| 1 | package host |
| 2 | |
| 3 | import ( |
| 4 | "context" |
| 5 | "errors" |
| 6 | "os" |
| 7 | "path/filepath" |
| 8 | "slices" |
| 9 | "sync" |
| 10 | "testing" |
| 11 | "time" |
| 12 | |
| 13 | "github.com/carlosframework/platform/internal/activator" |
| 14 | "github.com/carlosframework/platform/internal/hostbox" |
| 15 | "github.com/carlosframework/platform/internal/hostclaim" |
| 16 | "github.com/carlosframework/platform/internal/hostguard" |
| 17 | "github.com/carlosframework/platform/internal/stopevidence" |
| 18 | ) |
| 19 | |
| 20 | type writersJournalFake struct { |
| 21 | mu sync.Mutex |
| 22 | inv hostbox.LaunchInventory |
| 23 | settled []string |
| 24 | reads int |
| 25 | err error |
| 26 | } |
| 27 | |
| 28 | func (j *writersJournalFake) Read(_ context.Context, t hostguard.FenceTarget) (hostbox.LaunchInventory, error) { |
| 29 | j.mu.Lock() |
| 30 | defer j.mu.Unlock() |
| 31 | j.reads++ |
| 32 | if j.err != nil { |
| 33 | return hostbox.LaunchInventory{}, j.err |
| 34 | } |
| 35 | if !hostbox.SameLaunchOwner(j.inv.Target, t) { |
| 36 | return hostbox.LaunchInventory{}, hostbox.ErrChanged |
| 37 | } |
| 38 | out := j.inv |
| 39 | out.Reservations = slices.Clone(j.inv.Reservations) |
| 40 | return out, nil |
| 41 | } |
| 42 | |
| 43 | func (j *writersJournalFake) Settled(_ context.Context, r hostbox.LaunchReservation) error { |
| 44 | j.mu.Lock() |
| 45 | defer j.mu.Unlock() |
| 46 | for i := range j.inv.Reservations { |
| 47 | if j.inv.Reservations[i].ID == r.ID { |
| 48 | j.inv.Reservations[i].State = "settled" |
| 49 | j.settled = append(j.settled, r.ID) |
| 50 | return nil |
| 51 | } |
| 52 | } |
| 53 | return hostbox.ErrChanged |
| 54 | } |
| 55 | |
| 56 | type writersCgroupFake struct { |
| 57 | mu sync.Mutex |
| 58 | populated bool |
| 59 | kills int |
| 60 | killErr error |
| 61 | } |
| 62 | |
| 63 | func (c *writersCgroupFake) Populated() (bool, error) { |
| 64 | c.mu.Lock() |
| 65 | defer c.mu.Unlock() |
| 66 | return c.populated, nil |
| 67 | } |
| 68 | |
| 69 | func (c *writersCgroupFake) KillAndEmpty(context.Context) error { |
| 70 | c.mu.Lock() |
| 71 | defer c.mu.Unlock() |
| 72 | c.kills++ |
| 73 | if c.killErr != nil { |
| 74 | return c.killErr |
| 75 | } |
| 76 | c.populated = false |
| 77 | return nil |
| 78 | } |
| 79 | |
| 80 | // producerProbe is one registered producer: done closes when it finished. |
| 81 | type producerProbe struct { |
| 82 | ctx context.Context |
| 83 | done chan struct{} |
| 84 | } |
| 85 | |
| 86 | // runProducer registers id and runs it until its context ends, as a role |
| 87 | // launch, a timer delivery or the replicator does. |
| 88 | func runProducer(t *testing.T, p *ClaimProducers, target hostguard.FenceTarget, id string) producerProbe { |
| 89 | t.Helper() |
| 90 | ctx, finish, err := p.Register(context.Background(), target, id) |
| 91 | if err != nil { |
| 92 | t.Fatalf("register %s: %v", id, err) |
| 93 | } |
| 94 | done := make(chan struct{}) |
| 95 | go func() { |
| 96 | <-ctx.Done() |
| 97 | finish(nil) |
| 98 | close(done) |
| 99 | }() |
| 100 | return producerProbe{ctx: ctx, done: done} |
| 101 | } |
| 102 | |
| 103 | type writersFixture struct { |
| 104 | wake *ClaimWakeAssembly |
