |
| 42 | replaced []replaceCall |
| 43 | worker func(a activator.Attempt, fallback *release.StagedIdentity) activator.SwapResult |
| 44 | // mode, when set, is the effective mode Swap reports (SwapResult.Mode): |
| 45 | // plan 3 runs an overlap attempt on a foreign-layout row as a hold. |
| 46 | mode activator.SwapMode |
| 47 | // reason, when set, is a failed Swap's diagnostic text in place of |
| 48 | // "health check failed". |
| 49 | reason string |
| 50 | // onHeldPublish, when set, runs on every PublishHeldServing: a test's |
| 51 | // view of the moment the scheduler releases a held worker publication. |
| 52 | onHeldPublish func(host string) |
| 53 | // executed, when set, is Swap's execution evidence (SwapResult.Executed) |
| 54 | // for an attempt that ended out. Unset, only an abandoned attempt ran |
| 55 | // nothing. |
| 56 | executed func(a activator.Attempt, out activator.SwapOutcome) bool |
| 57 | } |
| 58 | |
| 59 | // PublishHeldServing runs onHeldPublish, when set. |
| 60 | func (f *fakeAct) PublishHeldServing(host string) { |
| 61 | f.mu.Lock() |
| 62 | hook := f.onHeldPublish |
| 63 | f.mu.Unlock() |
| 64 | if hook != nil { |
| 65 | hook(host) |
| 66 | } |
| 67 | } |
| 68 | |
| 69 | // EffectiveMode is the mode Swap will run m in: mode when a test set it (the |
| 70 | // foreign-layout override it models), m otherwise. |
| 71 | func (f *fakeAct) EffectiveMode(_ string, m activator.SwapMode) activator.SwapMode { |
| 72 | f.mu.Lock() |
| 73 | defer f.mu.Unlock() |
| 74 | if f.mode != "" { |
| 75 | return f.mode |
| 76 | } |
| 77 | return m |
| 78 | } |
| 79 | |
| 80 | func newFakeAct() *fakeAct { |
| 81 | return &fakeAct{gens: map[string]activator.Gen{}, owned: map[string]bool{}, gates: map[string]chan struct{}{}} |
| 82 | } |
| 83 | |
| 84 | // probeAct is the Activator the harness hands its manager: the fake, with |
| 85 | // a hook a test may install at any moment, run before every Owns and |
| 86 | // Current. It is wired in at harness build, before any scheduler starts, |
| 87 | // and the hook is swapped atomically (final review round 1, finding 7, |
| 88 | // Astra): a test that replaced Deps.Act after a Pass raced the scheduler |
| 89 | // goroutines that read it. |
| 90 | type probeAct struct { |
| 91 | *fakeAct |
| 92 | hook atomic.Pointer[func(method, host string)] |
| 93 | } |
| 94 | |
| 95 | // onAsk installs f as the hook; nil removes it. |
| 96 | func (p *probeAct) onAsk(f func(method, host string)) { |
| 97 | if f == nil { |
| 98 | p.hook.Store(nil) |
| 99 | return |
| 100 | } |
| 101 | p.hook.Store(&f) |
| 102 | } |
| 103 | |
| 104 | func (p *probeAct) ask(method, host string) { |
| 105 | if f := p.hook.Load(); f != nil { |
| 106 | (*f)(method, host) |
| 107 | } |
| 108 | } |
| 109 | |
| 110 | func (p *probeAct) Owns(host string) bool { |
| 111 | p.ask("Owns", host) |
| 112 | return p.fakeAct.Owns(host) |
| 113 | } |
| 114 | |
| 115 | func (p *probeAct) Current(host string) (activator.Gen, bool) { |
| 116 | p.ask("Current", host) |
| 117 | return p.fakeAct.Current(host) |
| 118 | } |
| 119 | |
