| 1 | //go:build linux |
| 2 | |
| 3 | package host |
| 4 | |
| 5 | import ( |
| 6 | "context" |
| 7 | "strings" |
| 8 | "sync" |
| 9 | "testing" |
| 10 | "time" |
| 11 | |
| 12 | "github.com/carlosframework/platform/internal/activator" |
| 13 | "github.com/carlosframework/platform/internal/lease" |
| 14 | "github.com/carlosframework/platform/internal/moves" |
| 15 | "github.com/carlosframework/platform/internal/provision" |
| 16 | ) |
| 17 | |
| 18 | // The pre-copy (offline-staging), through offlineRound over the seal world: |
| 19 | // the S3 wire protocol, real litestream replicas, the real lease package and |
| 20 | // the production edges-file writer. A stage round copies the replica into |
| 21 | // the target's staging prefix with the instance serving; the seal's round |
| 22 | // then copies only what changed since; abort from staging removes it all. |
| 23 | |
| 24 | // stagerEdges publishes b's edges file as an agent with an armed, capable |
| 25 | // moves worker does: the hibernate coordinates and the offline capabilities |
| 26 | // (offline-move-1, offline-stage-1). The stager is chosen from these. |
| 27 | func (sw *sealWorld) stagerEdges(b *offBox) { |
| 28 | sw.t.Helper() |
| 29 | ident := edgesFile{Box: b.name, Placement: b.region, DataBucket: sealIE, DataBucketRegion: sealIERegion, |
| 30 | HibernationConfigured: true, ServesSharedPool: true, Capabilities: withOfflineCapability(nil, b.w.cfg.OfflineCapable)} |
| 31 | if err := edgesFileWrite(context.Background(), sw.s, b.reg, ident, "", &b.last, false); err != nil { |
| 32 | sw.t.Fatal(err) |
| 33 | } |
| 34 | } |
| 35 | |
| 36 | // stagePass is sw.pass with the edges file stagerEdges writes. |
| 37 | func (sw *sealWorld) stagePass(b *offBox) error { |
| 38 | sw.t.Helper() |
| 39 | b.reconcile() |
| 40 | sw.stagerEdges(b) |
| 41 | ctx := context.Background() |
| 42 | refreshOfflineView(ctx, sw.s, b.w.cfg.Offline) |
| 43 | return b.w.offlineRound(ctx, sw.s, sw.record()) |
| 44 | } |
| 45 | |
| 46 | func (sw *sealWorld) mustStagePass(b *offBox) { |
| 47 | sw.t.Helper() |
| 48 | if err := sw.stagePass(b); err != nil { |
| 49 | sw.t.Fatal(err) |
| 50 | } |
| 51 | } |
| 52 | |
| 53 | func (sw *sealWorld) stage() { sw.stageBy("ie-a") } |
| 54 | |
| 55 | // stageBy stages with the stager the door pinned. |
| 56 | func (sw *sealWorld) stageBy(stager string) { |
| 57 | sw.move(provision.OfflineStage(offID, "", "uk", "op@example.com", provision.StageTarget{Stager: stager, Bucket: sealUK, Region: sealUKRegion}, sw.clock())) |
| 58 | } |
| 59 | |
| 60 | // restage is a second carlos move offline stage: one more round. |
| 61 | func (sw *sealWorld) restage() { |
| 62 | sw.advance(time.Second) |
| 63 | sw.move(provision.OfflineRestage(offID, sw.clock())) |
| 64 | } |
| 65 | |
| 66 | // copySpy records the source key of every server-side copy the fake S3 |
| 67 | // performs from now on. |
| 68 | type copySpy struct { |
| 69 | mu sync.Mutex |
| 70 | keys []string |
| 71 | } |
| 72 | |
| 73 | func (sw *sealWorld) spyCopies() *copySpy { |
| 74 | c := ©Spy{} |
| 75 | sw.fake.FailOn = func(op, key string) error { |
| 76 | if op == "copy" { |
| 77 | c.mu.Lock() |
| 78 | c.keys = append(c.keys, key) |
| 79 | c.mu.Unlock() |
| 80 | } |
| 81 | return nil |
| 82 | } |
| 83 | sw.t.Cleanup(func() { sw.fake.FailOn = nil }) |