| 105 | attempt *claimWakeAttempt |
| 106 | target hostguard.FenceTarget |
| 107 | journal *writersJournalFake |
| 108 | cgroup *writersCgroupFake |
| 109 | main producerProbe |
| 110 | timer producerProbe |
| 111 | rep producerProbe |
| 112 | pub producerProbe |
| 113 | producer *ClaimProducers |
| 114 | } |
| 115 | |
| 116 | func newWritersFixture(t *testing.T) *writersFixture { |
| 117 | t.Helper() |
| 118 | target := hostguard.FenceTarget{Host: "shop.example", Incarnation: "inc-1", Stamp: 11, Session: 13, Box: "box-1", Enrollment: "enrollment-1", BootID: "boot-1", ControllerID: "controller-1", DirectoryRevision: "7"} |
| 119 | claim := &hostclaim.Claim{V: 2, Host: target.Host, Gen: 4, |
| 120 | Holder: &hostclaim.Holder{Incarnation: "inc-1", Stamp: 11, Session: 13, WriterBox: "box-1", Generation: "g-13", Location: "loc"}, |
| 121 | Op: hostclaim.Op{Kind: hostclaim.KindServing}} |
| 122 | observed := hostclaim.Observed{Host: target.Host, Claim: claim, ETag: "etag-1"} |
| 123 | producers, err := NewClaimProducers(target, "start-1", func(context.Context, hostguard.FenceTarget, string) error { return nil }) |
| 124 | if err != nil { |
| 125 | t.Fatal(err) |
| 126 | } |
| 127 | journal := &writersJournalFake{inv: hostbox.LaunchInventory{Target: target, Reservations: []hostbox.LaunchReservation{ |
| 128 | {ID: "attempt-1:main", Role: "main", State: "started", Target: target, PID: 77}, |
| 129 | {ID: "attempt-0:main", Role: "main", State: "settled", Target: target}, |
| 130 | }}} |
| 131 | cg := &writersCgroupFake{populated: true} |
| 132 | wake := &ClaimWakeAssembly{Producers: func(context.Context, hostclaim.Observed) (*ClaimProducers, error) { return producers, nil }} |
| 133 | live, cancel := context.WithCancel(context.Background()) |
| 134 | t.Cleanup(cancel) |
| 135 | a := &claimWakeAttempt{owner: wake, observed: observed, inputs: ClaimLaunchInputs{DBPath: "/var/lib/carlos/shop.example/data.db"}, |
| 136 | prelaunch: &activator.RoleLaunch{Target: target, ParentPath: "carlos/claim-1.local"}, writersJournal: journal, writersCgroup: cg, |
| 137 | ctx: live, cancel: cancel, done: make(chan struct{})} |
| 138 | f := &writersFixture{wake: wake, attempt: a, target: target, journal: journal, cgroup: cg, producer: producers} |
| 139 | f.main = runProducer(t, producers, target, "main") |
| 140 | f.timer = runProducer(t, producers, target, "timer-1") |
| 141 | f.rep = runProducer(t, producers, target, ClaimProducerReplicator) |
| 142 | f.pub = runProducer(t, producers, target, ClaimProducerPublisher) |
| 143 | return f |
| 144 | } |
| 145 | |
| 146 | func closed(c <-chan struct{}) bool { |
| 147 | select { |
| 148 | case <-c: |
| 149 | return true |
| 150 | default: |
| 151 | return false |
| 152 | } |
| 153 | } |
| 154 | |
| 155 | func TestStopWritersDrainsWritersAndKeepsTheReplicator(t *testing.T) { |
| 156 | f := newWritersFixture(t) |
| 157 | drained, err := f.attempt.StopWriters(t.Context()) |
| 158 | if err != nil { |
| 159 | t.Fatal(err) |
| 160 | } |
| 161 | if !closed(f.main.done) || !closed(f.timer.done) { |
| 162 | t.Fatal("the role launch or the timer delivery was not joined") |
| 163 | } |
| 164 | if f.rep.ctx.Err() != nil || f.pub.ctx.Err() != nil || f.attempt.ctx.Err() != nil { |
