| 1 | package host |
| 2 | |
| 3 | import ( |
| 4 | "context" |
| 5 | "database/sql" |
| 6 | "encoding/json" |
| 7 | "errors" |
| 8 | "fmt" |
| 9 | "io" |
| 10 | "io/fs" |
| 11 | "log/slog" |
| 12 | "os" |
| 13 | "path/filepath" |
| 14 | "slices" |
| 15 | "sort" |
| 16 | "strings" |
| 17 | "sync" |
| 18 | "testing" |
| 19 | "time" |
| 20 | |
| 21 | "github.com/benbjohnson/litestream" |
| 22 | lsfile "github.com/benbjohnson/litestream/file" |
| 23 | "github.com/carlosframework/platform/internal/activator" |
| 24 | "github.com/carlosframework/platform/internal/appconfig" |
| 25 | "github.com/carlosframework/platform/internal/dbverify" |
| 26 | "github.com/carlosframework/platform/internal/hostbox" |
| 27 | "github.com/carlosframework/platform/internal/hostclaim" |
| 28 | "github.com/carlosframework/platform/internal/hostfiles" |
| 29 | "github.com/carlosframework/platform/internal/hostguard" |
| 30 | "github.com/carlosframework/platform/internal/hostwrite" |
| 31 | "github.com/carlosframework/platform/internal/lease" |
| 32 | "github.com/carlosframework/platform/internal/moves" |
| 33 | "github.com/carlosframework/platform/internal/moves/archive" |
| 34 | "github.com/carlosframework/platform/internal/park" |
| 35 | "github.com/carlosframework/platform/internal/provision" |
| 36 | "github.com/carlosframework/platform/internal/registry" |
| 37 | "github.com/carlosframework/platform/internal/release" |
| 38 | "github.com/superfly/ltx" |
| 39 | ) |
| 40 | |
| 41 | // The landing harness: the cut harness with a REAL litestream replica in |
| 42 | // the source's lineage (so the target's build restores a real SQLite file |
| 43 | // that passes its check), driven to the flip, and a target box (uk-box) |
| 44 | // with its own data directory, the held row reconcile would have minted, |
| 45 | // the landing's hooks faked where main has no production provider yet |
| 46 | // (ruling G17) and real where it has one: park.Local and hostguard.Guard |
| 47 | // over the target's data directory, dbverify over the store, the hostbox |
| 48 | // marker writer, moves.BuildGeneration over the fake versioned bucket. |
| 49 | // |
| 50 | // fakeParkLeases is one lease table for the landing's own-location lease |
| 51 | // and the park engine's: a second acquisition of a held location is |
| 52 | // refused. The landing holds its lease from checking through published |
| 53 | // (spec §7.6), so its residue parks run under it through the claim's |
| 54 | // AggregateLeases and park.Engine.RunHeld (M5); an Engine.Run that took the |
| 55 | // lock again would fail the test. |
| 56 | |
| 57 | type landingHarness struct { |
| 58 | *cutHarness |
| 59 | // carry is the target's moveCfg.CarryRequest (nil: none), for the return |
| 60 | // lane's request carry (buildstate_carry_test.go). |
| 61 | carry func(ctx context.Context, host, bucket, prefix string) error |
| 62 | tdir string // the target's data directory |
| 63 | leases *fakeParkLeases |
| 64 | lw *movesWorker |
| 65 | lcfg moveCfg |
| 66 | route registry.Route |
| 67 | rows int // rows the source database holds |
| 68 | n uint64 |
| 69 | replaced [][]registry.Timer |
| 70 | repl []string |
| 71 | nextDue []int64 |
| 72 | renders int |
| 73 | render func() (appconfig.HeldRender, error) |
| 74 | // renderHook, when set, is the whole held render: it gets the |
| 75 | // before-write callback the installer passes (fix round 1, I1). |
| 76 | renderHook func(before func(string) error) (appconfig.HeldRender, error) |
| 77 | release moves.Check |
| 78 | landAt string |
| 79 | landHits []string |
| 80 | landDone bool |
| 81 | parkAt string |
| 82 | parkHits []string |
| 83 | parkDone bool |
| 84 | parkView func(context.Context, hostclaim.Observed) (park.View, error) |
| 85 | landFault func(point string) error |
| 86 | store lease.VersionedStore // when set, the target's versioned store |
| 87 | |
| 88 | // renderPin, when set, sees the pins each render is asked to hold to |
| 89 | // (moves Task 7). |
| 90 | renderPin func(pin *provision.MoveInputs) |
| 91 | // releaseLive, when set, is the configured step's release |
| 92 | // (ReleaseLive, moves Task 7). |
| 93 | releaseLive func(ctx context.Context, s release.Store, rec provision.Record, version string) moves.Check |
| 94 | |
| 95 | // probe, when set, is the target's probe wake (Plan 3 Task 6). |
| 96 | probe func(context.Context, activator.ProbeConfig) (activator.ProbeResult, error) |
| 97 | // probeLive, when set, is the activator's live probe entry. |
| 98 | probeLive func(string) (string, activator.ProbeResult, bool) |
| 99 | // openLog records the target's open (Plan 3 Task 7): the probe entry's |
| 100 | // mark and the hold's release, in order, with the record and claim |
| 101 | // state each saw. |
| 102 | openLog []string |
| 103 | // noRelease leaves ReleaseHeld unwired (the open unarmed). |
| 104 | noRelease bool |
| 105 | // holdProbed stops every pass at probed, before the commit (Task 6's |
| 106 | // tests, which end there). |
| 107 | holdProbed bool |
| 108 | // releaseNoop makes ReleaseHeld match nothing and change nothing. |
| 109 | releaseNoop bool |
| 110 | // releaseErr makes ReleaseHeld fail. |
| 111 | releaseErr error |
| 112 | // ownAuthorize drops the park authorizer the target would inherit from |
| 113 | // the source's configuration (the finalization harness sets one), so |
| 114 | // the target's parks use the allowing fake behind their own checks |
| 115 | // (Plan 5 Task 3). |
| 116 | ownAuthorize bool |
| 117 | // parkLeases, runHeld and ownLease, when set, replace the park fakes' |
| 118 | // leases, the claim's RunHeld (park.Engine.RunHeld, M5) and the |
| 119 | // landing's own-location lease. |
| 120 | parkLeases park.Leases |
| 121 | runHeld func(ctx context.Context, e park.Engine, w park.Work, held park.HeldLease) error |
| 122 | ownLease func(ctx context.Context, bucket, location string) (park.HeldLease, error) |
| 123 | } |
| 124 | |
| 125 | // fakeParkLeases refuses a second acquisition of a held location. |
| 126 | type fakeParkLeases struct { |
| 127 | mu sync.Mutex |
| 128 | held map[string]bool |
| 129 | leases map[string]*fakeHeldLease |
| 130 | acquired []string |
| 131 | } |
| 132 | |
| 133 | // fakeHeldLease is a park.HeldLease with the Context and LeaseScope methods |
| 134 | // hostbox's renewingHeld has: a context derived from it is cancelled when |
| 135 | // the lease is lost or released, and the claim's AggregateLeases lends it to |
| 136 | // RunHeld for its own bucket and scope. |
| 137 | type fakeHeldLease struct { |
| 138 | l *fakeParkLeases |
| 139 | key string |
| 140 | bucket, scope string |
| 141 | ctx context.Context |
| 142 | cancel context.CancelFunc |
| 143 | } |
| 144 | |
| 145 | // LeaseScope is renewingHeld.LeaseScope: the bucket and scope it holds. |
| 146 | func (h *fakeHeldLease) LeaseScope() (string, string) { return h.bucket, h.scope } |
| 147 | |
| 148 | func (l *fakeParkLeases) Acquire(_ context.Context, bucket, scope string) (park.HeldLease, error) { |
| 149 | l.mu.Lock() |
| 150 | defer l.mu.Unlock() |
| 151 | key := bucket + "/" + scope |
| 152 | if l.held[key] { |
| 153 | return nil, &lease.HeldError{Owner: "the landing"} |
| 154 | } |
| 155 | if l.held == nil { |
| 156 | l.held = map[string]bool{} |
| 157 | } |
| 158 | l.held[key] = true |
| 159 | l.acquired = append(l.acquired, key) |
| 160 | ctx, cancel := context.WithCancel(context.Background()) |
| 161 | h := &fakeHeldLease{l: l, key: key, bucket: bucket, scope: scope, ctx: ctx, cancel: cancel} |
| 162 | if l.leases == nil { |
| 163 | l.leases = map[string]*fakeHeldLease{} |
| 164 | } |
| 165 | l.leases[key] = h |
| 166 | return h, nil |
| 167 | } |
| 168 | |
| 169 | // Context is renewingHeld.Context: ctx, cancelled too when the lease is. |
| 170 | func (h *fakeHeldLease) Context(ctx context.Context) context.Context { |
| 171 | c, cancel := context.WithCancel(ctx) |
| 172 | stop := context.AfterFunc(h.ctx, cancel) |
| 173 | context.AfterFunc(c, func() { stop() }) |
| 174 | return c |
| 175 | } |
| 176 | |
| 177 | func (l *fakeParkLeases) holding() []string { |
| 178 | l.mu.Lock() |
| 179 | defer l.mu.Unlock() |
| 180 | var out []string |
| 181 | for k, on := range l.held { |
| 182 | if on { |
| 183 | out = append(out, k) |
| 184 | } |
| 185 | } |
| 186 | sort.Strings(out) |
| 187 | return out |
| 188 | } |
| 189 | |
| 190 | func (h *fakeHeldLease) Check(context.Context) error { |
| 191 | h.l.mu.Lock() |
| 192 | defer h.l.mu.Unlock() |
| 193 | if !h.l.held[h.key] { |
| 194 | return errors.New("fake lease: released") |
| 195 | } |
| 196 | return nil |
| 197 | } |
| 198 | func (h *fakeHeldLease) Release(context.Context) error { |
| 199 | h.l.mu.Lock() |
| 200 | defer h.l.mu.Unlock() |
| 201 | delete(h.l.held, h.key) |
| 202 | h.cancel() |
| 203 | return nil |
| 204 | } |
| 205 | func (h *fakeHeldLease) Close() { _ = h.Release(context.Background()) } |
| 206 | |
| 207 | // realReplica writes a small SQLite database with rows rows over syncs |
| 208 | // litestream syncs (a level-9 snapshot too) and returns the file replica |
| 209 | // client and the newest level-0 position, as Task 8's test does. |
| 210 | func realReplica(t *testing.T, syncs int) (litestream.ReplicaClient, uint64, int) { |
| 211 | t.Helper() |
| 212 | dir := t.TempDir() |
| 213 | path := filepath.Join(dir, "src.db") |
| 214 | rows := 0 |
| 215 | write := func(n int) { |
| 216 | d, err := sql.Open("sqlite", path) |
| 217 | if err != nil { |
| 218 | t.Fatal(err) |
| 219 | } |
| 220 | defer d.Close() |
| 221 | if _, err := d.Exec(`CREATE TABLE IF NOT EXISTS t (id INTEGER PRIMARY KEY, body TEXT)`); err != nil { |
| 222 | t.Fatal(err) |
| 223 | } |
| 224 | for i := 0; i < n; i++ { |
| 225 | if _, err := d.Exec(`INSERT INTO t (body) VALUES (?)`, strings.Repeat("x", 100)); err != nil { |
| 226 | t.Fatal(err) |
| 227 | } |
| 228 | } |
| 229 | rows += n |
| 230 | } |
| 231 | write(1) |
| 232 | client := lsfile.NewReplicaClient(filepath.Join(dir, "replica")) |
| 233 | db := litestream.NewDB(path) |
| 234 | db.Logger = slog.New(slog.NewTextHandler(io.Discard, nil)) |
| 235 | db.SetMetaPath(filepath.Join(dir, "meta")) |
| 236 | db.MonitorInterval = time.Hour |
| 237 | db.Replica = litestream.NewReplicaWithClient(db, client) |
| 238 | db.Replica.SyncInterval = time.Hour |
| 239 | if err := db.Open(); err != nil { |
| 240 | t.Fatal(err) |
| 241 | } |
| 242 | ctx := context.Background() |
| 243 | for i := 0; i < syncs; i++ { |
| 244 | if err := db.SyncAndWait(ctx); err != nil { |
| 245 | t.Fatal(err) |
| 246 | } |
| 247 | write(2) |
| 248 | } |
| 249 | if err := db.SyncAndWait(ctx); err != nil { |
| 250 | t.Fatal(err) |
| 251 | } |
| 252 | if _, err := db.Snapshot(ctx); err != nil { |
| 253 | t.Fatal(err) |
| 254 | } |
| 255 | if err := db.Close(ctx); err != nil { |
| 256 | t.Fatal(err) |
| 257 | } |
| 258 | var n uint64 |
| 259 | for _, f := range replicaLevelFiles(t, client, 0) { |
| 260 | if uint64(f.MaxTXID) > n { |
| 261 | n = uint64(f.MaxTXID) |
| 262 | } |
| 263 | } |
| 264 | if n == 0 { |
| 265 | t.Fatal("the replica has no level-0 files") |
| 266 | } |
| 267 | return client, n, rows |
| 268 | } |
| 269 | |
| 270 | func replicaLevelFiles(t *testing.T, client litestream.ReplicaClient, level int) []*ltx.FileInfo { |
| 271 | t.Helper() |
| 272 | itr, err := client.LTXFiles(context.Background(), level, 0, false) |
| 273 | if err != nil { |
| 274 | t.Fatal(err) |
| 275 | } |
| 276 | infos, err := ltx.SliceFileIterator(itr) |
| 277 | if err != nil { |
| 278 | t.Fatal(err) |
| 279 | } |
| 280 | return infos |
| 281 | } |
| 282 | |