| 84 | return c |
| 85 | } |
| 86 | |
| 87 | func (c *copySpy) take() []string { |
| 88 | c.mu.Lock() |
| 89 | defer c.mu.Unlock() |
| 90 | out := c.keys |
| 91 | c.keys = nil |
| 92 | return out |
| 93 | } |
| 94 | |
| 95 | // replicaRel is the replica's listing by key relative to its prefix, lock |
| 96 | // aside; stagedRel the staging prefix's the same way. |
| 97 | func (sw *sealWorld) replicaRel() map[string]lease.MirrorObject { |
| 98 | prefix := sealPrefix + "/" + offHost + "/" |
| 99 | out := map[string]lease.MirrorObject{} |
| 100 | for k, o := range sw.list(sealIE, prefix) { |
| 101 | if rel := strings.TrimPrefix(k, prefix); rel != "lock.json" { |
| 102 | out[rel] = o |
| 103 | } |
| 104 | } |
| 105 | return out |
| 106 | } |
| 107 | |
| 108 | func (sw *sealWorld) stagedRel() map[string]lease.MirrorObject { |
| 109 | out := map[string]lease.MirrorObject{} |
| 110 | for k, o := range sw.list(sealUK, activator.MovesPrefix(offHost)) { |
| 111 | out[strings.TrimPrefix(k, activator.MovesPrefix(offHost))] = o |
| 112 | } |
| 113 | return out |
| 114 | } |
| 115 | |
| 116 | func sameSizes(t *testing.T, staged, replica map[string]lease.MirrorObject) { |
| 117 | t.Helper() |
| 118 | if len(staged) != len(replica) { |
| 119 | t.Fatalf("staged %d objects, replica %d", len(staged), len(replica)) |
| 120 | } |
| 121 | for rel, o := range replica { |
| 122 | if s, ok := staged[rel]; !ok || s.Size != o.Size { |
| 123 | t.Fatalf("%s: staged %+v (%v), replica %+v", rel, s, ok, o) |
| 124 | } |
| 125 | } |
| 126 | } |
| 127 | |
| 128 | func TestOfflineStageThenStartThenSealSkipsThePreCopy(t *testing.T) { |
| 129 | sw := newSealWorld(t) |
| 130 | ie := sw.ie |
| 131 | ieB := sw.sourceBox("ie-b") |
| 132 | sw.instance(ie) |
| 133 | ctx := context.Background() |
| 134 | // A stale staged prefix an earlier aborted move left behind: a file |
| 135 | // under a replica key with other bytes, and one the replica never had. |
| 136 | // Neither may survive into the seal. |
| 137 | for rel := range sw.replicaRel() { |
| 138 | sw.fake.Put(sealUK, activator.MovesPrefix(offHost)+rel, []byte("stale bytes of an earlier move")) |
| 139 | break |
| 140 | } |
| 141 | sw.fake.Put(sealUK, activator.MovesPrefix(offHost)+"0000/stale-extra.ltx", []byte("stale")) |
| 142 | owner0, expires0, gen0 := sw.lock() |
| 143 | |
| 144 | sw.stage() |
| 145 | spy := sw.spyCopies() |
| 146 | sw.stagerEdges(ie) // both boxes publish the capability, |
| 147 | sw.mustStagePass(ieB) // and ie-b is not the stager: it does nothing |
| 148 | if _, ok := sw.reportIf(moves.OfflineRoleStager, "ie-b"); ok || len(spy.take()) != 0 { |
| 149 | t.Fatal("a box that is not the stager staged") |
| 150 | } |
| 151 | sw.mustStagePass(ie) |
| 152 | first := spy.take() |
| 153 | replica := sw.replicaRel() |
| 154 | sameSizes(t, sw.stagedRel(), replica) |
| 155 | r := sw.report(moves.OfflineRoleStager, "ie-a") |
| 156 | if r.Phase != moves.OfflinePhaseStaged || r.ReplicaObjects != len(replica) || r.StagedObjects != len(replica) || r.RoundBytes == 0 || r.StagedBytes != r.ReplicaBytes || r.StageError != "" { |
| 157 | t.Fatalf("stager report %+v (replica %d objects)", r, len(replica)) |
| 158 | } |
| 159 | if len(first) != len(replica) { |
| 160 | t.Fatalf("the first round made %d copies, want %d", len(first), len(replica)) |
