| 1 | package release |
| 2 | |
| 3 | import ( |
| 4 | "bytes" |
| 5 | "context" |
| 6 | "crypto/ed25519" |
| 7 | "encoding/base64" |
| 8 | "encoding/json" |
| 9 | "errors" |
| 10 | "os" |
| 11 | "path/filepath" |
| 12 | "slices" |
| 13 | "strconv" |
| 14 | "strings" |
| 15 | "testing" |
| 16 | "time" |
| 17 | |
| 18 | "github.com/carlosframework/platform/internal/caps" |
| 19 | "github.com/carlosframework/platform/internal/registry" |
| 20 | ) |
| 21 | |
| 22 | func testRegistry(t *testing.T) *registry.Registry { |
| 23 | t.Helper() |
| 24 | reg, err := registry.Open(filepath.Join(t.TempDir(), "registry.db")) |
| 25 | if err != nil { |
| 26 | t.Fatalf("open registry: %v", err) |
| 27 | } |
| 28 | return reg |
| 29 | } |
| 30 | |
| 31 | func addRoute(t *testing.T, reg *registry.Registry, host, app, channel string, hibernate bool) { |
| 32 | t.Helper() |
| 33 | if err := reg.Upsert(registry.Route{ |
| 34 | Host: host, Kind: "instance", App: app, Channel: channel, Hibernate: hibernate, |
| 35 | }); err != nil { |
| 36 | t.Fatalf("upsert %s: %v", host, err) |
| 37 | } |
| 38 | } |
| 39 | |
| 40 | func TestPassRepointsAwakeAndHibernating(t *testing.T) { |
| 41 | s := NewDirStore(t.TempDir()) |
| 42 | ctx := context.Background() |
| 43 | priv, pub := testKeys(t) |
| 44 | opt := t.TempDir() |
| 45 | reg := testRegistry(t) |
| 46 | |
| 47 | // canary/* has no bake window, so the pass applies immediately. |
| 48 | shipRelease(t, s, "bab", "myapp", "v1", adoptPlatform, []byte("build one")) |
| 49 | if err := Promote(ctx, s, priv, "bab", "myapp", "canary/x", "v1"); err != nil { |
| 50 | t.Fatalf("Promote: %v", err) |
| 51 | } |
| 52 | addRoute(t, reg, "awake.example.com", "myapp", "canary/x", false) |
| 53 | addRoute(t, reg, "sleepy.example.com", "myapp", "canary/x", true) |
| 54 | |
| 55 | var restarted []string |
| 56 | restart := func(host string) error { restarted = append(restarted, host); return nil } |
| 57 | if err := Pass(ctx, reg, s, pub, opt, restart); err != nil { |
| 58 | t.Fatalf("Pass: %v", err) |
| 59 | } |
| 60 | |
| 61 | for _, host := range []string{"awake.example.com", "sleepy.example.com"} { |
| 62 | if v, _ := LiveVersion(opt, host); v != "v1" { |
| 63 | t.Fatalf("%s live = %q, want v1", host, v) |
| 64 | } |
| 65 | rt, err := reg.Get(host) |
| 66 | if err != nil || rt == nil || rt.Version != "v1" { |
| 67 | t.Fatalf("%s registry version = %+v, %v", host, rt, err) |
| 68 | } |
| 69 | } |
| 70 | // Only the awake host was restarted; the hibernating host is NEVER woken |
| 71 | // — its next wake execs the new binary via the live symlink. |
| 72 | if len(restarted) != 1 || restarted[0] != "awake.example.com" { |
| 73 | t.Fatalf("restarted = %v, want only awake.example.com", restarted) |
| 74 | } |
| 75 | // The staged artifact matches the shipped bytes (sha256-gated download). |
| 76 | body, err := os.ReadFile(VersionPath(opt, "bab", "v1", "myapp")) |
| 77 | if err != nil || string(body) != "build one" { |
| 78 | t.Fatalf("staged bytes = %q, %v", body, err) |
| 79 | } |
| 80 | } |
| 81 | |
| 82 | // TestPassRestartHibernatingOptIn pins both sides of the RestartHibernating |
| 83 | // door. Default (RunAdopt's shape, no option): hibernate-flagged routes get |
| 84 | // the pure repoint and the callback never sees them — byte-identical to the |
| 85 | // behavior TestPassRepointsAwakeAndHibernating above pins. Opted in (the |
| 86 | // agent's in-process loops): the callback is consulted for sleepers too, so a |
| 87 | // WARM exec child — minted Hibernate:true, but running right now — can be |
| 88 | // cycled onto the binary this pass just pointed it at. |
| 89 | // |
| 90 | // The pass itself still starts nothing either way: all the option changes is |
| 91 | // whether the callback is asked. Not waking a sleeper is the CALLBACK's |
| 92 | // promise (Bounce returns false when nothing is awake; try-restart is a no-op |
| 93 | // on a stopped unit), which is why handing it a sleeping host is safe. |
| 94 | func TestPassRestartHibernatingOptIn(t *testing.T) { |
| 95 | for _, tc := range []struct { |
| 96 | name string |
| 97 | optIn bool |
| 98 | want []string |
| 99 | reason string |
| 100 | }{ |
| 101 | {"default: sleepers are not offered to the callback", false, |
| 102 | []string{"awake.example.com"}, |
| 103 | "the root adopt pass has no children of its own to cycle"}, |
| 104 | {"opted in: sleepers are offered too", true, |
| 105 | []string{"awake.example.com", "sleepy.example.com"}, |
| 106 | "a warm exec child must bounce onto its new binary"}, |
| 107 | } { |
| 108 | t.Run(tc.name, func(t *testing.T) { |
| 109 | s := NewDirStore(t.TempDir()) |
| 110 | ctx := context.Background() |
| 111 | priv, pub := testKeys(t) |
| 112 | opt := t.TempDir() |
| 113 | reg := testRegistry(t) |
| 114 | |
| 115 | shipRelease(t, s, "bab", "myapp", "v1", adoptPlatform, []byte("build one")) |
| 116 | if err := Promote(ctx, s, priv, "bab", "myapp", "canary/x", "v1"); err != nil { |
| 117 | t.Fatalf("Promote: %v", err) |
| 118 | } |
| 119 | addRoute(t, reg, "awake.example.com", "myapp", "canary/x", false) |
| 120 | addRoute(t, reg, "sleepy.example.com", "myapp", "canary/x", true) |
| 121 | |
| 122 | var restarted []string |
| 123 | restart := func(host string) error { restarted = append(restarted, host); return nil } |
| 124 | var opts []PassOption |
| 125 | if tc.optIn { |
| 126 | opts = append(opts, RestartHibernating(restart)) |
| 127 | } |
| 128 | if err := Pass(ctx, reg, s, pub, opt, restart, opts...); err != nil { |
| 129 | t.Fatalf("Pass: %v", err) |
| 130 | } |
| 131 | if strings.Join(restarted, ",") != strings.Join(tc.want, ",") { |
| 132 | t.Fatalf("restarted = %v, want %v (%s)", restarted, tc.want, tc.reason) |
| 133 | } |
| 134 | // Either way the repoint happened for both — the option changes who |
| 135 | // is offered to the callback, never which hosts adopt. |
| 136 | for _, host := range []string{"awake.example.com", "sleepy.example.com"} { |
| 137 | if v, _ := LiveVersion(opt, host); v != "v1" { |
| 138 | t.Fatalf("%s live = %q, want v1", host, v) |
| 139 | } |
| 140 | } |
| 141 | }) |
| 142 | } |
| 143 | } |
| 144 | |
| 145 | // TestPassRestartHibernatingKeepsSleepersLast pins the ordering the option |
| 146 | // must not disturb: awake hosts are still adopted BEFORE any sleeper, so an |
| 147 | // awake host can discover a boot-broken build and poison it before a |
| 148 | // hibernating host is moved onto it. Handing sleepers a callback must not |
| 149 | // promote them into the first loop. |
| 150 | func TestPassRestartHibernatingKeepsSleepersLast(t *testing.T) { |
| 151 | s := NewDirStore(t.TempDir()) |
| 152 | ctx := context.Background() |
| 153 | priv, pub := testKeys(t) |
| 154 | opt := t.TempDir() |
| 155 | reg := testRegistry(t) |
| 156 | |
| 157 | shipRelease(t, s, "bab", "myapp", "v1", adoptPlatform, []byte("build one")) |
| 158 | if err := Promote(ctx, s, priv, "bab", "myapp", "canary/x", "v1"); err != nil { |
| 159 | t.Fatalf("Promote: %v", err) |
| 160 | } |
| 161 | // Named so a plain registry ordering (alphabetical by host) would put the |
| 162 | // sleeper first — only the awake/hibernating split can produce the order |
| 163 | // this test wants. |
| 164 | addRoute(t, reg, "aaa-sleepy.example.com", "myapp", "canary/x", true) |
| 165 | addRoute(t, reg, "zzz-awake.example.com", "myapp", "canary/x", false) |
| 166 | |
| 167 | var order []string |
| 168 | restart := func(host string) error { order = append(order, host); return nil } |
| 169 | if err := Pass(ctx, reg, s, pub, opt, restart, RestartHibernating(restart)); err != nil { |
| 170 | t.Fatalf("Pass: %v", err) |
| 171 | } |
| 172 | want := []string{"zzz-awake.example.com", "aaa-sleepy.example.com"} |
| 173 | if strings.Join(order, ",") != strings.Join(want, ",") { |
| 174 | t.Fatalf("restart order = %v, want %v (awake hosts must still go first)", order, want) |
| 175 | } |
| 176 | } |
| 177 | |
| 178 | // TestPassRestartHibernatingPoisonsLikeAnyOtherHost: a sleeper's failed |
| 179 | // restart takes the same rollback the awake path takes — the symlink goes |
| 180 | // back and the version is poisoned for the rest of the pass. Nothing about |
| 181 | // hibernation makes a broken build safe to leave pointed at. |
| 182 | func TestPassRestartHibernatingPoisonsLikeAnyOtherHost(t *testing.T) { |
| 183 | s := NewDirStore(t.TempDir()) |
| 184 | ctx := context.Background() |
| 185 | priv, pub := testKeys(t) |
| 186 | opt := t.TempDir() |
| 187 | reg := testRegistry(t) |
| 188 | |
| 189 | shipRelease(t, s, "bab", "myapp", "v1", adoptPlatform, []byte("build one")) |
| 190 | shipRelease(t, s, "bab", "myapp", "v2", adoptPlatform, []byte("build two")) |
| 191 | if err := Promote(ctx, s, priv, "bab", "myapp", "canary/x", "v1"); err != nil { |
| 192 | t.Fatalf("Promote v1: %v", err) |
| 193 | } |
| 194 | addRoute(t, reg, "sleepy.example.com", "myapp", "canary/x", true) |
| 195 | ok := func(string) error { return nil } |
| 196 | if err := Pass(ctx, reg, s, pub, opt, ok, RestartHibernating(ok)); err != nil { |
| 197 | t.Fatalf("Pass v1: %v", err) |
| 198 | } |
| 199 | |
| 200 | if err := Promote(ctx, s, priv, "bab", "myapp", "canary/x", "v2"); err != nil { |
| 201 | t.Fatalf("Promote v2: %v", err) |
| 202 | } |
| 203 | boom := func(string) error { return errors.New("bounce failed") } |
| 204 | if err := Pass(ctx, reg, s, pub, opt, boom, RestartHibernating(boom)); err != nil { |
| 205 | t.Fatalf("Pass v2: %v", err) |
| 206 | } |
| 207 | if v, _ := LiveVersion(opt, "sleepy.example.com"); v != "v1" { |
| 208 | t.Fatalf("live = %q, want the symlink rolled back to v1", v) |
| 209 | } |
| 210 | } |
| 211 | |
| 212 | // TestPassBakeDirMovesTheClock proves the BakeDir option's whole point |
| 213 | // (issue #13): the bake clock lands under the given dir, not optDir's |