| 283 | func countRowsForTest(t *testing.T, path string) int { |
| 284 | t.Helper() |
| 285 | d, err := sql.Open("sqlite", "file:"+path+"?mode=ro") |
| 286 | if err != nil { |
| 287 | t.Fatal(err) |
| 288 | } |
| 289 | defer d.Close() |
| 290 | var n int |
| 291 | if err := d.QueryRow(`SELECT count(*) FROM t`).Scan(&n); err != nil { |
| 292 | t.Fatal(err) |
| 293 | } |
| 294 | return n |
| 295 | } |
| 296 | |
| 297 | func newLandingHarness(t *testing.T, opts ...p1Option) *landingHarness { |
| 298 | t.Helper() |
| 299 | h := &landingHarness{cutHarness: newCutHarness(t, opts...), leases: &fakeParkLeases{}, release: moves.Check{OK: true}} |
| 300 | // The real replica replaces the harness's fake level files under the |
| 301 | // source's lineage; the final checkpoint adds nothing (the database is |
| 302 | // quiet at N). |
| 303 | client, n, rows := realReplica(t, 2) |
| 304 | h.n, h.rows = n, rows |
| 305 | loc := "instances/" + p1Host + "/" + h.inc + "/" |
| 306 | for k := range h.f.buckets[h.srcBucket] { |
| 307 | if strings.HasPrefix(k, loc+"g-cur/") && strings.HasSuffix(k, ".ltx") { |
| 308 | delete(h.f.buckets[h.srcBucket], k) |
| 309 | } |
| 310 | } |
| 311 | for level := 0; level <= litestream.SnapshotLevel; level++ { |
| 312 | for _, info := range replicaLevelFiles(t, client, level) { |
| 313 | rc, err := client.OpenLTXFile(context.Background(), info.Level, info.MinTXID, info.MaxTXID, 0, 0) |
| 314 | if err != nil { |
| 315 | t.Fatal(err) |
| 316 | } |
| 317 | body, err := io.ReadAll(rc) |
| 318 | rc.Close() |
| 319 | if err != nil { |
| 320 | t.Fatal(err) |
| 321 | } |
| 322 | key := loc + "g-cur/" + fmt.Sprintf("%04x/", level) + ltx.FormatFilename(info.MinTXID, info.MaxTXID) |
| 323 | h.f.store(h.srcBucket, key, fakeMoveObject{lease.MirrorObject{ETag: "r-" + filepath.Base(key), Size: int64(len(body))}, string(body)}) |
| 324 | } |
| 325 | } |
| 326 | h.localPos = n |
| 327 | h.cfg.FinalCheckpoint = func(ctx context.Context, _, _ string, _ hostclaim.Holder, check func(context.Context) error) error { |
| 328 | return check(ctx) |
| 329 | } |
| 330 | // The target's data directory, with the two directories the guard and |
| 331 | // park.Local need, and the held row reconcile minted. |
| 332 | h.tdir = t.TempDir() |
| 333 | for _, d := range []string{"hostguard", ".parked"} { |
| 334 | if err := os.Mkdir(filepath.Join(h.tdir, d), 0o770); err != nil { |
| 335 | t.Fatal(err) |
| 336 | } |
| 337 | } |
| 338 | h.route = registry.Route{Host: p1Host, Kind: "instance", App: "app", Account: "bab", Hibernate: true, Status: "landing", HoldOp: p1ID, |
| 339 | Socket: provision.DefaultSocket(p1Host), DBPath: provision.DefaultDBPath(h.tdir, p1Host), Provisioned: true} |
| 340 | h.render = func() (appconfig.HeldRender, error) { |
| 341 | in := h.read(t, p1Host).Move.Inputs |
| 342 | if in == nil { |
| 343 | return appconfig.HeldRender{}, errors.New("no pinned inputs") |
| 344 | } |
| 345 | return appconfig.HeldRender{Path: filepath.Join(h.tdir, p1Host+".json"), SHA256: "x", BundleETags: in.BundleETags, OverlayETag: in.OverlayETag}, nil |
| 346 | } |
| 347 | // The restore-verification record names the host: local removes it. |
| 348 | if err := dbverify.Publish(context.Background(), h.s, &dbverify.Record{V: 1, Box: h.dstBox, CheckedAt: h.clock.Now(), |
| 349 | Hosts: map[string]dbverify.Verdict{p1Host: {OK: true, VerifiedAt: h.clock.Now()}, "other.example": {OK: true, VerifiedAt: h.clock.Now()}}}); err != nil { |
| 350 | t.Fatal(err) |
| 351 | } |
| 352 | h.lcfg = h.landingCfg(t) |
| 353 | return h |
| 354 | } |
| 355 | |
| 356 | // landingCfg is the target's configuration with the landing armed. |
| 357 | func (h *landingHarness) landingCfg(t *testing.T) moveCfg { |
| 358 | t.Helper() |
| 359 | c := h.target() |
| 360 | c.DataDir = h.tdir |
| 361 | c.CarryRequest = h.carry |
| 362 | c.OwnsRoute = func(host string) bool { return host == p1Host } |
| 363 | c.Route = func(host string) (*registry.Route, error) { |
| 364 | if host != p1Host { |
| 365 | return nil, nil |
| 366 | } |
| 367 | rt := h.route |
| 368 | return &rt, nil |
| 369 | } |
| 370 | c.FilesDir = func(host string) (string, string) { |
| 371 | return filepath.Join(h.tdir, host), filepath.Join(h.tdir, host+".env") |
| 372 | } |
| 373 | c.Guard = hostguard.Guard{DataDir: h.tdir, LockDir: t.TempDir()} |
| 374 | c.Local = park.Local{DataDir: h.tdir, Box: h.dstBox} |
| 375 | inherited := c.ParkAuthorize |
| 376 | armParkFakes(&c, h.s, h.leases, park.PlacementRegion(c.Region), roleTarget) |
| 377 | if h.parkView != nil { |
| 378 | c.ParkView = func(ctx context.Context, obs hostclaim.Observed) (park.View, error) { return h.parkView(ctx, obs) } |
| 379 | } |
| 380 | if inherited != nil && !h.ownAuthorize { |
| 381 | c.ParkAuthorize = inherited |
| 382 | } |
| 383 | if h.parkLeases != nil { |
| 384 | c.ParkLeases = h.parkLeases |
| 385 | } |
| 386 | if h.runHeld != nil { |
| 387 | c.ParkRunHeld = h.runHeld |
| 388 | } |
| 389 | c.OwnLease = h.leases.Acquire |
| 390 | if h.ownLease != nil { |
| 391 | c.OwnLease = h.ownLease |
| 392 | } |
| 393 | if h.store != nil { |
| 394 | c.Versioned = func(context.Context, string, string) (lease.VersionedStore, error) { return h.store, nil } |
| 395 | } |
| 396 | c.RenderHeld = func(_ context.Context, _ release.Store, _ registry.Route, _ string, pin *provision.MoveInputs, before func(string) error) (appconfig.HeldRender, error) { |
| 397 | h.renders++ |
| 398 | if h.renderPin != nil { |
| 399 | h.renderPin(pin) |
| 400 | } |
| 401 | if h.renderHook != nil { |
| 402 | return h.renderHook(before) |
| 403 | } |
| 404 | hr, err := h.render() |
| 405 | if hr.SHA256 != "" { |
| 406 | if berr := before(hr.SHA256); berr != nil { |
| 407 | return appconfig.HeldRender{}, berr |
| 408 | } |
| 409 | } |
| 410 | return hr, err |
| 411 | } |
| 412 | c.ReleaseReady = func(context.Context, release.Store, provision.Record, string) moves.Check { return h.release } |
| 413 | c.ReleaseLive = h.releaseLive |
| 414 | c.ReplaceTimers = func(_ context.Context, host string, rows []registry.Timer) error { |
| 415 | h.replaced = append(h.replaced, append([]registry.Timer(nil), rows...)) |
| 416 | return nil |
| 417 | } |
| 418 | c.SetReplication = func(host, class string) error { h.repl = append(h.repl, class); return nil } |
| 419 | c.SetNextDue = func(host string, due int64) error { h.nextDue = append(h.nextDue, due); return nil } |
| 420 | c.RemoveVerifyEntry = dbverify.RemoveHost |
| 421 | c.WriteGen = hostbox.WriteGenerationMarker |
| 422 | if h.probe != nil { |
| 423 | c.Probe = h.probe |
| 424 | } |
| 425 | if h.probeLive != nil { |
| 426 | c.ProbeLive = h.probeLive |
| 427 | } |
| 428 | c.ReleaseHeld = func(host, op string) (bool, error) { |
| 429 | h.openLog = append(h.openLog, "release:"+h.openState(t)) |
| 430 | if h.releaseNoop { |
| 431 | return false, nil |
| 432 | } |
| 433 | if h.releaseErr != nil { |
| 434 | return false, h.releaseErr |
| 435 | } |
| 436 | if host != p1Host || !registry.IsHold(h.route.Status) || h.route.HoldOp != op { |
| 437 | return false, nil |
| 438 | } |
| 439 | h.route.Status, h.route.HoldOp = "ready", "" |
| 440 | return true, nil |
| 441 | } |
| 442 | if h.noRelease { |
| 443 | c.ReleaseHeld = nil |
| 444 | } |
| 445 | c.ConfirmProbeOpen = func(host, op string) bool { |
| 446 | h.openLog = append(h.openLog, "confirm:"+h.openState(t)) |
| 447 | return host == p1Host && op == p1ID |
| 448 | } |
| 449 | c.MarkProbeOpen = func(host, op string) bool { |
| 450 | h.openLog = append(h.openLog, "mark:"+h.openState(t)) |
| 451 | return host == p1Host && op == p1ID |
| 452 | } |
| 453 | c.landFault = func(point string) error { |
| 454 | h.landHits = append(h.landHits, point) |
| 455 | if h.holdProbed && point == "before-commit" { |
| 456 | return errHoldProbed |
| 457 | } |
| 458 | if h.landFault != nil { |
| 459 | if err := h.landFault(point); err != nil { |
| 460 | return err |
| 461 | } |
| 462 | } |
| 463 | if point == h.landAt && !h.landDone { |
| 464 | h.landDone = true |
| 465 | return errCrash |
| 466 | } |
| 467 | return nil |
| 468 | } |
| 469 | c.parkFault = func(step string) error { |
| 470 | h.parkHits = append(h.parkHits, step) |
| 471 | if step == h.parkAt && !h.parkDone { |
| 472 | h.parkDone = true |
| 473 | return errCrash |
| 474 | } |
| 475 | return nil |
| 476 | } |
| 477 | return c |
| 478 | } |
| 479 | |
| 480 | // flip drives the cut to the flip: the record landing/checking in uk and |
| 481 | // the claim landing/checking under uk-box. |
| 482 | func (h *landingHarness) flip(t *testing.T) { |
| 483 | t.Helper() |
| 484 | h.ready(t) |
| 485 | h.cut(t) |
| 486 | h.pass(t) |
| 487 | if rec := h.read(t, p1Host); rec.Region != h.dst || rec.Move.State != provision.MoveLanding || rec.Move.LandingPhase != provision.LandChecking { |
| 488 | t.Fatalf("the fixture did not flip: %s %+v", rec.Region, rec.Move) |
| 489 | } |
| 490 | } |
| 491 | |
| 492 | // land runs one pass as the target box, the worker kept across passes. |
| 493 | func (h *landingHarness) land(t *testing.T) { |
| 494 | t.Helper() |
| 495 | if h.lw == nil { |
| 496 | h.lw = &movesWorker{mirrors: map[string]lease.MirrorStore{}} |
| 497 | } |
| 498 | h.lw.cfg = h.landingCfg(t) |
| 499 | if err := movesPass(context.Background(), h.s, h.lw); err != nil { |
| 500 | t.Fatal(err) |
| 501 | } |
| 502 | } |
| 503 | |
| 504 | // crashLand runs one target pass that must reach the landing crash point. |
| 505 | func (h *landingHarness) crashLand(t *testing.T) { |
| 506 | t.Helper() |
| 507 | h.land(t) |
| 508 | if !h.landDone { |
| 509 | t.Fatalf("the pass never reached %s (points reached: %v)", h.landAt, h.landHits) |
| 510 | } |
| 511 | } |
| 512 | |
| 513 | // freshTarget drops the target worker, as a restarted agent would, and |
| 514 | // with it every lease the dead process held: a real lease lapses at its |
| 515 | // TTL once nobody renews it, and the restarted worker acquires afresh. |
| 516 | func (h *landingHarness) freshTarget() { |
| 517 | h.lw = nil |
| 518 | h.leases.expire() |
| 519 | } |
| 520 | |
| 521 | // expire lapses every held lease, as the TTL does for a dead holder, and |
| 522 | // cancels every context derived from them. |
| 523 | func (l *fakeParkLeases) expire() { |
| 524 | l.mu.Lock() |
| 525 | defer l.mu.Unlock() |
| 526 | l.held = map[string]bool{} |
| 527 | for _, h := range l.leases { |
| 528 | h.cancel() |
| 529 | } |
| 530 | } |
| 531 | |
| 532 | func (h *landingHarness) landJournal(t *testing.T) *moves.Journal { |
| 533 | t.Helper() |
| 534 | j, err := moves.ReadJournal(h.tdir, p1Host, p1ID) |
| 535 | if err != nil { |
| 536 | t.Fatal(err) |
| 537 | } |
| 538 | return j |
| 539 | } |
| 540 | |
| 541 | func (h *landingHarness) filesGen(t *testing.T) (filesGenMarker, bool) { |
| 542 | t.Helper() |
| 543 | body, err := os.ReadFile(filepath.Join(h.tdir, p1Host, archive.MarkerName)) |
| 544 | if errors.Is(err, os.ErrNotExist) { |
| 545 | return filesGenMarker{}, false |
| 546 | } |
| 547 | if err != nil { |
| 548 | t.Fatal(err) |
| 549 | } |
| 550 | var m filesGenMarker |
| 551 | if err := json.Unmarshal(body, &m); err != nil { |
| 552 | t.Fatal(err) |
| 553 | } |
| 554 | return m, true |
| 555 | } |
| 556 | |
| 557 | // installed asserts the end state of a landing through restored. |
| 558 | func (h *landingHarness) installed(t *testing.T) moves.InstallManifest { |
| 559 | t.Helper() |
| 560 | rec := h.read(t, p1Host) |
| 561 | m := rec.Move |
| 562 | if m.State != provision.MoveLanding || m.LandingPhase != provision.LandRestored || m.Reason != "" { |
| 563 | t.Fatalf("record: %+v", m) |
| 564 | } |