| 161 | } |
| 162 | if staged := sw.record().Move.StagedAt; staged == nil || !r.StageRequest.Equal(*staged) { |
| 163 | t.Fatalf("the round answered %v, the request is %v", r.StageRequest, staged) |
| 164 | } |
| 165 | // One round per request: an idle pass copies, lists and writes nothing. |
| 166 | snap := sw.fake.Snapshot() |
| 167 | sw.mustStagePass(ie) |
| 168 | if len(spy.take()) != 0 || sw.fake.Snapshot() != snap { |
| 169 | t.Fatal("a pass with no new stage request ran a round") |
| 170 | } |
| 171 | if again := sw.report(moves.OfflineRoleStager, "ie-a"); !again.At.Equal(r.At) { |
| 172 | t.Fatal("a pass with no new stage request rewrote the stager report") |
| 173 | } |
| 174 | |
| 175 | // Staging held, fenced and leased nothing: the row serves, no fence, |
| 176 | // the lease is as it was, and a wake takes it. |
| 177 | if rt := ie.row(offHost); rt == nil || rt.HoldOp != "" { |
| 178 | t.Fatalf("row %+v during staging", rt) |
| 179 | } |
| 180 | if ie.svc.Fenced(offHost) { |
| 181 | t.Fatal("fenced during staging") |
| 182 | } |
| 183 | if owner, expires, gen := sw.lock(); owner != owner0 || !expires.Equal(expires0) || gen != gen0 { |
| 184 | t.Fatalf("the lease changed during staging: %s %s %d, was %s %s %d", owner, expires, gen, owner0, expires0, gen0) |
| 185 | } |
| 186 | h, err := acquireAsAWake(ctx) |
| 187 | if err != nil { |
| 188 | t.Fatalf("a wake during staging: %v", err) |
| 189 | } |
| 190 | h.Release(ctx) |
| 191 | |
| 192 | // The instance keeps writing; a second round copies only what is new. |
| 193 | sw.writeRows(ie, sw.dbPath(ie), 20, 40, true) |
| 194 | before := sw.stagedRel() |
| 195 | sw.restage() |
| 196 | sw.mustStagePass(ie) |
| 197 | second := spy.take() |
| 198 | fresh := 0 |
| 199 | for rel, o := range sw.replicaRel() { |
| 200 | if s, ok := before[rel]; !ok || s.Size != o.Size { |
| 201 | fresh++ |
| 202 | } |
| 203 | } |
| 204 | if fresh == 0 || len(second) != fresh { |
| 205 | t.Fatalf("the second round made %d copies, want the %d new objects", len(second), fresh) |
| 206 | } |
| 207 | sameSizes(t, sw.stagedRel(), sw.replicaRel()) |
| 208 | |
| 209 | // A replica file that came and went between rounds (retention): the |
| 210 | // seal's round removes it from staging. |
| 211 | retired := sealPrefix + "/" + offHost + "/scratch/retired.bin" |
| 212 | sw.fake.Put(sealIE, retired, []byte("gone before the seal")) |
| 213 | sw.restage() |
| 214 | sw.mustStagePass(ie) |
| 215 | spy.take() |
| 216 | if _, ok := sw.stagedRel()["scratch/retired.bin"]; !ok { |
| 217 | t.Fatal("fixture: the retired file was not staged") |
| 218 | } |
| 219 | sw.fake.BareDelete(sealIE, retired) |
| 220 | |
| 221 | // Start: the stage's objects are in place, and the seal's round copies |
| 222 | // only what changed since the last stage round. |
| 223 | preStaged := sw.stagedRel() |
| 224 | preReplica := sw.replicaRel() |
| 225 | sw.move(provision.OfflineStartStaged(offID, "op@example.com", sw.clock())) |
| 226 | ie.tick() |
| 227 | ieB.tick() |
| 228 | sw.move(provision.OfflineSeal(offID, "ie-a", []string{"ie-a", "ie-b"})) |
| 229 | spy.take() |
| 230 | sw.mustPass(ie) |
| 231 | if st := sw.record().Move.State; st != provision.OfflineSealed { |
| 232 | t.Fatalf("state %s, want sealed", st) |