| 120 | func (f *fakeAct) Current(host string) (activator.Gen, bool) { |
| 121 | f.mu.Lock() |
| 122 | defer f.mu.Unlock() |
| 123 | g, ok := f.gens[host] |
| 124 | return g, ok |
| 125 | } |
| 126 | |
| 127 | func (f *fakeAct) Owns(host string) bool { |
| 128 | f.mu.Lock() |
| 129 | defer f.mu.Unlock() |
| 130 | _, ok := f.gens[host] |
| 131 | return ok || f.owned[host] |
| 132 | } |
| 133 | |
| 134 | func (f *fakeAct) serve(host, build string) { |
| 135 | f.mu.Lock() |
| 136 | defer f.mu.Unlock() |
| 137 | f.n++ |
| 138 | f.gens[host] = activator.Gen{N: f.n, Build: build, StagedPath: "/opt/" + build, Digest: "sha-" + build} |
| 139 | } |
| 140 | |
| 141 | func (f *fakeAct) hold(host string) chan struct{} { |
| 142 | f.mu.Lock() |
| 143 | defer f.mu.Unlock() |
| 144 | ch := make(chan struct{}) |
| 145 | f.gates[host] = ch |
| 146 | return ch |
| 147 | } |
| 148 | |
| 149 | func (f *fakeAct) Swap(_ context.Context, a activator.Attempt, newest func(string) uint64) activator.SwapResult { |
| 150 | f.mu.Lock() |
| 151 | f.inFlight++ |
| 152 | f.maxFlight = max(f.maxFlight, f.inFlight) |
| 153 | f.started = append(f.started, a.Host+":"+a.Build) |
| 154 | gate := f.gates[a.Host] |
| 155 | delete(f.gates, a.Host) |
| 156 | f.mu.Unlock() |
| 157 | if gate != nil { |
| 158 | <-gate |
| 159 | } |
| 160 | out := activator.SwapDone |
| 161 | if newest(a.Host) != a.SchedRev { |
| 162 | out = activator.SwapAbandoned |
| 163 | } else if f.outcome != nil { |
| 164 | out = f.outcome(a) |
| 165 | } |
| 166 | f.mu.Lock() |
| 167 | defer f.mu.Unlock() |
| 168 | f.inFlight-- |
| 169 | f.calls = append(f.calls, swapCall{a: a, outcome: out}) |
| 170 | executed := out != activator.SwapAbandoned |
| 171 | if f.executed != nil { |
| 172 | executed = f.executed(a, out) |
| 173 | } |
| 174 | if out != activator.SwapDone { |
| 175 | reason := "health check failed" |
| 176 | if f.reason != "" { |
| 177 | reason = f.reason |
| 178 | } |
| 179 | return activator.SwapResult{Outcome: out, Reason: reason, Mode: f.mode, Executed: executed} |
| 180 | } |
| 181 | f.n++ |
| 182 | g := activator.Gen{N: f.n, Build: a.Build, StagedPath: a.StagedPath, Digest: a.Digest, Demand: a.Demand} |
| 183 | f.gens[a.Host] = g |
| 184 | return activator.SwapResult{Outcome: out, Gen: g, Executed: executed} |
| 185 | } |
| 186 | |
| 187 | // ReplaceWorker models plan 5's same-entry socketless attempt (Task 11): |
| 188 | // worker decides the outcome, and every call is recorded with the fallback |
| 189 | // it was handed. The default is a clean replacement onto the attempt's build. |
| 190 | func (f *fakeAct) ReplaceWorker(_ context.Context, a activator.Attempt, fallback *release.StagedIdentity, newest func(string) uint64) activator.SwapResult { |
| 191 | f.mu.Lock() |
| 192 | f.replaced = append(f.replaced, replaceCall{a: a, fallback: fallback}) |
| 193 | decide := f.worker |
| 194 | f.mu.Unlock() |
| 195 | if newest(a.Host) != a.SchedRev { |
| 196 | return activator.SwapResult{Outcome: activator.SwapAbandoned, Reason: "superseded"} |
| 197 | } |
| 198 | res := activator.SwapResult{Outcome: activator.SwapDone} |