| 565 | c := h.claim(t) |
| 566 | if c.Op.Kind != hostclaim.KindLanding || c.Op.Phase != provision.LandRestored || c.Op.Exec != h.dstBox || !containsString(c.Touched.Boxes, h.dstBox) { |
| 567 | t.Fatalf("claim: %+v touched %+v", c.Op, c.Touched) |
| 568 | } |
| 569 | j := h.landJournal(t) |
| 570 | var im moves.InstallManifest |
| 571 | if err := json.Unmarshal([]byte(j.Steps[landStepInstall]), &im); err != nil || im.Op != p1ID || im.N != h.n { |
| 572 | t.Fatalf("install manifest in the journal: %v %+v", err, im) |
| 573 | } |
| 574 | if m.Landing == nil || m.Landing.InstalledDBSHA256 != "sha256:"+im.DBSHA256 { |
| 575 | t.Fatalf("installed_db_sha256: %+v want %s", m.Landing, im.DBSHA256) |
| 576 | } |
| 577 | dbPath := provision.DefaultDBPath(h.tdir, p1Host) |
| 578 | if got, err := fileSHA256(dbPath); err != nil || got != "sha256:"+im.DBSHA256 { |
| 579 | t.Fatalf("the installed database: %v %s want %s", err, got, im.DBSHA256) |
| 580 | } |
| 581 | if got := countRowsForTest(t, dbPath); got != h.rows { |
| 582 | t.Fatalf("the installed database holds %d rows, the source wrote %d", got, h.rows) |
| 583 | } |
| 584 | if gen, err := hostbox.ReadGenerationMarker(dbPath + ".gen"); err != nil || gen != p1ID { |
| 585 | t.Fatalf("%s.gen: %q %v", dbPath, gen, err) |
| 586 | } |
| 587 | if fg, ok := h.filesGen(t); !ok || fg.Op != p1ID || fg.Incarnation != h.inc { |
| 588 | t.Fatalf(".files-gen: %+v present=%v", fg, ok) |
| 589 | } |
| 590 | if body, err := os.ReadFile(filepath.Join(h.tdir, p1Host, "upload.txt")); err != nil || string(body) != "hello" { |
| 591 | t.Fatalf("the extracted files: %q %v", body, err) |
| 592 | } |
| 593 | if body, err := os.ReadFile(filepath.Join(h.tdir, p1Host+".env")); err != nil || string(body) != h.srcEnvFile { |
| 594 | t.Fatalf("the extracted env: %q %v", body, err) |
| 595 | } |
| 596 | if held := h.leases.holding(); len(held) != 0 { |
| 597 | t.Fatalf("the landing still holds %v", held) |
| 598 | } |
| 599 | if _, err := os.Stat(filepath.Join(h.tdir, ".moves", p1Host, p1ID, "build", p1Host+".db")); !errors.Is(err, os.ErrNotExist) { |
| 600 | t.Fatalf("the build scratch was not renamed away: %v", err) |
| 601 | } |
| 602 | return im |
| 603 | } |
| 604 | |
| 605 | // The end to end landing through restored: the generation built from the |
| 606 | // seal, the archive extracted, timers replaced with their registrations, |
| 607 | // the carried columns set, the verify entry gone, .files-gen and |
| 608 | // <dbPath>.gen written, installed_db_sha256 recorded, the lease released. |
| 609 | func TestLandingBuildsInstallsAndRestores(t *testing.T) { |
| 610 | t.Parallel() |
| 611 | ctx := context.Background() |
| 612 | h := newLandingHarness(t) |
| 613 | h.timers = []registry.Timer{{Host: p1Host, Name: "nightly", Kind: "cron", Spec: "0 3 * * *", NextAt: 1000, Attempts: 2, Registration: "reg-1"}, |
| 614 | {Host: p1Host, Name: "once", Kind: "once", NextAt: 2000, DueAt: 2000, Registration: "reg-2"}} |
| 615 | h.flip(t) |
| 616 | h.route.Replication = "" |
| 617 | h.land(t) |
| 618 | im := h.installed(t) |
| 619 | // The generation holds exactly the build's four objects, at usable |
| 620 | // versions, in this box's own bucket. |
| 621 | prefix := moves.GenerationPrefix(cutRoot, p1ID) |
| 622 | var keys []string |
| 623 | for k := range h.f.buckets["data-uk"] { |
| 624 | if strings.HasPrefix(k, prefix) { |
| 625 | keys = append(keys, strings.TrimPrefix(k, prefix)) |
| 626 | } |
| 627 | } |
| 628 | sort.Strings(keys) |
| 629 | want := []string{"0000/" + ltx.FormatFilename(ltx.TXID(h.n), ltx.TXID(h.n)), "0009/" + ltx.FormatFilename(1, ltx.TXID(h.n)), "build-intent.json", "install.json"} |
| 630 | if strings.Join(keys, " ") != strings.Join(want, " ") { |
| 631 | t.Fatalf("generation objects %v, want %v", keys, want) |
| 632 | } |
| 633 | if !lease.UsableVersionID(im.SnapshotVersion) || !lease.UsableVersionID(im.AnchorVersion) { |
| 634 | t.Fatalf("manifest versions %q %q", im.SnapshotVersion, im.AnchorVersion) |
| 635 | } |
| 636 | // Timers: the sealed snapshot's rows, registrations included, once. |
| 637 | if len(h.replaced) != 1 || len(h.replaced[0]) != 2 { |
| 638 | t.Fatalf("ReplaceTimers calls: %+v", h.replaced) |
| 639 | } |
| 640 | if moves.TimersDigest(h.replaced[0]) != h.read(t, p1Host).Move.Source.TimersSHA256 { |
| 641 | t.Fatalf("the replaced rows do not digest to the record's: %+v", h.replaced[0]) |
| 642 | } |
| 643 | byName := map[string]registry.Timer{} |
| 644 | for _, r := range h.replaced[0] { |
| 645 | byName[r.Name] = r |
| 646 | } |
| 647 | if byName["nightly"].Registration != "reg-1" || byName["nightly"].Attempts != 2 || byName["once"].Registration != "reg-2" || byName["once"].DueAt != 2000 { |
| 648 | t.Fatalf("registrations and bookkeeping not preserved: %+v", h.replaced[0]) |
| 649 | } |
| 650 | src := h.read(t, p1Host).Move.Source |
| 651 | if len(h.repl) != 1 || h.repl[0] != src.Replication || len(h.nextDue) != 1 || h.nextDue[0] != src.NextDue { |
| 652 | t.Fatalf("columns: replication %v next_due %v, source %+v", h.repl, h.nextDue, src) |
| 653 | } |
| 654 | dv, err := dbverify.Load(ctx, h.s) |
| 655 | if err != nil || dv == nil { |
| 656 | t.Fatal(err) |
| 657 | } |
| 658 | if _, ok := dv.Hosts[p1Host]; ok || len(dv.Hosts) != 1 { |
| 659 | t.Fatalf("the verify entry: %+v", dv.Hosts) |
| 660 | } |
| 661 | // The inventory names the four flat paths under F, the flip's stamp. |
| 662 | c := h.claim(t) |
| 663 | inv := c.Inventory["uk-box"] |
| 664 | if inv.Layout != "flat" || len(inv.Paths) != 4 { |
| 665 | t.Fatalf("inventory: %+v", inv) |
| 666 | } |
| 667 | for _, p := range []string{p1Host + ".db", p1Host + ".db.gen", p1Host, p1Host + ".env"} { |
| 668 | if inv.Paths[p] != c.Holder.Stamp { |
| 669 | t.Fatalf("inventory path %s at %d, want F %d", p, inv.Paths[p], c.Holder.Stamp) |
| 670 | } |
| 671 | } |
| 672 | captured, err := hostguard.Capture(h.tdir, p1Host) |
| 673 | if err != nil || captured.Stamp != c.Holder.Stamp || captured.Incarnation != h.inc { |
| 674 | t.Fatalf("the installation record: %+v %v, want stamp %d", captured, err, c.Holder.Stamp) |
| 675 | } |
| 676 | // No residue, so no park chores; and no park-local (no earlier stay). |
| 677 | if len(c.Obligations) != 0 { |
| 678 | t.Fatalf("obligations: %+v", c.Obligations) |
| 679 | } |
| 680 | // The journal records every sub-step, the prepared manifest as skipped. |
| 681 | j := h.landJournal(t) |
| 682 | for _, step := range []string{landStepResidue, landStepInstall, landStepPublished, landStepRetired, landStepExtracted, landStepInventory, landStepTimers, landStepColumns, landStepVerify, landStepFilesManifest, landStepFilesGen} { |
| 683 | if !j.Done(step) { |
| 684 | t.Fatalf("journal lacks %s: %v", step, j.Steps) |
| 685 | } |
| 686 | } |
| 687 | if h.renders != 1 { |
| 688 | t.Fatalf("renders %d", h.renders) |
| 689 | } |
| 690 | // Restored is Plan 3's: another pass changes nothing. |
| 691 | before := h.read(t, p1Host).Move.UpdatedAt |
| 692 | h.land(t) |
| 693 | if after := h.read(t, p1Host).Move; after.LandingPhase != provision.LandRestored || !after.UpdatedAt.Equal(before) { |
| 694 | t.Fatalf("a pass at restored wrote the record: %+v", after) |
| 695 | } |
| 696 | } |
| 697 | |
| 698 | // checking (§7.6): the claim must be this move's landing op with this box |
| 699 | // executing, citing complete source evidence, and this box is added to |
| 700 | // touched.boxes before any local write; the installation record is |
| 701 | // activated at F. A claim the source left at moving/sealed (a crash after |
| 702 | // the record's flip) is flipped by the target itself (ruling G12). |
| 703 | func TestLandingCheckingRequiresClaimEvidenceAndTouch(t *testing.T) { |
| 704 | ctx := context.Background() |
| 705 | t.Run("the target writes the flip's claim half", func(t *testing.T) { |
| 706 | h := newLandingHarness(t) |
| 707 | h.ready(t) |
| 708 | h.crashAt = "after-record-flip" |
| 709 | h.cut(t) |
| 710 | h.crashPass(t) |
| 711 | if c := h.claim(t); c.Op.Kind != hostclaim.KindMoving || c.Op.Phase != hostclaim.PhaseSealed { |
| 712 | t.Fatalf("fixture: %+v", c.Op) |
| 713 | } |
| 714 | h.landAt = "after-target-flip" |
| 715 | h.crashLand(t) |
| 716 | c := h.claim(t) |
| 717 | if c.Op.Kind != hostclaim.KindLanding || c.Op.Phase != provision.LandChecking || c.Op.Exec != "uk-box" { |
| 718 | t.Fatalf("after the target's flip: %+v", c.Op) |
| 719 | } |
| 720 | if c.Op.Journal[hostclaim.JournalSourceStamp] != "1" || c.Op.Journal[hostclaim.JournalSrcGen] != "cur" { |
| 721 | t.Fatalf("the target's flip did not journal the source's stamp and generation: %+v", c.Op.Journal) |
| 722 | } |
| 723 | h.land(t) |
| 724 | h.installed(t) |
| 725 | }) |
| 726 | t.Run("another executor is refused", func(t *testing.T) { |
| 727 | h := newLandingHarness(t) |
| 728 | h.flip(t) |
| 729 | // Refused before any claim write: with uk-box taken out of touched, |
| 730 | // the K12 Touch that follows the executor check must not run (the |
| 731 | // guard would refuse the activation too, but later and locally). |
| 732 | rewriteP1Claim(t, h.s, func(c *hostclaim.Claim) { c.Op.Exec = "other-box"; c.Touched.Boxes = []string{"ie-box"} }) |
| 733 | h.land(t) |
| 734 | if m := h.read(t, p1Host).Move; m.LandingPhase != provision.LandChecking { |
| 735 | t.Fatalf("advanced under another executor: %+v", m) |
| 736 | } |
| 737 | if _, err := hostguard.Capture(h.tdir, p1Host); !errors.Is(err, hostguard.ErrNoRecord) { |
| 738 | t.Fatalf("the installation record was written: %v", err) |
| 739 | } |
| 740 | if c := h.claim(t); containsString(c.Touched.Boxes, "uk-box") { |
| 741 | t.Fatal("a claim write under another executor") |
| 742 | } |
| 743 | }) |
| 744 | t.Run("missing evidence is refused", func(t *testing.T) { |
| 745 | h := newLandingHarness(t) |
| 746 | h.flip(t) |
| 747 | if err := h.s.Delete(ctx, h.stoppedKey()); err != nil { |
| 748 | t.Fatal(err) |
| 749 | } |
| 750 | h.land(t) |
| 751 | if m := h.read(t, p1Host).Move; m.LandingPhase != provision.LandChecking { |
| 752 | t.Fatalf("advanced without the source's evidence: %+v", m) |
| 753 | } |
| 754 | }) |
| 755 | t.Run("touch and the installation record", func(t *testing.T) { |
| 756 | h := newLandingHarness(t) |
| 757 | h.flip(t) |
| 758 | // Plan 1's target report touched uk-box at its first staging write; |
| 759 | // checking's Touch is the idempotent K12 re-add, so take it away |
| 760 | // first to see it written. |
| 761 | rewriteP1Claim(t, h.s, func(c *hostclaim.Claim) { c.Touched.Boxes = []string{"ie-box"} }) |
| 762 | h.landAt = "after-record-parking" |
| 763 | h.crashLand(t) |
| 764 | c := h.claim(t) |
| 765 | if !containsString(c.Touched.Boxes, "uk-box") || !containsString(c.Touched.Regions, "ie") { |
| 766 | t.Fatalf("touched: %+v", c.Touched) |
| 767 | } |
| 768 | captured, err := hostguard.Capture(h.tdir, p1Host) |
| 769 | if err != nil || captured.Stamp != c.Holder.Stamp || captured.Incarnation != h.inc { |
| 770 | t.Fatalf("active_stamp %+v %v, want F %d", captured, err, c.Holder.Stamp) |
| 771 | } |
| 772 | // Spec §7.6 checking (Plan 5 Task 9a, M5): the location's lease in |
| 773 | // the target's own bucket, taken once and held into parking. |
| 774 | want := "data-uk/" + strings.TrimSuffix(c.Holder.Location, "/") |
| 775 | if len(h.leases.acquired) != 1 || h.leases.acquired[0] != want { |
| 776 | t.Fatalf("checking took %v, want the location's lease %s", h.leases.acquired, want) |
| 777 | } |
| 778 | if held := h.leases.holding(); len(held) != 1 || held[0] != want { |
| 779 | t.Fatalf("after checking the landing holds %v", held) |
