| 1 | package console |
| 2 | |
| 3 | import ( |
| 4 | "context" |
| 5 | "crypto/ed25519" |
| 6 | "encoding/json" |
| 7 | "errors" |
| 8 | "net/url" |
| 9 | "strconv" |
| 10 | "strings" |
| 11 | "testing" |
| 12 | "time" |
| 13 | |
| 14 | "github.com/carlosframework/platform/internal/hostactors" |
| 15 | "github.com/carlosframework/platform/internal/hostops" |
| 16 | "github.com/carlosframework/platform/internal/lease/leasetest" |
| 17 | "github.com/carlosframework/platform/internal/release" |
| 18 | ) |
| 19 | |
| 20 | // Every console in these tests is built by the full production constructor |
| 21 | // with the flag on from the start, exactly as cmd/console builds it: |
| 22 | // OpenGated (triggers armed before NewServer), NewServer with HostClaim and |
| 23 | // the RAW bucket, raw versioned store, signing key and console id in Config, |
| 24 | // then ConfigureClaimRuntime, which reads the signed roster, builds the |
| 25 | // control writer, the production data and versioned boundaries and the |
| 26 | // authority, and replaces every raw handle in Config. Nothing is set on the |
| 27 | // authority by hand and nothing is switched on after construction. Nothing |
| 28 | // below writes SQL except through the console's own code or the test-only |
| 29 | // seed scope, and the raw bucket is only written to stage the world |
| 30 | // (roster, markers, the initial admission). |
| 31 | |
| 32 | const ( |
| 33 | crAcctID = int64(OpsAccountID) // NewServer's backfill creates it, sqid "bab" |
| 34 | crSqid = "bab" |
| 35 | crApp = "kass" |
| 36 | crAppID = "incarnation-a" |
| 37 | crOwner = "paul@keymail.dev" |
| 38 | crBucket = "deploy-bucket" |
| 39 | crNS = "platform" |
| 40 | crConsole = "console-a" |
| 41 | crOther = "console-b" |
| 42 | ) |
| 43 | |
| 44 | var crEnrollment = map[string]string{crConsole: strings.Repeat("a", 32), crOther: strings.Repeat("b", 32)} |
| 45 | |
| 46 | type claimRestoreWorld struct { |
| 47 | ctx context.Context |
| 48 | bucket *release.DirStore |
| 49 | a, b *Server |
| 50 | } |
| 51 | |
| 52 | // publishRestoreRoster signs the host-actor roster naming both consoles |
| 53 | // into the raw bucket, as the operator's roster publication does. |
| 54 | func publishRestoreRoster(t *testing.T, raw release.Store, key ed25519.PrivateKey) { |
| 55 | t.Helper() |
| 56 | consoles := map[string]hostactors.Console{} |
| 57 | for id, enrollment := range crEnrollment { |
| 58 | consoles[id] = hostactors.Console{Enrollment: enrollment, InstanceID: id, Namespaces: []string{crNS}, DeploymentBucket: crBucket} |
| 59 | } |
| 60 | roster := hostactors.Roster{V: 1, DeploymentBucket: crBucket, Revision: 1, Boxes: map[string]hostactors.Box{}, Consoles: consoles, |
| 61 | Holds: map[string]hostactors.Reservation{}, Activations: map[string]hostactors.Reservation{}} |
| 62 | envelope, err := release.SignRootEnvelope(release.RootSigner(key), roster) |
| 63 | if err != nil { |
| 64 | t.Fatal(err) |
| 65 | } |
| 66 | if _, err := raw.Put(context.Background(), hostactors.Key, envelope, ""); err != nil { |
| 67 | t.Fatal(err) |
| 68 | } |
| 69 | } |
| 70 | |
| 71 | func newEnabledRestoreConsole(t *testing.T, raw release.Store, key ed25519.PrivateKey, name string) *Server { |
| 72 | t.Helper() |
| 73 | st := openGatedStore(t) |
| 74 | if !st.claimGated() { |
| 75 | t.Fatal("OpenGated returned an ungated store") |
| 76 | } |
| 77 | versions := leasetest.NewVersionedBucket() |
| 78 | versions.AddBucket(crBucket) |
| 79 | srv, err := NewServer(Config{Origin: "https://console.example", Keymail: "https://keymail.example", HostClaim: true, |
| 80 | HostConsoleID: name, Bucket: crBucket, Releases: raw, DBVersions: versions, |
| 81 | Key: key, ReleasePub: key.Public().(ed25519.PublicKey)}, st) |
| 82 | if err != nil { |
| 83 | t.Fatal(err) |
| 84 | } |