| 214 | // legacy /opt-dir location. |
| 215 | func TestPassBakeDirMovesTheClock(t *testing.T) { |
| 216 | s := NewDirStore(t.TempDir()) |
| 217 | ctx := context.Background() |
| 218 | priv, pub := testKeys(t) |
| 219 | opt := t.TempDir() |
| 220 | bakeDir := t.TempDir() |
| 221 | reg := testRegistry(t) |
| 222 | |
| 223 | shipRelease(t, s, "bab", "myapp", "v1", adoptPlatform, []byte("build one")) |
| 224 | if err := Promote(ctx, s, priv, "bab", "myapp", "canary/x", "v1"); err != nil { |
| 225 | t.Fatalf("Promote: %v", err) |
| 226 | } |
| 227 | addRoute(t, reg, "awake.example.com", "myapp", "canary/x", false) |
| 228 | |
| 229 | if err := Pass(ctx, reg, s, pub, opt, func(string) error { return nil }, BakeDir(bakeDir)); err != nil { |
| 230 | t.Fatalf("Pass: %v", err) |
| 231 | } |
| 232 | if _, err := os.Stat(filepath.Join(bakeDir, "channel-first-seen.json")); err != nil { |
| 233 | t.Fatalf("BakeDir option did not land the clock in bakeDir: %v", err) |
| 234 | } |
| 235 | if _, err := os.Stat(FirstSeenPath(opt)); !os.IsNotExist(err) { |
| 236 | t.Fatalf("bake clock also written to the legacy opt dir despite BakeDir (stat err %v)", err) |
| 237 | } |
| 238 | } |
| 239 | |
| 240 | // TestPassMigratesBakeClockDurably proves the migration off the legacy path |
| 241 | // completes in ONE pass even when nothing new was observed (2026-08-03 |
| 242 | // review, IMPORTANT): Save is changed-gated, so with the head already |
| 243 | // recorded in the legacy clock, Observe reports no change this pass and the |
| 244 | // ordinary "tick and save" path would never write the new location — a box |
| 245 | // could run for days with the new path absent, and deleting the legacy file |
| 246 | // in that window would silently reset the channel's bake window. The legacy |
| 247 | // clock here already has this exact head, so the ONLY way the new path ends |
| 248 | // up populated is the migration forcing its own first Save. |
| 249 | func TestPassMigratesBakeClockDurably(t *testing.T) { |
| 250 | s := NewDirStore(t.TempDir()) |
| 251 | ctx := context.Background() |
| 252 | priv, pub := testKeys(t) |
| 253 | opt := t.TempDir() |
| 254 | bakeDir := t.TempDir() |
| 255 | reg := testRegistry(t) |
| 256 | |
| 257 | shipRelease(t, s, "bab", "myapp", "v1", adoptPlatform, []byte("build one")) |
| 258 | if err := Promote(ctx, s, priv, "bab", "myapp", "canary/x", "v1"); err != nil { |
| 259 | t.Fatalf("Promote: %v", err) |
| 260 | } |
| 261 | addRoute(t, reg, "awake.example.com", "myapp", "canary/x", false) |
| 262 | |
| 263 | legacy := LoadFirstSeen(FirstSeenPath(opt)) |
| 264 | legacy.Observe("bab", "myapp", "canary/x", "v1", time.Now()) |
| 265 | if err := legacy.Save(); err != nil { |
| 266 | t.Fatalf("seed legacy clock: %v", err) |
| 267 | } |
| 268 | |
| 269 | if err := Pass(ctx, reg, s, pub, opt, func(string) error { return nil }, BakeDir(bakeDir)); err != nil { |
| 270 | t.Fatalf("Pass: %v", err) |
| 271 | } |
| 272 | |
| 273 | st := LoadFirstSeen(filepath.Join(bakeDir, "channel-first-seen.json")) |
| 274 | if _, ok := st.EligibleAt("bab", "myapp", "canary/x", "v1", DefaultHolds); !ok { |
| 275 | t.Fatal("migrated content did not land in the new path after one pass with no new head observed") |
| 276 | } |
| 277 | } |
| 278 | |
| 279 | func TestPassPoisonedBuildSkipsRemainingHosts(t *testing.T) { |
| 280 | s := NewDirStore(t.TempDir()) |
| 281 | ctx := context.Background() |
| 282 | priv, pub := testKeys(t) |
| 283 | opt := t.TempDir() |
| 284 | reg := testRegistry(t) |
| 285 | |
| 286 | // Both hosts start on v1 (staged locally), channel head moves to v2. |
| 287 | shipRelease(t, s, "bab", "myapp", "v1", adoptPlatform, []byte("good")) |
| 288 | shipRelease(t, s, "bab", "myapp", "v2", adoptPlatform, []byte("bad")) |
| 289 | if err := StageArtifact(opt, "bab", "v1", "myapp", []byte("good")); err != nil { |
| 290 | t.Fatal(err) |
| 291 | } |
| 292 | for _, h := range []string{"a.example.com", "b.example.com"} { |
| 293 | addRoute(t, reg, h, "myapp", "canary/x", false) |
| 294 | if err := PointLive(opt, h, "bab", "v1", "myapp"); err != nil { |
| 295 | t.Fatal(err) |
| 296 | } |
| 297 | } |
| 298 | if err := Promote(ctx, s, priv, "bab", "myapp", "canary/x", "v2"); err != nil { |
| 299 | t.Fatalf("Promote v2: %v", err) |
| 300 | } |
| 301 | |
| 302 | // v2 fails its restart on the first host — it must be poisoned for the |
| 303 | // rest of the pass, never cycling every host in turn. |
| 304 | var calls int |
| 305 | restart := func(host string) error { calls++; return errors.New("unit failed") } |
| 306 | if err := Pass(ctx, reg, s, pub, opt, restart); err != nil { |
| 307 | t.Fatalf("Pass: %v", err) |
| 308 | } |
| 309 | if calls != 1 { |
| 310 | t.Fatalf("restart called %d times, want 1 (poisoned build must not bounce every host)", calls) |
| 311 | } |
| 312 | // The failed host was rolled back to what it ran; the second never moved. |
| 313 | for _, h := range []string{"a.example.com", "b.example.com"} { |
| 314 | if v, _ := LiveVersion(opt, h); v != "v1" { |
| 315 | t.Fatalf("%s live = %q, want v1", h, v) |
| 316 | } |
| 317 | } |
| 318 | } |
| 319 | |
| 320 | func TestPassStableHeldByBakeWindow(t *testing.T) { |
| 321 | s := NewDirStore(t.TempDir()) |
| 322 | ctx := context.Background() |
| 323 | priv, pub := testKeys(t) |
| 324 | opt := t.TempDir() |
| 325 | reg := testRegistry(t) |
| 326 | |
| 327 | shipRelease(t, s, "bab", "myapp", "v1", adoptPlatform, []byte("bin")) |
| 328 | if err := Promote(ctx, s, priv, "bab", "myapp", "stable", "v1"); err != nil { |
| 329 | t.Fatalf("Promote: %v", err) |
| 330 | } |
| 331 | // The app is DECLARED production (2026-08-04 spec §2): the window is no |
| 332 | // longer unconditional — it is what the flag buys. |
| 333 | if err := SetProduction(ctx, s, priv, "bab", "myapp", true, "test"); err != nil { |
| 334 | t.Fatalf("SetProduction: %v", err) |
| 335 | } |
| 336 | addRoute(t, reg, "h.example.com", "myapp", "", false) // "" = stable |
| 337 | |
| 338 | if err := Pass(ctx, reg, s, pub, opt, func(string) error { return nil }); err != nil { |
| 339 | t.Fatalf("Pass: %v", err) |
| 340 | } |
| 341 | // A brand-new stable head is inside its 72h supply-chain window: the |
| 342 | // clock started this pass, nothing installs. |
| 343 | if v, _ := LiveVersion(opt, "h.example.com"); v != "" { |
| 344 | t.Fatalf("live = %q, want unmoved (bake hold)", v) |
| 345 | } |
| 346 | // The pass persisted the first sighting so the window is measured from |
| 347 | // the box's own clock. |
| 348 | st := LoadFirstSeen(FirstSeenPath(opt)) |
| 349 | if _, ok := st.EligibleAt("bab", "myapp", "stable", "v1", DefaultHolds); !ok { |
| 350 | t.Fatalf("Pass did not tick the first-seen clock") |
| 351 | } |
| 352 | } |
| 353 | |
| 354 | func TestPassSkipsUnadoptableRoutes(t *testing.T) { |
| 355 | s := NewDirStore(t.TempDir()) |
| 356 | ctx := context.Background() |
| 357 | priv, pub := testKeys(t) |
| 358 | opt := t.TempDir() |
| 359 | reg := testRegistry(t) |
| 360 | |
| 361 | shipRelease(t, s, "bab", "myapp", "v1", adoptPlatform, []byte("bin")) |
| 362 | if err := Promote(ctx, s, priv, "bab", "myapp", "canary/x", "v1"); err != nil { |
| 363 | t.Fatalf("Promote: %v", err) |
| 364 | } |
| 365 | // A static route, an app-less instance, and a mid-provision instance: |
| 366 | // none may be moved by an adoption pass. |
| 367 | if err := reg.Upsert(registry.Route{Host: "site.example.com", Kind: "static", App: "myapp", Channel: "canary/x"}); err != nil { |
| 368 | t.Fatal(err) |
| 369 | } |
| 370 | addRoute(t, reg, "noapp.example.com", "", "canary/x", false) |
| 371 | if err := reg.Upsert(registry.Route{Host: "prov.example.com", Kind: "instance", App: "myapp", Channel: "canary/x", Status: "creating"}); err != nil { |
| 372 | t.Fatal(err) |
| 373 | } |
| 374 | if err := Pass(ctx, reg, s, pub, opt, func(string) error { return nil }); err != nil { |
| 375 | t.Fatalf("Pass: %v", err) |
| 376 | } |
| 377 | for _, h := range []string{"site.example.com", "noapp.example.com", "prov.example.com"} { |
| 378 | if v, _ := LiveVersion(opt, h); v != "" { |
| 379 | t.Fatalf("%s was moved (%q) but is not adoptable", h, v) |
| 380 | } |
| 381 | } |
| 382 | } |
| 383 | |
| 384 | // TestPassSkipsPlatformMintedRoutes: a platform domain is an ALIAS for an |
| 385 | // app's canonical route, not a lifecycle host of its own. Adopting one would |
| 386 | // stage a second copy of the artifacts, mint a live/<alias> symlink no unit |
| 387 | // ever execs, try-restart a carlos-app@<alias> unit that doesn't exist, and |
| 388 | // stamp Version onto a row the console then renders as if it were a real |
| 389 | // instance. The canonical route it aliases is adopted normally — that IS the |
| 390 | // deploy. |
| 391 | func TestPassSkipsPlatformMintedRoutes(t *testing.T) { |
| 392 | s := NewDirStore(t.TempDir()) |
| 393 | ctx := context.Background() |
| 394 | priv, pub := testKeys(t) |
| 395 | opt := t.TempDir() |
| 396 | reg := testRegistry(t) |
| 397 | |
| 398 | shipRelease(t, s, "bab", "myapp", "v1", adoptPlatform, []byte("bin")) |
| 399 | if err := Promote(ctx, s, priv, "bab", "myapp", "canary/x", "v1"); err != nil { |
| 400 | t.Fatalf("Promote: %v", err) |
| 401 | } |
| 402 | // The canonical route and its platform alias, identical but for the flag. |
| 403 | addRoute(t, reg, "myapp.example.com", "myapp", "canary/x", false) |
| 404 | if err := reg.Upsert(registry.Route{Host: "myapp.bab.oncarlos.test", Kind: "instance", |
| 405 | App: "myapp", Channel: "canary/x", Status: "ready", Platform: true}); err != nil { |
| 406 | t.Fatal(err) |
| 407 | } |
| 408 | var restarted []string |
| 409 | if err := Pass(ctx, reg, s, pub, opt, func(h string) error { |
| 410 | restarted = append(restarted, h) |