| 780 | } |
| 781 | }) |
| 782 | t.Run("a nested row is refused for a rollback", func(t *testing.T) { |
| 783 | h := newLandingHarness(t) |
| 784 | h.flip(t) |
| 785 | h.route.DBPath = filepath.Join(h.tdir, "instances", p1Host, "db", "instance.db") |
| 786 | h.land(t) |
| 787 | if m := h.read(t, p1Host).Move; m.Reason != provision.RollbackLandingLayout || m.LandingPhase != provision.LandChecking || m.State != provision.MoveLanding { |
| 788 | t.Fatalf("a nested row: %+v", m) |
| 789 | } |
| 790 | }) |
| 791 | t.Run("host claims on refuse at checking", func(t *testing.T) { |
| 792 | h := newLandingHarness(t) |
| 793 | h.flip(t) |
| 794 | t.Setenv("CARLOS_HOST_CLAIM", "1") |
| 795 | h.land(t) |
| 796 | if m := h.read(t, p1Host).Move; m.Reason != provision.RollbackClaimWired || m.LandingPhase != provision.LandChecking { |
| 797 | t.Fatalf("under CARLOS_HOST_CLAIM: %+v", m) |
| 798 | } |
| 799 | }) |
| 800 | } |
| 801 | |
| 802 | // residue plants an earlier stay of this incarnation at the holder's |
| 803 | // location in the target's bucket: two generations (old and the origin) |
| 804 | // and a generation.json, each at a usable version. |
| 805 | func (h *landingHarness) residue(t *testing.T) []string { |
| 806 | t.Helper() |
| 807 | ctx := context.Background() |
| 808 | keys := []string{ |
| 809 | cutRoot + "/g-old/0000/0000000000000001-0000000000000001.ltx", |
| 810 | cutRoot + "/g-old/0009/0000000000000001-0000000000000001.ltx", |
| 811 | cutRoot + "/0000/0000000000000001-0000000000000001.ltx", |
| 812 | cutRoot + "/generation.json", |
| 813 | } |
| 814 | for _, k := range keys { |
| 815 | if _, err := h.vs.PutIfAbsent(ctx, h.dstBucket, k, []byte("residue "+k)); err != nil { |
| 816 | t.Fatal(err) |
| 817 | } |
| 818 | } |
| 819 | return keys |
| 820 | } |
| 821 | |
| 822 | // parking (§7.6, D6): an earlier stay's objects under holder.location are |
| 823 | // archived under parked/<inc>/<host>/<ret>/instances/... by one |
| 824 | // park-generation chore per generation before building; a crash inside the |
| 825 | // engine leaves the phase at parking and the replay finishes; the chores |
| 826 | // stay pending until the horizon; after the probe makes the new generation |
| 827 | // live here, Engine.Run on them acknowledges rather than refusing. |
| 828 | func TestLandingParkingWaitsForTheManifest(t *testing.T) { |
| 829 | t.Parallel() |
| 830 | ctx := context.Background() |
| 831 | h := newLandingHarness(t) |
| 832 | h.flip(t) |
| 833 | keys := h.residue(t) |
| 834 | h.parkAt = "copy" |
| 835 | h.land(t) |
| 836 | if !h.parkDone { |
| 837 | t.Fatalf("the engine never copied (%v)", h.parkHits) |
| 838 | } |
| 839 | if m := h.read(t, p1Host).Move; m.LandingPhase != provision.LandParking { |
| 840 | t.Fatalf("after the crash: %+v", m) |
| 841 | } |
| 842 | c := h.claim(t) |
| 843 | var gens []string |
| 844 | for _, o := range c.Obligations { |
| 845 | if o.Op == p1ID && o.Kind == "park-generation" && o.Region == "uk" && o.Park != nil && o.Park.Location == cutRoot && o.Park.EvidenceAt.Equal(h.read(t, p1Host).Move.FlippedAt.UTC()) { |
| 846 | gens = append(gens, o.Generation) |
| 847 | } |
| 848 | } |
| 849 | sort.Strings(gens) |
| 850 | if strings.Join(gens, " ") != hostclaim.RootGeneration+" old" { |
| 851 | t.Fatalf("residue chores: %v (%+v)", gens, c.Obligations) |
| 852 | } |
| 853 | if len(h.leases.acquired) == 0 { |
| 854 | t.Fatal("the engine took no lease") |
| 855 | } |
| 856 | // The replay archives and writes the manifest, but one delete the store |
| 857 | // accepts does not land (its listing still shows the version): the |
| 858 | // manifest exists and its deletion set is not gone, so Parked is false |
| 859 | // and the phase waits, chores queued, nothing built. |
| 860 | stuck := keys[0] |
| 861 | h.vs.skipDelete = func(key string) bool { return key == stuck } |
| 862 | h.land(t) |
| 863 | if m := h.read(t, p1Host).Move; m.LandingPhase != provision.LandParking { |
| 864 | t.Fatalf("advanced with a deletion outstanding: %+v", m) |
| 865 | } |
| 866 | if _, ok := h.f.buckets["data-uk"][stuck]; !ok { |
| 867 | t.Fatalf("fixture: %s was deleted", stuck) |
| 868 | } |
| 869 | manifests := 0 |
| 870 | for k := range h.f.buckets["data-uk"] { |
| 871 | if strings.HasPrefix(k, "parked/"+h.inc+"/"+p1Host+"/") && strings.HasSuffix(k, "/manifest.json") { |
| 872 | manifests++ |
| 873 | } |
| 874 | } |
| 875 | if manifests != 2 { |
| 876 | t.Fatalf("%d park manifests after the second pass", manifests) |
| 877 | } |
| 878 | if n := len(h.vs.callsOn(moves.GenerationPrefix(cutRoot, p1ID))); n != 0 { |
| 879 | t.Fatalf("the build ran while parking waited: %d calls", n) |
| 880 | } |
| 881 | // The deletes land: the deletion set is gone, the landing goes on to |
| 882 | // restored. The chores are still owed. |
| 883 | h.vs.skipDelete = nil |
| 884 | h.land(t) |
| 885 | h.installed(t) |
| 886 | c = h.claim(t) |
| 887 | var retirements []string |
| 888 | for _, o := range c.Obligations { |
| 889 | if o.Kind == "park-generation" { |
| 890 | retirements = append(retirements, o.Retirement) |
| 891 | } |
| 892 | } |
| 893 | if len(retirements) != 2 { |
| 894 | t.Fatalf("chores after the landing: %+v", c.Obligations) |
| 895 | } |
| 896 | // The marker is the location's one reused key: it is not parked (the |
| 897 | // probe supersedes it), every other object is. |
| 898 | marker := cutRoot + "/generation.json" |
| 899 | if _, ok := h.f.buckets["data-uk"][marker]; !ok { |
| 900 | t.Fatal("the earlier stay's generation.json was parked") |
| 901 | } |
| 902 | for _, k := range keys { |
| 903 | if _, ok := h.f.buckets["data-uk"][k]; ok && k != marker { |
| 904 | t.Fatalf("%s is still at the location", k) |
| 905 | } |
| 906 | } |
| 907 | var archived []string |
| 908 | for k := range h.f.buckets["data-uk"] { |
| 909 | if strings.HasPrefix(k, "parked/"+h.inc+"/"+p1Host+"/") { |
| 910 | archived = append(archived, k) |
| 911 | } |
| 912 | } |
| 913 | sort.Strings(archived) |
| 914 | if len(archived) != len(keys)-1+2 { // every object but the marker, plus two manifests |
| 915 | t.Fatalf("archived: %v", archived) |
| 916 | } |
| 917 | for _, k := range keys { |
| 918 | if k == marker { |
| 919 | continue |
| 920 | } |
| 921 | rel := strings.TrimPrefix(k, "instances/") |
| 922 | found := false |
| 923 | for _, a := range archived { |
| 924 | if strings.Contains(a, "/instances/"+strings.TrimPrefix(rel, p1Host+"/")+"@") { |
| 925 | found = true |
| 926 | } |
| 927 | } |
| 928 | if !found { |
| 929 | t.Fatalf("%s was not archived under parked/<inc>/<host>/<ret>/instances/ (%v)", k, archived) |
| 930 | } |
| 931 | } |
| 932 | // The probe's admission (Plan 3) makes g-<move id> live here: the |
| 933 | // holder's generation is the move id, the marker names it and pins the |
| 934 | // snapshot, and uk is authoritative. Past the horizon the chores |
| 935 | // acknowledge: their generations are not the live one. |
| 936 | j := h.landJournal(t) |
| 937 | var im moves.InstallManifest |
| 938 | if err := json.Unmarshal([]byte(j.Steps[landStepInstall]), &im); err != nil { |
| 939 | t.Fatal(err) |
| 940 | } |
| 941 | // The probe writes the marker If-Match over the earlier stay's: a new |
| 942 | // current version at the same key. |
| 943 | markerVer := h.f.nextVersion() |
| 944 | h.f.store("data-uk", marker, fakeMoveObject{lease.MirrorObject{ETag: "etag-" + markerVer, Size: 20}, `{"op":"` + p1ID + `"}`}) |
| 945 | h.f.setVersion("data-uk", marker, markerVer) |
| 946 | rewriteP1Claim(t, h.s, func(c *hostclaim.Claim) { c.Holder.Generation = p1ID }) |
| 947 | h.clock.Advance(park.SignatureAge + landingRequestLimit + time.Minute) |
| 948 | view := func(_ context.Context, obs hostclaim.Observed) (park.View, error) { |
| 949 | return park.View{ClaimETag: obs.ETag, Region: "uk", AuthoritativeRegion: "uk", MarkerKnown: true, MarkerGeneration: p1ID, |
| 950 | Marker: park.Ref{Key: cutRoot + "/generation.json", VersionID: markerVer}, |
| 951 | Pinned: map[park.Ref]bool{{Key: im.SnapshotKey, VersionID: im.SnapshotVersion}: true}, |
| 952 | GenerationOf: func(r park.Ref) (string, bool) { return landingGenerationOf(cutRoot, r.Key) }}, nil |
| 953 | } |
| 954 | l := &landingRun{w: h.lw, s: h.s, rec: h.read(t, p1Host), g: provision.MoveGuard{ID: p1ID, Incarnation: h.inc}} |
| 955 | eng := park.Engine{Store: h.vs, Claims: h.s, Bucket: "data-uk", Region: "uk", Leases: h.leases, Now: h.clock.Now, NewID: obligationID, View: view, Authorize: l.authorizeObligation} |
| 956 | for _, o := range c.Obligations { |
| 957 | if o.Kind != "park-generation" { |
| 958 | continue |
| 959 | } |
| 960 | if err := eng.Run(ctx, park.Work{Host: p1Host, ID: o.ID, Op: p1ID}); err != nil { |
| 961 | t.Fatalf("acknowledge generation %s: %v", o.Generation, err) |
| 962 | } |
| 963 | } |
| 964 | if c := h.claim(t); len(c.Obligations) != 0 { |
| 965 | t.Fatalf("chores after the acknowledgement: %+v", c.Obligations) |
| 966 | } |
| 967 | prefix := moves.GenerationPrefix(cutRoot, p1ID) |
| 968 | n := 0 |
| 969 | for k := range h.f.buckets["data-uk"] { |
| 970 | if strings.HasPrefix(k, prefix) { |
| 971 | n++ |
| 972 | } |
| 973 | } |
| 974 | if n != 4 { |
| 975 | t.Fatalf("the acknowledgement touched the live generation: %d objects left", n) |
| 976 | } |
| 977 | if held := h.leases.holding(); len(held) != 0 { |
| 978 | t.Fatalf("leases held after the acknowledgement: %v", held) |
| 979 | } |
| 980 | } |
| 981 | |
| 982 | // Phase catch-up (§7.5, round-2 finding 11): a crash after the record's CAS |
| 983 | // and before the claim's leaves the claim one step behind; the next pass |
| 984 | // walks the claim forward and continues. A claim several steps behind is |
| 985 | // walked step by step, with each transition written separately. |
| 986 | func TestLandingPhaseCatchUp(t *testing.T) { |
| 987 | t.Parallel() |
| 988 | ctx := context.Background() |
| 989 | h := newLandingHarness(t) |
| 990 | h.flip(t) |
| 991 | h.landAt = "after-record-parking" |
| 992 | h.crashLand(t) |
| 993 | if m := h.read(t, p1Host).Move; m.LandingPhase != provision.LandParking { |
| 994 | t.Fatalf("record: %+v", m) |
| 995 | } |
| 996 | if c := h.claim(t); c.Op.Phase != provision.LandChecking { |
| 997 | t.Fatalf("the claim moved with the record: %+v", c.Op) |
| 998 | } |
| 999 | h.freshTarget() |
| 1000 | h.landAt = "after-residue-queued" // never reached: no residue |
| 1001 | h.landDone = false |
| 1002 | h.land(t) |
| 1003 | h.installed(t) |
| 1004 | |
| 1005 | // A claim left at checking with the record at restored. |
| 1006 | rewriteClaimOp(t, h.s, func(op *hostclaim.Op) { op.Phase = provision.LandChecking }) |
| 1007 | l := &landingRun{w: h.lw, s: h.s, rec: h.read(t, p1Host), g: provision.MoveGuard{ID: p1ID, Incarnation: h.inc}} |
| 1008 | obs, err := l.syncClaimPhase(ctx) |
| 1009 | if err != nil { |
| 1010 | t.Fatal(err) |
| 1011 | } |
| 1012 | if obs.Claim.Op.Phase != provision.LandRestored || h.claim(t).Op.Phase != provision.LandRestored { |
| 1013 | t.Fatalf("after the catch-up: %+v", obs.Claim.Op) |
| 1014 | } |
| 1015 | // A claim ahead of the record is left alone. |
| 1016 | rewriteClaimOp(t, h.s, func(op *hostclaim.Op) { op.Phase = provision.LandProbing }) |
| 1017 | if obs, err := l.syncClaimPhase(ctx); err != nil || obs.Claim.Op.Phase != provision.LandProbing { |
| 1018 | t.Fatalf("a claim ahead: %v %+v", err, obs.Claim.Op) |
| 1019 | } |
| 1020 | // A claim at another kind is an error. |
| 1021 | rewriteClaimOp(t, h.s, func(op *hostclaim.Op) { op.Kind = hostclaim.KindRestoring; op.Phase = "released" }) |
| 1022 | if _, err := l.syncClaimPhase(ctx); err == nil { |