| 85 | if err := srv.ConfigureClaimRuntime(t.Context()); err != nil { |
| 86 | t.Fatalf("ConfigureClaimRuntime: %v", err) |
| 87 | } |
| 88 | a := srv.cfg.ClaimAuthority |
| 89 | if a == nil || a.Control == nil || a.ConsoleID != name || a.Enrollment != crEnrollment[name] || srv.cfg.Releases == raw { |
| 90 | t.Fatalf("production constructor did not install the authority over the raw bucket: %+v", a) |
| 91 | } |
| 92 | if _, _, err := srv.claimRestoreRuntime(); err != nil { |
| 93 | t.Fatalf("restore runtime: %v", err) |
| 94 | } |
| 95 | return srv |
| 96 | } |
| 97 | |
| 98 | // seedClaimRow writes this console's row for the incarnation, under the |
| 99 | // test-only seed scope: the world before the test, not a door under test. |
| 100 | func seedClaimRow(t *testing.T, st *Store, trashedAt time.Time) { |
| 101 | t.Helper() |
| 102 | scope := []claimSQLScope{{AccountID: crAcctID, App: crApp, AppID: crAppID, Mode: claimModeEffect}} |
| 103 | err := st.withClaimSQL(gateCtx(t), scope, allowAllClaimSQL, func(v *Store) error { |
| 104 | if _, err := v.CreateAppIn(crAcctID, crApp, crOwner, trashedAt.Add(-time.Hour)); err != nil { |
| 105 | return err |
| 106 | } |
| 107 | if err := v.BindClaimAppIdentity(crAcctID, crApp, crAppID); err != nil { |
| 108 | return err |
| 109 | } |
| 110 | return v.TrashAppRow(crAcctID, crApp, trashedAt) |
| 111 | }) |
| 112 | if err != nil { |
| 113 | t.Fatal(err) |
| 114 | } |
| 115 | } |
| 116 | |
| 117 | func newClaimRestoreWorld(t *testing.T) (*claimRestoreWorld, time.Time) { |
| 118 | t.Helper() |
| 119 | w := &claimRestoreWorld{ctx: t.Context(), bucket: release.NewDirStore(t.TempDir())} |
| 120 | a := hostops.AppAdmission{V: 1, AppID: crAppID, Revision: 1, Epoch: 1, State: "open", Registrations: map[string]hostops.Registration{}, Effects: map[string]hostops.Effect{}} |
| 121 | body, _ := json.Marshal(a) |
| 122 | if _, err := w.bucket.Put(w.ctx, "apps/"+crSqid+"/"+crApp+"/admission.json", body, ""); err != nil { |
| 123 | t.Fatal(err) |
| 124 | } |
| 125 | if _, err := w.bucket.Put(w.ctx, "apps/"+crSqid+"/"+crApp+"/releases/r1.json", []byte(`{}`), ""); err != nil { |
| 126 | t.Fatal(err) |
| 127 | } |
| 128 | trashed := time.Now().Add(-time.Hour).UTC() |
| 129 | if err := release.TrashApp(w.ctx, w.bucket, crSqid, crApp, crOwner, trashed); err != nil { |
| 130 | t.Fatal(err) |
| 131 | } |
| 132 | _, key, err := ed25519.GenerateKey(nil) |
| 133 | if err != nil { |
| 134 | t.Fatal(err) |
| 135 | } |
| 136 | publishRestoreRoster(t, w.bucket, key) |
| 137 | w.a = newEnabledRestoreConsole(t, w.bucket, key, crConsole) |
| 138 | w.b = newEnabledRestoreConsole(t, w.bucket, key, crOther) |
| 139 | seedClaimRow(t, w.a.store, trashed) |
| 140 | seedClaimRow(t, w.b.store, trashed) |
| 141 | return w, trashed |
| 142 | } |
| 143 | |
| 144 | func (w *claimRestoreWorld) account(t *testing.T, s *Server) *Account { |
| 145 | t.Helper() |
| 146 | acct, err := s.store.AccountBySqid(crSqid) |
| 147 | if err != nil || acct == nil { |
| 148 | t.Fatalf("account: %+v %v", acct, err) |
| 149 | } |
| 150 | return acct |
| 151 | } |
| 152 | |
| 153 | func (w *claimRestoreWorld) admission(t *testing.T) hostops.AdmissionObserved { |
| 154 | t.Helper() |
| 155 | a, err := hostops.ReadAdmission(w.ctx, w.bucket, crSqid, crApp, crAppID) |
| 156 | if err != nil { |
| 157 | t.Fatal(err) |
| 158 | } |
| 159 | return a |
| 160 | } |
| 161 | |
| 162 | func (w *claimRestoreWorld) marker(t *testing.T) (*release.AppTombstone, string) { |
| 163 | t.Helper() |
| 164 | tomb, tag, err := readTombstoneVersion(w.ctx, w.bucket, crSqid, crApp) |
| 165 | if err != nil { |
| 166 | t.Fatal(err) |
| 167 | } |
| 168 | return tomb, tag |
| 169 | } |
| 170 | |
| 171 | func requireLife(t *testing.T, s *Server, trashed, restoring bool) { |
| 172 | t.Helper() |
