| 1 | package release |
| 2 | |
| 3 | import ( |
| 4 | "encoding/json" |
| 5 | "errors" |
| 6 | "os" |
| 7 | "path/filepath" |
| 8 | "reflect" |
| 9 | "testing" |
| 10 | ) |
| 11 | |
| 12 | func rootRestartFixture(t *testing.T) (*RootRestarts, *rootRestartSyncStore, RestartBinding, RestartSnapshot) { |
| 13 | t.Helper() |
| 14 | s := &rootRestartSyncStore{healthyRetryStore: &healthyRetryStore{dir: t.TempDir(), fault: errors.New("injected directory sync failure")}} |
| 15 | b := RestartBinding{Box: "box-a", Account: "bab", App: "console"} |
| 16 | in := RestartSnapshot{Complete: true, AppID: "initial", AppBody: []byte(`{"schema":1}`), Scopes: ScopedRestartGeneration{Schema: 1}, Targets: []RestartTarget{ |
| 17 | {Host: "one.example", Source: "record-one", Installation: "installation-one", Unit: "one.service", Environment: "prod"}, |
| 18 | {Host: "two.example", Source: "record-two", Installation: "installation-two", Unit: "two.service", Environment: "test"}, |
| 19 | }} |
| 20 | return &RootRestarts{Store: s}, s, b, in |
| 21 | } |
| 22 | |
| 23 | func nextRootRestart(in RestartSnapshot) RestartSnapshot { |
| 24 | in.Scopes.Requests = []RestartRequest{{ID: "request-1", RequestedAt: "2026-10-13T12:00:00Z", Scope: RestartScope{Kind: RestartScopeEnvironment, Value: "prod"}}} |
| 25 | return in |
| 26 | } |
| 27 | |
| 28 | func TestPlan14RootRestartConsumption(t *testing.T) { |
| 29 | for _, attack := range []string{"delete", "rename", "forge"} { |
| 30 | t.Run(attack, func(t *testing.T) { |
| 31 | h, _, b, in := rootRestartFixture(t) |
| 32 | if p, err := h.Observe(b, in); err != nil || p != nil { |
| 33 | t.Fatalf("baseline: %v %v", p, err) |
| 34 | } |
| 35 | in = nextRootRestart(in) |
| 36 | p, err := h.Observe(b, in) |
| 37 | if err != nil || p == nil || len(p.Targets) != 1 || p.Targets[0].Target.Host != "one.example" { |
| 38 | t.Fatalf("exact scope: %+v %v", p, err) |
| 39 | } |
| 40 | legacy := filepath.Join(t.TempDir(), "restart-marks.json") |
| 41 | if err := os.WriteFile(legacy, []byte(`{"schema":1,"apps":{},"scopes":{"bab|console":"request-1"}}`), 0600); err != nil { |
| 42 | t.Fatal(err) |
| 43 | } |
| 44 | switch attack { |
| 45 | case "delete": |
| 46 | err = os.Remove(legacy) |
| 47 | case "rename": |
| 48 | err = os.Rename(filepath.Dir(legacy), filepath.Dir(legacy)+"-moved") |
| 49 | t.Cleanup(func() { os.RemoveAll(filepath.Dir(legacy) + "-moved") }) |
| 50 | case "forge": |
| 51 | err = os.WriteFile(legacy, []byte(`{"schema":1,"apps":{},"scopes":{}}`), 0600) |
| 52 | } |
| 53 | if err != nil { |
| 54 | t.Fatal(err) |
| 55 | } |
| 56 | target := p.Targets[0].Target |
| 57 | if err := h.Begin(b, p.ID, target, "boot-a/invocation-a"); err != nil { |
| 58 | t.Fatal(err) |
| 59 | } |
| 60 | if err := h.Complete(b, p.ID, target, "boot-a/invocation-b"); err != nil { |
| 61 | t.Fatal(err) |
| 62 | } |
| 63 | if err := h.Finish(b, p.ID); err != nil { |