| 233 | } |
| 234 | sealCopies := spy.take() |
| 235 | // Litestream's compaction decides how many files the checkpoint adds, |
| 236 | // so the count of the seal's copies is not fixed. What is: at least one |
| 237 | // object the stage copied is still in the replica, unchanged, and the |
| 238 | // seal copied none of those (the loop below). |
| 239 | kept := 0 |
| 240 | for rel, o := range sw.replicaRel() { |
| 241 | if s, ok := preStaged[rel]; ok && s.Size == o.Size && preReplica[rel].ETag == o.ETag { |
| 242 | kept++ |
| 243 | } |
| 244 | } |
| 245 | if kept == 0 { |
| 246 | t.Fatal("fixture: no pre-copied object survived to the seal, so nothing shows a skip") |
| 247 | } |
| 248 | t.Logf("the seal copied %d of %d objects; %d were pre-copied and skipped", len(sealCopies), len(sw.replicaRel()), kept) |
| 249 | for _, k := range sealCopies { |
| 250 | rel := strings.TrimPrefix(k, sealPrefix+"/"+offHost+"/") |
| 251 | s, staged := preStaged[rel] |
| 252 | o, listed := preReplica[rel] |
| 253 | if staged && listed && s.Size == o.Size && s.ETag == o.ETag { |
| 254 | t.Fatalf("the seal recopied %s, which the stage had copied", rel) |
| 255 | } |
| 256 | } |
| 257 | seal, _, ok := sw.sealRecord() |
| 258 | if !ok { |
| 259 | t.Fatal("no seal") |
| 260 | } |
| 261 | if _, ok := seal.Staged["scratch/retired.bin"]; ok { |
| 262 | t.Fatal("the seal kept a staged file the replica no longer has") |
| 263 | } |
| 264 | if _, ok := seal.Staged["0000/stale-extra.ltx"]; ok { |
| 265 | t.Fatal("the seal kept the stale prefix's extra file") |
| 266 | } |
| 267 | sameSizes(t, sw.stagedRel(), sw.replicaRel()) |
| 268 | // The staged copy restores to every row written while staging. |
| 269 | restored := t.TempDir() + "/restored.db" |
| 270 | if ok, err := lease.Restore(ctx, restored, sealUK, strings.TrimSuffix(activator.MovesPrefix(offHost), "/"), sealUKRegion); err != nil || !ok { |
| 271 | t.Fatalf("restore the staged copy: %v %v", ok, err) |
| 272 | } |
| 273 | if n := countRows(t, restored); n != 40 { |
| 274 | t.Fatalf("the staged copy holds %d rows, want 40", n) |
| 275 | } |
| 276 | } |
| 277 | |
| 278 | // Abort from staging: the stager deletes the staged copy and the receipts |
| 279 | // before it reports reverted, every other source box reports reverted as |
| 280 | // for any abort, and finish ends a move that was never offline. Nothing the |
| 281 | // move did not write is touched. |
| 282 | func TestOfflineAbortFromStagingRemovesTheStagedCopy(t *testing.T) { |
| 283 | sw := newSealWorld(t) |
| 284 | ie := sw.ie |
| 285 | sw.instance(ie) |
| 286 | sw.stage() |
| 287 | sw.mustStagePass(ie) |
| 288 | if len(sw.stagedRel()) == 0 { |
| 289 | t.Fatal("fixture: nothing staged") |
| 290 | } |
| 291 | // Receipts, as a round over a multipart source leaves them, and |
| 292 | // neighbours the abort must not touch. |
| 293 | sw.fake.Put(sealIE, activator.MoveReceiptsKey(offHost), []byte(`{"v":1}`)) |
| 294 | sw.fake.Put(sealUK, "moves/other.example.com/0000/x.ltx", []byte("x")) |
| 295 | replica := sw.replicaRel() |
| 296 | |
| 297 | sw.abort() |
| 298 | sw.mustStagePass(ie) |
| 299 | if n := len(sw.stagedRel()); n != 0 { |
| 300 | t.Fatalf("%d staged objects left after the abort", n) |
| 301 | } |
| 302 | if _, ok := sw.fake.Current(sealIE, activator.MoveReceiptsKey(offHost)); ok { |