| 173 | life, err := s.store.AppLife(crAcctID, crApp) |
| 174 | if err != nil || life == nil { |
| 175 | t.Fatalf("life: %+v %v", life, err) |
| 176 | } |
| 177 | if life.Trashed() != trashed || life.Restoring != restoring { |
| 178 | t.Fatalf("row trashed=%v restoring=%v, want %v %v", life.Trashed(), life.Restoring, trashed, restoring) |
| 179 | } |
| 180 | } |
| 181 | |
| 182 | func requireSettledReceipt(t *testing.T, a hostops.AdmissionObserved) hostops.RestoreMark { |
| 183 | t.Helper() |
| 184 | v := a.Value |
| 185 | if v.State != "open" || v.Op != nil || v.LastRestore == nil || v.LastRestore.State != "complete" || v.LastRestore.AppID != crAppID || v.LastRestore.Epoch == 0 || v.LastRestore.Epoch > v.Epoch { |
| 186 | t.Fatalf("restore not settled: %+v receipt %+v", v, v.LastRestore) |
| 187 | } |
| 188 | for id, e := range v.Effects { |
| 189 | if e.State != "complete" { |
| 190 | t.Fatalf("effect %s still %s", id, e.State) |
| 191 | } |
| 192 | } |
| 193 | return *v.LastRestore |
| 194 | } |
| 195 | |
| 196 | func (w *claimRestoreWorld) materialize(t *testing.T, s *Server) { |
| 197 | t.Helper() |
| 198 | if err := MaterializeApps(w.ctx, s.cfg.Releases, s.store, true); err != nil { |
| 199 | t.Fatal(err) |
| 200 | } |
| 201 | } |
| 202 | |
| 203 | // Issuer crash points, in restoreClaimApp's order. |
| 204 | const ( |
| 205 | crashAfterReceipt = iota // PrepareRestore saved, nothing issued |
| 206 | crashAfterIssue // effect issued, restoring slot set, marker still there |
| 207 | crashAfterDelete // the exact-ETag delete ran, receipt still prepared |
| 208 | ) |
| 209 | |
| 210 | // crashAt drives console A's own issuer steps up to a crash point, through |
| 211 | // the same control writer, permit, grant and data boundary restoreClaimApp |
| 212 | // uses, in its order (BeginExclusive, PrepareRestore, Issue, SQL, delete), |
| 213 | // and stops without settling. |
| 214 | func (w *claimRestoreWorld) crashAt(t *testing.T, point int) (op, etag string) { |
| 215 | t.Helper() |
| 216 | control, self, err := w.a.claimRestoreRuntime() |
| 217 | if err != nil { |
| 218 | t.Fatal(err) |
| 219 | } |
| 220 | tomb, etag := w.marker(t) |
| 221 | op = "restore-crashed" |
| 222 | owner := self |
| 223 | owner.Controller = op |
| 224 | now := time.Now() |
| 225 | h, err := control.BeginExclusive(w.ctx, crSqid, crApp, crAppID, hostops.AppOperation{ID: op, Kind: hostops.OpAppMutation, AppID: crAppID, Phase: "issued", |
| 226 | Intent: map[string]string{"action": "restore", hostops.RestoreActionStarted: strconv.FormatInt(now.Unix(), 10)}}, owner) |
| 227 | if err != nil { |
| 228 | t.Fatal(err) |
| 229 | } |
| 230 | if err := h.PrepareRestore(w.ctx, etag, tomb.DeletedAt.Unix()); err != nil { |
| 231 | t.Fatal(err) |
| 232 | } |
| 233 | if point == crashAfterReceipt { |
| 234 | return op, etag |
| 235 | } |
| 236 | pctx, _, err := h.Runner(time.Now).Issue(w.ctx, hostops.Effect{ID: op, Resource: hostops.OpAppMutation, Action: "restore"}) |
| 237 | if err != nil { |
| 238 | t.Fatal(err) |
| 239 | } |
| 240 | scope := claimSQLScope{AccountID: crAcctID, App: crApp, AppID: crAppID} |
| 241 | err = w.a.store.WithAppMutation(w.ctx, []sqlGrant{permitGrant(pctx, w.a.cfg.Releases, crSqid, scope)}, func(v *Store) error { |
| 242 | _, err := v.SetClaimRestoring(crAcctID, crApp, crAppID, true) |
| 243 | return err |
| 244 | }) |
| 245 | if err != nil { |
| 246 | t.Fatal(err) |
| 247 | } |
| 248 | if point == crashAfterDelete { |
| 249 | if err := release.DeleteIfMatch(pctx, w.a.cfg.Releases, release.AppTombstoneKey(crSqid, crApp), etag); err != nil { |
| 250 | t.Fatal(err) |
| 251 | } |
| 252 | } |
| 253 | return op, etag |
| 254 | } |
| 255 | |
| 256 | // closeAdmission stages a foreign exclusive operation (as a purge would |
| 257 | // hold) straight into the bucket. |