| 411 | return nil |
| 412 | }); err != nil { |
| 413 | t.Fatalf("Pass: %v", err) |
| 414 | } |
| 415 | if v, _ := LiveVersion(opt, "myapp.bab.oncarlos.test"); v != "" { |
| 416 | t.Errorf("platform alias was adopted (live = %q) — aliases are not lifecycle hosts", v) |
| 417 | } |
| 418 | for _, h := range restarted { |
| 419 | if h == "myapp.bab.oncarlos.test" { |
| 420 | t.Error("adopt try-restarted a platform alias's nonexistent unit") |
| 421 | } |
| 422 | } |
| 423 | if rt, _ := reg.Get("myapp.bab.oncarlos.test"); rt != nil && rt.Version != "" { |
| 424 | t.Errorf("adopt stamped Version %q onto a platform alias row", rt.Version) |
| 425 | } |
| 426 | // The canonical route still adopts: the alias must not have stolen its |
| 427 | // pass (canaries bypass bake, so this lands immediately). |
| 428 | if v, _ := LiveVersion(opt, "myapp.example.com"); v != "v1" { |
| 429 | t.Errorf("canonical route live = %q, want v1 — skipping aliases must not skip real hosts", v) |
| 430 | } |
| 431 | } |
| 432 | |
| 433 | // The test that proves per-account app names actually work end to end: two |
| 434 | // routes, the SAME app name, different accounts, different channel pointers. |
| 435 | // Each host must adopt its own account's version. Before accounts, adopt |
| 436 | // grouped by (app, channel) and would have installed one of them on both. |
| 437 | func TestPassResolvesPerAccount(t *testing.T) { |
| 438 | s := NewDirStore(t.TempDir()) |
| 439 | ctx := context.Background() |
| 440 | priv, pub := testKeys(t) |
| 441 | opt := t.TempDir() |
| 442 | reg := testRegistry(t) |
| 443 | |
| 444 | shipRelease(t, s, "bab", "blog", "v1", adoptPlatform, []byte("ops build")) |
| 445 | shipRelease(t, s, "kqmv", "blog", "v2", adoptPlatform, []byte("kass build")) |
| 446 | if err := Promote(ctx, s, priv, "bab", "blog", "canary/x", "v1"); err != nil { |
| 447 | t.Fatal(err) |
| 448 | } |
| 449 | if err := Promote(ctx, s, priv, "kqmv", "blog", "canary/x", "v2"); err != nil { |
| 450 | t.Fatal(err) |
| 451 | } |
| 452 | for _, rt := range []registry.Route{ |
| 453 | {Host: "ops.example.com", Kind: "instance", App: "blog", Channel: "canary/x", Account: "bab", Status: "ready"}, |
| 454 | {Host: "kass.example.com", Kind: "instance", App: "blog", Channel: "canary/x", Account: "kqmv", Status: "ready"}, |
| 455 | } { |
| 456 | if err := reg.Upsert(rt); err != nil { |
| 457 | t.Fatal(err) |
| 458 | } |
| 459 | } |
| 460 | if err := Pass(ctx, reg, s, pub, opt, func(string) error { return nil }); err != nil { |
| 461 | t.Fatalf("Pass: %v", err) |
| 462 | } |
| 463 | for host, want := range map[string]string{ |
| 464 | "ops.example.com": "v1", |
| 465 | "kass.example.com": "v2", |
| 466 | } { |
| 467 | if got, _ := LiveVersion(opt, host); got != want { |
| 468 | t.Errorf("%s live = %q, want %q — each account follows its OWN pointer", host, got, want) |
| 469 | } |
| 470 | } |
| 471 | } |
| 472 | |
| 473 | // Same app name, same VERSION STRING, different accounts, different bytes. |
| 474 | // The on-box version store must keep them apart, or one account's binary gets |
| 475 | // executed under another account's host — the same collision in a different |
| 476 | // layer from the channel-pointer one above. |
| 477 | func TestPassStagesPerAccountForIdenticalVersions(t *testing.T) { |
| 478 | s := NewDirStore(t.TempDir()) |
| 479 | ctx := context.Background() |
| 480 | priv, pub := testKeys(t) |
| 481 | opt := t.TempDir() |
| 482 | reg := testRegistry(t) |
| 483 | |
| 484 | shipRelease(t, s, "bab", "blog", "v1", adoptPlatform, []byte("ops bytes")) |
| 485 | shipRelease(t, s, "kqmv", "blog", "v1", adoptPlatform, []byte("kass bytes")) |
| 486 | for _, acct := range []string{"bab", "kqmv"} { |
| 487 | if err := Promote(ctx, s, priv, acct, "blog", "canary/x", "v1"); err != nil { |
| 488 | t.Fatal(err) |
| 489 | } |
| 490 | } |
| 491 | for _, rt := range []registry.Route{ |
| 492 | {Host: "ops.example.com", Kind: "instance", App: "blog", Channel: "canary/x", Account: "bab", Status: "ready"}, |
| 493 | {Host: "kass.example.com", Kind: "instance", App: "blog", Channel: "canary/x", Account: "kqmv", Status: "ready"}, |
| 494 | } { |
| 495 | if err := reg.Upsert(rt); err != nil { |
| 496 | t.Fatal(err) |
| 497 | } |
| 498 | } |
| 499 | if err := Pass(ctx, reg, s, pub, opt, func(string) error { return nil }); err != nil { |
| 500 | t.Fatalf("Pass: %v", err) |
| 501 | } |
| 502 | for host, want := range map[string]string{ |
| 503 | "ops.example.com": "ops bytes", |
| 504 | "kass.example.com": "kass bytes", |
| 505 | } { |
| 506 | target, err := os.Readlink(LiveLink(opt, host)) |
| 507 | if err != nil { |
| 508 | t.Fatalf("%s: read live link: %v", host, err) |
| 509 | } |
| 510 | got, err := os.ReadFile(target) |
| 511 | if err != nil { |
| 512 | t.Fatalf("%s: read staged binary: %v", host, err) |
| 513 | } |
| 514 | if string(got) != want { |
| 515 | t.Errorf("%s runs %q, want %q — one account's build was staged over another's", host, got, want) |
| 516 | } |
| 517 | } |
| 518 | } |
| 519 | |
| 520 | // The ops account keeps the pre-accounts on-disk layout byte for byte, so |
| 521 | // rolling this code out does not restage and restart every existing host. |
| 522 | func TestOpsVersionPathIsUnchanged(t *testing.T) { |
| 523 | opt := "/opt/carlos" |
| 524 | if got := VersionPath(opt, "bab", "v1", "kass"); got != "/opt/carlos/versions/v1/kass" { |
| 525 | t.Errorf("ops VersionPath = %q, want the pre-accounts path", got) |
| 526 | } |
| 527 | if got := VersionPath(opt, "kqmv", "v1", "kass"); got != "/opt/carlos/versions/v1/kqmv/kass" { |
| 528 | t.Errorf("non-ops VersionPath = %q", got) |
| 529 | } |
| 530 | } |
| 531 | |
| 532 | func TestPassMissingPlatformArtifactSkips(t *testing.T) { |
| 533 | s := NewDirStore(t.TempDir()) |
| 534 | ctx := context.Background() |
| 535 | priv, pub := testKeys(t) |
| 536 | opt := t.TempDir() |
| 537 | reg := testRegistry(t) |
| 538 | |
| 539 | // Shipped for some other platform only: this box must skip, never |
| 540 | // install the wrong arch. |
| 541 | shipRelease(t, s, "bab", "myapp", "v1", "plan9-mips", []byte("bin")) |
| 542 | if err := Promote(ctx, s, priv, "bab", "myapp", "canary/x", "v1"); err != nil { |
| 543 | t.Fatalf("Promote: %v", err) |
| 544 | } |
| 545 | addRoute(t, reg, "h.example.com", "myapp", "canary/x", false) |
| 546 | if err := Pass(ctx, reg, s, pub, opt, func(string) error { return nil }); err != nil { |
| 547 | t.Fatalf("Pass: %v", err) |
| 548 | } |
| 549 | if v, _ := LiveVersion(opt, "h.example.com"); v != "" { |
| 550 | t.Fatalf("live = %q, want unmoved (no artifact for this platform)", v) |
| 551 | } |
| 552 | } |
| 553 | |
| 554 | func TestPassFilterOnlyTouchesRoutesExtraAccepts(t *testing.T) { |
| 555 | s := NewDirStore(t.TempDir()) |
| 556 | ctx := context.Background() |
| 557 | priv, pub := testKeys(t) |
| 558 | opt := t.TempDir() |
| 559 | reg := testRegistry(t) |
| 560 | |
| 561 | shipRelease(t, s, "bab", "myapp", "v1", adoptPlatform, []byte("build one")) |
| 562 | if err := Promote(ctx, s, priv, "bab", "myapp", "canary/x", "v1"); err != nil { |
| 563 | t.Fatalf("Promote canary/x: %v", err) |
| 564 | } |
| 565 | if err := Promote(ctx, s, priv, "bab", "myapp", "stable", "v1"); err != nil { |
| 566 | t.Fatalf("Promote stable: %v", err) |
| 567 | } |
| 568 | addRoute(t, reg, "canary.example.com", "myapp", "canary/x", false) |
| 569 | addRoute(t, reg, "stable.example.com", "myapp", "stable", false) |
| 570 | |
| 571 | canaryOnly := func(rt registry.Route) bool { return rt.Channel == "canary/x" } |
| 572 | restart := func(string) error { return nil } |
| 573 | if err := PassFilter(ctx, reg, s, pub, opt, restart, canaryOnly); err != nil { |
| 574 | t.Fatalf("PassFilter: %v", err) |
| 575 | } |
| 576 | |
| 577 | if v, _ := LiveVersion(opt, "canary.example.com"); v != "v1" { |
| 578 | t.Errorf("canary.example.com live = %q, want v1 (extra accepts it)", v) |
| 579 | } |
| 580 | if v, _ := LiveVersion(opt, "stable.example.com"); v != "" { |
| 581 | t.Errorf("stable.example.com live = %q, want unmoved — PassFilter must not touch a route extra rejects", v) |
| 582 | } |
| 583 | } |
| 584 | |
| 585 | // WithPointerGate (S3 change-index design §5): a group the gate skips is out |
| 586 | // of the pass, its pointer unread; every other group is read as before and |
| 587 | // its outcome reported once. |
| 588 | func TestPassFilterPointerGate(t *testing.T) { |
| 589 | s := NewDirStore(t.TempDir()) |
| 590 | ctx := context.Background() |
| 591 | priv, pub := testKeys(t) |
| 592 | opt := t.TempDir() |
| 593 | reg := testRegistry(t) |
| 594 | |
| 595 | shipRelease(t, s, "bab", "myapp", "v1", adoptPlatform, []byte("build one")) |
| 596 | for _, ch := range []string{"canary/x", "stable"} { |
| 597 | if err := Promote(ctx, s, priv, "bab", "myapp", ch, "v1"); err != nil { |
| 598 | t.Fatalf("Promote %s: %v", ch, err) |
| 599 | } |
| 600 | } |
| 601 | addRoute(t, reg, "canary.example.com", "myapp", "canary/x", false) |
| 602 | addRoute(t, reg, "stable.example.com", "myapp", "stable", false) |
| 603 | addRoute(t, reg, "beta.example.com", "myapp", "beta", false) // nothing promoted |
| 604 | |
| 605 | asked := map[string][]string{} |
| 606 | read := map[string]string{} |
| 607 | gate := func(acct, app, channel string, routes []registry.Route) (bool, func(*ChannelPointer, error)) { |
| 608 | for _, rt := range routes { |
| 609 | asked[channel] = append(asked[channel], rt.Host) |
| 610 | } |
| 611 | if channel == "canary/x" { |
| 612 | return true, nil |
| 613 | } |
| 614 | return false, func(ptr *ChannelPointer, err error) { |
| 615 | switch { |
| 616 | case err != nil: |
| 617 | read[channel] += "error;" |
| 618 | default: |
| 619 | read[channel] += ptr.Version + ";" |
| 620 | } |
| 621 | } |
| 622 | } |