| 64 | t.Fatal(err) |
| 65 | } |
| 66 | if p, err := h.Observe(b, in); err != nil || p != nil { |
| 67 | t.Fatalf("replayed after %s: %+v %v", attack, p, err) |
| 68 | } |
| 69 | }) |
| 70 | } |
| 71 | } |
| 72 | |
| 73 | func TestPlan14RootRestartPendingSurvivesRollover(t *testing.T) { |
| 74 | h, _, b, in := rootRestartFixture(t) |
| 75 | if _, err := h.Observe(b, in); err != nil { |
| 76 | t.Fatal(err) |
| 77 | } |
| 78 | in = nextRootRestart(in) |
| 79 | p, err := h.Observe(b, in) |
| 80 | if err != nil { |
| 81 | t.Fatal(err) |
| 82 | } |
| 83 | // A held target is not begun. Neither a new request nor vanished routes may erase it. |
| 84 | in.Targets = nil |
| 85 | in.AppID = "later" |
| 86 | in.Scopes.Requests = []RestartRequest{{ID: "rolled-over", Scope: RestartScope{Kind: RestartScopeHost, Value: "other.example"}}} |
| 87 | again, err := h.Observe(b, in) |
| 88 | if err != nil || !reflect.DeepEqual(p, again) { |
| 89 | t.Fatalf("pending work changed: %+v %v", again, err) |
| 90 | } |
| 91 | if err := h.Complete(b, p.ID, p.Targets[0].Target, "unrelated"); err == nil { |
| 92 | t.Fatal("unperformed request consumed") |
| 93 | } |
| 94 | } |
| 95 | |
| 96 | func TestPlan14RootRestartAmbiguousDoesNotReplay(t *testing.T) { |
| 97 | h, _, b, in := rootRestartFixture(t) |
| 98 | if _, err := h.Observe(b, in); err != nil { |
| 99 | t.Fatal(err) |
| 100 | } |
| 101 | in = nextRootRestart(in) |
| 102 | p, err := h.Observe(b, in) |
| 103 | if err != nil { |
| 104 | t.Fatal(err) |
| 105 | } |
| 106 | target := p.Targets[0].Target |
| 107 | if err := h.Begin(b, p.ID, target, "boot-a/invocation-a"); err != nil { |
| 108 | t.Fatal(err) |
| 109 | } |
| 110 | if err := h.Begin(b, p.ID, target, "boot-a/invocation-a"); err == nil { |
| 111 | t.Fatal("ambiguous attempt granted a second dispatch") |
| 112 | } |
| 113 | for _, change := range []string{"same-invocation", "empty", "source", "installation", "unit", "request"} { |
| 114 | t.Run(change, func(t *testing.T) { |
| 115 | target := target |
| 116 | id := p.ID |
| 117 | after := "boot-a/invocation-b" |
| 118 | switch change { |
| 119 | case "same-invocation": |
| 120 | after = "boot-a/invocation-a" |
| 121 | case "empty": |
| 122 | after = "" |
| 123 | case "source": |
| 124 | target.Source = "changed" |
| 125 | case "installation": |
| 126 | target.Installation = "changed" |
| 127 | case "unit": |
| 128 | target.Unit = "foreign.service" |
| 129 | case "request": |
| 130 | id = "other" |
| 131 | } |
| 132 | if err := h.Complete(b, id, target, after); err == nil { |
| 133 | t.Fatal("ambiguous or foreign evidence consumed request") |
| 134 | } |
| 135 | }) |
| 136 | } |
| 137 | if err := h.Complete(b, p.ID, target, "boot-a/invocation-b"); err != nil { |
| 138 | t.Fatal(err) |
| 139 | } |
| 140 | } |
| 141 | |
| 142 | func TestPlan14RootRestartVisibleSync(t *testing.T) { |
| 143 | for _, boundary := range []string{"baseline", "pending", "begin", "complete"} { |