| 258 | func (w *claimRestoreWorld) closeAdmission(t *testing.T) hostops.AdmissionObserved { |
| 259 | t.Helper() |
| 260 | a := w.admission(t) |
| 261 | v := a.Value |
| 262 | v.State = "closed" |
| 263 | v.Op = &hostops.AppOperation{ID: "purge", Kind: hostops.OpPurging, AppID: crAppID} |
| 264 | v.Revision++ |
| 265 | body, _ := json.Marshal(v) |
| 266 | if _, err := w.bucket.Put(w.ctx, a.Key, body, a.ETag); err != nil { |
| 267 | t.Fatal(err) |
| 268 | } |
| 269 | return w.admission(t) |
| 270 | } |
| 271 | |
| 272 | func TestClaimRestoreGateIsArmedOnEnabledConsoles(t *testing.T) { |
| 273 | w, _ := newClaimRestoreWorld(t) |
| 274 | for _, s := range []*Server{w.a, w.b} { |
| 275 | _, err := s.store.db.Exec(`UPDATE apps SET deleted_at = 0 WHERE app = ?`, crApp) |
| 276 | requireDenied(t, "ungated restore write", err) |
| 277 | requireLife(t, s, true, false) |
| 278 | } |
| 279 | if _, err := NewServer(Config{Origin: "https://console.example", Keymail: "https://keymail.example", Releases: w.bucket, HostClaim: true}, openTestStore(t)); err == nil { |
| 280 | t.Fatal("NewServer accepted HostClaim on an ungated database") |
| 281 | } |
| 282 | ungated, err := NewServer(Config{Origin: "https://console.example", Keymail: "https://keymail.example", Releases: w.bucket}, openTestStore(t)) |
| 283 | if err != nil { |
| 284 | t.Fatal(err) |
| 285 | } |
| 286 | if ungated.store.claimGated() { |
| 287 | t.Fatal("flag-off console was gated") |
| 288 | } |
| 289 | } |
| 290 | |
| 291 | func TestClaimRestoreAcrossTwoGatedConsoles(t *testing.T) { |
| 292 | w, _ := newClaimRestoreWorld(t) |
| 293 | _, oldTag := w.marker(t) |
| 294 | if err := w.a.restoreApp(w.ctx, w.account(t, w.a), crApp, time.Now()); err != nil { |
| 295 | t.Fatalf("restore: %v", err) |
| 296 | } |
| 297 | if tomb, _ := w.marker(t); tomb != nil { |
| 298 | t.Fatal("marker survived the restore") |
| 299 | } |
| 300 | receipt := requireSettledReceipt(t, w.admission(t)) |
| 301 | if receipt.TombstoneETag != oldTag { |
| 302 | t.Fatalf("receipt names %q, want the deleted version %q", receipt.TombstoneETag, oldTag) |
| 303 | } |
| 304 | requireLife(t, w.a, false, false) |
| 305 | // Console B consumes the same incarnation-bound receipt and writes no |
| 306 | // canonical state. |
| 307 | before := w.admission(t) |
| 308 | requireLife(t, w.b, true, false) |
| 309 | w.materialize(t, w.b) |
| 310 | requireLife(t, w.b, false, false) |
| 311 | if after := w.admission(t); after.Value.AppID != before.Value.AppID || after.Value.Op != nil || *after.Value.LastRestore != *before.Value.LastRestore { |
| 312 | t.Fatal("projection changed the restore evidence or left admission closed") |
| 313 | } |
| 314 | // A later trash wins although the old receipt survives, on both consoles. |
| 315 | if err := release.TrashApp(w.ctx, w.bucket, crSqid, crApp, crOwner, time.Now().Add(time.Minute)); err != nil { |
| 316 | t.Fatal(err) |
| 317 | } |
| 318 | w.materialize(t, w.b) |
| 319 | w.materialize(t, w.a) |
| 320 | requireLife(t, w.b, true, false) |
| 321 | requireLife(t, w.a, true, false) |
| 322 | // The old receipt cannot be replayed against the new marker. |
| 323 | scope := claimSQLScope{AccountID: crAcctID, App: crApp, AppID: crAppID} |
| 324 | err := w.b.store.WithAppMutation(w.ctx, []sqlGrant{restoreReceiptGrant(w.ctx, w.bucket, crSqid, scope, receipt)}, func(v *Store) error { |
| 325 | _, err := v.FinishClaimRestoreRow(crAcctID, crApp, crAppID, receipt.DeletedAt, time.Now()) |
| 326 | return err |
| 327 | }) |
| 328 | if !errors.Is(err, ErrClaimSQLDenied) { |
| 329 | t.Fatalf("old receipt under a new marker: %v", err) |
| 330 | } |
| 331 | requireLife(t, w.b, true, false) |
| 332 | } |
| 333 | |
| 334 | // The receipt precedes Issue, so both crash windows before the delete |
| 335 | // (no effect yet, and an issued effect) are finished from the receipt and |