| 623 | if err := PassFilter(ctx, reg, s, pub, opt, func(string) error { return nil }, nil, WithPointerGate(gate)); err != nil { |
| 624 | t.Fatalf("PassFilter: %v", err) |
| 625 | } |
| 626 | want := map[string][]string{"canary/x": {"canary.example.com"}, "stable": {"stable.example.com"}, "beta": {"beta.example.com"}} |
| 627 | if len(asked) != len(want) { |
| 628 | t.Fatalf("the gate was asked about %v, want %v", asked, want) |
| 629 | } |
| 630 | for ch, hosts := range want { |
| 631 | if !slices.Equal(asked[ch], hosts) { |
| 632 | t.Fatalf("the gate was asked about %s with %v, want %v", ch, asked[ch], hosts) |
| 633 | } |
| 634 | } |
| 635 | if v, _ := LiveVersion(opt, "canary.example.com"); v != "" { |
| 636 | t.Errorf("a skipped group adopted: canary.example.com live = %q", v) |
| 637 | } |
| 638 | if v, _ := LiveVersion(opt, "stable.example.com"); v != "v1" { |
| 639 | t.Errorf("stable.example.com live = %q, want v1", v) |
| 640 | } |
| 641 | if read["stable"] != "v1;" || read["beta"] != "error;" || read["canary/x"] != "" { |
| 642 | t.Errorf("read outcomes %v, want stable v1 once, beta's error once, nothing for the skipped group", read) |
| 643 | } |
| 644 | } |
| 645 | |
| 646 | // writeBackdatedPointer signs and writes a channel pointer with an arbitrary |
| 647 | // PromotedAt, bypassing Promote/signPointer's own "stamp now" behavior — the |
| 648 | // only way to get an EXPIRED canary pointer into a store under test, since |
| 649 | // every public promotion path stamps the real clock. Kept in this internal |
| 650 | // test package (never exported) precisely because forging a pointer's lease |
| 651 | // is not something production code should ever be able to do. |
| 652 | func writeBackdatedPointer(t *testing.T, s Store, priv ed25519.PrivateKey, acct, app, channel, version string, promotedAt time.Time) { |
| 653 | t.Helper() |
| 654 | p := &ChannelPointer{Schema: 1, App: app, Channel: channel, Version: version, PromotedAt: promotedAt} |
| 655 | payload, err := json.MarshalIndent(p, "", " ") |
| 656 | if err != nil { |
| 657 | t.Fatalf("marshal backdated pointer: %v", err) |
| 658 | } |
| 659 | b64 := base64.StdEncoding.EncodeToString(payload) |
| 660 | sig := ed25519.Sign(priv, []byte(b64)) |
| 661 | wrapper, err := json.MarshalIndent(signedWrapper{ |
| 662 | Payload: b64, Sig: base64.StdEncoding.EncodeToString(sig), |
| 663 | }, "", " ") |
| 664 | if err != nil { |
| 665 | t.Fatalf("marshal backdated wrapper: %v", err) |
| 666 | } |
| 667 | if _, err := s.Put(context.Background(), ChannelKey(acct, app, channel), wrapper, ""); err != nil { |
| 668 | t.Fatalf("write backdated pointer: %v", err) |
| 669 | } |
| 670 | } |
| 671 | |
| 672 | // TestPassAdoptsAnExpiredCanary pins spec §5's core promise: expiry is |
| 673 | // VISIBILITY, not enforcement. A canary/* route whose pointer is well past |
| 674 | // CanaryTTL must still adopt exactly like a fresh one — nothing about |
| 675 | // expiry may block a repoint, a restart, or a rollback/promote (the same |
| 676 | // signed-pointer plumbing every other channel uses; expiry checks nothing |
| 677 | // about the ladder or the CAS path). |
| 678 | func TestPassAdoptsAnExpiredCanary(t *testing.T) { |
| 679 | s := NewDirStore(t.TempDir()) |
| 680 | ctx := context.Background() |
| 681 | priv, pub := testKeys(t) |
| 682 | opt := t.TempDir() |
| 683 | reg := testRegistry(t) |
| 684 | |
| 685 | shipRelease(t, s, "bab", "myapp", "v1", adoptPlatform, []byte("build one")) |
| 686 | writeBackdatedPointer(t, s, priv, "bab", "myapp", "canary/x", "v1", |
| 687 | time.Now().Add(-(CanaryTTL + 30*24*time.Hour))) // long expired |
| 688 | addRoute(t, reg, "old-canary.example.com", "myapp", "canary/x", false) |
| 689 | |
| 690 | var restarted []string |
| 691 | restart := func(host string) error { restarted = append(restarted, host); return nil } |
| 692 | if err := Pass(ctx, reg, s, pub, opt, restart); err != nil { |
| 693 | t.Fatalf("Pass: %v", err) |
| 694 | } |
| 695 | |
| 696 | if v, _ := LiveVersion(opt, "old-canary.example.com"); v != "v1" { |
| 697 | t.Fatalf("an expired canary route did not adopt: live = %q, want v1", v) |
| 698 | } |
| 699 | if len(restarted) != 1 || restarted[0] != "old-canary.example.com" { |
| 700 | t.Fatalf("an expired canary route was not restarted: restarted = %v", restarted) |
| 701 | } |
| 702 | rt, err := reg.Get("old-canary.example.com") |
| 703 | if err != nil || rt == nil || rt.Version != "v1" { |
| 704 | t.Fatalf("registry version after adopting an expired canary = %+v, %v", rt, err) |
| 705 | } |
| 706 | |
| 707 | // Promotion off (or onto) an expired canary channel is unaffected too — |
| 708 | // expiry is computed from PromotedAt at read time, never stored, so |
| 709 | // nothing about a pointer's own expiry state can gate a write to it. |
| 710 | shipRelease(t, s, "bab", "myapp", "v2", adoptPlatform, []byte("build two")) |
| 711 | if err := Promote(ctx, s, priv, "bab", "myapp", "canary/x", "v2"); err != nil { |
| 712 | t.Fatalf("promoting a new version onto an expired canary channel: %v", err) |
| 713 | } |
| 714 | p, err := Channel(ctx, s, pub, "bab", "myapp", "canary/x") |
| 715 | if err != nil { |
| 716 | t.Fatalf("Channel after re-promoting an expired canary: %v", err) |
| 717 | } |
| 718 | if p.Version != "v2" { |
| 719 | t.Fatalf("channel = %q after promote, want v2", p.Version) |
| 720 | } |
| 721 | // And the re-promotion refreshed the lease — CanaryExpired.go's whole |
| 722 | // point: an actively-updated canary never expires. |
| 723 | if CanaryExpired(p, time.Now()) { |
| 724 | t.Error("re-promoting an expired canary must refresh its lease, not leave it expired") |
| 725 | } |
| 726 | } |
| 727 | |
| 728 | // TestPassFilterDoesNotNagWithoutTheOption pins the fast-loop half of the |
| 729 | // nag's scoping (2026-08-03 review): internal/host/agent.go's fastConvergeLoop |
| 730 | // calls PassFilter every 2-3 seconds for its fast set, and NEVER |
| 731 | // passes NagExpiredCanaries — so an expired canary that is sitting still |
| 732 | // must produce ZERO nag lines from that path, not one every few seconds. |
| 733 | // This is PassFilter called the way that loop calls it (without the nag |
| 734 | // option) against the same expired fixture TestPassAdoptsAnExpiredCanary |
| 735 | // uses. |
| 736 | func TestPassFilterDoesNotNagWithoutTheOption(t *testing.T) { |
| 737 | s := NewDirStore(t.TempDir()) |
| 738 | ctx := context.Background() |
| 739 | priv, pub := testKeys(t) |
| 740 | opt := t.TempDir() |
| 741 | reg := testRegistry(t) |
| 742 | |
| 743 | shipRelease(t, s, "bab", "myapp", "v1", adoptPlatform, []byte("build one")) |
| 744 | writeBackdatedPointer(t, s, priv, "bab", "myapp", "canary/x", "v1", |
| 745 | time.Now().Add(-(CanaryTTL + 30*24*time.Hour))) |
| 746 | addRoute(t, reg, "old-canary.example.com", "myapp", "canary/x", false) |
| 747 | |
| 748 | var buf bytes.Buffer |
| 749 | withCapturedStdout(t, &buf, func() { |
| 750 | if err := PassFilter(ctx, reg, s, pub, opt, func(string) error { return nil }, isCanaryTestRoute); err != nil { |
| 751 | t.Fatalf("PassFilter: %v", err) |
| 752 | } |
| 753 | }) |
| 754 | if strings.Contains(buf.String(), "expired") { |
| 755 | t.Errorf("PassFilter without NagExpiredCanaries printed a nag line:\n%s", buf.String()) |
| 756 | } |
| 757 | } |
| 758 | |
| 759 | // TestPassNagsOncePerExpiredCanaryRoute pins the root path's half: RunAdopt |
| 760 | // (carlos-adopt.service, hourly) passes NagExpiredCanaries(os.Stdout) to |
| 761 | // Pass, and that call must print ONE nag line PER EXPIRED CANARY ROUTE it |
| 762 | // sees — not one per hourly pass regardless of how many routes follow the |
| 763 | // expired channel. Three routes (two awake, one hibernating — the awake and |
| 764 | // hibernating lists are walked separately in Pass/PassFilter, so this is |
| 765 | // also the regression test for "hibernating routes get skipped" and |
| 766 | // "a route is visited twice") on ONE expired canary channel must produce |
| 767 | // exactly three lines, each naming its own host. |
| 768 | // |
| 769 | // (2026-08-03 review, MINOR 6: this test used to be named |
| 770 | // TestPassNagsExactlyOnceWithTheOption and only ever exercised one route, |
| 771 | // so "not once per host revisit" was asserted by name only, never proven.) |
| 772 | func TestPassNagsOncePerExpiredCanaryRoute(t *testing.T) { |
| 773 | s := NewDirStore(t.TempDir()) |
| 774 | ctx := context.Background() |
| 775 | priv, pub := testKeys(t) |
| 776 | opt := t.TempDir() |
| 777 | reg := testRegistry(t) |
| 778 | |
| 779 | shipRelease(t, s, "bab", "myapp", "v1", adoptPlatform, []byte("build one")) |
| 780 | writeBackdatedPointer(t, s, priv, "bab", "myapp", "canary/x", "v1", |
| 781 | time.Now().Add(-(CanaryTTL + 30*24*time.Hour))) |
| 782 | addRoute(t, reg, "awake-a.example.com", "myapp", "canary/x", false) |
| 783 | addRoute(t, reg, "awake-b.example.com", "myapp", "canary/x", false) |
| 784 | addRoute(t, reg, "sleepy.example.com", "myapp", "canary/x", true) |
| 785 | |
| 786 | var buf bytes.Buffer |
| 787 | if err := Pass(ctx, reg, s, pub, opt, func(string) error { return nil }, NagExpiredCanaries(&buf)); err != nil { |
| 788 | t.Fatalf("Pass: %v", err) |
| 789 | } |
| 790 | out := buf.String() |
| 791 | if got := strings.Count(out, "expired"); got != 3 { |
| 792 | t.Errorf("nag lines containing \"expired\" = %d, want exactly 3 (one per route, awake and hibernating alike):\n%s", got, out) |
| 793 | } |
| 794 | for _, host := range []string{"awake-a.example.com", "awake-b.example.com", "sleepy.example.com"} { |
| 795 | if !strings.Contains(out, host) { |
| 796 | t.Errorf("nag output does not name %s:\n%s", host, out) |
| 797 | } |
| 798 | } |
| 799 | } |
| 800 | |