| 199 | if decide != nil { |
| 200 | res = decide(a, fallback) |
| 201 | } |
| 202 | f.mu.Lock() |
| 203 | defer f.mu.Unlock() |
| 204 | switch { |
| 205 | case res.Outcome == activator.SwapDone && res.CleanExit: |
| 206 | // A one-shot's child exited: the result names its generation, but |
| 207 | // the activator serves nothing for the host afterwards. |
| 208 | f.n++ |
| 209 | res.Gen = activator.Gen{N: f.n, Build: a.Build, StagedPath: a.StagedPath, Digest: a.Digest, Demand: a.Demand} |
| 210 | delete(f.gens, a.Host) |
| 211 | case res.Outcome == activator.SwapDone: |
| 212 | f.n++ |
| 213 | res.Gen = activator.Gen{N: f.n, Build: a.Build, StagedPath: a.StagedPath, Digest: a.Digest, Demand: a.Demand} |
| 214 | f.gens[a.Host] = res.Gen |
| 215 | case res.Outcome == activator.SwapFailed && fallback != nil: |
| 216 | f.n++ |
| 217 | res.Gen = activator.Gen{N: f.n, Build: fallback.Version, StagedPath: fallback.StagedPath, Digest: fallback.Digest, Demand: fallback.Demand} |
| 218 | f.gens[a.Host] = res.Gen |
| 219 | case res.Outcome == activator.SwapDown: |
| 220 | delete(f.gens, a.Host) |
| 221 | } |
| 222 | return res |
| 223 | } |
| 224 | |
| 225 | func (f *fakeAct) replacements() []replaceCall { |
| 226 | f.mu.Lock() |
| 227 | defer f.mu.Unlock() |
| 228 | return append([]replaceCall(nil), f.replaced...) |
| 229 | } |
| 230 | |
| 231 | func (f *fakeAct) snapshot() ([]swapCall, []string, int) { |
| 232 | f.mu.Lock() |
| 233 | defer f.mu.Unlock() |
| 234 | return append([]swapCall(nil), f.calls...), append([]string(nil), f.started...), f.maxFlight |
| 235 | } |
| 236 | |
| 237 | type fakeRoutes struct { |
| 238 | mu sync.Mutex |
| 239 | rows map[string]registry.Route |
| 240 | // getHook, when set, runs at the start of every Get, with no lock held: |
| 241 | // a test's window into the moment the scheduler reads a row. getErr, |
| 242 | // when set and answering an error, fails that Get. |
| 243 | getHook func(host string) |
| 244 | getErr func(host string) error |
| 245 | // setGoodHook, when set, runs at the start of every SetGoodBuild, with |
| 246 | // no lock held. setServingHook does the same for SetServingVersion. |
| 247 | setGoodHook func(host string) |
| 248 | setServingHook func(host string) |
| 249 | } |
| 250 | |
| 251 | func (f *fakeRoutes) Get(host string) (*registry.Route, error) { |
| 252 | f.mu.Lock() |
| 253 | hook, fail := f.getHook, f.getErr |
| 254 | f.mu.Unlock() |
| 255 | if hook != nil { |
| 256 | hook(host) |
| 257 | } |
| 258 | if fail != nil { |
| 259 | if err := fail(host); err != nil { |
| 260 | return nil, err |
| 261 | } |
| 262 | } |
| 263 | f.mu.Lock() |
| 264 | defer f.mu.Unlock() |
| 265 | r, ok := f.rows[host] |
| 266 | if !ok { |
| 267 | return nil, nil |
| 268 | } |
| 269 | return &r, nil |
| 270 | } |
| 271 | |
| 272 | func (f *fakeRoutes) List() ([]registry.Route, error) { |
| 273 | f.mu.Lock() |
| 274 | defer f.mu.Unlock() |
| 275 | out := make([]registry.Route, 0, len(f.rows)) |
| 276 | for _, r := range f.rows { |
| 277 | out = append(out, r) |
| 278 | } |
| 279 | sort.Slice(out, func(i, j int) bool { return out[i].Host < out[j].Host }) |