| 336 | // never abandoned. The no-effect window needs ControlWriter.RecoverMutation |
| 337 | // to accept a prepared receipt with no effect (control-work adaptation). |
| 338 | func TestClaimRestoreCrashAfterReceiptBeforeDelete(t *testing.T) { |
| 339 | for _, c := range []struct { |
| 340 | name string |
| 341 | point int |
| 342 | finisher string |
| 343 | }{ |
| 344 | {"receipt-only/sweep", crashAfterReceipt, "sweep"}, |
| 345 | {"receipt-only/retry", crashAfterReceipt, "retry"}, |
| 346 | {"issued/sweep", crashAfterIssue, "sweep"}, |
| 347 | {"issued/retry", crashAfterIssue, "retry"}, |
| 348 | } { |
| 349 | finisher := c.finisher |
| 350 | t.Run(c.name, func(t *testing.T) { |
| 351 | w, _ := newClaimRestoreWorld(t) |
| 352 | op, etag := w.crashAt(t, c.point) |
| 353 | if _, tag := w.marker(t); tag != etag { |
| 354 | t.Fatal("fixture is not the crash window") |
| 355 | } |
| 356 | requireLife(t, w.a, true, c.point == crashAfterIssue) |
| 357 | // Projection never restores from a prepared receipt. |
| 358 | w.materialize(t, w.b) |
| 359 | requireLife(t, w.b, true, false) |
| 360 | switch finisher { |
| 361 | case "sweep": // the other console's admission sweep |
| 362 | if err := w.b.reconcileClaimRestore(w.ctx, crSqid, crApp, crAppID); err != nil { |
| 363 | t.Fatal(err) |
| 364 | } |
| 365 | w.materialize(t, w.a) |
| 366 | case "retry": // the person presses Restore again on console A |
| 367 | if err := w.a.restoreApp(w.ctx, w.account(t, w.a), crApp, time.Now()); err != nil { |
| 368 | t.Fatalf("retry: %v", err) |
| 369 | } |
| 370 | } |
| 371 | if tomb, _ := w.marker(t); tomb != nil { |
| 372 | t.Fatal("receipt was not replayed") |
| 373 | } |
| 374 | if r := requireSettledReceipt(t, w.admission(t)); r.ID != op || r.TombstoneETag != etag { |
| 375 | t.Fatalf("settled the wrong receipt: %+v", r) |
| 376 | } |
| 377 | w.materialize(t, w.b) |
| 378 | requireLife(t, w.a, false, false) |
| 379 | requireLife(t, w.b, false, false) |
| 380 | }) |
| 381 | } |
| 382 | } |
| 383 | |
| 384 | func TestClaimRestoreCrashAfterDeleteBeforeCompletion(t *testing.T) { |
| 385 | w, _ := newClaimRestoreWorld(t) |
| 386 | op, etag := w.crashAt(t, crashAfterDelete) |
| 387 | if tomb, _ := w.marker(t); tomb != nil { |
| 388 | t.Fatal("fixture is not the crash window") |
| 389 | } |
| 390 | if a := w.admission(t).Value; a.Op == nil || a.Op.ID != op || a.LastRestore.State != "prepared" { |
| 391 | t.Fatalf("fixture admission: %+v", a) |
| 392 | } |
| 393 | // While the operation is open, no console projects anything. |
| 394 | w.materialize(t, w.b) |
| 395 | requireLife(t, w.b, true, false) |
| 396 | // B's own restore door finishes A's receipt first (step 0), then answers |
| 397 | // it as a restore already done. |
| 398 | if err := w.b.restoreApp(w.ctx, w.account(t, w.b), crApp, time.Now()); err != nil { |
| 399 | t.Fatalf("B's restore should finish A's receipt: %v", err) |
| 400 | } |
| 401 | if r := requireSettledReceipt(t, w.admission(t)); r.ID != op || r.TombstoneETag != etag { |
| 402 | t.Fatalf("settled the wrong receipt: %+v", r) |
| 403 | } |
| 404 | requireLife(t, w.b, false, false) |
| 405 | w.materialize(t, w.a) |
| 406 | requireLife(t, w.a, false, false) |
| 407 | } |
| 408 | |
| 409 | func TestClaimRestoreSQLFailureIsLocalAndReplayable(t *testing.T) { |
| 410 | w, _ := newClaimRestoreWorld(t) |
| 411 | // A real SQL failure on console A only: a persistent trigger written on |
| 412 | // A's trusted boot handle after arming. |
| 413 | if _, err := w.a.store.bootDB().Exec(`CREATE TRIGGER fail_restore BEFORE UPDATE OF deleted_at ON apps WHEN NEW.deleted_at = 0 |
| 414 | BEGIN SELECT RAISE(ABORT, 'test restore SQL failure'); END`); err != nil { |
| 415 | t.Fatal(err) |
| 416 | } |