| 801 | // TestNagReportsTimePastTTLNotTotalAge pins the 2026-08-03 review's |
| 802 | // IMPORTANT 2 fix: the nag must report how long PAST CanaryTTL a canary is, |
| 803 | // not its total age. A pointer promoted CanaryTTL+1h ago has been EXPIRED |
| 804 | // for about an hour, not 169 hours (7 days 1 hour, its total age) — the |
| 805 | // second number is real but answers a different question, and printing it |
| 806 | // under "has been expired" is the bug the review caught. It must also |
| 807 | // disagree with nothing the console shows for the same pointer |
| 808 | // (internal/console/ladder_test.go's TestLadderExpiredCanaryIsBadgedWithAge |
| 809 | // exercises the same release.CanaryAge/FormatCanaryAge pair). |
| 810 | func TestNagReportsTimePastTTLNotTotalAge(t *testing.T) { |
| 811 | s := NewDirStore(t.TempDir()) |
| 812 | ctx := context.Background() |
| 813 | priv, pub := testKeys(t) |
| 814 | opt := t.TempDir() |
| 815 | reg := testRegistry(t) |
| 816 | |
| 817 | shipRelease(t, s, "bab", "myapp", "v1", adoptPlatform, []byte("build one")) |
| 818 | writeBackdatedPointer(t, s, priv, "bab", "myapp", "canary/x", "v1", |
| 819 | time.Now().Add(-(CanaryTTL + time.Hour))) |
| 820 | addRoute(t, reg, "old-canary.example.com", "myapp", "canary/x", false) |
| 821 | |
| 822 | var buf bytes.Buffer |
| 823 | if err := Pass(ctx, reg, s, pub, opt, func(string) error { return nil }, NagExpiredCanaries(&buf)); err != nil { |
| 824 | t.Fatalf("Pass: %v", err) |
| 825 | } |
| 826 | out := buf.String() |
| 827 | if !strings.Contains(out, "1h") { |
| 828 | t.Errorf("nag does not say 1h (time past TTL) for a canary expired ~1h ago:\n%s", out) |
| 829 | } |
| 830 | if strings.Contains(out, "169h") { |
| 831 | t.Errorf("nag reports total age (169h) instead of time past TTL:\n%s", out) |
| 832 | } |
| 833 | } |
| 834 | |
| 835 | // TestPassDoesNotNagAFreshCanaryEvenWithTheOption: the option opts a CALL |
| 836 | // into nagging; it does not turn every canary into a nag target. A fresh |
| 837 | // canary must stay silent whether or not NagExpiredCanaries is set. |
| 838 | func TestPassDoesNotNagAFreshCanaryEvenWithTheOption(t *testing.T) { |
| 839 | s := NewDirStore(t.TempDir()) |
| 840 | ctx := context.Background() |
| 841 | priv, pub := testKeys(t) |
| 842 | opt := t.TempDir() |
| 843 | reg := testRegistry(t) |
| 844 | |
| 845 | shipRelease(t, s, "bab", "myapp", "v1", adoptPlatform, []byte("build one")) |
| 846 | if err := Promote(ctx, s, priv, "bab", "myapp", "canary/x", "v1"); err != nil { |
| 847 | t.Fatalf("Promote: %v", err) |
| 848 | } |
| 849 | addRoute(t, reg, "fresh-canary.example.com", "myapp", "canary/x", false) |
| 850 | |
| 851 | var buf bytes.Buffer |
| 852 | if err := Pass(ctx, reg, s, pub, opt, func(string) error { return nil }, NagExpiredCanaries(&buf)); err != nil { |
| 853 | t.Fatalf("Pass: %v", err) |
| 854 | } |
| 855 | if strings.Contains(buf.String(), "expired") { |
| 856 | t.Errorf("a freshly-promoted canary produced a nag line:\n%s", buf.String()) |
| 857 | } |
| 858 | } |
| 859 | |
| 860 | // isCanaryTestRoute mirrors host.isCanaryRoute (unexported, different |
| 861 | // package) for the one fast-loop-shaped test above that needs a canary-only |
| 862 | // filter without importing internal/host. |
| 863 | func isCanaryTestRoute(rt registry.Route) bool { return strings.HasPrefix(rt.Channel, "canary/") } |
| 864 | |
| 865 | // withCapturedStdout redirects os.Stdout to buf for the duration of fn — |
| 866 | // used ONLY to prove PassFilter's default (no NagExpiredCanaries) path |
| 867 | // prints nothing canary-related even via the package's ordinary fmt.Printf |
| 868 | // progress lines, which is the honest thing to assert against (adoptHost's |
| 869 | // other progress lines still go to os.Stdout; NagExpiredCanaries is what |
| 870 | // redirects the NAG line specifically to an injected writer instead). |
| 871 | func withCapturedStdout(t *testing.T, buf *bytes.Buffer, fn func()) { |
| 872 | t.Helper() |
| 873 | r, w, err := os.Pipe() |
| 874 | if err != nil { |
| 875 | t.Fatalf("os.Pipe: %v", err) |
| 876 | } |
| 877 | orig := os.Stdout |
| 878 | os.Stdout = w |
| 879 | defer func() { os.Stdout = orig }() |
| 880 | |
| 881 | done := make(chan struct{}) |
| 882 | go func() { |
| 883 | buf.ReadFrom(r) |
| 884 | close(done) |
| 885 | }() |
| 886 | |
| 887 | fn() |
| 888 | w.Close() |
| 889 | <-done |
| 890 | } |
| 891 | |
| 892 | // adoptRefusesManifest runs the whole fail-closed scenario for a stored |
| 893 | // manifest whose capability block this binary will not accept: a box already |
| 894 | // serving v1, a channel that has moved to v2, and v2's manifest bytes written |
| 895 | // DIRECTLY to the store (bypassing Ship, which is exactly what a newer |
| 896 | // platform binary would leave behind). It asserts the four things that make |
| 897 | // the refusal safe — the live symlink unmoved, nothing staged, no restart, |
| 898 | // the registry unchanged — and returns what the pass printed. |
| 899 | // |
| 900 | // Shared by the two tests below because the two refusals arrive by DIFFERENT |
| 901 | // routes and both have to land here: one from caps.Set's strict decoder |
| 902 | // (json.Unmarshal fails), one from ValidCapabilities (the block decodes |
| 903 | // perfectly and is only then refused). A gate that caught the first and not |
| 904 | // the second would look correct in a single test. |
| 905 | func adoptRefusesManifest(t *testing.T, capsJSON string) string { |
| 906 | t.Helper() |
| 907 | s := NewDirStore(t.TempDir()) |
| 908 | ctx := context.Background() |
| 909 | priv, pub := testKeys(t) |
| 910 | opt := t.TempDir() |
| 911 | reg := testRegistry(t) |
| 912 | |
| 913 | // The box is up and running on v1. |
| 914 | shipRelease(t, s, "bab", "myapp", "v1", adoptPlatform, []byte("good")) |
| 915 | if err := StageArtifact(opt, "bab", "v1", "myapp", []byte("good")); err != nil { |
| 916 | t.Fatal(err) |
| 917 | } |
| 918 | addRoute(t, reg, "h.example.com", "myapp", "canary/x", false) |
| 919 | if err := PointLive(opt, "h.example.com", "bab", "v1", "myapp"); err != nil { |
| 920 | t.Fatal(err) |
| 921 | } |
| 922 | |
| 923 | // v2 ships and the channel moves onto it while it is still a manifest this |
| 924 | // binary understands... |
| 925 | body := []byte("newer") |
| 926 | shipRelease(t, s, "bab", "myapp", "v2", adoptPlatform, body) |
| 927 | if err := Promote(ctx, s, priv, "bab", "myapp", "canary/x", "v2"); err != nil { |
| 928 | t.Fatalf("Promote v2: %v", err) |
| 929 | } |
| 930 | // ...and is then replaced by the bytes a NEWER binary would have written: |
| 931 | // same artifacts, same sha, plus a capability block this one refuses. |
| 932 | key := ManifestKey("bab", "myapp", "v2") |
| 933 | _, etag, err := s.Get(ctx, key) |
| 934 | if err != nil { |
| 935 | t.Fatalf("read v2 manifest: %v", err) |
| 936 | } |
| 937 | future := []byte(`{ |
| 938 | "schema": 1, |
| 939 | "app": "myapp", |
| 940 | "version": "v2", |
| 941 | "kind": "binary", |
| 942 | "artifacts": { "` + adoptPlatform + `": { "file": "` + adoptPlatform + `", "sha256": "` + etagOf(body) + `", "size": ` + strconv.Itoa(len(body)) + ` } }, |
| 943 | "capabilities": ` + capsJSON + ` |
| 944 | }`) |
| 945 | if _, err := s.Put(ctx, key, future, etag); err != nil { |
| 946 | t.Fatalf("replace v2 manifest: %v", err) |
| 947 | } |
| 948 | |
| 949 | var out bytes.Buffer |
| 950 | var restarted []string |
| 951 | withCapturedStdout(t, &out, func() { |
| 952 | if err := Pass(ctx, reg, s, pub, opt, func(h string) error { |
| 953 | restarted = append(restarted, h) |
| 954 | return nil |
| 955 | }); err != nil { |
| 956 | t.Fatalf("Pass: %v", err) |
| 957 | } |
| 958 | }) |
| 959 | |
| 960 | if v, _ := LiveVersion(opt, "h.example.com"); v != "v1" { |
| 961 | t.Fatalf("live = %q, want v1 — the box must keep serving what it has", v) |
| 962 | } |
| 963 | if _, err := os.Stat(VersionPath(opt, "bab", "v2", "myapp")); !os.IsNotExist(err) { |
| 964 | t.Fatalf("v2 was staged (stat err = %v); a refused version must never reach disk", err) |
| 965 | } |
| 966 | if len(restarted) != 0 { |
| 967 | t.Fatalf("restarted = %v, want none", restarted) |
| 968 | } |
| 969 | if rt, _ := reg.Get("h.example.com"); rt == nil || rt.Version == "v2" { |
| 970 | t.Fatalf("registry recorded v2 as adopted: %+v", rt) |
| 971 | } |
| 972 | return out.String() |
| 973 | } |
| 974 | |
| 975 | // TestAdoptRefusesTheVersionWhenAManifestDeclaresAnUnknownName is the |
| 976 | // fail-closed direction, and it is the test that matters most in this task. |
| 977 | // |
| 978 | // The box's answer to a capability name from a vocabulary it does not have |
| 979 | // must be to refuse the VERSION: nothing staged, the live symlink unmoved, |
| 980 | // one printed line naming the capability and the fix. The box keeps serving |
| 981 | // what it has, which for a daemon is the safe direction; the alternative — |
| 982 | // materializing a unit that is missing something the app says it needs, and |
| 983 | // looking healthy — is the config-that-lies failure mode this whole design |
| 984 | // exists to avoid. |
| 985 | // |
| 986 | // This refusal comes from caps.Set's strict DECODER: json.Unmarshal never |
| 987 | // completes, so the manifest never becomes a value at all. |
| 988 | func TestAdoptRefusesTheVersionWhenAManifestDeclaresAnUnknownName(t *testing.T) { |
| 989 | line := adoptRefusesManifest(t, `{ "v": 1, "daemon": true, "gpu": true }`) |
| 990 | if !strings.Contains(line, "gpu") { |
| 991 | t.Fatalf("printed %q, want a line naming the capability", line) |
| 992 | } |
| 993 | if !strings.Contains(line, "update the platform binary") { |
| 994 | t.Fatalf("printed %q, want a line naming the fix", line) |
| 995 | } |
| 996 | } |
| 997 | |