| 303 | t.Fatal("the receipts are still there") |
| 304 | } |
| 305 | if _, ok := sw.fake.Current(sealUK, "moves/other.example.com/0000/x.ltx"); !ok { |
| 306 | t.Fatal("the abort deleted another host's staging") |
| 307 | } |
| 308 | if got := sw.replicaRel(); len(got) != len(replica) { |
| 309 | t.Fatalf("the replica changed: %d objects, was %d", len(got), len(replica)) |
| 310 | } |
| 311 | if r := sw.report(moves.OfflineRoleStager, "ie-a"); r.Phase != moves.OfflinePhaseReverted { |
| 312 | t.Fatalf("stager report %+v", r) |
| 313 | } |
| 314 | if r := sw.report(moves.OfflineRoleSource, "ie-a"); r.Phase != moves.OfflinePhaseReverted { |
| 315 | t.Fatalf("source report %+v", r) |
| 316 | } |
| 317 | // Again: nothing in the buckets changes. |
| 318 | snap := sw.fake.Snapshot() |
| 319 | sw.mustStagePass(ie) |
| 320 | if sw.fake.Snapshot() != snap { |
| 321 | t.Fatal("a second revert pass wrote to the buckets") |
| 322 | } |
| 323 | rec := sw.move(provision.OfflineFinish(offID, sw.clock())) |
| 324 | if rec.Move != nil || rec.LastMove == nil || rec.LastMove.Outcome != provision.OfflineOutcomeAborted || rec.LastMove.OfflineS != 0 { |
| 325 | t.Fatalf("finish: %+v", rec.LastMove) |
| 326 | } |
| 327 | if rt := ie.row(offHost); rt == nil || rt.HoldOp != "" { |
| 328 | t.Fatalf("row %+v after the abort", rt) |
| 329 | } |
| 330 | } |
| 331 | |
| 332 | // A staging bucket that is the source's own would have the serving instance |
| 333 | // restore from the staged copy (the activator's staging wins). The stage |
| 334 | // door refuses it (MV-O191); a record that names it anyway (written by |
| 335 | // hand) gets a refusing round that writes nothing, and a cleanup that |
| 336 | // deletes and acknowledges nothing. |
| 337 | func TestOfflineStageRefusesTheSourcesOwnBucket(t *testing.T) { |
| 338 | sw := newSealWorld(t) |
| 339 | ie := sw.ie |
| 340 | sw.instance(ie) |
| 341 | sw.publishTarget(sealIE, sealIERegion) |
| 342 | sw.move(provision.OfflineStage(offID, "", "uk", "op@example.com", provision.StageTarget{Stager: "ie-a", Bucket: sealIE, Region: sealIERegion}, sw.clock())) |
| 343 | if err := sw.stagePass(ie); err == nil { |
| 344 | t.Fatal("staged into the source's own bucket") |
| 345 | } |
| 346 | if n := len(sw.list(sealIE, "moves/")); n != 0 { |
| 347 | t.Fatalf("%d objects staged into the source's bucket", n) |
| 348 | } |
| 349 | r := sw.report(moves.OfflineRoleStager, "ie-a") |
| 350 | if r.Phase != moves.OfflinePhaseFailed || !strings.Contains(r.StageError, "own bucket") { |
| 351 | t.Fatalf("stager report %+v", r) |
| 352 | } |
| 353 | sw.abort() |
| 354 | if err := sw.stagePass(ie); err == nil { |
| 355 | t.Fatal("an abort acknowledged a cleanup of the source's own bucket") |
| 356 | } |
| 357 | if r := sw.report(moves.OfflineRoleStager, "ie-a"); r.Phase == moves.OfflinePhaseReverted { |
| 358 | t.Fatal("reverted written for a staging bucket that was never cleaned") |
| 359 | } |
| 360 | } |
| 361 | |
| 362 | // Astra round 1, finding 1: a round that copied and then never wrote its |
| 363 | // report (the box crashed, or the report write failed) still leaves its |
| 364 | // copy to the abort. Cleanup follows the record's staged_at, never the |
| 365 | // reports. |