| 1023 | t.Fatal("a restoring claim was accepted") |
| 1024 | } |
| 1025 | } |
| 1026 | |
| 1027 | // earlierStay plants an earlier stay of this incarnation on the target box: |
| 1028 | // the installation record at stamp 1, the inventory listing the four flat |
| 1029 | // paths at that stamp, and the files at the live paths. |
| 1030 | func (h *landingHarness) earlierStay(t *testing.T) { |
| 1031 | t.Helper() |
| 1032 | rec := fmt.Sprintf(`{"v":1,"host":%q,"incarnation":%q,"active_stamp":1,"retired_stamps":[]}`, p1Host, h.inc) |
| 1033 | if err := os.WriteFile(filepath.Join(h.tdir, "hostguard", p1Host+".json"), []byte(rec), 0o660); err != nil { |
| 1034 | t.Fatal(err) |
| 1035 | } |
| 1036 | rewriteP1Claim(t, h.s, func(c *hostclaim.Claim) { |
| 1037 | c.Inventory = map[string]hostclaim.InventoryEntry{h.dstBox: {Layout: "flat", Paths: map[string]uint64{p1Host + ".db": 1, p1Host + ".db.gen": 1, p1Host: 1, p1Host + ".env": 1}}} |
| 1038 | }) |
| 1039 | if err := os.MkdirAll(filepath.Join(h.tdir, p1Host), 0o755); err != nil { |
| 1040 | t.Fatal(err) |
| 1041 | } |
| 1042 | for name, body := range map[string]string{p1Host + ".db": "old database", p1Host + ".db.gen": "old\n", p1Host + ".env": "OLD=1\n", filepath.Join(p1Host, "old.txt"): "old files"} { |
| 1043 | if err := os.WriteFile(filepath.Join(h.tdir, name), []byte(body), 0o644); err != nil { |
| 1044 | t.Fatal(err) |
| 1045 | } |
| 1046 | } |
| 1047 | } |
| 1048 | |
| 1049 | // local's order (round-2 finding 8, rulings G5 and G15): an earlier stay's |
| 1050 | // artifacts tagged S' are parked into .parked/ and gone from the live paths |
| 1051 | // before the new installation is registered under F; registering first is |
| 1052 | // what the real RecordInventory refuses. A park-local is queued at checking |
| 1053 | // only because the inventory has paths at S'. |
| 1054 | func TestLandingInventoryOrder(t *testing.T) { |
| 1055 | t.Parallel() |
| 1056 | ctx := context.Background() |
| 1057 | h := newLandingHarness(t) |
| 1058 | h.flip(t) |
| 1059 | h.earlierStay(t) |
| 1060 | // Registering first is refused: with the record activated at F and the |
| 1061 | // old paths still tagged 1, RecordInventory says ErrLocalInventory. |
| 1062 | h.landAt = "after-activate" |
| 1063 | h.crashLand(t) |
| 1064 | c := h.claim(t) |
| 1065 | if !hostclaim.InventoryHasStamp(c, "uk-box", 1) { |
| 1066 | t.Fatal("fixture: no S' paths") |
| 1067 | } |
| 1068 | if j := h.landJournal(t); j.Steps[landStepPrevious] != "1" { |
| 1069 | t.Fatalf("previous: %v", j.Steps) |
| 1070 | } |
| 1071 | captured, err := hostguard.Capture(h.tdir, p1Host) |
| 1072 | if err != nil || captured.Stamp != c.Holder.Stamp { |
| 1073 | t.Fatalf("captured %+v %v", captured, err) |
| 1074 | } |
| 1075 | local := park.Local{Guard: hostguard.Guard{DataDir: h.tdir, LockDir: t.TempDir(), Claims: hostguard.StoreReader{Store: h.s}}, Claims: h.s, DataDir: h.tdir, Box: "uk-box"} |
| 1076 | check := func(c *hostclaim.Claim) error { return nil } |
| 1077 | if err := local.RecordInventory(ctx, p1Host, captured, "flat", []string{p1Host + ".db"}, check); !errors.Is(err, park.ErrLocalInventory) { |
| 1078 | t.Fatalf("registering over S' paths: %v", err) |
| 1079 | } |
| 1080 | // The landing: the park-local chore is queued, local retires then registers. |
| 1081 | h.land(t) |
| 1082 | h.installed(t) |
| 1083 | c = h.claim(t) |
| 1084 | if len(c.Obligations) != 0 { |
| 1085 | t.Fatalf("the park-local was not completed: %+v", c.Obligations) |
| 1086 | } |
| 1087 | inv := c.Inventory["uk-box"] |
| 1088 | for p, s := range inv.Paths { |
| 1089 | if s != c.Holder.Stamp { |
| 1090 | t.Fatalf("inventory %s at %d, want F %d", p, s, c.Holder.Stamp) |
| 1091 | } |
| 1092 | } |
| 1093 | // The old artifacts are under .parked/<inc>/<host>/<ret>/artifacts/, and |
| 1094 | // the live paths hold the new installation. |
| 1095 | var parked []string |
| 1096 | filepath.WalkDir(filepath.Join(h.tdir, ".parked", h.inc, p1Host), func(p string, d os.DirEntry, err error) error { |
| 1097 | if err == nil && !d.IsDir() && !strings.HasSuffix(p, ".journal.json") { |
| 1098 | rel, _ := filepath.Rel(filepath.Join(h.tdir, ".parked", h.inc, p1Host), p) |
| 1099 | parked = append(parked, strings.Join(strings.Split(rel, string(filepath.Separator))[1:], "/")) |
| 1100 | } |
| 1101 | return nil |
| 1102 | }) |
| 1103 | sort.Strings(parked) |
| 1104 | wantParked := []string{"artifacts/" + p1Host + ".db", "artifacts/" + p1Host + ".db.gen", "artifacts/" + p1Host + ".env", "artifacts/" + p1Host + "/old.txt"} |
| 1105 | if strings.Join(parked, " ") != strings.Join(wantParked, " ") { |
| 1106 | t.Fatalf("parked: %v, want %v", parked, wantParked) |
| 1107 | } |
| 1108 | if _, err := os.Stat(filepath.Join(h.tdir, p1Host, "old.txt")); !errors.Is(err, os.ErrNotExist) { |
| 1109 | t.Fatalf("the old files are still live: %v", err) |
| 1110 | } |
| 1111 | if body, err := os.ReadFile(filepath.Join(h.tdir, p1Host+".env")); err != nil || string(body) != "APP_MODE=prod\n" { |
| 1112 | t.Fatalf("env: %q %v", body, err) |
| 1113 | } |
| 1114 | } |
| 1115 | |
| 1116 | // No earlier stay queues no park-local and runs no retirement (ruling G5), |
| 1117 | // and the first landing on a box completes end to end with the real |
| 1118 | // park.Local and hostguard.Guard. |
| 1119 | func TestLandingFirstStayQueuesNoLocalPark(t *testing.T) { |
| 1120 | t.Parallel() |
| 1121 | h := newLandingHarness(t) |
| 1122 | h.flip(t) |
| 1123 | h.landAt = "after-record-parking" |
| 1124 | h.crashLand(t) |
| 1125 | if c := h.claim(t); len(c.Obligations) != 0 { |
| 1126 | t.Fatalf("a first landing queued %+v", c.Obligations) |
| 1127 | } |
| 1128 | if j := h.landJournal(t); j.Done(landStepPrevious) { |
| 1129 | t.Fatalf("a first landing noted a previous stay: %v", j.Steps) |
| 1130 | } |
| 1131 | h.land(t) |
| 1132 | h.installed(t) |
| 1133 | } |
| 1134 | |
| 1135 | // The markers are written last (witness wMarkersLast, codex R2-6): a crash |
| 1136 | // before local's final sub-step leaves no .files-gen with everything else |
| 1137 | // durable; the replay finishes local and the landing. |
| 1138 | func TestLandingMarkersAreWrittenLast(t *testing.T) { |
| 1139 | t.Parallel() |
| 1140 | h := newLandingHarness(t) |
| 1141 | h.flip(t) |
| 1142 | h.landAt = "before-files-gen" |
| 1143 | h.crashLand(t) |
| 1144 | if _, ok := h.filesGen(t); ok { |
| 1145 | t.Fatal(".files-gen exists before the last sub-step") |
| 1146 | } |
| 1147 | if m := h.read(t, p1Host).Move; m.LandingPhase != provision.LandLocal { |
| 1148 | t.Fatalf("record: %+v", m) |
| 1149 | } |
| 1150 | if body, err := os.ReadFile(filepath.Join(h.tdir, p1Host, "upload.txt")); err != nil || string(body) != "hello" { |
| 1151 | t.Fatalf("the files were not extracted before the marker: %q %v", body, err) |
| 1152 | } |
| 1153 | if _, err := os.Stat(provision.DefaultDBPath(h.tdir, p1Host) + ".gen"); !errors.Is(err, os.ErrNotExist) { |
| 1154 | t.Fatalf("<dbPath>.gen before restored: %v", err) |
| 1155 | } |
| 1156 | h.freshTarget() |
| 1157 | h.land(t) |
| 1158 | h.installed(t) |
| 1159 | if len(h.replaced) != 1 { |
| 1160 | t.Fatalf("the replay replaced the timers again: %d", len(h.replaced)) |
| 1161 | } |
| 1162 | } |
| 1163 | |
| 1164 | // configured (§7.6, Review Focus 4): the held render's bundle and overlay |
| 1165 | // etags must be the pinned inputs; a difference marks the rollback with |
| 1166 | // "configuration changed during the move" and the landing stops there. The |
| 1167 | // release gate waits. |
| 1168 | func TestLandingConfiguredPinsInputs(t *testing.T) { |
| 1169 | t.Parallel() |
| 1170 | h := newLandingHarness(t) |
| 1171 | h.flip(t) |
| 1172 | h.render = func() (appconfig.HeldRender, error) { |
| 1173 | return appconfig.HeldRender{BundleETags: map[string]string{"default": "e2", "environment": ""}}, nil |
| 1174 | } |
| 1175 | h.land(t) |
| 1176 | m := h.read(t, p1Host).Move |
| 1177 | if m.Reason != provision.RollbackConfigChanged || m.LandingPhase != provision.LandConfigured || m.State != provision.MoveLanding { |
| 1178 | t.Fatalf("after the render: %+v", m) |
| 1179 | } |
| 1180 | if _, err := os.Stat(provision.DefaultDBPath(h.tdir, p1Host)); !errors.Is(err, os.ErrNotExist) { |
| 1181 | t.Fatalf("the database was installed under a mismatch: %v", err) |
| 1182 | } |
| 1183 | // The next pass rolls the marked landing back (Plan 4's decision, J10), |
| 1184 | // installing nothing. |
| 1185 | h.land(t) |
| 1186 | rolledBack(t, h, provision.RollbackConfigChanged, provision.LandConfigured) |
| 1187 | if _, err := os.Stat(provision.DefaultDBPath(h.tdir, p1Host)); !errors.Is(err, os.ErrNotExist) { |
| 1188 | t.Fatalf("the rollback installed the database: %v", err) |
| 1189 | } |
| 1190 | |
| 1191 | h2 := newLandingHarness(t) |
| 1192 | h2.flip(t) |
| 1193 | h2.release = moves.Check{Reason: t7Waiting} |
| 1194 | h2.land(t) |
| 1195 | if m := h2.read(t, p1Host).Move; m.LandingPhase != provision.LandConfigured || m.Reason != "" { |
| 1196 | t.Fatalf("waiting on the release: %+v", m) |
| 1197 | } |
| 1198 | h2.release = moves.Check{OK: true} |
| 1199 | h2.land(t) |
| 1200 | h2.installed(t) |
| 1201 | } |
| 1202 | |
| 1203 | // landing_deadline passed rolls the landing back (Plan 4's decision, J11), |
| 1204 | // at whatever phase the landing is, restored included. |
| 1205 | func TestLandingDeadlineRollsBack(t *testing.T) { |
| 1206 | t.Parallel() |
| 1207 | h := newLandingHarness(t) |
| 1208 | h.flip(t) |
| 1209 | h.clock.Advance(cutLandingWindow + time.Minute) |
| 1210 | h.land(t) |
| 1211 | rolledBack(t, h, provision.RollbackLandingDeadline, provision.LandChecking) |
| 1212 | |
| 1213 | h2 := newLandingHarness(t) |
| 1214 | h2.flip(t) |
| 1215 | h2.release = moves.Check{Reason: t7NotProven} |
| 1216 | h2.land(t) |
| 1217 | if m := h2.read(t, p1Host).Move; m.LandingPhase != provision.LandConfigured || m.Reason != "" { |
| 1218 | t.Fatalf("waiting on the proof: %+v", m) |
| 1219 | } |
| 1220 | h2.clock.Advance(cutLandingWindow + time.Minute) |
| 1221 | h2.land(t) |
| 1222 | rolledBack(t, h2, provision.RollbackLandingDeadline, provision.LandConfigured) |
| 1223 | |
| 1224 | h3 := newLandingHarness(t) |
| 1225 | h3.flip(t) |
| 1226 | h3.land(t) |
| 1227 | h3.installed(t) |
| 1228 | h3.clock.Advance(cutLandingWindow + time.Minute) |
| 1229 | h3.land(t) |
| 1230 | rolledBack(t, h3, provision.RollbackLandingDeadline, provision.LandRestored) |
| 1231 | } |
| 1232 | |
| 1233 | // Every phase is replay-safe: a crash after each durable side effect, then |
| 1234 | // a fresh worker to restored, yields one object per generation key, one |
| 1235 | // timers replacement, one install manifest, and the same end state. |
| 1236 | func TestLandingReplaysArePhaseIdempotent(t *testing.T) { |
| 1237 | slots := make(chan struct{}, 4) |
| 1238 | points := []string{"after-previous", "after-activate", "after-record-parking", "after-residue-queued", "after-record-building", "after-build", |
| 1239 | "after-record-published", "after-record-local", "after-retired", "after-staged", "after-files-rename", "after-files_placed", "after-extracted", "after-inventory", "after-timers", |
| 1240 | "after-columns", "after-verify", "after-files_manifest", "before-files-gen", "after-files_gen", "after-record-configured", "after-release-ready", "after-db-rename"} |
| 1241 | for _, point := range points { |
| 1242 | t.Run(point, func(t *testing.T) { |
| 1243 | t.Parallel() |
| 1244 | slots <- struct{}{} |
| 1245 | defer func() { <-slots }() |
| 1246 | h := newLandingHarness(t) |
| 1247 | h.flip(t) |
| 1248 | h.earlierStay(t) |