| 280 | return out, nil |
| 281 | } |
| 282 | |
| 283 | func (f *fakeRoutes) SetGoodBuild(host string, expect, next registry.GoodBuild) error { |
| 284 | f.mu.Lock() |
| 285 | hook := f.setGoodHook |
| 286 | f.mu.Unlock() |
| 287 | if hook != nil { |
| 288 | hook(host) |
| 289 | } |
| 290 | f.mu.Lock() |
| 291 | defer f.mu.Unlock() |
| 292 | r := f.rows[host] |
| 293 | if r.Good() != expect { |
| 294 | return registry.ErrGoodBuildChanged |
| 295 | } |
| 296 | r.GoodVersion, r.GoodDigest, r.GoodDemand = next.Version, next.Digest, next.Demand |
| 297 | f.rows[host] = r |
| 298 | return nil |
| 299 | } |
| 300 | |
| 301 | func (f *fakeRoutes) SetServingVersion(host, ver string) error { |
| 302 | f.mu.Lock() |
| 303 | hook := f.setServingHook |
| 304 | f.mu.Unlock() |
| 305 | if hook != nil { |
| 306 | hook(host) |
| 307 | } |
| 308 | f.mu.Lock() |
| 309 | defer f.mu.Unlock() |
| 310 | r := f.rows[host] |
| 311 | r.ServingVersion = ver |
| 312 | f.rows[host] = r |
| 313 | return nil |
| 314 | } |
| 315 | |
| 316 | func (f *fakeRoutes) put(r registry.Route) { |
| 317 | f.mu.Lock() |
| 318 | defer f.mu.Unlock() |
| 319 | f.rows[r.Host] = r |
| 320 | } |
| 321 | |
| 322 | func (f *fakeRoutes) get(host string) registry.Route { |
| 323 | f.mu.Lock() |
| 324 | defer f.mu.Unlock() |
| 325 | return f.rows[host] |
| 326 | } |
| 327 | |
| 328 | type eventLog struct { |
| 329 | mu sync.Mutex |
| 330 | lines []string |
| 331 | } |
| 332 | |
| 333 | func (e *eventLog) App(string, string, string, string) {} |
| 334 | |
| 335 | func (e *eventLog) Event(_, _, host, _, msg string) { |
| 336 | e.mu.Lock() |
| 337 | defer e.mu.Unlock() |
| 338 | e.lines = append(e.lines, host+": "+msg) |
| 339 | } |
| 340 | |
| 341 | func (e *eventLog) has(substr string) bool { |
| 342 | e.mu.Lock() |
| 343 | defer e.mu.Unlock() |
| 344 | for _, l := range e.lines { |
| 345 | if strings.Contains(l, substr) { |
| 346 | return true |
| 347 | } |
| 348 | } |
| 349 | return false |
| 350 | } |
| 351 | |
| 352 | // count is how many logged lines contain substr. |
| 353 | func (e *eventLog) count(substr string) int { |
| 354 | e.mu.Lock() |
| 355 | defer e.mu.Unlock() |
| 356 | n := 0 |
| 357 | for _, l := range e.lines { |
| 358 | if strings.Contains(l, substr) { |
| 359 | n++ |
| 360 | } |
| 361 | } |
| 362 | return n |
| 363 | } |
| 364 | |
| 365 | // harness is a manager over the fakes, for app "blog" of account "bab", |
| 366 | // every host awake on v1, its good build. |
| 367 | type harness struct { |
| 368 | t *testing.T |
| 369 | act *fakeAct |
| 370 | probe *probeAct // the manager's Activator: act behind a test hook |
| 371 | rows *fakeRoutes |
| 372 | book *FailureBook |
| 373 | m *Manager |
| 374 | opt, data string |
| 375 | mu sync.Mutex |
| 376 | now time.Time |
| 377 | concurrency int |
| 378 | swap string |
| 379 | demands map[string]string |
| 380 | mismatch map[string]bool |
| 381 | demandErr map[string]error |
| 382 | noDigest map[string]bool |
| 383 | pointers map[string]release.ChannelPointer // channel -> its signed pointer |