| 366 | func TestOfflineAbortCleansARoundThatNeverReported(t *testing.T) { |
| 367 | sw := newSealWorld(t) |
| 368 | ie := sw.ie |
| 369 | sw.instance(ie) |
| 370 | sw.stage() |
| 371 | sw.stagerEdges(ie) |
| 372 | // The round's copies land; the box dies before its report. |
| 373 | if _, err := ie.w.offlineStageRound(context.Background(), sw.s, sw.record()); err != nil { |
| 374 | t.Fatal(err) |
| 375 | } |
| 376 | if len(sw.stagedRel()) == 0 { |
| 377 | t.Fatal("fixture: nothing staged") |
| 378 | } |
| 379 | if _, ok := sw.reportIf(moves.OfflineRoleStager, "ie-a"); ok { |
| 380 | t.Fatal("fixture: the round reported") |
| 381 | } |
| 382 | sw.abort() |
| 383 | sw.mustStagePass(ie) |
| 384 | if n := len(sw.stagedRel()); n != 0 { |
| 385 | t.Fatalf("%d staged objects left by a round that never reported", n) |
| 386 | } |
| 387 | if r := sw.report(moves.OfflineRoleStager, "ie-a"); r.Phase != moves.OfflinePhaseReverted { |
| 388 | t.Fatalf("stager report %+v", r) |
| 389 | } |
| 390 | } |
| 391 | |
| 392 | // Astra round 1, finding 2: a box that joins mid-move, lexically before the |
| 393 | // stager, never stages and never cleans for it. Only the pinned stager |
| 394 | // writes the staging prefix, so its cleanup, after its own last round, is |
| 395 | // the one that counts. |
| 396 | func TestOfflineAbortWaitsForThePinnedStager(t *testing.T) { |
| 397 | sw := newSealWorld(t) |
| 398 | sw.instance(sw.ie) |
| 399 | ieB := sw.sourceBox("ie-b") |
| 400 | sw.stageBy("ie-b") |
| 401 | sw.mustStagePass(ieB) |
| 402 | if r := sw.report(moves.OfflineRoleStager, "ie-b"); r.Phase != moves.OfflinePhaseStaged { |
| 403 | t.Fatalf("stager report %+v", r) |
| 404 | } |
| 405 | // ie-a joins, publishing the capability, and the operator stages again. |
| 406 | sw.stagerEdges(sw.ie) |
| 407 | sw.restage() |
| 408 | spy := sw.spyCopies() |
| 409 | sw.mustStagePass(sw.ie) |
| 410 | if _, ok := sw.reportIf(moves.OfflineRoleStager, "ie-a"); ok || len(spy.take()) != 0 { |
| 411 | t.Fatal("a box that is not the pinned stager staged") |
| 412 | } |
| 413 | sw.abort() |
| 414 | sw.mustStagePass(sw.ie) // ie-a's revert: not its copy to remove |
| 415 | if _, ok := sw.reportIf(moves.OfflineRoleStager, "ie-a"); ok { |
| 416 | t.Fatal("a box that is not the pinned stager acknowledged the cleanup") |
| 417 | } |
| 418 | // ie-b's round, which began before the abort, lands its copies now. |
| 419 | sw.fake.Put(sealIE, sealPrefix+"/"+offHost+"/scratch/late.bin", []byte("late")) |
| 420 | if _, err := ieB.w.offlineStageRound(context.Background(), sw.s, sw.record()); err != nil { |
| 421 | t.Fatal(err) |
| 422 | } |
| 423 | if len(sw.stagedRel()) == 0 { |
| 424 | t.Fatal("fixture: the late round staged nothing") |
| 425 | } |
| 426 | sw.mustStagePass(ieB) // the pinned stager's revert, after its last round |
| 427 | if n := len(sw.stagedRel()); n != 0 { |
| 428 | t.Fatalf("%d staged objects left after the pinned stager's cleanup", n) |
| 429 | } |
| 430 | if r := sw.report(moves.OfflineRoleStager, "ie-b"); r.Phase != moves.OfflinePhaseReverted { |
| 431 | t.Fatalf("stager report %+v", r) |
| 432 | } |
| 433 | } |
| 434 | |
| 435 | // ---- Astra round 2 ---- |
| 436 | |
| 437 | // publishTarget republishes the target box's edges file with bucket in |
| 438 | // region: a target replaced or reconfigured after the stage. |