| 998 | // TestAdoptRefusesTheVersionWhenAManifestIsSemanticallyInvalid closes the |
| 999 | // other half of the gate, and it is the half a decoder-only test would miss: |
| 1000 | // this block decodes PERFECTLY — every name is known, "v" is 1, the shape is |
| 1001 | // right — and is refused only by ValidCapabilities, on a port that is not a |
| 1002 | // capability at all. It is reached because validateManifest runs on every |
| 1003 | // GetManifest, not only at ship; delete that one line and this test goes red |
| 1004 | // while the unknown-name test above stays green. |
| 1005 | func TestAdoptRefusesTheVersionWhenAManifestIsSemanticallyInvalid(t *testing.T) { |
| 1006 | line := adoptRefusesManifest(t, `{ "v": 1, "daemon": true, "ports": [ { "port": 8080 } ] }`) |
| 1007 | if !strings.Contains(line, "8080") { |
| 1008 | t.Fatalf("printed %q, want a line naming the offending port", line) |
| 1009 | } |
| 1010 | if !strings.Contains(line, "capabilities") { |
| 1011 | t.Fatalf("printed %q, want a line naming what was refused", line) |
| 1012 | } |
| 1013 | } |
| 1014 | |
| 1015 | // TestEnsureStagedWritesTheSiblingManifest pins the fresh-download half of |
| 1016 | // Task 3: after ensureStaged fetches and stages a version for the first |
| 1017 | // time, ReadStagedManifest returns a manifest whose App/Version match and |
| 1018 | // whose Capabilities round-trip byte for byte — this is what lets the |
| 1019 | // converge pass read a running version's capabilities off disk, with no |
| 1020 | // bucket round-trip on every tick. |
| 1021 | func TestEnsureStagedWritesTheSiblingManifest(t *testing.T) { |
| 1022 | s := NewDirStore(t.TempDir()) |
| 1023 | ctx := context.Background() |
| 1024 | opt := t.TempDir() |
| 1025 | |
| 1026 | m := Manifest{ |
| 1027 | App: "keymail", Version: "v1abc", Kind: "binary", |
| 1028 | Artifacts: map[string]Artifact{adoptPlatform: {}}, |
| 1029 | Capabilities: &caps.Set{V: 1, Command: "relay", Daemon: true, RuntimeDir: "keymail"}, |
| 1030 | } |
| 1031 | if err := Ship(ctx, s, "bab", m, map[string][]byte{adoptPlatform: []byte("relay binary")}); err != nil { |
| 1032 | t.Fatalf("Ship: %v", err) |
| 1033 | } |
| 1034 | |
| 1035 | inHand, err := ensureStaged(ctx, s, opt, "bab", "keymail", "v1abc", "keymail.example.com") |
| 1036 | if err != nil { |
| 1037 | t.Fatalf("ensureStaged: %v", err) |
| 1038 | } |
| 1039 | // The manifest ensureStaged HAS IN HAND is what adoptHost's converge |
| 1040 | // builds its plan from (Task 7's first pin) — not a re-read of the |
| 1041 | // sibling it just wrote. |
| 1042 | if inHand == nil || inHand.Capabilities == nil || inHand.Capabilities.Command != "relay" { |
| 1043 | t.Fatalf("ensureStaged returned %+v, want the fetched manifest with its capability block", inHand) |
| 1044 | } |
| 1045 | |
| 1046 | got, err := ReadStagedManifest(opt, "bab", "v1abc", "keymail") |
| 1047 | if err != nil { |
| 1048 | t.Fatalf("ReadStagedManifest: %v", err) |
| 1049 | } |
| 1050 | if got == nil || got.App != "keymail" || got.Version != "v1abc" { |
| 1051 | t.Fatalf("got = %+v, want App/Version to match", got) |
| 1052 | } |
| 1053 | if got.Capabilities == nil || got.Capabilities.Command != "relay" || !got.Capabilities.Daemon || got.Capabilities.RuntimeDir != "keymail" { |
| 1054 | t.Fatalf("Capabilities did not round-trip: %+v", got.Capabilities) |
| 1055 | } |
| 1056 | } |
| 1057 | |
| 1058 | // TestEnsureStagedBackfillsASiblingForAnAlreadyStagedBinary pins the named |
| 1059 | // finding: a box that staged a version's binary before this feature existed |
| 1060 | // (or before the sibling write otherwise happened) must have the sibling |
| 1061 | // back-filled on the next pass WITHOUT the artifact being re-downloaded — |
| 1062 | // otherwise the converge pass is permanently blind on exactly the versions |
| 1063 | // already running, since a staged binary never touches ensureStaged's |
| 1064 | // download branch again on its own. |
| 1065 | func TestEnsureStagedBackfillsASiblingForAnAlreadyStagedBinary(t *testing.T) { |
| 1066 | s := NewDirStore(t.TempDir()) |
| 1067 | ctx := context.Background() |
| 1068 | opt := t.TempDir() |
| 1069 | |
| 1070 | m := Manifest{ |
| 1071 | App: "keymail", Version: "v1abc", Kind: "binary", |
| 1072 | Artifacts: map[string]Artifact{adoptPlatform: {}}, |
| 1073 | Capabilities: &caps.Set{V: 1, Daemon: true, RuntimeDir: "keymail"}, |
| 1074 | } |
| 1075 | if err := Ship(ctx, s, "bab", m, map[string][]byte{adoptPlatform: []byte("relay binary")}); err != nil { |
| 1076 | t.Fatalf("Ship: %v", err) |
| 1077 | } |
| 1078 | |
| 1079 | // Stage the binary BY HAND, as a pre-Task-3 box would have — no sibling. |
| 1080 | if err := StageArtifact(opt, "bab", "v1abc", "keymail", []byte("relay binary")); err != nil { |
| 1081 | t.Fatalf("StageArtifact: %v", err) |
| 1082 | } |
| 1083 | binPath := VersionPath(opt, "bab", "v1abc", "keymail") |
| 1084 | beforeInfo, err := os.Stat(binPath) |
| 1085 | if err != nil { |
| 1086 | t.Fatalf("stat before backfill: %v", err) |
| 1087 | } |
| 1088 | beforeBytes, err := os.ReadFile(binPath) |
| 1089 | if err != nil { |
| 1090 | t.Fatalf("read before backfill: %v", err) |
| 1091 | } |
| 1092 | if m2, err := ReadStagedManifest(opt, "bab", "v1abc", "keymail"); err != nil || m2 != nil { |
| 1093 | t.Fatalf("sibling should be absent before backfill: %+v, %v", m2, err) |
| 1094 | } |
| 1095 | |
| 1096 | inHand, err := ensureStaged(ctx, s, opt, "bab", "keymail", "v1abc", "keymail.example.com") |
| 1097 | if err != nil { |
| 1098 | t.Fatalf("ensureStaged (backfill): %v", err) |
| 1099 | } |
| 1100 | if inHand == nil || inHand.Capabilities == nil || !inHand.Capabilities.Daemon { |
| 1101 | t.Fatalf("back-fill returned %+v, want the fetched manifest in hand", inHand) |
| 1102 | } |
| 1103 | |
| 1104 | got, err := ReadStagedManifest(opt, "bab", "v1abc", "keymail") |
| 1105 | if err != nil { |
| 1106 | t.Fatalf("ReadStagedManifest after backfill: %v", err) |
| 1107 | } |
| 1108 | if got == nil || got.Capabilities == nil || !got.Capabilities.Daemon || got.Capabilities.RuntimeDir != "keymail" { |
| 1109 | t.Fatalf("backfilled sibling missing capabilities: %+v", got) |
| 1110 | } |
| 1111 | |
| 1112 | afterInfo, err := os.Stat(binPath) |
| 1113 | if err != nil { |
| 1114 | t.Fatalf("stat after backfill: %v", err) |
| 1115 | } |
| 1116 | afterBytes, err := os.ReadFile(binPath) |
| 1117 | if err != nil { |
| 1118 | t.Fatalf("read after backfill: %v", err) |
| 1119 | } |
| 1120 | if !beforeInfo.ModTime().Equal(afterInfo.ModTime()) { |
| 1121 | t.Fatalf("backfill re-touched the binary: mtime %v -> %v", beforeInfo.ModTime(), afterInfo.ModTime()) |
| 1122 | } |
| 1123 | if string(beforeBytes) != string(afterBytes) { |
| 1124 | t.Fatalf("backfill re-downloaded the binary bytes") |
| 1125 | } |
| 1126 | } |
| 1127 | |
| 1128 | // TestEnsureStagedSiblingWriteFailureIsNotFatalToStagingTheBinary is |
| 1129 | // killing test 2 from the 2026-08-03 review: a mutant that makes the |
| 1130 | // marshal/StageManifest failure at the end of ensureStaged's fresh-download |
| 1131 | // path fatal must be caught. The binary must stage successfully even when |
| 1132 | // its sibling cannot be written — a missing sibling is the safe leave-alone |
| 1133 | // state, and backfillManifest gets another chance at it every later pass. |
| 1134 | // |
| 1135 | // StageManifest's own write is blocked by pre-creating its final rename |
| 1136 | // target AS A DIRECTORY — rename then fails to replace it, without ever |
| 1137 | // touching StageArtifact's own (different) binary path. |
| 1138 | func TestEnsureStagedSiblingWriteFailureIsNotFatalToStagingTheBinary(t *testing.T) { |
| 1139 | s := NewDirStore(t.TempDir()) |
| 1140 | ctx := context.Background() |
| 1141 | opt := t.TempDir() |
| 1142 | |
| 1143 | m := Manifest{ |
| 1144 | App: "keymail", Version: "v1abc", Kind: "binary", |
| 1145 | Artifacts: map[string]Artifact{adoptPlatform: {}}, |
| 1146 | Capabilities: &caps.Set{V: 1, Daemon: true, RuntimeDir: "keymail"}, |
| 1147 | } |
| 1148 | if err := Ship(ctx, s, "bab", m, map[string][]byte{adoptPlatform: []byte("relay binary")}); err != nil { |
| 1149 | t.Fatalf("Ship: %v", err) |
| 1150 | } |
| 1151 | |
| 1152 | tmpBlock := StagedManifestPath(opt, "bab", "v1abc", "keymail") |
| 1153 | if err := os.MkdirAll(tmpBlock, 0o755); err != nil { |
| 1154 | t.Fatalf("pre-create tmp block: %v", err) |
| 1155 | } |
| 1156 | |
| 1157 | inHand, err := ensureStaged(ctx, s, opt, "bab", "keymail", "v1abc", "h.example.com") |
| 1158 | if err != nil { |
| 1159 | t.Fatalf("ensureStaged returned an error for a failed SIBLING write: %v — the binary staged fine and that must not be fatal", err) |
| 1160 | } |
| 1161 | // Task 7's first pin, in its sharpest form: the sibling write FAILED, so |
| 1162 | // a converge that re-read the sibling would see (nil, nil) and leave the |
| 1163 | // previous version's drop-ins in place under a new binary. The manifest |
| 1164 | // this function returns is the fetched one and is unaffected. |
| 1165 | if inHand == nil || inHand.Capabilities == nil || !inHand.Capabilities.Daemon { |
| 1166 | t.Fatalf("ensureStaged returned %+v after a failed sibling write, want the fetched manifest still in hand", inHand) |
| 1167 | } |
| 1168 | |
| 1169 | body, err := os.ReadFile(VersionPath(opt, "bab", "v1abc", "keymail")) |
| 1170 | if err != nil || string(body) != "relay binary" { |
| 1171 | t.Fatalf("binary did not stage despite the sibling write failing: %q, %v", body, err) |
| 1172 | } |
| 1173 | if got, err := ReadStagedManifest(opt, "bab", "v1abc", "keymail"); err == nil || got != nil { |
| 1174 | t.Fatalf("sibling = %+v, %v; want blocked (the write genuinely failed)", got, err) |
| 1175 | } |
| 1176 | } |
| 1177 | |