| 1249 | h.residue(t) |
| 1250 | h.landAt = point |
| 1251 | h.crashLand(t) |
| 1252 | h.freshTarget() |
| 1253 | h.land(t) |
| 1254 | h.installed(t) |
| 1255 | if len(h.replaced) != 1 { |
| 1256 | t.Fatalf("timers replaced %d times", len(h.replaced)) |
| 1257 | } |
| 1258 | prefix := moves.GenerationPrefix(cutRoot, p1ID) |
| 1259 | n := 0 |
| 1260 | for k := range h.f.buckets["data-uk"] { |
| 1261 | if strings.HasPrefix(k, prefix) { |
| 1262 | n++ |
| 1263 | } |
| 1264 | } |
| 1265 | if n != 4 { |
| 1266 | t.Fatalf("%d generation objects", n) |
| 1267 | } |
| 1268 | c := h.claim(t) |
| 1269 | gens := 0 |
| 1270 | for _, o := range c.Obligations { |
| 1271 | switch o.Kind { |
| 1272 | case "park-generation": |
| 1273 | gens++ |
| 1274 | default: |
| 1275 | t.Fatalf("obligation left: %+v", o) |
| 1276 | } |
| 1277 | } |
| 1278 | if gens != 2 { |
| 1279 | t.Fatalf("%d residue chores", gens) |
| 1280 | } |
| 1281 | for p, s := range c.Inventory["uk-box"].Paths { |
| 1282 | if s != c.Holder.Stamp { |
| 1283 | t.Fatalf("inventory %s at %d", p, s) |
| 1284 | } |
| 1285 | } |
| 1286 | }) |
| 1287 | } |
| 1288 | } |
| 1289 | |
| 1290 | // landedPreparedManifest reads the prepared manifest the landing's local |
| 1291 | // phase journalled (Task 6, C8): the binding from the op journal, the |
| 1292 | // manifest at its key in the target's bucket, decoded. |
| 1293 | func landedPreparedManifest(t *testing.T, h *landingHarness) (preparedManifest, hostfiles.Manifest, string) { |
| 1294 | t.Helper() |
| 1295 | raw, ok := h.landJournal(t).Steps[landStepFilesManifest] |
| 1296 | if !ok || raw == landingFilesManifestSkipped { |
| 1297 | t.Fatalf("files_manifest: %q", raw) |
| 1298 | } |
| 1299 | var pm preparedManifest |
| 1300 | if err := json.Unmarshal([]byte(raw), &pm); err != nil { |
| 1301 | t.Fatalf("files_manifest %q: %v", raw, err) |
| 1302 | } |
| 1303 | obj, ok := h.f.buckets[h.dstBucket][pm.Key] |
| 1304 | if !ok { |
| 1305 | t.Fatalf("no manifest at %s in the target's bucket", pm.Key) |
| 1306 | } |
| 1307 | if got := sha([]byte(obj.body)); got != pm.SHA256 { |
| 1308 | t.Fatalf("the manifest at %s is not the journalled bytes: %s", pm.Key, got) |
| 1309 | } |
| 1310 | var m hostfiles.Manifest |
| 1311 | if err := json.Unmarshal([]byte(obj.body), &m); err != nil { |
| 1312 | t.Fatal(err) |
| 1313 | } |
| 1314 | return pm, m, obj.body |
| 1315 | } |
| 1316 | |
| 1317 | // The prepared files manifest (spec rev 7.22 §7.6, R19-1; C8): local |
| 1318 | // writes it under files/g-<op>/ in the location, bound to {location, |
| 1319 | // g-<op>, op, stamp} in the op journal, at sequence 0, while |
| 1320 | // holder.generation is still the source's. It carries the placed files, |
| 1321 | // <host>.env and the wake state the landing installed. |
| 1322 | func TestLocalBindsThePreparedManifestToTheNewGeneration(t *testing.T) { |
| 1323 | t.Parallel() |
| 1324 | h := newLandingHarness(t) |
| 1325 | h.flip(t) |
| 1326 | h.land(t) |
| 1327 | h.installed(t) |
| 1328 | c := h.claim(t) |
| 1329 | if c.Holder.Generation != "cur" { |
| 1330 | t.Fatalf("holder.generation moved before the probe: %q", c.Holder.Generation) |
| 1331 | } |
| 1332 | pm, m, _ := landedPreparedManifest(t, h) |
| 1333 | prefix := movesFilesPrefix(c.Holder.Location, p1ID) |
| 1334 | if pm.Key != hostfiles.ManifestKey(prefix) || pm.Op != p1ID || pm.Stamp != c.Holder.Stamp { |
| 1335 | t.Fatalf("binding: %+v, want key %s op %s stamp %d", pm, hostfiles.ManifestKey(prefix), p1ID, c.Holder.Stamp) |
| 1336 | } |
| 1337 | vs, err := h.vs.ListVersions(context.Background(), "data-uk", pm.Key) |
| 1338 | if err != nil { |
| 1339 | t.Fatal(err) |
| 1340 | } |
| 1341 | exact := "" |
| 1342 | for _, v := range vs { |
| 1343 | if v.Key == pm.Key && v.IsLatest { |
| 1344 | exact = v.VersionID |
| 1345 | } |
| 1346 | } |
| 1347 | if pm.Version == "" || pm.Version != exact { |
| 1348 | t.Fatalf("the binding's version %q is not the manifest's exact version %q", pm.Version, exact) |
| 1349 | } |
| 1350 | if m.Seq != 0 { |
| 1351 | t.Fatalf("prepared manifest at seq %d", m.Seq) |
| 1352 | } |
| 1353 | if err := hostfiles.ValidateManifest(m, []string{prefix}); err != nil { |
| 1354 | t.Fatal(err) |
| 1355 | } |
| 1356 | dir := filepath.Join(h.tdir, p1Host) |
| 1357 | placed := 0 |
| 1358 | if err := filepath.WalkDir(dir, func(path string, d fs.DirEntry, err error) error { |
| 1359 | if err != nil || d.IsDir() { |
| 1360 | return err |
| 1361 | } |
| 1362 | rel, _ := filepath.Rel(dir, path) |
| 1363 | if rel == archive.MarkerName { |
| 1364 | return nil // .files-gen is written after the manifest, so it is not in it |
| 1365 | } |
| 1366 | body, err := os.ReadFile(path) |
| 1367 | if err != nil { |
| 1368 | return err |
| 1369 | } |
| 1370 | placed++ |
| 1371 | if e := m.Entries[filepath.ToSlash(rel)]; e.SHA256 != sha(body) || e.Seq != 0 { |
| 1372 | t.Errorf("%s: %+v", rel, e) |
| 1373 | } |
| 1374 | return nil |
| 1375 | }); err != nil { |
| 1376 | t.Fatal(err) |
| 1377 | } |
| 1378 | if placed == 0 || len(m.Entries) != placed { |
| 1379 | t.Fatalf("%d entries for %d placed files: %+v", len(m.Entries), placed, m.Entries) |
| 1380 | } |
| 1381 | env, err := os.ReadFile(filepath.Join(h.tdir, p1Host+".env")) |
| 1382 | if err != nil { |
| 1383 | t.Fatal(err) |
| 1384 | } |
| 1385 | if e := m.Platform[hostfiles.EnvItem]; e.SHA256 != sha(env) { |
| 1386 | t.Fatalf("env item %+v is not <host>.env", e) |
| 1387 | } |
| 1388 | wobj, ok := h.f.buckets["data-uk"][m.Platform[hostfiles.WakeItem].Object] |
| 1389 | if !ok { |
| 1390 | t.Fatal("no wake state object") |
| 1391 | } |
| 1392 | var ws hostfiles.WakeState |
| 1393 | if err := json.Unmarshal([]byte(wobj.body), &ws); err != nil { |
| 1394 | t.Fatal(err) |
| 1395 | } |
| 1396 | if ws.V != 1 || ws.Host != p1Host || len(h.replaced) != 1 || len(ws.Timers) != len(h.replaced[0]) || ws.NextDue != h.nextDue[0] || ws.Replication != h.repl[0] { |
| 1397 | t.Fatalf("wake state %+v, timers replaced %v, next_due %v, replication %v", ws, h.replaced, h.nextDue, h.repl) |
| 1398 | } |
| 1399 | } |
| 1400 | |
| 1401 | // A journal that records the step as skipped (written before C8 existed) |
| 1402 | // counts as not done (ruling Q3 item 2): a replay builds the manifest and |
| 1403 | // the landing goes on with the binding in its journal. |
| 1404 | func TestSkippedManifestStepIsBuiltOnReplay(t *testing.T) { |
| 1405 | t.Parallel() |
| 1406 | h := newLandingHarness(t) |
| 1407 | h.flip(t) |
| 1408 | h.landAt = "after-verify" |
| 1409 | h.crashLand(t) |
| 1410 | j := h.landJournal(t) |
| 1411 | j.Mark(landStepFilesManifest, landingFilesManifestSkipped) |
| 1412 | if err := j.Commit(h.tdir); err != nil { |
| 1413 | t.Fatal(err) |
| 1414 | } |
| 1415 | if got := h.landJournal(t).Steps[landStepFilesManifest]; got != landingFilesManifestSkipped { |
| 1416 | t.Fatalf("the skip was not planted: %q", got) |
| 1417 | } |
| 1418 | h.freshTarget() |
| 1419 | h.land(t) |
| 1420 | h.installed(t) |
| 1421 | pm, m, _ := landedPreparedManifest(t, h) |
| 1422 | if pm.Op != p1ID || m.Seq != 0 { |
| 1423 | t.Fatalf("replay did not build the manifest: %+v seq %d", pm, m.Seq) |
| 1424 | } |
| 1425 | } |
| 1426 | |
| 1427 | // A journalled files_manifest value that is not this install's binding |
| 1428 | // (fix round 1, finding 3): empty, malformed, or a binding for another op, |
| 1429 | // stamp or key. The install neither skips the build over it nor rebuilds |
| 1430 | // over it: the replay refuses, the record stays at local, nothing is |
| 1431 | // written under the files prefix and the journal keeps the value for an |
| 1432 | // operator to see. |
| 1433 | func TestLocalRefusesADamagedFilesManifestJournal(t *testing.T) { |
| 1434 | t.Parallel() |
| 1435 | good := func(h *landingHarness) preparedManifest { |
| 1436 | c := h.claim(t) |
| 1437 | return preparedManifest{Key: hostfiles.ManifestKey(movesFilesPrefix(c.Holder.Location, p1ID)), SHA256: sha([]byte("x")), Version: "v1", Op: p1ID, Stamp: c.Holder.Stamp} |
| 1438 | } |
| 1439 | for name, plant := range map[string]func(h *landingHarness) string{ |
| 1440 | "empty": func(*landingHarness) string { return "" }, |
| 1441 | "malformed": func(*landingHarness) string { return `{"key":` }, |
| 1442 | "another op": func(h *landingHarness) string { pm := good(h); pm.Op = "mv_other"; return string(mustJSON(t, pm)) }, |
| 1443 | "another stamp": func(h *landingHarness) string { pm := good(h); pm.Stamp++; return string(mustJSON(t, pm)) }, |
| 1444 | "another key": func(h *landingHarness) string { |
| 1445 | pm := good(h) |
| 1446 | pm.Key = hostfiles.ManifestKey(movesFilesPrefix(h.claim(t).Holder.Location, "mv_other")) |
| 1447 | return string(mustJSON(t, pm)) |
| 1448 | }, |
| 1449 | "no version": func(h *landingHarness) string { pm := good(h); pm.Version = ""; return string(mustJSON(t, pm)) }, |
| 1450 | } { |
| 1451 | t.Run(name, func(t *testing.T) { |
| 1452 | h := newLandingHarness(t) |
| 1453 | h.flip(t) |
| 1454 | h.landAt = "after-verify" |
| 1455 | h.crashLand(t) |
| 1456 | value := plant(h) |
| 1457 | j := h.landJournal(t) |
| 1458 | j.Mark(landStepFilesManifest, value) |
| 1459 | if err := j.Commit(h.tdir); err != nil { |
| 1460 | t.Fatal(err) |
| 1461 | } |
| 1462 | h.freshTarget() |
| 1463 | h.land(t) |
| 1464 | if m := h.read(t, p1Host).Move; m.LandingPhase != provision.LandLocal || m.Reason != "" { |
| 1465 | t.Fatalf("the record moved over a damaged journal: %+v", m) |
| 1466 | } |
| 1467 | j = h.landJournal(t) |
| 1468 | if got := j.Steps[landStepFilesManifest]; got != value { |
| 1469 | t.Fatalf("the journal value changed: %q", got) |
| 1470 | } |
| 1471 | if _, done := j.Steps[landStepFilesGen]; done { |
| 1472 | t.Fatal("the install went on past the damaged step") |
| 1473 | } |
| 1474 | for k := range h.f.buckets["data-uk"] { |
| 1475 | if strings.HasPrefix(k, cutRoot+"/files/") { |
| 1476 | t.Fatalf("written over a damaged journal: %s", k) |
| 1477 | } |
| 1478 | } |
| 1479 | }) |
| 1480 | } |
| 1481 | } |
| 1482 | |
| 1483 | // The placed tree or the placed env file vanishes after placement and |
| 1484 | // before the manifest is built (fix round 1, finding 4): the install |
| 1485 | // refuses rather than publishing an empty or partial image. The record |
| 1486 | // stays at local, nothing is written under the files prefix and the step |
| 1487 | // is not journalled. |
| 1488 | func TestLocalRefusesAVanishedTreeBeforeTheManifest(t *testing.T) { |
| 1489 | t.Parallel() |
| 1490 | for name, vanish := range map[string]func(h *landingHarness) error{ |
| 1491 | "the files directory": func(h *landingHarness) error { return os.RemoveAll(filepath.Join(h.tdir, p1Host)) }, |
| 1492 | "the env file": func(h *landingHarness) error { return os.Remove(filepath.Join(h.tdir, p1Host+".env")) }, |
| 1493 | } { |
| 1494 | t.Run(name, func(t *testing.T) { |
| 1495 | h := newLandingHarness(t) |
| 1496 | h.flip(t) |
| 1497 | h.landAt = "after-verify" |
| 1498 | h.crashLand(t) |
| 1499 | if err := vanish(h); err != nil { |
| 1500 | t.Fatal(err) |
| 1501 | } |
| 1502 | h.freshTarget() |
| 1503 | h.land(t) |
| 1504 | if m := h.read(t, p1Host).Move; m.LandingPhase != provision.LandLocal || m.Reason != "" { |
| 1505 | t.Fatalf("the record moved: %+v", m) |
| 1506 | } |
| 1507 | if _, ok := h.landJournal(t).Steps[landStepFilesManifest]; ok { |
| 1508 | t.Fatal("the step was journalled over a vanished input") |
| 1509 | } |
| 1510 | for k := range h.f.buckets["data-uk"] { |
| 1511 | if strings.HasPrefix(k, cutRoot+"/files/") { |
| 1512 | t.Fatalf("written over a vanished input: %s", k) |
| 1513 | } |