| 165 | t.Fatal("the replicator, the publisher or the attempt was cancelled") |
| 166 | } |
| 167 | if !drained.FenceClosed || !drained.SpawnsSettled || !drained.WorkDrained || !drained.LaunchJoined || !drained.CgroupEmpty { |
| 168 | t.Fatalf("facts = %+v", drained) |
| 169 | } |
| 170 | if drained.Target != f.target || drained.Generation != "g-13" || drained.Cgroup != "carlos/claim-1.local" || !slices.Equal(drained.DBPaths, []string{"/var/lib/carlos/shop.example/data.db"}) { |
| 171 | t.Fatalf("witness = %+v", drained) |
| 172 | } |
| 173 | if !slices.Equal(drained.Drained, []string{"main", "timer-1"}) || !slices.Equal(drained.Retained, []string{ClaimProducerPublisher, ClaimProducerReplicator}) { |
| 174 | t.Fatalf("drained %v retained %v", drained.Drained, drained.Retained) |
| 175 | } |
| 176 | if !slices.Equal(f.journal.settled, []string{"attempt-1:main"}) || f.cgroup.kills != 1 { |
| 177 | t.Fatalf("settled %v kills %d", f.journal.settled, f.cgroup.kills) |
| 178 | } |
| 179 | if err := drained.For(f.attempt.observed); err != nil { |
| 180 | t.Fatalf("witness does not bind its own session: %v", err) |
| 181 | } |
| 182 | } |
| 183 | |
| 184 | func TestStopWritersClosesAdmission(t *testing.T) { |
| 185 | f := newWritersFixture(t) |
| 186 | if _, err := f.attempt.StopWriters(t.Context()); err != nil { |
| 187 | t.Fatal(err) |
| 188 | } |
| 189 | for _, id := range []string{"main-2", "timer-2"} { |
| 190 | if _, _, err := f.producer.Register(context.Background(), f.target, id); !errors.Is(err, ErrClaimWritersStopped) { |
| 191 | t.Fatalf("%s registered after the writers stopped: %v", id, err) |
| 192 | } |
| 193 | if err := f.producer.Launch(context.Background(), f.target, id, func(context.Context) error { return nil }); !errors.Is(err, ErrClaimWritersStopped) { |
| 194 | t.Fatalf("%s launched after the writers stopped: %v", id, err) |
| 195 | } |
| 196 | } |
| 197 | // The files publisher is retained: it may still register (L13). |
| 198 | if _, finish, err := f.producer.Register(context.Background(), f.target, ClaimProducerFiles); err != nil { |
| 199 | t.Fatalf("files publisher refused: %v", err) |
| 200 | } else { |
| 201 | finish(nil) |
| 202 | } |
| 203 | if err := f.attempt.admitsWriters(); !errors.Is(err, ErrClaimWritersStopped) { |
| 204 | t.Fatalf("the attempt still admits writers: %v", err) |
| 205 | } |
| 206 | } |
| 207 | |
| 208 | func TestStopWritersIsIdempotent(t *testing.T) { |
| 209 | f := newWritersFixture(t) |
| 210 | first, err := f.attempt.StopWriters(t.Context()) |
| 211 | if err != nil { |
| 212 | t.Fatal(err) |
| 213 | } |
| 214 | reads, kills := f.journal.reads, f.cgroup.kills |
| 215 | second, err := f.attempt.StopWriters(t.Context()) |
| 216 | if err != nil { |
| 217 | t.Fatal(err) |
| 218 | } |
| 219 | if !slices.Equal(first.Drained, second.Drained) || first.Target != second.Target || first.CheckedAt != second.CheckedAt { |
| 220 | t.Fatalf("second = %+v first = %+v", second, first) |
| 221 | } |
| 222 | if f.journal.reads != reads || f.cgroup.kills != kills { |
| 223 | t.Fatal("a stopped attempt repeated its effects") |
| 224 | } |
| 225 | } |
| 226 | |
| 227 | func TestStopWritersRetriesAfterAFailure(t *testing.T) { |