| 1178 | // TestEnsureStagedBackfillFetchFailureIsNotFatal is killing test 3 / I2's |
| 1179 | // regression test (2026-08-03 review): ensureStaged's back-fill fetch (a |
| 1180 | // GetManifest for an ALREADY-STAGED binary) must not be fatal to adoption. |
| 1181 | // The store here has no release shipped at all, so GetManifest is |
| 1182 | // guaranteed to fail — simulating a bucket hiccup for a version that is |
| 1183 | // already running and needs no adoption. A mutant that makes this fetch |
| 1184 | // failure fatal again would hard-fail (and, on a canary route's 2-3-second |
| 1185 | // fast loop, journal-flood) adoption of a binary that doesn't need it. |
| 1186 | func TestEnsureStagedBackfillFetchFailureIsNotFatal(t *testing.T) { |
| 1187 | s := NewDirStore(t.TempDir()) // empty — nothing ever shipped |
| 1188 | ctx := context.Background() |
| 1189 | opt := t.TempDir() |
| 1190 | |
| 1191 | if err := StageArtifact(opt, "bab", "v1abc", "keymail", []byte("relay binary")); err != nil { |
| 1192 | t.Fatalf("StageArtifact: %v", err) |
| 1193 | } |
| 1194 | |
| 1195 | inHand, err := ensureStaged(ctx, s, opt, "bab", "keymail", "v1abc", "h.example.com") |
| 1196 | if err != nil { |
| 1197 | t.Fatalf("ensureStaged returned an error for an already-staged binary's failed back-fill fetch: %v — adoption of an already-running version must proceed", err) |
| 1198 | } |
| 1199 | // This is the ONE path with no manifest to give. It stays (nil, nil) so |
| 1200 | // ordinary adoption proceeds; a capability-armed pass turns that nil into |
| 1201 | // a refusal of the version rather than a silent leave-alone — see |
| 1202 | // TestArmedPassRefusesAVersionWhoseManifestItCouldNotObtain. |
| 1203 | if inHand != nil { |
| 1204 | t.Fatalf("ensureStaged returned %+v, want nil (no manifest could be obtained)", inHand) |
| 1205 | } |
| 1206 | |
| 1207 | if got, err := ReadStagedManifest(opt, "bab", "v1abc", "keymail"); err != nil || got != nil { |
| 1208 | t.Fatalf("sibling = %+v, %v; want absent (the fetch failed, nothing to write)", got, err) |
| 1209 | } |
| 1210 | } |
| 1211 | |
| 1212 | // TestPassBackfillsTheLiveVersionsSiblingOnAnUnchangedPass pins C1, the |
| 1213 | // critical finding from the 2026-08-03 review: ensureStaged's own back-fill |
| 1214 | // only runs from the call inside adoptHost that sits BEHIND the cur == ver |
| 1215 | // short-circuit, so a version that is already live — with no repoint due |
| 1216 | // this pass — never reaches it on its own. Since the migration path is |
| 1217 | // entirely "a box's CURRENT version was staged before this feature |
| 1218 | // existed," without a separate per-pass repair call the sibling for |
| 1219 | // exactly the version actually running would stay missing forever. This |
| 1220 | // test exercises the real call site (Pass, not ensureStaged directly): the |
| 1221 | // route is already on v1abc, nothing repoints, and the sibling must still |
| 1222 | // appear. |
| 1223 | func TestPassBackfillsTheLiveVersionsSiblingOnAnUnchangedPass(t *testing.T) { |
| 1224 | s := NewDirStore(t.TempDir()) |
| 1225 | ctx := context.Background() |
| 1226 | priv, pub := testKeys(t) |
| 1227 | opt := t.TempDir() |
| 1228 | reg := testRegistry(t) |
| 1229 | |
| 1230 | m := Manifest{ |
| 1231 | App: "keymail", Version: "v1abc", Kind: "binary", |
| 1232 | Artifacts: map[string]Artifact{adoptPlatform: {}}, |
| 1233 | Capabilities: &caps.Set{V: 1, Daemon: true, RuntimeDir: "keymail"}, |
| 1234 | } |
| 1235 | if err := Ship(ctx, s, "bab", m, map[string][]byte{adoptPlatform: []byte("relay binary")}); err != nil { |
| 1236 | t.Fatalf("Ship: %v", err) |
| 1237 | } |
| 1238 | if err := Promote(ctx, s, priv, "bab", "keymail", "canary/x", "v1abc"); err != nil { |
| 1239 | t.Fatalf("Promote: %v", err) |
| 1240 | } |
| 1241 | |
| 1242 | // The binary is staged and pointed live BY HAND, exactly as a |
| 1243 | // pre-Task-3 box would be — no sibling, and no version change due this |
| 1244 | // pass (the route is already on v1abc). |
| 1245 | if err := StageArtifact(opt, "bab", "v1abc", "keymail", []byte("relay binary")); err != nil { |
| 1246 | t.Fatalf("StageArtifact: %v", err) |
| 1247 | } |
| 1248 | addRoute(t, reg, "h.example.com", "keymail", "canary/x", false) |
| 1249 | if err := PointLive(opt, "h.example.com", "bab", "v1abc", "keymail"); err != nil { |
| 1250 | t.Fatalf("PointLive: %v", err) |
| 1251 | } |
| 1252 | |
| 1253 | if m2, err := ReadStagedManifest(opt, "bab", "v1abc", "keymail"); err != nil || m2 != nil { |
| 1254 | t.Fatalf("sibling should be absent before the pass: %+v, %v", m2, err) |
| 1255 | } |
| 1256 | |
| 1257 | if err := Pass(ctx, reg, s, pub, opt, func(string) error { return nil }); err != nil { |
| 1258 | t.Fatalf("Pass: %v", err) |
| 1259 | } |
| 1260 | |
| 1261 | got, err := ReadStagedManifest(opt, "bab", "v1abc", "keymail") |
| 1262 | if err != nil { |
| 1263 | t.Fatalf("ReadStagedManifest after Pass: %v", err) |
| 1264 | } |
| 1265 | if got == nil || got.Capabilities == nil || !got.Capabilities.Daemon { |
| 1266 | t.Fatalf("Pass did not back-fill the live version's sibling: %+v", got) |
| 1267 | } |
| 1268 | if v, _ := LiveVersion(opt, "h.example.com"); v != "v1abc" { |
| 1269 | t.Fatalf("live version moved unexpectedly: %q", v) |
| 1270 | } |
| 1271 | } |
| 1272 | |
| 1273 | // ---- Task 7: the capability converge inside adoptHost ----------------------- |
| 1274 | |
| 1275 | // capsRecorder is a tiny in-memory systemd for the capability tests in this |
| 1276 | // package, plus the one thing internal/caps's own fake cannot observe: WHERE |
| 1277 | // THE LIVE SYMLINK POINTED at the moment each verb ran. That is how the |
| 1278 | // ordering rule — converge before PointLive — is asserted as a fact about |
| 1279 | // this pass rather than as a reading of the code. |
| 1280 | type capsRecorder struct { |
| 1281 | opt, host string |
| 1282 | calls []string |
| 1283 | live []string // live[i] is LiveVersion when calls[i] ran |
| 1284 | liveAt map[string]string // verb+unit -> LiveVersion at its FIRST call |
| 1285 | active map[string]bool |
| 1286 | enabled map[string]bool |
| 1287 | fail map[string]error // "verb unit" -> error |
| 1288 | onCall func(call string) // optional: fires after a call is recorded |
| 1289 | } |
| 1290 | |
| 1291 | func newCapsRecorder(opt, host string) *capsRecorder { |
| 1292 | return &capsRecorder{ |
| 1293 | opt: opt, host: host, |
| 1294 | liveAt: map[string]string{}, |
| 1295 | active: map[string]bool{}, |
| 1296 | enabled: map[string]bool{}, |
| 1297 | fail: map[string]error{}, |
| 1298 | } |
| 1299 | } |
| 1300 | |
| 1301 | func (r *capsRecorder) run(args ...string) error { |
| 1302 | call := strings.Join(args, " ") |
| 1303 | v, _ := LiveVersion(r.opt, r.host) |
| 1304 | r.calls = append(r.calls, call) |
| 1305 | r.live = append(r.live, v) |
| 1306 | if _, seen := r.liveAt[call]; !seen { |
| 1307 | r.liveAt[call] = v |
| 1308 | } |
| 1309 | if r.onCall != nil { |
| 1310 | r.onCall(call) |
| 1311 | } |
| 1312 | if err := r.fail[call]; err != nil { |
| 1313 | return err |
| 1314 | } |
| 1315 | if len(args) >= 2 { |
| 1316 | switch args[0] { |
| 1317 | case "start", "restart": |
| 1318 | r.active[args[1]] = true |
| 1319 | case "stop": |
| 1320 | r.active[args[1]] = false |
| 1321 | case "enable": |
| 1322 | r.enabled[args[1]] = true |
| 1323 | } |
| 1324 | } |
| 1325 | return nil |
| 1326 | } |
| 1327 | |
| 1328 | func (r *capsRecorder) query(args ...string) bool { |
| 1329 | if len(args) < 2 { |
| 1330 | return false |
| 1331 | } |
| 1332 | switch args[0] { |
| 1333 | case "is-active": |
| 1334 | return r.active[args[1]] |
| 1335 | case "is-enabled": |
| 1336 | return r.enabled[args[1]] |
| 1337 | } |
| 1338 | return false |
| 1339 | } |
| 1340 | |
| 1341 | func (r *capsRecorder) called(call string) bool { |
| 1342 | for _, c := range r.calls { |
| 1343 | if c == call { |
| 1344 | return true |
| 1345 | } |
| 1346 | } |
| 1347 | return false |
| 1348 | } |
| 1349 | |
| 1350 | // lastLiveAt is where the live symlink pointed at the LAST occurrence of a |
| 1351 | // call. liveAt records only the first, which is blind to everything a |
| 1352 | // rollback does after the same verb has already run once — the ordering |
| 1353 | // inside the rollback (symlink first, drop-ins second) is exactly what needs |
| 1354 | // pinning, so it needs the last one. |
| 1355 | func (r *capsRecorder) lastLiveAt(call string) (string, bool) { |
| 1356 | for i := len(r.calls) - 1; i >= 0; i-- { |
| 1357 | if r.calls[i] == call { |
| 1358 | return r.live[i], true |
| 1359 | } |
| 1360 | } |
| 1361 | return "", false |
| 1362 | } |
| 1363 | |
| 1364 | // serviceCycles is every verb that STARTS the service unit — the count that |
| 1365 | // says whether a version change cycled the app once or twice. |
| 1366 | func (r *capsRecorder) serviceCycles(host string) (n int, lastLive string) { |
| 1367 | for i, c := range r.calls { |
| 1368 | if c == "start "+caps.ServiceUnit(host) || c == "restart "+caps.ServiceUnit(host) { |
| 1369 | n++ |
| 1370 | lastLive = r.live[i] |
| 1371 | } |
| 1372 | } |
| 1373 | return n, lastLive |
| 1374 | } |
| 1375 | |
| 1376 | // reset clears the call log between passes. |
| 1377 | func (r *capsRecorder) reset() { |
| 1378 | r.calls, r.live = nil, nil |
| 1379 | r.liveAt = map[string]string{} |
| 1380 | } |
| 1381 | |
| 1382 | // capsTestOps wires every caps seam at a temp dir, so a Pass in this package |
| 1383 | // touches no systemd, no /etc, no /run and no /var/lib. Logf is left nil on |
| 1384 | // purpose: converge's own lines then go to os.Stdout with every other adopt |
| 1385 | // line, which is what withCapturedStdout asserts against. |
| 1386 | func capsTestOps(t *testing.T, r *capsRecorder) (caps.Ops, string) { |
| 1387 | t.Helper() |