| 144 | t.Run(boundary, func(t *testing.T) { |
| 145 | h, s, b, in := rootRestartFixture(t) |
| 146 | if boundary == "baseline" { |
| 147 | s.failAt = s.writes + 1 |
| 148 | if _, err := h.Observe(b, in); !errors.Is(err, s.fault) { |
| 149 | t.Fatalf("baseline: %v", err) |
| 150 | } |
| 151 | } |
| 152 | if _, err := h.Observe(b, in); err != nil { |
| 153 | t.Fatal(err) |
| 154 | } |
| 155 | in = nextRootRestart(in) |
| 156 | if boundary == "pending" { |
| 157 | s.failAt = s.writes + 2 |
| 158 | if _, err := h.Observe(b, in); !errors.Is(err, s.fault) { |
| 159 | t.Fatalf("pending: %v", err) |
| 160 | } |
| 161 | } |
| 162 | p, err := h.Observe(b, in) |
| 163 | if err != nil { |
| 164 | t.Fatal(err) |
| 165 | } |
| 166 | target := p.Targets[0].Target |
| 167 | if boundary == "begin" { |
| 168 | s.failAt = s.writes + 2 |
| 169 | if err := h.Begin(b, p.ID, target, "boot-a/invocation-a"); !errors.Is(err, s.fault) { |
| 170 | t.Fatalf("begin: %v", err) |
| 171 | } |
| 172 | if err := h.Begin(b, p.ID, target, "boot-a/invocation-a"); err == nil { |
| 173 | t.Fatal("visible intent dispatched twice") |
| 174 | } |
| 175 | return |
| 176 | } |
| 177 | if err := h.Begin(b, p.ID, target, "boot-a/invocation-a"); err != nil { |
| 178 | t.Fatal(err) |
| 179 | } |
| 180 | if boundary == "complete" { |
| 181 | s.failAt = s.writes + 2 |
| 182 | if err := h.Complete(b, p.ID, target, "boot-a/invocation-b"); !errors.Is(err, s.fault) { |
| 183 | t.Fatalf("complete: %v", err) |
| 184 | } |
| 185 | } |
| 186 | if err := h.Complete(b, p.ID, target, "boot-a/invocation-b"); err != nil { |
| 187 | t.Fatal(err) |
| 188 | } |
| 189 | if err := h.Finish(b, p.ID); err != nil { |
| 190 | t.Fatal(err) |
| 191 | } |
| 192 | if p, err := h.Observe(b, in); err != nil || p != nil { |
| 193 | t.Fatalf("completion not durable: %+v %v", p, err) |
| 194 | } |
| 195 | }) |
| 196 | } |
| 197 | } |
| 198 | |
| 199 | func TestPlan14RootRestartStrictState(t *testing.T) { |
| 200 | for _, change := range []string{"schema", "box", "complete", "revision", "extra", "duplicate", "trailing", "null"} { |
| 201 | t.Run(change, func(t *testing.T) { |
| 202 | h, s, b, in := rootRestartFixture(t) |
| 203 | h.Observe(b, in) |
| 204 | raw, err := s.Read(rootRestartKey(b)) |
| 205 | if err != nil { |
| 206 | t.Fatal(err) |
| 207 | } |
| 208 | var r rootRestartRecord |
| 209 | if err := json.Unmarshal(raw, &r); err != nil { |
| 210 | t.Fatal(err) |
| 211 | } |
| 212 | switch change { |
| 213 | case "schema": |
| 214 | r.Schema++ |
| 215 | case "box": |
| 216 | r.Binding.Box = "other" |
| 217 | case "complete": |
| 218 | r.Complete = 0 |
| 219 | case "revision": |
| 220 | r.Revision = 0 |
| 221 | } |
| 222 | raw, _ = json.Marshal(r) |
| 223 | switch change { |
| 224 | case "extra": |
| 225 | raw = append(raw[:len(raw)-1], []byte(`,"extra":true}`)...) |
| 226 | case "duplicate": |