| 228 | f := newWritersFixture(t) |
| 229 | f.cgroup.killErr = errors.New("cgroup busy") |
| 230 | if _, err := f.attempt.StopWriters(t.Context()); err == nil { |
| 231 | t.Fatal("a cgroup that would not empty passed") |
| 232 | } |
| 233 | if !closed(f.main.done) { |
| 234 | t.Fatal("the role was not joined before the cgroup step") |
| 235 | } |
| 236 | f.cgroup.killErr = nil |
| 237 | drained, err := f.attempt.StopWriters(t.Context()) |
| 238 | if err != nil || !drained.CgroupEmpty { |
| 239 | t.Fatalf("retry = %+v %v", drained, err) |
| 240 | } |
| 241 | } |
| 242 | |
| 243 | func TestStopWritersRefuses(t *testing.T) { |
| 244 | t.Run("a pending role intention", func(t *testing.T) { |
| 245 | f := newWritersFixture(t) |
| 246 | f.journal.inv.Reservations = append(f.journal.inv.Reservations, hostbox.LaunchReservation{ID: "attempt-2:main", State: "intended", Target: f.target}) |
| 247 | if _, err := f.attempt.StopWriters(t.Context()); !errors.Is(err, ErrClaimStopInventory) { |
| 248 | t.Fatalf("got %v", err) |
| 249 | } |
| 250 | }) |
| 251 | t.Run("a populated cgroup after the kill", func(t *testing.T) { |
| 252 | f := newWritersFixture(t) |
| 253 | f.cgroup.killErr = nil |
| 254 | f.attempt.writersCgroup = stuckCgroup{} |
| 255 | if _, err := f.attempt.StopWriters(t.Context()); !errors.Is(err, ErrClaimStopInventory) { |
| 256 | t.Fatalf("got %v", err) |
| 257 | } |
| 258 | }) |
| 259 | t.Run("another session", func(t *testing.T) { |
| 260 | f := newWritersFixture(t) |
| 261 | next := *f.attempt.observed.Claim |
| 262 | holder := *next.Holder |
| 263 | holder.Session = 14 |
| 264 | next.Holder = &holder |
| 265 | f.attempt.observed.Claim = &next |
| 266 | if _, err := f.attempt.StopWriters(t.Context()); !errors.Is(err, hostbox.ErrChanged) { |
| 267 | t.Fatalf("got %v", err) |
| 268 | } |
| 269 | if closed(f.main.done) { |
| 270 | t.Fatal("a mismatched session drained the writers") |
| 271 | } |
| 272 | }) |
| 273 | t.Run("no launch record", func(t *testing.T) { |
| 274 | f := newWritersFixture(t) |
| 275 | f.attempt.prelaunch = nil |
| 276 | if _, err := f.attempt.StopWriters(t.Context()); err == nil { |
| 277 | t.Fatal("an attempt with no launch record stopped writers") |
| 278 | } |
| 279 | }) |
| 280 | t.Run("after the full drain", func(t *testing.T) { |
| 281 | f := newWritersFixture(t) |
| 282 | req := StopRequest{Host: f.target.Host, Incarnation: f.target.Incarnation, InstallationStamp: f.target.Stamp, WriterSession: f.target.Session, Box: f.target.Box, Enrollment: f.target.Enrollment, BootID: f.target.BootID, ControllerID: f.target.ControllerID} |
| 283 | if _, err := f.producer.DrainClaim(t.Context(), req, "fence-1"); err != nil { |
| 284 | t.Fatal(err) |
| 285 | } |
| 286 | if _, err := f.attempt.StopWriters(t.Context()); !errors.Is(err, ErrClaimStopIdentity) { |
| 287 | t.Fatalf("got %v", err) |
| 288 | } |
| 289 | }) |
| 290 | } |
| 291 | |
| 292 | type stuckCgroup struct{} |
| 293 | |
| 294 | func (stuckCgroup) Populated() (bool, error) { return true, nil } |
| 295 | func (stuckCgroup) KillAndEmpty(context.Context) error { return nil } |
| 296 | |
| 297 | func TestWritersWitnessBindsTheExactSession(t *testing.T) { |
| 298 | f := newWritersFixture(t) |