| 1514 | } |
| 1515 | }) |
| 1516 | } |
| 1517 | } |
| 1518 | |
| 1519 | // A cross-region landing's prepared manifest names destination-local |
| 1520 | // objects it materialized from the sealed archive (claim P6-D3): every |
| 1521 | // object lies under the manifest's own prefix in the target's bucket, and |
| 1522 | // nothing is written under the source's files prefix. |
| 1523 | func TestCrossRegionLandingManifestNamesDestinationObjects(t *testing.T) { |
| 1524 | t.Parallel() |
| 1525 | h := newLandingHarness(t) |
| 1526 | h.flip(t) |
| 1527 | h.land(t) |
| 1528 | h.installed(t) |
| 1529 | c := h.claim(t) |
| 1530 | pm, m, _ := landedPreparedManifest(t, h) |
| 1531 | prefix := movesFilesPrefix(c.Holder.Location, p1ID) |
| 1532 | if !strings.HasPrefix(pm.Key, prefix) { |
| 1533 | t.Fatalf("manifest key %s is not under %s", pm.Key, prefix) |
| 1534 | } |
| 1535 | for _, entries := range []map[string]hostfiles.Entry{m.Entries, m.Platform} { |
| 1536 | for name, e := range entries { |
| 1537 | if !strings.HasPrefix(e.Object, prefix+"objects/") { |
| 1538 | t.Fatalf("%s names %s, outside the destination prefix %s", name, e.Object, prefix) |
| 1539 | } |
| 1540 | obj, ok := h.f.buckets["data-uk"][e.Object] |
| 1541 | if !ok || sha([]byte(obj.body)) != e.SHA256 { |
| 1542 | t.Fatalf("%s names %s, which the target's bucket does not hold at that digest", name, e.Object) |
| 1543 | } |
| 1544 | } |
| 1545 | } |
| 1546 | for k := range h.f.buckets["data-ie"] { |
| 1547 | if strings.Contains(k, "/files/") { |
| 1548 | t.Fatalf("the landing wrote into the source region's files: %s", k) |
| 1549 | } |
| 1550 | } |
| 1551 | } |
| 1552 | |
| 1553 | // A files manifest already at P that is not this install's (another |
| 1554 | // writer's, or a damaged one) is not the landing's to replace: local |
| 1555 | // refuses and the move rolls back (RollbackManifestOccupied), as it does |
| 1556 | // for an occupied files directory. |
| 1557 | func TestLocalRefusesAForeignPreparedManifest(t *testing.T) { |
| 1558 | t.Parallel() |
| 1559 | h := newLandingHarness(t) |
| 1560 | h.flip(t) |
| 1561 | c := h.claim(t) |
| 1562 | key := hostfiles.ManifestKey(movesFilesPrefix(c.Holder.Location, p1ID)) |
| 1563 | body := `{"v":2,"seq":0,"entries":{"x.txt":{"object":"` + movesFilesPrefix(c.Holder.Location, p1ID) + `objects/0000000000000000-` + sha([]byte("x")) + `","sha256":"` + sha([]byte("x")) + `","size":1,"seq":0}},"platform":{}}` |
| 1564 | h.f.store("data-uk", key, fakeMoveObject{lease.MirrorObject{ETag: "foreign", Size: int64(len(body))}, body}) |
| 1565 | h.f.setVersion("data-uk", key, h.f.nextVersion()) |
| 1566 | h.land(t) |
| 1567 | if m := h.read(t, p1Host).Move; m.Reason != provision.RollbackManifestOccupied || m.LandingPhase != provision.LandLocal { |
| 1568 | t.Fatalf("not refused for %q at local: %+v", provision.RollbackManifestOccupied, m) |
| 1569 | } |
| 1570 | if got := h.f.buckets["data-uk"][key].body; got != body { |
| 1571 | t.Fatal("the foreign manifest was replaced") |
| 1572 | } |
| 1573 | if _, ok := h.landJournal(t).Steps[landStepFilesManifest]; ok { |
| 1574 | t.Fatal("the refused step was journalled as done") |
| 1575 | } |
| 1576 | } |
| 1577 | |
| 1578 | // moveAuthority hands the landing's op to exec in the claim, as a |
| 1579 | // replacement or a return of the executor would. |
| 1580 | func moveAuthority(t *testing.T, h *landingHarness, exec string) { |
| 1581 | t.Helper() |
| 1582 | obs, err := hostclaim.Read(context.Background(), h.s, p1Host) |
| 1583 | if err != nil { |
| 1584 | t.Fatal(err) |
| 1585 | } |
| 1586 | if _, err := hostclaim.Update(context.Background(), h.s, obs, func(c *hostclaim.Claim) error { |
| 1587 | c.Op.Exec = exec |
| 1588 | if !slices.Contains(c.Op.Participants, exec) { |
| 1589 | c.Op.Participants = append(c.Op.Participants, exec) |
| 1590 | } |
| 1591 | return nil |
| 1592 | }); err != nil { |
| 1593 | t.Fatal(err) |
| 1594 | } |
| 1595 | } |
| 1596 | |
| 1597 | // The prepared build re-checks the installation's authority (the op at |
| 1598 | // phase local with this box executing, the holder at the captured stamp) |
| 1599 | // immediately before every bucket write, retries included (fix round 1, |
| 1600 | // finding 1; region HA's authorityStore). Authority moves to another |
| 1601 | // executor between the first object upload and the next write: the stale |
| 1602 | // writer writes nothing more, no manifest exists, and the step is not |
| 1603 | // journalled. Once authority returns, the installer completes, reusing |
| 1604 | // the one object the stale writer left. |
| 1605 | func TestPreparedBuildRechecksAuthorityBeforeEveryWrite(t *testing.T) { |
| 1606 | t.Parallel() |
| 1607 | h := newLandingHarness(t) |
| 1608 | h.flip(t) |
| 1609 | c := h.claim(t) |
| 1610 | prefix := movesFilesPrefix(c.Holder.Location, p1ID) |
| 1611 | moved := false |
| 1612 | h.vs.deny = func(op, bucket, key string) error { |
| 1613 | if op == "put-if-absent" && strings.HasPrefix(key, prefix+"objects/") && !moved { |
| 1614 | moved = true |
| 1615 | moveAuthority(t, h, "other-box") |
| 1616 | } |
| 1617 | return nil |
| 1618 | } |
| 1619 | h.land(t) |
| 1620 | if !moved { |
| 1621 | t.Fatal("the build never uploaded an object") |
| 1622 | } |
| 1623 | objects := 0 |
| 1624 | for k := range h.f.buckets["data-uk"] { |
| 1625 | if strings.HasPrefix(k, prefix+"objects/") { |
| 1626 | objects++ |
| 1627 | } |
| 1628 | if k == hostfiles.ManifestKey(prefix) { |
| 1629 | t.Fatal("the stale writer published the manifest") |
| 1630 | } |
| 1631 | } |
| 1632 | if objects != 1 { |
| 1633 | t.Fatalf("%d objects after authority moved, want the one uploaded before it", objects) |
| 1634 | } |
| 1635 | if _, ok := h.landJournal(t).Steps[landStepFilesManifest]; ok { |
| 1636 | t.Fatal("the step was journalled without the manifest") |
| 1637 | } |
| 1638 | if m := h.read(t, p1Host).Move; m.LandingPhase != provision.LandLocal || m.Reason != "" { |
| 1639 | t.Fatalf("the record moved: %+v", m) |
| 1640 | } |
| 1641 | moveAuthority(t, h, "uk-box") |
| 1642 | h.vs.deny = nil |
| 1643 | h.freshTarget() |
| 1644 | h.land(t) |
| 1645 | h.installed(t) |
| 1646 | pm, m, _ := landedPreparedManifest(t, h) |
| 1647 | if pm.Op != p1ID || m.Seq != 0 { |
| 1648 | t.Fatalf("the authorized installer did not complete: %+v", pm) |
| 1649 | } |
| 1650 | vs, err := h.vs.ListVersions(context.Background(), "data-uk", prefix+"objects/") |
| 1651 | if err != nil { |
| 1652 | t.Fatal(err) |
| 1653 | } |
| 1654 | seen := map[string]int{} |
| 1655 | for _, v := range vs { |
| 1656 | seen[v.Key]++ |
| 1657 | } |
| 1658 | for k, n := range seen { |
| 1659 | if n != 1 { |
| 1660 | t.Fatalf("%s has %d versions: the replay did not reuse it", k, n) |
| 1661 | } |
| 1662 | } |
| 1663 | } |
| 1664 | |
| 1665 | // Nothing is written under the location's files prefix before the |
| 1666 | // files_manifest step: the manifest follows the placed files, the env |
| 1667 | // file, the timers and the columns (the step order of local, §7.6). |
| 1668 | func TestLocalWritesNoFilesManifestBeforeItsStep(t *testing.T) { |
| 1669 | t.Parallel() |
| 1670 | h := newLandingHarness(t) |
| 1671 | h.flip(t) |
| 1672 | h.landAt = "after-verify" |
| 1673 | h.crashLand(t) |
| 1674 | for k := range h.f.buckets["data-uk"] { |
| 1675 | if strings.HasPrefix(k, cutRoot+"/files/") { |
| 1676 | t.Fatalf("a files manifest was written before the files_manifest step: %s", k) |
| 1677 | } |
| 1678 | } |
| 1679 | } |
| 1680 | |
| 1681 | // Ruling G17: with the park engine's hooks unwired, as armPhase1 leaves |
| 1682 | // them, the landing is not armed: the pass logs and writes nothing. |
| 1683 | func TestLandingWaitsUnarmedInProduction(t *testing.T) { |
| 1684 | t.Parallel() |
| 1685 | h := newLandingHarness(t) |
| 1686 | h.flip(t) |
| 1687 | h.lw = &movesWorker{mirrors: map[string]lease.MirrorStore{}} |
| 1688 | c := h.landingCfg(t) |
| 1689 | c.ParkLeases = nil |
| 1690 | h.lw.cfg = c |
| 1691 | if landingArmed(c) { |
| 1692 | t.Fatal("armed without park leases") |
| 1693 | } |
| 1694 | if err := movesPass(context.Background(), h.s, h.lw); err != nil { |
| 1695 | t.Fatal(err) |
| 1696 | } |
| 1697 | if m := h.read(t, p1Host).Move; m.LandingPhase != provision.LandChecking { |
| 1698 | t.Fatalf("an unarmed landing advanced: %+v", m) |
| 1699 | } |
| 1700 | if _, err := hostguard.Capture(h.tdir, p1Host); !errors.Is(err, hostguard.ErrNoRecord) { |
| 1701 | t.Fatalf("an unarmed landing wrote the installation record: %v", err) |
| 1702 | } |
| 1703 | // Production's own wiring: everything but the park engine's hooks. |
| 1704 | var prod moveCfg |
| 1705 | hib := activator.Config{Bucket: "data-uk", Region: "eu-west-2", Prefix: "instances"} |
| 1706 | armPhase1(&prod, &movesWorker{mirrors: map[string]lease.MirrorStore{}}, hib, h.tdir, nil, nil, nil, hostwrite.Config{}, func(string) string { return "" }) |
| 1707 | if prod.Guard.DataDir != h.tdir || prod.Local.DataDir != h.tdir || prod.OwnLease == nil || prod.RenderHeld == nil || prod.RemoveVerifyEntry == nil || prod.WriteGen == nil { |
| 1708 | t.Fatalf("production wiring: %+v", prod) |
| 1709 | } |
| 1710 | if prod.ParkLeases != nil || prod.ParkView != nil || prod.ParkAuthorize != nil || landingArmed(prod) { |
| 1711 | t.Fatal("production armed the landing") |
| 1712 | } |
| 1713 | } |
| 1714 | |
| 1715 | // The generation classifier the stand-in view and parking share. |
| 1716 | func TestLandingGenerationOf(t *testing.T) { |
| 1717 | loc := "instances/h" |
| 1718 | cases := map[string]struct { |
| 1719 | gen string |
| 1720 | ok bool |
| 1721 | }{ |
| 1722 | loc + "/g-old/0000/x.ltx": {"old", true}, |
| 1723 | loc + "/g-mv_1/install.json": {"mv_1", true}, |
| 1724 | loc + "/files/g-mv_1/manifest": {"mv_1", true}, |
| 1725 | loc + "/0000/x.ltx": {"", true}, |
| 1726 | loc + "/0009/x.ltx": {"", true}, |
| 1727 | loc + "/generation.json": {landingMarkerGeneration, true}, |
| 1728 | loc + "/g-/x": {"", false}, |
| 1729 | loc + "/stray.json": {"", false}, |
| 1730 | loc + "/zzzz/x.ltx": {"", false}, |
| 1731 | "instances/other/0000/x.ltx": {"", false}, |
| 1732 | loc + "/" + strings.Repeat("a", 64) + "/0000/x.ltx": {"", false}, |
| 1733 | } |
| 1734 | for key, want := range cases { |
| 1735 | gen, ok := landingGenerationOf(loc, key) |
| 1736 | if gen != want.gen || ok != want.ok { |
| 1737 | t.Errorf("%s: %q %v, want %q %v", key, gen, ok, want.gen, want.ok) |
| 1738 | } |
| 1739 | } |
| 1740 | } |
| 1741 | |
| 1742 | // The target report pass keeps a landing's refusal and verdict live (G26). |
| 1743 | func TestTargetLiveSetCoversTheCutAndTheLanding(t *testing.T) { |
| 1744 | t.Parallel() |
| 1745 | h := newLandingHarness(t) |
| 1746 | h.flip(t) |
| 1747 | w := &movesWorker{cfg: h.landingCfg(t), mirrors: map[string]lease.MirrorStore{}, archives: map[string]verifiedArchive{p1ID: {}, "mv_other": {}}} |
| 1748 | recs, err := provision.ListAllRecords(context.Background(), h.s) |
| 1749 | if err != nil { |
| 1750 | t.Fatal(err) |
| 1751 | } |
| 1752 | fleet, err := loadMoveFleet(context.Background(), h.s) |
| 1753 | if err != nil { |
| 1754 | t.Fatal(err) |
| 1755 | } |
| 1756 | w.targetReportPass(context.Background(), h.s, recs, fleet) |
| 1757 | if _, ok := w.archives[p1ID]; !ok { |
| 1758 | t.Fatal("the landing's verdict was evicted") |
| 1759 | } |
| 1760 | if _, ok := w.archives["mv_other"]; ok { |
| 1761 | t.Fatal("a finished move's verdict was kept") |
| 1762 | } |
| 1763 | } |
| 1764 | |
| 1765 | // Fix round 1, item 1: the pass catches the claim up at restored before it |