| 417 | err := w.a.restoreApp(w.ctx, w.account(t, w.a), crApp, time.Now()) |
| 418 | var rf *restoreFailure |
| 419 | if !errors.As(err, &rf) || rf.Marker || !strings.Contains(err.Error(), "test restore SQL failure") { |
| 420 | t.Fatalf("restore = %v, want the row-only failure", err) |
| 421 | } |
| 422 | // The global effect settled; only A's projection is behind. |
| 423 | receipt := requireSettledReceipt(t, w.admission(t)) |
| 424 | if tomb, _ := w.marker(t); tomb != nil { |
| 425 | t.Fatal("marker survived") |
| 426 | } |
| 427 | requireLife(t, w.a, true, true) |
| 428 | // Console B replays the same incarnation-bound terminal receipt without |
| 429 | // opening anything. |
| 430 | before := w.admission(t) |
| 431 | w.materialize(t, w.b) |
| 432 | requireLife(t, w.b, false, false) |
| 433 | after := w.admission(t) |
| 434 | if after.Value.AppID != before.Value.AppID || after.Value.Op != nil || *after.Value.LastRestore != *before.Value.LastRestore { |
| 435 | t.Fatal("replay changed the restore evidence or left admission closed") |
| 436 | } |
| 437 | // A replays it too once its SQL works again. |
| 438 | w.materialize(t, w.a) |
| 439 | requireLife(t, w.a, true, true) // still failing: nothing reopened |
| 440 | if _, err := w.a.store.bootDB().Exec(`DROP TRIGGER fail_restore`); err != nil { |
| 441 | t.Fatal(err) |
| 442 | } |
| 443 | w.materialize(t, w.a) |
| 444 | requireLife(t, w.a, false, false) |
| 445 | if r := requireSettledReceipt(t, w.admission(t)); r != receipt { |
| 446 | t.Fatal("receipt changed during replay") |
| 447 | } |
| 448 | } |
| 449 | |
| 450 | func TestClaimRestoreReplayNeverReopensClosedAdmission(t *testing.T) { |
| 451 | w, _ := newClaimRestoreWorld(t) |
| 452 | if err := w.a.restoreApp(w.ctx, w.account(t, w.a), crApp, time.Now()); err != nil { |
| 453 | t.Fatal(err) |
| 454 | } |
| 455 | receipt := requireSettledReceipt(t, w.admission(t)) |
| 456 | // Another operation now owns the app (as a purge would), with no marker |
| 457 | // present. B still has its stale trashed row. |
| 458 | closed := w.closeAdmission(t) |
| 459 | w.materialize(t, w.b) |
| 460 | requireLife(t, w.b, true, false) |
| 461 | scope := claimSQLScope{AccountID: crAcctID, App: crApp, AppID: crAppID} |
| 462 | err := w.b.store.WithAppMutation(w.ctx, []sqlGrant{restoreReceiptGrant(w.ctx, w.bucket, crSqid, scope, receipt)}, func(v *Store) error { |
| 463 | _, err := v.FinishClaimRestoreRow(crAcctID, crApp, crAppID, receipt.DeletedAt, time.Now()) |
| 464 | return err |
| 465 | }) |
| 466 | if !errors.Is(err, ErrClaimSQLDenied) { |
| 467 | t.Fatalf("receipt replayed under a foreign operation: %v", err) |
| 468 | } |
| 469 | // A restore door on B neither starts an operation nor reopens this one. |
| 470 | var rf *restoreFailure |
| 471 | if err := w.b.restoreApp(w.ctx, w.account(t, w.b), crApp, time.Now()); !errors.As(err, &rf) || !rf.Marker { |
| 472 | t.Fatalf("restore under a foreign operation = %v", err) |
| 473 | } |
| 474 | if now := w.admission(t); now.ETag != closed.ETag { |
| 475 | t.Fatal("canonical admission changed") |
| 476 | } |
| 477 | requireLife(t, w.b, true, false) |
| 478 | } |
| 479 | |
| 480 | func TestClaimRestoreSuccessorIncarnationIsRefused(t *testing.T) { |
| 481 | w, _ := newClaimRestoreWorld(t) |
| 482 | if err := w.a.restoreApp(w.ctx, w.account(t, w.a), crApp, time.Now()); err != nil { |
| 483 | t.Fatal(err) |
| 484 | } |
| 485 | receipt := requireSettledReceipt(t, w.admission(t)) |
| 486 | // A grant for a different incarnation cannot touch this row. |
| 487 | other := claimSQLScope{AccountID: crAcctID, App: crApp, AppID: "incarnation-b"} |
| 488 | r := receipt |
| 489 | r.AppID = "incarnation-b" |
| 490 | err := w.b.store.WithAppMutation(w.ctx, []sqlGrant{restoreReceiptGrant(w.ctx, w.bucket, crSqid, other, r)}, func(v *Store) error { |