| 299 | drained, err := f.attempt.StopWriters(t.Context()) |
| 300 | if err != nil { |
| 301 | t.Fatal(err) |
| 302 | } |
| 303 | for name, change := range map[string]func(*hostclaim.Holder){ |
| 304 | "session": func(h *hostclaim.Holder) { h.Session++ }, |
| 305 | "stamp": func(h *hostclaim.Holder) { h.Stamp++ }, |
| 306 | "incarnation": func(h *hostclaim.Holder) { h.Incarnation = "inc-2" }, |
| 307 | "generation": func(h *hostclaim.Holder) { h.Generation = "g-14" }, |
| 308 | "writer box": func(h *hostclaim.Holder) { h.WriterBox = "box-2" }, |
| 309 | } { |
| 310 | o := f.attempt.observed |
| 311 | c := *o.Claim |
| 312 | h := *c.Holder |
| 313 | change(&h) |
| 314 | c.Holder = &h |
| 315 | o.Claim = &c |
| 316 | if err := drained.For(o); !errors.Is(err, hostbox.ErrChanged) { |
| 317 | t.Fatalf("%s: %v", name, err) |
| 318 | } |
| 319 | } |
| 320 | if err := (ClaimWritersDrained{}).For(f.attempt.observed); !errors.Is(err, hostbox.ErrChanged) { |
| 321 | t.Fatalf("a zero witness passed: %v", err) |
| 322 | } |
| 323 | } |
| 324 | |
| 325 | func TestOwnedReplicatorOutlivesTheActivatorCloseOnceWritersStop(t *testing.T) { |
| 326 | f := newWritersFixture(t) |
| 327 | if _, err := f.attempt.StopWriters(t.Context()); err != nil { |
| 328 | t.Fatal(err) |
| 329 | } |
| 330 | rep := &claimOwnedRep{attempt: f.attempt, owner: f.wake, host: f.target.Host} |
| 331 | ctx, cancel := context.WithTimeout(t.Context(), time.Second) |
| 332 | defer cancel() |
| 333 | if err := rep.Close(ctx); err != nil { |
| 334 | t.Fatal(err) |
| 335 | } |
| 336 | if f.attempt.ctx.Err() != nil || f.rep.ctx.Err() != nil { |
| 337 | t.Fatal("the activator's close released a writers-stopped attempt") |
| 338 | } |
| 339 | } |
| 340 | |
| 341 | // R3's agent half: the replicator's descriptors are the difference between |
| 342 | // this process's own table before and after it opens. |
| 343 | func TestRecordReplicatorHintsKeepsOnlyTheNewDescriptors(t *testing.T) { |
| 344 | dir := t.TempDir() |
| 345 | db := filepath.Join(dir, "data.db") |
| 346 | if err := os.WriteFile(db, nil, 0o600); err != nil { |
| 347 | t.Fatal(err) |
| 348 | } |
| 349 | earlier, err := os.Open(db) |
| 350 | if err != nil { |
| 351 | t.Fatal(err) |
| 352 | } |
| 353 | defer earlier.Close() |
| 354 | var opened *os.File |
| 355 | defer func() { |
| 356 | if opened != nil { |
| 357 | opened.Close() |
| 358 | } |
| 359 | }() |
| 360 | hints, err := recordReplicatorHints(t.Context(), "/proc", os.Getpid(), db, func() error { |
| 361 | var err error |
| 362 | opened, err = os.Open(db) |
| 363 | return err |
| 364 | }) |
| 365 | if err != nil { |
| 366 | t.Fatal(err) |
| 367 | } |
| 368 | if len(hints) != 1 || hints[0].FD != int(opened.Fd()) { |
| 369 | t.Fatalf("hints = %+v, want only fd %d", hints, opened.Fd()) |
| 370 | } |
| 371 | failed := errors.New("open failed") |
| 372 | if _, err := recordReplicatorHints(t.Context(), "/proc", os.Getpid(), db, func() error { return failed }); !errors.Is(err, failed) { |
| 373 | t.Fatalf("got %v", err) |
| 374 | } |
| 375 | } |
| 376 | |
| 377 | // The agent's whole step: StopWriters, then root's observation of the |
| 378 | // witnessed session at the current claim, carrying the recorded hint. |