| 1766 | // returns. A crash after the record's restored CAS, or a following claim |
| 1767 | // CAS that fails, leaves the claim behind; the next pass (a fresh worker) |
| 1768 | // converges through movesPass alone. |
| 1769 | func TestLandingRestoredCatchesTheClaimUp(t *testing.T) { |
| 1770 | t.Parallel() |
| 1771 | t.Run("crash after the record's restored", func(t *testing.T) { |
| 1772 | h := newLandingHarness(t) |
| 1773 | h.flip(t) |
| 1774 | h.landAt = "after-record-restored" |
| 1775 | h.crashLand(t) |
| 1776 | if c := h.claim(t); c.Op.Phase != provision.LandConfigured { |
| 1777 | t.Fatalf("fixture: claim %+v", c.Op) |
| 1778 | } |
| 1779 | h.freshTarget() |
| 1780 | h.land(t) |
| 1781 | h.installed(t) |
| 1782 | }) |
| 1783 | t.Run("the following claim CAS fails", func(t *testing.T) { |
| 1784 | h := newLandingHarness(t) |
| 1785 | h.flip(t) |
| 1786 | // At the boundary the claim is moved two steps back, so the mirror's |
| 1787 | // SetMovePhase(configured, restored) is refused and the claim stays |
| 1788 | // behind; the pass must not fail, and the next pass walks it up. |
| 1789 | h.landFault = func(point string) error { |
| 1790 | if point == "after-record-restored" { |
| 1791 | rewriteClaimOp(t, h.s, func(op *hostclaim.Op) { op.Phase = provision.LandBuilding }) |
| 1792 | } |
| 1793 | return nil |
| 1794 | } |
| 1795 | h.land(t) |
| 1796 | h.landFault = nil |
| 1797 | if !containsString(h.landHits, "after-record-restored") { |
| 1798 | t.Fatalf("the boundary was never reached: %v", h.landHits) |
| 1799 | } |
| 1800 | // The refused mirror is logged, and the same pass's restored step |
| 1801 | // walks the claim from building to restored before it returns. |
| 1802 | h.installed(t) |
| 1803 | h.freshTarget() |
| 1804 | h.land(t) |
| 1805 | h.installed(t) |
| 1806 | }) |
| 1807 | } |
| 1808 | |
| 1809 | // Fix round 1, item 2: a build that does not verify marks the rollback and |
| 1810 | // leaves the phase at configured: no transition to restored, no digest, no |
| 1811 | // database installed, no marker. |
| 1812 | func TestLandingCorruptBuildMarksRollback(t *testing.T) { |
| 1813 | t.Parallel() |
| 1814 | h := newLandingHarness(t) |
| 1815 | h.flip(t) |
| 1816 | h.landFault = func(point string) error { |
| 1817 | if point == "before-db-verify" { |
| 1818 | f, err := os.OpenFile(filepath.Join(h.tdir, ".moves", p1Host, p1ID, "build", p1Host+".db"), os.O_APPEND|os.O_WRONLY, 0o600) |
| 1819 | if err != nil { |
| 1820 | t.Fatal(err) |
| 1821 | } |
| 1822 | if _, err := f.Write([]byte("corrupt")); err != nil { |
| 1823 | t.Fatal(err) |
| 1824 | } |
| 1825 | f.Close() |
| 1826 | } |
| 1827 | return nil |
| 1828 | } |
| 1829 | h.land(t) |
| 1830 | if !containsString(h.landHits, "before-db-verify") { |
| 1831 | t.Fatalf("the verify point was never reached: %v", h.landHits) |
| 1832 | } |
| 1833 | m := h.read(t, p1Host).Move |
| 1834 | if m.Reason != provision.RollbackInstallMismatch || m.LandingPhase != provision.LandConfigured || m.State != provision.MoveLanding { |
| 1835 | t.Fatalf("after the corrupt build: %+v", m) |
| 1836 | } |
| 1837 | if m.Landing != nil && m.Landing.InstalledDBSHA256 != "" { |
| 1838 | t.Fatalf("a digest was recorded: %+v", m.Landing) |
| 1839 | } |
| 1840 | dbPath := provision.DefaultDBPath(h.tdir, p1Host) |
| 1841 | for _, p := range []string{dbPath, dbPath + ".gen"} { |
| 1842 | if _, err := os.Stat(p); !errors.Is(err, os.ErrNotExist) { |
| 1843 | t.Fatalf("%s exists after a refused install: %v", p, err) |
| 1844 | } |
| 1845 | } |
| 1846 | if c := h.claim(t); c.Op.Phase != provision.LandConfigured { |
| 1847 | t.Fatalf("claim: %+v", c.Op) |
| 1848 | } |
| 1849 | } |
| 1850 | |
| 1851 | // Fix round 1, item 3: the generation is written only under live lease |
| 1852 | // authority. A rebuild (the scratch lost after published released the |
| 1853 | // lease) takes the location's lease first, and waits while another holder |
| 1854 | // has it, writing nothing; a lease lost during the build stops the build |
| 1855 | // at the next generation write, and the replay completes. |
| 1856 | func TestLandingRebuildNeedsTheLease(t *testing.T) { |
| 1857 | t.Parallel() |
| 1858 | ctx := context.Background() |
| 1859 | h := newLandingHarness(t) |
| 1860 | h.flip(t) |
| 1861 | h.landAt = "after-record-configured" |
| 1862 | h.crashLand(t) |
| 1863 | if err := os.RemoveAll(filepath.Join(h.tdir, ".moves", p1Host, p1ID, "build")); err != nil { |
| 1864 | t.Fatal(err) |
| 1865 | } |
| 1866 | h.freshTarget() |
| 1867 | other, err := h.leases.Acquire(ctx, "data-uk", cutRoot) |
| 1868 | if err != nil { |
| 1869 | t.Fatal(err) |
| 1870 | } |
| 1871 | prefix := moves.GenerationPrefix(cutRoot, p1ID) |
| 1872 | before := len(h.vs.callsOn(prefix)) |
| 1873 | h.land(t) |
| 1874 | if m := h.read(t, p1Host).Move; m.LandingPhase != provision.LandConfigured || m.Reason != "" { |
| 1875 | t.Fatalf("installed without the lease: %+v", m) |
| 1876 | } |
| 1877 | if got := h.vs.callsOn(prefix); len(got) != before { |
| 1878 | t.Fatalf("generation requests while another holder had the lease: %v", got[before:]) |
| 1879 | } |
| 1880 | if _, err := os.Stat(provision.DefaultDBPath(h.tdir, p1Host)); !errors.Is(err, os.ErrNotExist) { |
| 1881 | t.Fatalf("a database was installed without the lease: %v", err) |
| 1882 | } |
| 1883 | if err := other.Release(ctx); err != nil { |
| 1884 | t.Fatal(err) |
| 1885 | } |
| 1886 | h.land(t) |
| 1887 | h.installed(t) |
| 1888 | } |
| 1889 | |
| 1890 | func TestLandingBuildStopsWhenTheLeaseIsLost(t *testing.T) { |
| 1891 | t.Parallel() |
| 1892 | h := newLandingHarness(t) |
| 1893 | h.flip(t) |
| 1894 | prefix := moves.GenerationPrefix(cutRoot, p1ID) |
| 1895 | lost := false |
| 1896 | h.vs.deny = func(op, bucket, key string) error { |
| 1897 | if op == "put-if-absent" && strings.HasPrefix(key, prefix) && !lost { |
| 1898 | lost = true |
| 1899 | h.leases.expire() // the holder went dark between two generation writes |
| 1900 | } |
| 1901 | return nil |
| 1902 | } |
| 1903 | h.land(t) |
| 1904 | h.vs.deny = nil |
| 1905 | if !lost { |
| 1906 | t.Fatal("the build never wrote") |
| 1907 | } |
| 1908 | if m := h.read(t, p1Host).Move; m.LandingPhase != provision.LandBuilding { |
| 1909 | t.Fatalf("the build completed on a lost lease: %+v", m) |
| 1910 | } |
| 1911 | if _, ok := h.f.buckets["data-uk"][moves.InstallManifestKey(prefix)]; ok { |
| 1912 | t.Fatal("install.json was written on a lost lease") |
| 1913 | } |
| 1914 | var written []string |
| 1915 | for k := range h.f.buckets["data-uk"] { |
| 1916 | if strings.HasPrefix(k, prefix) { |
| 1917 | written = append(written, strings.TrimPrefix(k, prefix)) |
| 1918 | } |
| 1919 | } |
| 1920 | if len(written) > 1 { |
| 1921 | t.Fatalf("writes after the lease was lost: %v", written) |
| 1922 | } |
| 1923 | h.freshTarget() |
| 1924 | h.land(t) |
| 1925 | h.installed(t) |
| 1926 | } |
| 1927 | |
| 1928 | // Fix round 1, item 4: landing_deadline is checked again right before the |
| 1929 | // install transition. Past it after the release gate, the rollback is marked |
| 1930 | // and nothing is installed; past it after the install itself, the rollback |
| 1931 | // is marked and restored is never written. |
| 1932 | func TestLandingDeadlineBeforeTheInstallTransition(t *testing.T) { |
| 1933 | t.Parallel() |
| 1934 | for _, point := range []string{"after-release-ready", "after-db-rename"} { |
| 1935 | t.Run(point, func(t *testing.T) { |
| 1936 | h := newLandingHarness(t) |
| 1937 | h.flip(t) |
| 1938 | h.landFault = func(p string) error { |
| 1939 | if p == point { |
| 1940 | h.clock.Advance(cutLandingWindow + time.Minute) |
| 1941 | } |
| 1942 | return nil |
| 1943 | } |
| 1944 | h.land(t) |
| 1945 | m := h.read(t, p1Host).Move |
| 1946 | if m.Reason != provision.RollbackLandingDeadline || m.LandingPhase != provision.LandConfigured { |
| 1947 | t.Fatalf("past the deadline at %s: %+v", point, m) |
| 1948 | } |
| 1949 | if m.Landing != nil && m.Landing.InstalledDBSHA256 != "" { |
| 1950 | t.Fatalf("a digest was recorded: %+v", m.Landing) |
| 1951 | } |
| 1952 | _, err := os.Stat(provision.DefaultDBPath(h.tdir, p1Host)) |
| 1953 | if point == "after-release-ready" && !errors.Is(err, os.ErrNotExist) { |
| 1954 | t.Fatalf("installed past the deadline: %v", err) |
| 1955 | } |
| 1956 | if c := h.claim(t); c.Op.Phase != provision.LandConfigured { |
| 1957 | t.Fatalf("claim: %+v", c.Op) |
| 1958 | } |
| 1959 | }) |
| 1960 | } |
| 1961 | } |
| 1962 | |
| 1963 | // retryingStore is a versioned store whose first generation write blocks |
| 1964 | // like a client retrying a request: the lease lapses while it is in flight, |
| 1965 | // and if nothing cancels the request its retry lands after the loss. |
| 1966 | type retryingStore struct { |
| 1967 | lease.VersionedStore |
| 1968 | h *landingHarness |
| 1969 | prefix string |
| 1970 | once sync.Once |
| 1971 | landed bool |
| 1972 | blocked bool |
| 1973 | } |
| 1974 | |
| 1975 | func (r *retryingStore) PutIfAbsent(ctx context.Context, bucket, key string, body []byte) (string, error) { |
| 1976 | block := false |
| 1977 | if strings.HasPrefix(key, r.prefix) { |
| 1978 | r.once.Do(func() { block = true }) |
| 1979 | } |
| 1980 | if !block { |
| 1981 | return r.VersionedStore.PutIfAbsent(ctx, bucket, key, body) |
| 1982 | } |
| 1983 | r.blocked = true |
| 1984 | r.h.leases.expire() // the holder goes dark while the request is in flight |
| 1985 | select { |
| 1986 | case <-ctx.Done(): |
| 1987 | return "", ctx.Err() |
| 1988 | case <-time.After(1500 * time.Millisecond): |
| 1989 | r.landed = true // the client's retry succeeds after the lease was lost |
| 1990 | return r.VersionedStore.PutIfAbsent(ctx, bucket, key, body) |
| 1991 | } |
| 1992 | } |
| 1993 | |
| 1994 | // Fix round 2: every generation request runs under a context derived from |
| 1995 | // the lease's, so a lease lost while a request is in flight cancels it and |
| 1996 | // its retries; nothing is recorded. The replay builds afresh. |
| 1997 | func TestLandingBuildIsCancelledWhenTheLeaseIsLostMidRequest(t *testing.T) { |
| 1998 | t.Parallel() |
| 1999 | h := newLandingHarness(t) |
| 2000 | h.flip(t) |
| 2001 | prefix := moves.GenerationPrefix(cutRoot, p1ID) |
| 2002 | rs := &retryingStore{VersionedStore: h.vs, h: h, prefix: prefix} |
| 2003 | h.store = rs |
| 2004 | h.land(t) |
| 2005 | h.store = nil |
| 2006 | if !rs.blocked { |
| 2007 | t.Fatal("the build never wrote under the generation") |
| 2008 | } |
| 2009 | if rs.landed { |
| 2010 | t.Fatal("the request landed after the lease was lost: it was not cancelled") |
| 2011 | } |
| 2012 | if m := h.read(t, p1Host).Move; m.LandingPhase != provision.LandBuilding { |
| 2013 | t.Fatalf("record: %+v", m) |
| 2014 | } |
| 2015 | for k := range h.f.buckets["data-uk"] { |
| 2016 | if strings.HasPrefix(k, prefix) { |
| 2017 | t.Fatalf("written on a lost lease: %s", k) |
| 2018 | } |
| 2019 | } |
| 2020 | if j := h.landJournal(t); j.Done(landStepInstall) { |
| 2021 | t.Fatal("the build was recorded on a lost lease") |
| 2022 | } |
| 2023 | h.freshTarget() |
| 2024 | h.land(t) |
| 2025 | h.installed(t) |
| 2026 | } |
| 2027 | |
| 2028 | // openState is the record's state, the claim's phase and the row's status |
| 2029 | // at a step of the open. |
| 2030 | func (h *landingHarness) openState(t *testing.T) string { |
| 2031 | return h.read(t, p1Host).Move.State + "/" + h.claim(t).Op.Phase + "/" + h.route.Status |
| 2032 | } |
| 2033 | |
| 2034 | // errHoldProbed is holdProbed's stop before the commit. |
| 2035 | var errHoldProbed = errors.New("held at probed by the test") |
| 2036 | |
amadan