| 491 | _, err := v.FinishClaimRestoreRow(crAcctID, crApp, crAppID, r.DeletedAt, time.Now()) |
| 492 | return err |
| 493 | }) |
| 494 | if err == nil { |
| 495 | t.Fatal("successor grant reached the old incarnation") |
| 496 | } |
| 497 | requireLife(t, w.b, true, false) |
| 498 | } |
| 499 | |
| 500 | // A restore's receipt carries the closed epoch its operation ran in. Later |
| 501 | // legitimate operations reopen admission and raise the epoch; a console that |
| 502 | // missed the restore still catches it up, through materialization and |
| 503 | // through its own restore door, because the receipt binds the same immutable |
| 504 | // app_id with receipt.Epoch <= admission.Epoch, no marker and no purge. |
| 505 | func TestClaimRestoreReplaysAfterLaterEpochBumps(t *testing.T) { |
| 506 | for _, via := range []string{"materialize", "door"} { |
| 507 | t.Run(via, func(t *testing.T) { |
| 508 | w, _ := newClaimRestoreWorld(t) |
| 509 | if err := w.a.restoreApp(w.ctx, w.account(t, w.a), crApp, time.Now()); err != nil { |
| 510 | t.Fatal(err) |
| 511 | } |
| 512 | receipt := requireSettledReceipt(t, w.admission(t)) |
| 513 | control, self, err := w.a.claimRestoreRuntime() |
| 514 | if err != nil { |
| 515 | t.Fatal(err) |
| 516 | } |
| 517 | for i := 0; i < 2; i++ { |
| 518 | op := "later-mutation-" + strconv.Itoa(i) |
| 519 | owner := self |
| 520 | owner.Controller = op |
| 521 | h, err := control.BeginExclusive(w.ctx, crSqid, crApp, crAppID, hostops.AppOperation{ID: op, Kind: hostops.OpAppMutation, AppID: crAppID, Phase: "issued", |
| 522 | Intent: map[string]string{"action": "settings"}}, owner) |
| 523 | if err != nil { |
| 524 | t.Fatal(err) |
| 525 | } |
| 526 | if err := h.FinishMutation(w.ctx); err != nil { |
| 527 | t.Fatal(err) |
| 528 | } |
| 529 | } |
| 530 | now := w.admission(t).Value |
| 531 | if now.Epoch < receipt.Epoch+3 || now.LastRestore == nil || *now.LastRestore != receipt { |
| 532 | t.Fatalf("fixture: epoch %d, receipt %+v", now.Epoch, now.LastRestore) |
| 533 | } |
| 534 | requireLife(t, w.b, true, false) |
| 535 | switch via { |
| 536 | case "materialize": |
| 537 | w.materialize(t, w.b) |
| 538 | case "door": |
| 539 | if err := w.b.restoreApp(w.ctx, w.account(t, w.b), crApp, time.Now()); err != nil { |
| 540 | t.Fatalf("B's restore door on a lagging row: %v", err) |
| 541 | } |
| 542 | } |
| 543 | requireLife(t, w.b, false, false) |
| 544 | if after := w.admission(t); after.Value.Epoch <= now.Epoch || after.Value.Op != nil || *after.Value.LastRestore != receipt { |
| 545 | t.Fatal("catching up failed to release its projection operation or changed the receipt") |
| 546 | } |
| 547 | }) |
| 548 | } |
| 549 | } |
| 550 | |
| 551 | // The terminal replay grants bind terminal permits: a callback holding a |
| 552 | // valid restore receipt or trash marker grant can still only write that |
| 553 | // projection's columns of that incarnation's apps row. INSERT, a column the |
| 554 | // projection does not own, DELETE and unbinding app_id are refused by SQLite. |
| 555 | func TestClaimRestoreTerminalGrantsOnlyWriteTheirProjection(t *testing.T) { |
| 556 | w, _ := newClaimRestoreWorld(t) |
| 557 | if err := w.a.restoreApp(w.ctx, w.account(t, w.a), crApp, time.Now()); err != nil { |
| 558 | t.Fatal(err) |
| 559 | } |
| 560 | receipt := requireSettledReceipt(t, w.admission(t)) |
| 561 | scope := claimSQLScope{AccountID: crAcctID, App: crApp, AppID: crAppID} |
| 562 | attempt := func(v *Store) { |
| 563 | t.Helper() |
| 564 | _, err := v.db.Exec(`INSERT INTO audit (app, address, action, created_at, account_id) VALUES (?, 'p', 'forged', 1, ?)`, crApp, crAcctID) |
| 565 | requireDenied(t, "insert", err) |
| 566 | _, err = v.db.Exec(`INSERT INTO apps (app, account_id, created_by, created_at) VALUES ('kass-twin', ?, 'x', 1)`, crAcctID) |