| 439 | func (sw *sealWorld) publishTarget(bucket, region string) { |
| 440 | sw.t.Helper() |
| 441 | var last []byte |
| 442 | if err := edgesFileWrite(context.Background(), sw.s, sw.uk.reg, edgesFile{Box: "uk-box", Placement: "uk", DataBucket: bucket, DataBucketRegion: region}, "", &last, false); err != nil { |
| 443 | sw.t.Fatal(err) |
| 444 | } |
| 445 | } |
| 446 | |
| 447 | // Finding 1: once the move has left staging, the stager never starts a |
| 448 | // round, even on a pass that read the record while it was still staging. |
| 449 | func TestOfflineStageNoRoundAfterTheMoveLeftStaging(t *testing.T) { |
| 450 | sw := newSealWorld(t) |
| 451 | ie := sw.ie |
| 452 | sw.instance(ie) |
| 453 | sw.stage() |
| 454 | sw.mustStagePass(ie) |
| 455 | sw.restage() |
| 456 | stale := sw.record() // a pass's listing, read while still staging |
| 457 | sw.move(provision.OfflineStartStaged(offID, "op@example.com", sw.clock())) |
| 458 | sw.fake.Put(sealIE, sealPrefix+"/"+offHost+"/scratch/new.bin", []byte("new")) |
| 459 | spy := sw.spyCopies() |
| 460 | if err := ie.w.offlineRound(context.Background(), sw.s, stale); err != nil { |
| 461 | t.Fatal(err) |
| 462 | } |
| 463 | if n := len(spy.take()); n != 0 { |
| 464 | t.Fatalf("a round made %d copies after the move left staging", n) |
| 465 | } |
| 466 | if r := sw.report(moves.OfflineRoleStager, "ie-a"); r.StageRequest.Equal(*stale.Move.StagedAt) { |
| 467 | t.Fatal("the stager answered a request after the move left staging") |
| 468 | } |
| 469 | } |
| 470 | |
| 471 | // Finding 1: a round the start overtakes deletes nothing. The delete phase |
| 472 | // reads the record again first, and a move that has left staging stops the |
| 473 | // round there: an object the seal's checkpoint added must never be removed |
| 474 | // by a stage round. |
| 475 | func TestOfflineStageRoundDeletesNothingOnceTheMoveLeftStaging(t *testing.T) { |
| 476 | sw := newSealWorld(t) |
| 477 | ie := sw.ie |
| 478 | sw.instance(ie) |
| 479 | sw.stage() |
| 480 | sw.mustStagePass(ie) |
| 481 | // An object only the staging prefix has (the round would delete it), |
| 482 | // and a new replica object (the round copies it, and the start lands |
| 483 | // during that copy). |
| 484 | extra := activator.MovesPrefix(offHost) + "0000/added-by-the-seal.ltx" |
| 485 | sw.fake.Put(sealUK, extra, []byte("the seal's")) |
| 486 | sw.fake.Put(sealIE, sealPrefix+"/"+offHost+"/scratch/new.bin", []byte("new")) |
| 487 | sw.restage() |
| 488 | var once sync.Once |
| 489 | sw.fake.FailOn = func(op, key string) error { |
| 490 | if op == "copy" { |
| 491 | once.Do(func() { |
| 492 | go func() { // the operator's start, while the copy is in flight |
| 493 | _, _ = provision.SetRecordMove(context.Background(), sw.s, sw.acct, "app", offHost, |
| 494 | provision.OfflineStartStaged(offID, "op@example.com", time.Now())) |
| 495 | }() |
| 496 | time.Sleep(200 * time.Millisecond) |
| 497 | }) |
| 498 | } |
| 499 | return nil |
| 500 | } |
| 501 | t.Cleanup(func() { sw.fake.FailOn = nil }) |
| 502 | _ = ie.w.offlineRound(context.Background(), sw.s, sw.record()) |
| 503 | if sw.record().Move.State != provision.OfflineStopping { |
| 504 | t.Fatal("fixture: the start did not land during the round") |
| 505 | } |
| 506 | if _, ok := sw.fake.Current(sealUK, extra); !ok { |