| 227 | raw = append(raw[:len(raw)-1], []byte(`,"schema":1}`)...) |
| 228 | case "trailing": |
| 229 | raw = append(raw, []byte(` {}`)...) |
| 230 | case "null": |
| 231 | raw = []byte(`null`) |
| 232 | } |
| 233 | if err := s.Write(rootRestartKey(b), raw); err != nil { |
| 234 | t.Fatal(err) |
| 235 | } |
| 236 | if _, err := h.Observe(b, nextRootRestart(in)); err == nil { |
| 237 | t.Fatal("invalid state granted consumption") |
| 238 | } |
| 239 | }) |
| 240 | } |
| 241 | } |
| 242 | |
| 243 | type rootRestartSyncStore struct { |
| 244 | *healthyRetryStore |
| 245 | writes int |
| 246 | failAt int |
| 247 | } |
| 248 | |
| 249 | func (s *rootRestartSyncStore) Write(name string, raw []byte) error { |
| 250 | s.writes++ |
| 251 | if s.writes == s.failAt { |
| 252 | s.fail = 1 |
| 253 | } |
| 254 | return s.healthyRetryStore.Write(name, raw) |
| 255 | } |
| 256 | |
| 257 | func TestPlan14RootRestartSelection(t *testing.T) { |
| 258 | for _, kind := range []string{"app", "host", "environment", "unknown", "retry", "hibernate", "rollover", "incomplete", "absent", "duplicate-target", "duplicate-request"} { |
| 259 | t.Run(kind, func(t *testing.T) { |
| 260 | h, _, b, in := rootRestartFixture(t) |
| 261 | if kind == "absent" { |
| 262 | in.AppID = RootRestartAbsent |
| 263 | in.AppBody = nil |
| 264 | } |
| 265 | if _, err := h.Observe(b, in); err != nil { |
| 266 | t.Fatal(err) |
| 267 | } |
| 268 | in = nextRootRestart(in) |
| 269 | want := 1 |
| 270 | switch kind { |
| 271 | case "app", "absent": |
| 272 | in.AppID = "new-app-request" |
| 273 | in.AppBody = []byte(`{"requested_at":"2026-10-13T12:00:00Z"}`) |
| 274 | want = 2 |
| 275 | case "host": |
| 276 | in.Scopes.Requests[0].Scope = RestartScope{Kind: RestartScopeHost, Value: "two.example"} |
| 277 | case "unknown": |
| 278 | in.Scopes.Requests[0].Scope = RestartScope{Kind: "future-selector", Value: "prod"} |
| 279 | want = 0 |
| 280 | case "retry": |
| 281 | in.Scopes.Requests[0].Scope = RestartScope{Kind: RestartScopeRetry, Value: "one.example"} |
| 282 | want = 0 |
| 283 | case "hibernate": |
| 284 | in.Targets[0].Hibernate = true |
| 285 | want = 0 |
| 286 | case "rollover": |
| 287 | if p, err := h.Observe(b, RestartSnapshot{Complete: true, AppID: in.AppID, AppBody: in.AppBody, Scopes: in.Scopes}); err != nil || p != nil { |
| 288 | t.Fatalf("first empty target request: %+v %v", p, err) |
| 289 | } |
| 290 | in.Scopes.Requests[0].ID = "new-position" |
| 291 | want = 0 |
| 292 | case "incomplete": |
| 293 | in.Complete = false |
| 294 | case "duplicate-target": |
| 295 | in.Targets = append(in.Targets, in.Targets[0]) |
| 296 | case "duplicate-request": |
| 297 | in.Scopes.Requests = append(in.Scopes.Requests, in.Scopes.Requests[0]) |
| 298 | } |
| 299 | p, err := h.Observe(b, in) |
| 300 | if kind == "incomplete" || kind == "duplicate-target" || kind == "duplicate-request" { |
| 301 | if err == nil { |