| 384 | failing atomic.Bool |
| 385 | touched atomic.Int32 |
| 386 | events *eventLog |
| 387 | // pointerHook, when set, runs at the start of every pointer read. |
| 388 | pointerHook func() |
| 389 | // demandHook, when set, runs at the start of every demand read, with no |
| 390 | // lock held: a test's window into the planner's slow store read. |
| 391 | demandHook func(ver string) |
| 392 | // policyHook, when set, runs at the start of every policy load, with no |
| 393 | // lock held: a test's window into the planner's policy read. |
| 394 | policyHook func() |
| 395 | } |
| 396 | |
| 397 | func newHarness(t *testing.T, hosts ...string) *harness { |
| 398 | t.Helper() |
| 399 | h := &harness{t: t, act: newFakeAct(), rows: &fakeRoutes{rows: map[string]registry.Route{}}, book: &FailureBook{}, |
| 400 | opt: t.TempDir(), data: t.TempDir(), now: t0, concurrency: 1, |
| 401 | demands: map[string]string{}, mismatch: map[string]bool{}, demandErr: map[string]error{}, noDigest: map[string]bool{}, |
| 402 | pointers: map[string]release.ChannelPointer{}, |
| 403 | events: &eventLog{}} |
| 404 | h.book.now = h.clock // a marker's At follows the harness clock |
| 405 | h.probe = &probeAct{fakeAct: h.act} |
| 406 | h.stage("v1") |
| 407 | for _, host := range hosts { |
| 408 | h.rows.put(registry.Route{Host: host, Kind: "instance", App: "blog", Account: "bab", Status: "ready", |
| 409 | Hibernate: true, Version: "v1", GoodVersion: "v1", GoodDigest: "sha-v1"}) |
| 410 | h.act.serve(host, "v1") |
| 411 | } |
| 412 | ctx, cancel := context.WithCancel(context.Background()) |
| 413 | t.Cleanup(cancel) |
| 414 | h.m = NewManager(ctx, Deps{ |
| 415 | Act: h.probe, Routes: h.rows, Book: h.book, OptDir: h.opt, DataDir: h.data, Box: "box-1", |
| 416 | Demand: h.demand, Policy: h.policy, Pointer: h.pointerFor, Events: h.events, |
| 417 | Touch: func() { h.touched.Add(1) }, Now: h.clock, Logf: func(string, ...any) {}, |
| 418 | }) |
| 419 | return h |
| 420 | } |
| 421 | |
| 422 | // pointer signs channel's pointer onto ver at promoted, as `carlos promote` |
| 423 | // (with or without --retry) would. |
| 424 | func (h *harness) pointer(channel, ver string, promoted time.Time) { |
| 425 | h.mu.Lock() |
| 426 | defer h.mu.Unlock() |
| 427 | h.pointers[channel] = release.ChannelPointer{Schema: 1, Account: "bab", App: "blog", Channel: channel, Version: ver, PromotedAt: promoted} |
| 428 | } |
| 429 | |
| 430 | func (h *harness) pointerFor(_ context.Context, _, _, channel string) (*release.ChannelPointer, error) { |
| 431 | h.mu.Lock() |
| 432 | hook := h.pointerHook |
| 433 | h.mu.Unlock() |
| 434 | if hook != nil { |
| 435 | hook() |
| 436 | } |
| 437 | h.mu.Lock() |
| 438 | defer h.mu.Unlock() |
| 439 | p, ok := h.pointers[channel] |
| 440 | if !ok { |
| 441 | return nil, nil |
| 442 | } |
| 443 | return &p, nil |
| 444 | } |
| 445 | |
| 446 | func (h *harness) clock() time.Time { |
| 447 | h.mu.Lock() |
| 448 | defer h.mu.Unlock() |
| 449 | return h.now |
| 450 | } |
| 451 | |
| 452 | func (h *harness) advance(d time.Duration) { |
| 453 | h.mu.Lock() |
| 454 | defer h.mu.Unlock() |