| 507 | t.Fatal("a stage round deleted from the staging prefix after the move left staging") |
| 508 | } |
| 509 | } |
| 510 | |
| 511 | // Finding 1: in stopping and on, the pinned stager says, on every pass and |
| 512 | // under its pass lock, that no round is in flight and none will start. |
| 513 | func TestOfflineStagerAcknowledgesItHasStopped(t *testing.T) { |
| 514 | sw := newSealWorld(t) |
| 515 | ie := sw.ie |
| 516 | sw.instance(ie) |
| 517 | sw.stage() |
| 518 | sw.mustStagePass(ie) |
| 519 | sw.move(provision.OfflineStartStaged(offID, "op@example.com", sw.clock())) |
| 520 | ie.tick() |
| 521 | r := sw.report(moves.OfflineRoleStager, "ie-a") |
| 522 | if r.Phase != moves.OfflinePhaseStopped || !r.At.Equal(sw.clock()) || r.StagedObjects == 0 { |
| 523 | t.Fatalf("stager report %+v, want stopped now with its progress kept", r) |
| 524 | } |
| 525 | } |
| 526 | |
| 527 | // Finding 2: cleanup deletes from the destination the stage fixed on the |
| 528 | // record, whatever the target publishes now, and says reverted only then. |
| 529 | func TestOfflineAbortCleansTheStoredDestination(t *testing.T) { |
| 530 | for _, c := range []struct { |
| 531 | name string |
| 532 | bucket, region string |
| 533 | }{ |
| 534 | {"a replaced target", "data-uk-new", sealUKRegion}, |
| 535 | {"a target that now publishes the source's own bucket", sealIE, sealIERegion}, |
| 536 | } { |
| 537 | t.Run(c.name, func(t *testing.T) { |
| 538 | sw := newSealWorld(t) |
| 539 | sw.fake.AddBucket("data-uk-new") |
| 540 | ie := sw.ie |
| 541 | sw.instance(ie) |
| 542 | sw.stage() |
| 543 | sw.mustStagePass(ie) |
| 544 | if len(sw.stagedRel()) == 0 { |
| 545 | t.Fatal("fixture: nothing staged") |
| 546 | } |
| 547 | sw.publishTarget(c.bucket, c.region) |
| 548 | sw.abort() |
| 549 | sw.mustStagePass(ie) |
| 550 | if n := len(sw.stagedRel()); n != 0 { |
| 551 | t.Fatalf("%d objects left in the stored staging bucket %s", n, sealUK) |
| 552 | } |
| 553 | if r := sw.report(moves.OfflineRoleStager, "ie-a"); r.Phase != moves.OfflinePhaseReverted { |
| 554 | t.Fatalf("stager report %+v", r) |
| 555 | } |
| 556 | if got := sw.replicaRel(); len(got) == 0 { |
| 557 | t.Fatal("the cleanup touched the source's replica") |
| 558 | } |
| 559 | }) |
| 560 | } |
| 561 | } |
| 562 | |
| 563 | // Finding 2: a stage round never follows a target that now publishes other |
| 564 | // coordinates than the stored ones. It refuses and writes nowhere. |
| 565 | func TestOfflineStageRoundRefusesChangedCoordinates(t *testing.T) { |
| 566 | sw := newSealWorld(t) |
| 567 | sw.fake.AddBucket("data-uk-new") |
| 568 | ie := sw.ie |
| 569 | sw.instance(ie) |
| 570 | sw.stage() |
| 571 | sw.publishTarget("data-uk-new", sealUKRegion) |
| 572 | if err := sw.stagePass(ie); err == nil { |
| 573 | t.Fatal("staged with changed coordinates") |
| 574 | } |
| 575 | if n := len(sw.list("data-uk-new", "moves/")); n != 0 { |
| 576 | t.Fatalf("%d objects staged into the new bucket", n) |
| 577 | } |
| 578 | if n := len(sw.stagedRel()); n != 0 { |
| 579 | t.Fatalf("%d objects staged at all", n) |
| 580 | } |
| 581 | if r := sw.report(moves.OfflineRoleStager, "ie-a"); r.Phase != moves.OfflinePhaseFailed { |
| 582 | t.Fatalf("stager report %+v", r) |
| 583 | } |
| 584 | } |
| 585 | |
amadan