| 302 | t.Fatal("invalid census/request accepted") |
| 303 | } |
| 304 | return |
| 305 | } |
| 306 | if err != nil { |
| 307 | t.Fatal(err) |
| 308 | } |
| 309 | if want == 0 { |
| 310 | if p != nil { |
| 311 | t.Fatalf("unexpected restart: %+v", p) |
| 312 | } |
| 313 | return |
| 314 | } |
| 315 | if p == nil || len(p.Targets) != want { |
| 316 | t.Fatalf("targets: %+v", p) |
| 317 | } |
| 318 | if kind == "host" && p.Targets[0].Target.Host != "two.example" { |
| 319 | t.Fatal("host selection widened") |
| 320 | } |
| 321 | if kind == "app" && string(p.AppBody) != string(in.AppBody) { |
| 322 | t.Fatal("lost exact app request") |
| 323 | } |
| 324 | if err := h.Finish(b, p.ID); err == nil { |
| 325 | t.Fatal("unperformed targets consumed") |
| 326 | } |
| 327 | if err := h.Begin(b, p.ID, p.Targets[0].Target, "old"); err != nil { |
| 328 | t.Fatal(err) |
| 329 | } |
| 330 | if err := h.Complete(b, p.ID, p.Targets[0].Target, "new"); err != nil { |
| 331 | t.Fatal(err) |
| 332 | } |
| 333 | if want == 2 { |
| 334 | if err := h.Finish(b, p.ID); err == nil { |
| 335 | t.Fatal("partial target set consumed") |
| 336 | } |
| 337 | } |
| 338 | }) |
| 339 | } |
| 340 | } |
| 341 | |
| 342 | func TestPlan14RootRestartFailureIsTerminal(t *testing.T) { |
| 343 | h, _, b, in := rootRestartFixture(t) |
| 344 | if _, err := h.Observe(b, in); err != nil { |
| 345 | t.Fatal(err) |
| 346 | } |
| 347 | in = nextRootRestart(in) |
| 348 | p, err := h.Observe(b, in) |
| 349 | if err != nil { |
| 350 | t.Fatal(err) |
| 351 | } |
| 352 | target := p.Targets[0].Target |
| 353 | if err := h.Fail(b, p.ID, target); err == nil { |
| 354 | t.Fatal("undispatched request consumed as failure") |
| 355 | } |
| 356 | if err := h.Begin(b, p.ID, target, "before"); err != nil { |
| 357 | t.Fatal(err) |
| 358 | } |
| 359 | if err := h.Fail(b, p.ID, target); err != nil { |
| 360 | t.Fatal(err) |
| 361 | } |
| 362 | if err := h.Begin(b, p.ID, target, "before"); err == nil { |
| 363 | t.Fatal("terminal failure retried") |
| 364 | } |
| 365 | if err := h.Complete(b, p.ID, target, "after"); err == nil { |
| 366 | t.Fatal("terminal failure changed to success") |
| 367 | } |
| 368 | if err := h.Finish(b, p.ID); err != nil { |
| 369 | t.Fatal(err) |
| 370 | } |
| 371 | if p, err := h.Observe(b, in); err != nil || p != nil { |
| 372 | t.Fatalf("failure replayed: %+v %v", p, err) |
| 373 | } |
| 374 | in.Scopes.Requests = append(in.Scopes.Requests, RestartRequest{ID: "retry-as-new-request", Scope: RestartScope{Kind: RestartScopeHost, Value: target.Host}}) |
| 375 | if p, err := h.Observe(b, in); err != nil || p == nil { |
| 376 | t.Fatalf("fresh explicit request refused: %+v %v", p, err) |
| 377 | } |
| 378 | } |
| 379 | |
| 380 | func TestPlan14RootRestartReadMustSync(t *testing.T) { |
| 381 | for _, step := range []string{"observe", "begin", "complete", "finish"} { |
| 382 | t.Run(step, func(t *testing.T) { |