| 455 | h.now = h.now.Add(d) |
| 456 | } |
| 457 | |
| 458 | func (h *harness) stage(vers ...string) { |
| 459 | for _, v := range vers { |
| 460 | if err := release.StageArtifact(h.opt, "bab", v, "blog", []byte(v)); err != nil { |
| 461 | h.t.Fatal(err) |
| 462 | } |
| 463 | } |
| 464 | } |
| 465 | |
| 466 | // demand is release.CanonicalDemand's contract: a mismatch wraps |
| 467 | // release.ErrStagedDigestMismatch (a staging refusal); demandErr is any other |
| 468 | // error, and noDigest is a store that could not compare, both of which must |
| 469 | // end as a hold attempt. |
| 470 | func (h *harness) demand(_ context.Context, _, _, ver, _ string) (string, string, error) { |
| 471 | h.mu.Lock() |
| 472 | hook := h.demandHook |
| 473 | h.mu.Unlock() |
| 474 | if hook != nil { |
| 475 | hook(ver) |
| 476 | } |
| 477 | h.mu.Lock() |
| 478 | defer h.mu.Unlock() |
| 479 | if h.mismatch[ver] { |
| 480 | return "", "", fmt.Errorf("%w: the staged executable's digest differs from the canonical manifest", release.ErrStagedDigestMismatch) |
| 481 | } |
| 482 | if err := h.demandErr[ver]; err != nil { |
| 483 | return "", "", err |
| 484 | } |
| 485 | if h.noDigest[ver] { |
| 486 | return "", "", nil |
| 487 | } |
| 488 | return h.demands[ver], "sha-" + ver, nil |
| 489 | } |
| 490 | |
| 491 | func (h *harness) policy(context.Context, string, string) (*appconfig.Manifest, error) { |
| 492 | h.mu.Lock() |
| 493 | hook := h.policyHook |
| 494 | h.mu.Unlock() |
| 495 | if hook != nil { |
| 496 | hook() |
| 497 | } |
| 498 | h.mu.Lock() |
| 499 | defer h.mu.Unlock() |
| 500 | return &appconfig.Manifest{Schema: 1, App: "blog", Swap: h.swap, SwapConcurrency: h.concurrency}, nil |
| 501 | } |
| 502 | |
| 503 | func (h *harness) install(ver string, hosts ...string) { |
| 504 | for _, host := range hosts { |
| 505 | r := h.rows.get(host) |
| 506 | r.Version = ver |
| 507 | h.rows.put(r) |
| 508 | } |
| 509 | } |
| 510 | |
| 511 | func (h *harness) pass() { |
| 512 | rs, _ := h.rows.List() |
| 513 | h.m.Pass(rs) |
| 514 | } |
| 515 | |
| 516 | func (h *harness) servingAll(build string, hosts ...string) bool { |
| 517 | for _, host := range hosts { |
| 518 | if g, ok := h.act.Current(host); !ok || g.Build != build { |
| 519 | return false |
| 520 | } |
| 521 | } |
| 522 | return true |
| 523 | } |
| 524 | |
| 525 | // flushAcks waits until the manager's ack writer has written every ack |
| 526 | // handed to it so far. Acks are written off the path that releases them |
| 527 | // (Task 11), so a test reads what was written only after this. |
| 528 | func (h *harness) flushAcks() { |
| 529 | h.t.Helper() |
| 530 | waitFor(h.t, "the ack writer to finish", h.m.acks.idle) |
| 531 | } |
| 532 | |
| 533 | func waitFor(t *testing.T, what string, cond func() bool) { |
| 534 | t.Helper() |
| 535 | deadline := time.Now().Add(5 * time.Second) |
| 536 | for time.Now().Before(deadline) { |
| 537 | if cond() { |
| 538 | return |
| 539 | } |
| 540 | time.Sleep(5 * time.Millisecond) |
| 541 | } |
| 542 | t.Fatalf("timed out waiting for %s", what) |
| 543 | } |
| 544 | |
amadan