| 567 | requireDenied(t, "insert apps (unbound adoption)", err) |
| 568 | _, err = v.db.Exec(`UPDATE apps SET placement = 'evil' WHERE account_id = ? AND app = ?`, crAcctID, crApp) |
| 569 | requireDenied(t, "foreign column", err) |
| 570 | _, err = v.db.Exec(`UPDATE apps SET app_id = '' WHERE account_id = ? AND app = ?`, crAcctID, crApp) |
| 571 | requireDenied(t, "unbinding app_id", err) |
| 572 | _, err = v.db.Exec(`DELETE FROM apps WHERE account_id = ? AND app = ?`, crAcctID, crApp) |
| 573 | requireDenied(t, "delete", err) |
| 574 | } |
| 575 | err := w.b.claimProjection(w.ctx, w.account(t, w.b), crApp, func(pctx context.Context) error { |
| 576 | return w.b.store.WithAppMutation(pctx, []sqlGrant{restoreReceiptGrant(pctx, w.bucket, crSqid, scope, receipt)}, func(v *Store) error { |
| 577 | attempt(v) |
| 578 | changed, err := v.FinishClaimRestoreRow(crAcctID, crApp, crAppID, receipt.DeletedAt, time.Now()) |
| 579 | if err == nil && !changed { |
| 580 | err = errors.New("the projection itself did not land") |
| 581 | } |
| 582 | return err |
| 583 | }) |
| 584 | }) |
| 585 | if err != nil { |
| 586 | t.Fatal(err) |
| 587 | } |
| 588 | requireLife(t, w.b, false, false) |
| 589 | // A later trash: the trash-marker grant hides the row and nothing else. |
| 590 | if err := release.TrashApp(w.ctx, w.bucket, crSqid, crApp, crOwner, time.Now().Add(time.Minute)); err != nil { |
| 591 | t.Fatal(err) |
| 592 | } |
| 593 | tomb, tag := w.marker(t) |
| 594 | err = w.b.claimProjection(w.ctx, w.account(t, w.b), crApp, func(pctx context.Context) error { |
| 595 | return w.b.store.WithAppMutation(pctx, []sqlGrant{tombstoneGrant(pctx, w.bucket, crSqid, scope, tag)}, func(v *Store) error { |
| 596 | attempt(v) |
| 597 | _, err := v.TrashClaimAppRow(crAcctID, crApp, crAppID, tomb.DeletedAt) |
| 598 | return err |
| 599 | }) |
| 600 | }) |
| 601 | if err != nil { |
| 602 | t.Fatal(err) |
| 603 | } |
| 604 | requireLife(t, w.b, true, false) |
| 605 | if n := countWhere(t, w.b.store, "apps", "app = ? AND placement = '' AND app_id = ?", crApp, crAppID); n != 1 { |
| 606 | t.Fatal("a terminal grant wrote outside its projection") |
| 607 | } |
| 608 | // While any operation owns the app, the marker grant refuses outright: a |
| 609 | // materializer cannot clear a live restore's restoring slot. |
| 610 | w.closeAdmission(t) |
| 611 | err = w.b.store.WithAppMutation(w.ctx, []sqlGrant{tombstoneGrant(w.ctx, w.bucket, crSqid, scope, tag)}, func(v *Store) error { |
| 612 | _, err := v.SetClaimRestoring(crAcctID, crApp, crAppID, false) |
| 613 | return err |
| 614 | }) |
| 615 | if !errors.Is(err, ErrClaimSQLDenied) { |
| 616 | t.Fatalf("trash-marker grant under an operation: %v", err) |
| 617 | } |
| 618 | } |
| 619 | |
| 620 | // The disabled branch is the base restoreApp, byte for byte: its separate |
| 621 | // SQL failure response is unchanged. |
| 622 | func TestClaimDisabledRestorePreservesSQLFailureResponse(t *testing.T) { |
| 623 | fx, c, csrf := trashPageFixture(t) |
| 624 | if _, err := fx.st.db.Exec(`CREATE TRIGGER fail_restore BEFORE UPDATE OF deleted_at ON apps WHEN NEW.deleted_at=0 BEGIN SELECT RAISE(ABORT,'test restore SQL failure'); END`); err != nil { |
| 625 | t.Fatal(err) |
| 626 | } |
| 627 | _, location, _ := postForm(t, c, fx.u("/trash/restore"), url.Values{"csrf": {csrf}, "app": {"kass"}}) |
| 628 | _, _, body := get(t, c, fx.ts.URL+location) |
| 629 | if !strings.Contains(body, "The trash marker was removed") || strings.Contains(body, "Could not remove the trash marker") { |
| 630 | t.Fatalf("wrong disabled SQL failure response: %s", body) |
| 631 | } |
| 632 | if tomb, err := release.ReadAppTombstone(t.Context(), fx.rel, fx.sqid, "kass"); err != nil || tomb != nil { |
| 633 | t.Fatalf("marker: %+v %v", tomb, err) |
| 634 | } |
| 635 | } |
| 636 | |
amadan