| 383 | h, s, b, in := rootRestartFixture(t) |
| 384 | if _, err := h.Observe(b, in); err != nil { |
| 385 | t.Fatal(err) |
| 386 | } |
| 387 | in = nextRootRestart(in) |
| 388 | p, err := h.Observe(b, in) |
| 389 | if err != nil { |
| 390 | t.Fatal(err) |
| 391 | } |
| 392 | target := p.Targets[0].Target |
| 393 | if step == "complete" || step == "finish" { |
| 394 | if err := h.Begin(b, p.ID, target, "before"); err != nil { |
| 395 | t.Fatal(err) |
| 396 | } |
| 397 | } |
| 398 | if step == "finish" { |
| 399 | if err := h.Complete(b, p.ID, target, "after"); err != nil { |
| 400 | t.Fatal(err) |
| 401 | } |
| 402 | } |
| 403 | s.failAt = s.writes + 1 |
| 404 | switch step { |
| 405 | case "observe": |
| 406 | _, err = h.Observe(b, in) |
| 407 | case "begin": |
| 408 | err = h.Begin(b, p.ID, target, "before") |
| 409 | case "complete": |
| 410 | err = h.Complete(b, p.ID, target, "after") |
| 411 | case "finish": |
| 412 | err = h.Finish(b, p.ID) |
| 413 | } |
| 414 | if !errors.Is(err, s.fault) { |
| 415 | t.Fatalf("%s granted without durable read: %v", step, err) |
| 416 | } |
| 417 | }) |
| 418 | } |
| 419 | } |
| 420 | |
| 421 | func TestPlan14RootRestartPendingStateTampering(t *testing.T) { |
| 422 | for _, change := range []string{"remove-target", "append-target", "source", "selector", "body", "id", "completed-without-begin", "failed-without-begin"} { |
| 423 | t.Run(change, func(t *testing.T) { |
| 424 | h, s, b, in := rootRestartFixture(t) |
| 425 | if _, err := h.Observe(b, in); err != nil { |
| 426 | t.Fatal(err) |
| 427 | } |
| 428 | if _, err := h.Observe(b, nextRootRestart(in)); err != nil { |
| 429 | t.Fatal(err) |
| 430 | } |
| 431 | raw, err := s.Read(rootRestartKey(b)) |
| 432 | if err != nil { |
| 433 | t.Fatal(err) |
| 434 | } |
| 435 | var r rootRestartRecord |
| 436 | if err := json.Unmarshal(raw, &r); err != nil { |
| 437 | t.Fatal(err) |
| 438 | } |
| 439 | switch change { |
| 440 | case "remove-target": |
| 441 | r.Pending.Targets = nil |
| 442 | case "append-target": |
| 443 | r.Pending.Targets = append(r.Pending.Targets, RestartProgress{Target: in.Targets[1]}) |
| 444 | case "source": |
| 445 | r.Pending.Targets[0].Target.Source = "other" |
| 446 | case "selector": |
| 447 | r.Pending.Requests[0].Scope.Value = "test" |
| 448 | case "body": |
| 449 | r.Pending.AppBody = []byte("other") |
| 450 | case "id": |
| 451 | r.Pending.ID = "other" |
| 452 | case "completed-without-begin": |
| 453 | r.Pending.Targets[0].After = "after" |
| 454 | case "failed-without-begin": |
| 455 | r.Pending.Targets[0].Failed = true |
| 456 | } |
| 457 | raw, _ = json.Marshal(r) |
| 458 | if err := s.Write(rootRestartKey(b), raw); err != nil { |
| 459 | t.Fatal(err) |
| 460 | } |
| 461 | if _, err := h.Observe(b, nextRootRestart(in)); err == nil { |
| 462 | t.Fatal("changed pending authority accepted") |
| 463 | } |
| 464 | }) |
| 465 | } |
| 466 | } |
| 467 | |
amadan