| 379 | func TestClaimAssemblyWritersStoppedAsksRootForTheWitnessedSession(t *testing.T) { |
| 380 | f := newWritersFixture(t) |
| 381 | f.attempt.replicatorFDs = []stopevidence.DescriptorHint{{FD: 7, Device: 1, Inode: 2}} |
| 382 | f.wake.pending = map[string]*claimWakeAttempt{f.target.Host: f.attempt} |
| 383 | current := f.attempt.observed |
| 384 | a := ClaimAssembly{Wake: f.wake, stopLocal: func(context.Context) (claimStopLocal, error) { |
| 385 | return claimStopLocal{Actor: hostbox.Actor{Box: "box-1", Enrollment: "enrollment-1", Region: "eu-west-1", Revision: 9, Binding: "9"}, BootID: "boot-1", ControllerID: "controller-1", PID: 42, ProcessStart: "start-42"}, nil |
| 386 | }} |
| 387 | read := func(context.Context, string) (hostclaim.Observed, error) { return current, nil } |
| 388 | var got WritersStoppedRequest |
| 389 | collect := func(_ context.Context, r WritersStoppedRequest) (WritersStoppedEvidence, error) { |
| 390 | got = r |
| 391 | return WritersStoppedEvidence{Host: r.Host, WriterSession: r.WriterSession}, nil |
| 392 | } |
| 393 | drained, ev, err := a.writersStopped(t.Context(), f.target.Host, read, collect) |
| 394 | if err != nil { |
| 395 | t.Fatal(err) |
| 396 | } |
| 397 | if !drained.Valid() || ev.Host != f.target.Host || f.rep.ctx.Err() != nil { |
| 398 | t.Fatalf("drained %+v ev %+v", drained, ev) |
| 399 | } |
| 400 | if got.WriterSession != 13 || got.ExpectedClaimETag != "etag-1" || got.ExpectedDirectoryRevision != "7" || got.Region != "eu-west-1" || got.BootID != "boot-1" || len(got.Replicator) != 1 { |
| 401 | t.Fatalf("request = %+v", got) |
| 402 | } |
| 403 | t.Run("a later session", func(t *testing.T) { |
| 404 | c := *current.Claim |
| 405 | h := *c.Holder |
| 406 | h.Session = 14 |
| 407 | c.Holder = &h |
| 408 | current.Claim = &c |
| 409 | if _, _, err := a.writersStopped(t.Context(), f.target.Host, read, collect); !errors.Is(err, hostbox.ErrChanged) { |
| 410 | t.Fatalf("got %v", err) |
| 411 | } |
| 412 | }) |
| 413 | t.Run("another controller", func(t *testing.T) { |
| 414 | current = f.attempt.observed |
| 415 | b := a |
| 416 | b.stopLocal = func(context.Context) (claimStopLocal, error) { |
| 417 | return claimStopLocal{Actor: hostbox.Actor{Box: "box-1", Enrollment: "enrollment-1"}, BootID: "boot-1", ControllerID: "controller-2"}, nil |
| 418 | } |
| 419 | if _, _, err := b.writersStopped(t.Context(), f.target.Host, read, collect); !errors.Is(err, hostbox.ErrChanged) { |
| 420 | t.Fatalf("got %v", err) |
| 421 | } |
| 422 | }) |
| 423 | t.Run("no attempt", func(t *testing.T) { |
| 424 | if _, _, err := a.writersStopped(t.Context(), "other.example", read, collect); !errors.Is(err, ErrClaimStopUnavailable) { |
| 425 | t.Fatalf("got %v", err) |
| 426 | } |
| 427 | }) |
| 428 | } |
| 429 | |
| 430 | func (j *writersJournalFake) CloseWriters(_ context.Context, target hostguard.FenceTarget) error { |
| 431 | j.mu.Lock() |
| 432 | defer j.mu.Unlock() |
| 433 | if j.err != nil { |
| 434 | return j.err |
| 435 | } |
| 436 | if !hostbox.SameLaunchOwner(j.inv.Target, target) { |
| 437 | return hostbox.ErrChanged |
| 438 | } |
| 439 | j.inv.WritersClosed = true |
| 440 | return nil |
| 441 | } |
| 442 | |
amadan