| 1388 | etc := t.TempDir() |
| 1389 | return caps.Ops{ |
| 1390 | EtcDir: etc, |
| 1391 | StateRoot: t.TempDir(), |
| 1392 | RunDir: t.TempDir(), |
| 1393 | Systemctl: r.run, |
| 1394 | Query: r.query, |
| 1395 | EnsureUser: func(string) error { return nil }, |
| 1396 | }, etc |
| 1397 | } |
| 1398 | |
| 1399 | // addCapRoute adds a route with a backing and a granted capability column — |
| 1400 | // addRoute plus the two columns Task 7 reads. |
| 1401 | func addCapRoute(t *testing.T, reg *registry.Registry, host, app, channel, backing, grants string, hibernate, platform bool) { |
| 1402 | t.Helper() |
| 1403 | if err := reg.Upsert(registry.Route{ |
| 1404 | Host: host, Kind: "instance", App: app, Channel: channel, |
| 1405 | Hibernate: hibernate, Backing: backing, Caps: grants, Platform: platform, |
| 1406 | }); err != nil { |
| 1407 | t.Fatalf("upsert %s: %v", host, err) |
| 1408 | } |
| 1409 | } |
| 1410 | |
| 1411 | // shipCapRelease is shipRelease with a capability block. |
| 1412 | func shipCapRelease(t *testing.T, s Store, acct, app, version string, set *caps.Set, body []byte) { |
| 1413 | t.Helper() |
| 1414 | m := Manifest{ |
| 1415 | App: app, Version: version, Kind: "binary", |
| 1416 | Artifacts: map[string]Artifact{adoptPlatform: {}}, |
| 1417 | Capabilities: set, |
| 1418 | } |
| 1419 | if err := Ship(context.Background(), s, acct, m, map[string][]byte{adoptPlatform: body}); err != nil { |
| 1420 | t.Fatalf("Ship %s/%s: %v", app, version, err) |
| 1421 | } |
| 1422 | } |
| 1423 | |
| 1424 | func relaySet() *caps.Set { |
| 1425 | return &caps.Set{ |
| 1426 | V: 1, Command: "relay", Daemon: true, RuntimeDir: "keymail", |
| 1427 | Ports: []caps.Port{{Port: 25, Name: "smtp"}}, |
| 1428 | } |
| 1429 | } |
| 1430 | |
| 1431 | // TestAdoptConvergesCapabilitiesBeforePointLive is the ordering rule of spec |
| 1432 | // §4d, asserted where it can actually fail: the drop-ins are written and |
| 1433 | // daemon-reloaded while the live symlink still names the OLD version, and the |
| 1434 | // daemon's restart happens after it names the new one. Converge after |
| 1435 | // PointLive would leave a window where the new binary is live under the old |
| 1436 | // unit shape; converge without a reload would leave bytes systemd never read. |
| 1437 | func TestAdoptConvergesCapabilitiesBeforePointLive(t *testing.T) { |
| 1438 | s := NewDirStore(t.TempDir()) |
| 1439 | ctx := context.Background() |
| 1440 | priv, pub := testKeys(t) |
| 1441 | opt := t.TempDir() |
| 1442 | reg := testRegistry(t) |
| 1443 | host := "keymail.dev" |
| 1444 | |
| 1445 | // The box is up on v1, which declares nothing. |
| 1446 | shipRelease(t, s, "bab", "keymail", "v1", adoptPlatform, []byte("old relay")) |
| 1447 | if err := StageArtifact(opt, "bab", "v1", "keymail", []byte("old relay")); err != nil { |
| 1448 | t.Fatal(err) |
| 1449 | } |
| 1450 | addCapRoute(t, reg, host, "keymail", "canary/x", registry.BackingUnit, "port:25", false, false) |
| 1451 | if err := PointLive(opt, host, "bab", "v1", "keymail"); err != nil { |
| 1452 | t.Fatal(err) |
| 1453 | } |
| 1454 | |
| 1455 | set := relaySet() |
| 1456 | shipCapRelease(t, s, "bab", "keymail", "v2", set, []byte("new relay")) |
| 1457 | if err := Promote(ctx, s, priv, "bab", "keymail", "canary/x", "v2"); err != nil { |
| 1458 | t.Fatalf("Promote v2: %v", err) |
| 1459 | } |
| 1460 | |
| 1461 | rec := newCapsRecorder(opt, host) |
| 1462 | o, etc := capsTestOps(t, rec) |
| 1463 | var restarted []string |
| 1464 | restart := func(h string) error { restarted = append(restarted, h); return nil } |
| 1465 | if err := Pass(ctx, reg, s, pub, opt, restart, WithCapabilities(o)); err != nil { |
| 1466 | t.Fatalf("Pass: %v", err) |
| 1467 | } |
| 1468 | |
| 1469 | if v, _ := LiveVersion(opt, host); v != "v2" { |
| 1470 | t.Fatalf("live = %q, want v2", v) |
| 1471 | } |
| 1472 | if v, seen := rec.liveAt["daemon-reload"]; !seen || v != "v1" { |
| 1473 | t.Fatalf("daemon-reload observed live=%q (seen=%v), want v1 — the drop-ins must be loaded BEFORE the symlink moves", v, seen) |
| 1474 | } |
| 1475 | |
| 1476 | // The bytes are the renderer's, byte for byte — this test pins the |
| 1477 | // wiring, not a second copy of the drop-in format. |
| 1478 | g := caps.Grants{Ports: map[int]bool{25: true}} |
| 1479 | svcPath, sockPath := caps.DropInPaths(etc, host) |
| 1480 | wantSvc, err := caps.RenderService("keymail", set, g) |
| 1481 | if err != nil { |
| 1482 | t.Fatalf("RenderService: %v", err) |
| 1483 | } |
| 1484 | gotSvc, err := os.ReadFile(svcPath) |
| 1485 | if err != nil || !bytes.Equal(gotSvc, wantSvc) { |
| 1486 | t.Fatalf("service drop-in = %q (%v), want %q", gotSvc, err, wantSvc) |
| 1487 | } |
| 1488 | wantSock, err := caps.RenderSocket(set, g) |
| 1489 | if err != nil { |
| 1490 | t.Fatalf("RenderSocket: %v", err) |
| 1491 | } |
| 1492 | gotSock, err := os.ReadFile(sockPath) |
| 1493 | if err != nil || !bytes.Equal(gotSock, wantSock) { |
| 1494 | t.Fatalf("socket drop-in = %q (%v), want %q", gotSock, err, wantSock) |
| 1495 | } |
| 1496 | |
| 1497 | // A converged DAEMON's restart is caps.Restart, not the caller's |
| 1498 | // try-restart callback — that is what makes its failure real enough to |
| 1499 | // roll back (restartRoute swallows every error it sees). |
| 1500 | if len(restarted) != 0 { |
| 1501 | t.Fatalf("restart callback called for %v; a converged daemon's restart is caps.Restart's", restarted) |
| 1502 | } |
| 1503 | svcRestart := "restart " + caps.ServiceUnit(host) |
| 1504 | if v, seen := rec.liveAt[svcRestart]; !seen || v != "v2" { |
| 1505 | t.Fatalf("%q observed live=%q (seen=%v), want v2 — the daemon restarts onto the NEW binary", svcRestart, v, seen) |
| 1506 | } |
| 1507 | // And exactly ONCE. Converge runs before the symlink moves, so any start |
| 1508 | // it performs itself boots the OLD binary under the NEW unit shape — a |
| 1509 | // combination nobody shipped — and then the caller cycles the app a |
| 1510 | // second time. Plan.DeferRestart is what collapses that to one cycle. |
| 1511 | if n, last := rec.serviceCycles(host); n != 1 || last != "v2" { |
| 1512 | t.Fatalf("service started/restarted %d time(s), last at live=%q; want exactly 1, at v2 (calls: %v)", n, last, rec.calls) |
| 1513 | } |
| 1514 | } |
| 1515 | |
| 1516 | // TestRestartFailureRestoresBothTheSymlinkAndTheDropIns is the rollback that |
| 1517 | // keeps a bad build off a mail server: when the daemon will not come back on |
| 1518 | // the new version, the box must end the pass exactly where it started — old |
| 1519 | // symlink, old unit text, the version poisoned for the rest of the pass. |
| 1520 | // Restoring one without the other is the divergence this pairing exists to |
| 1521 | // prevent (bytes that say :465 under a binary that never asked for it). |
| 1522 | func TestRestartFailureRestoresBothTheSymlinkAndTheDropIns(t *testing.T) { |
| 1523 | s := NewDirStore(t.TempDir()) |
| 1524 | ctx := context.Background() |
| 1525 | priv, pub := testKeys(t) |
| 1526 | opt := t.TempDir() |
| 1527 | reg := testRegistry(t) |
| 1528 | host := "a.example.com" |
| 1529 | |
| 1530 | v1 := &caps.Set{V: 1, Command: "relay", Daemon: true, Ports: []caps.Port{{Port: 25}}} |
| 1531 | v2 := &caps.Set{V: 1, Command: "relay", Daemon: true, Ports: []caps.Port{{Port: 25}, {Port: 465}}} |
| 1532 | shipCapRelease(t, s, "bab", "keymail", "v1", v1, []byte("good relay")) |
| 1533 | shipCapRelease(t, s, "bab", "keymail", "v2", v2, []byte("bad relay")) |
| 1534 | if err := Promote(ctx, s, priv, "bab", "keymail", "canary/x", "v1"); err != nil { |
| 1535 | t.Fatalf("Promote v1: %v", err) |
| 1536 | } |
| 1537 | addCapRoute(t, reg, host, "keymail", "canary/x", registry.BackingUnit, "port:25,port:465", false, false) |
| 1538 | // b is exec-backed (materializes nothing) and exists only to prove the |
| 1539 | // poison: it must never be moved onto the version a's restart rejected. |
| 1540 | addCapRoute(t, reg, "b.example.com", "keymail", "canary/x", "", "", false, false) |
| 1541 | |
| 1542 | rec := newCapsRecorder(opt, host) |
| 1543 | o, etc := capsTestOps(t, rec) |
| 1544 | if err := Pass(ctx, reg, s, pub, opt, func(string) error { return nil }, WithCapabilities(o)); err != nil { |
| 1545 | t.Fatalf("Pass v1: %v", err) |
| 1546 | } |
| 1547 | svcPath, sockPath := caps.DropInPaths(etc, host) |
| 1548 | wantSvc, err := os.ReadFile(svcPath) |
| 1549 | if err != nil { |
| 1550 | t.Fatalf("read converged service drop-in: %v", err) |
| 1551 | } |
| 1552 | wantSock, err := os.ReadFile(sockPath) |
| 1553 | if err != nil { |
| 1554 | t.Fatalf("read converged socket drop-in: %v", err) |
| 1555 | } |
| 1556 | if !strings.Contains(string(wantSock), "ListenStream=25") || strings.Contains(string(wantSock), "465") { |
| 1557 | t.Fatalf("v1 socket drop-in = %q, want :25 only", wantSock) |
| 1558 | } |
| 1559 | |
| 1560 | if err := Promote(ctx, s, priv, "bab", "keymail", "canary/x", "v2"); err != nil { |
| 1561 | t.Fatalf("Promote v2: %v", err) |
| 1562 | } |
| 1563 | rec.fail["restart "+caps.ServiceUnit(host)] = errors.New("relay failed to start") |
| 1564 | |
| 1565 | var out bytes.Buffer |
| 1566 | withCapturedStdout(t, &out, func() { |
| 1567 | if err := Pass(ctx, reg, s, pub, opt, func(string) error { return nil }, WithCapabilities(o)); err != nil { |
| 1568 | t.Fatalf("Pass v2: %v", err) |
| 1569 | } |
| 1570 | }) |
| 1571 | |
| 1572 | if v, _ := LiveVersion(opt, host); v != "v1" { |
| 1573 | t.Fatalf("live = %q, want v1 — the symlink must be rolled back", v) |
| 1574 | } |
| 1575 | gotSvc, _ := os.ReadFile(svcPath) |
| 1576 | gotSock, _ := os.ReadFile(sockPath) |
| 1577 | if !bytes.Equal(gotSvc, wantSvc) { |
| 1578 | t.Fatalf("service drop-in = %q, want v1's %q", gotSvc, wantSvc) |
| 1579 | } |
| 1580 | if !bytes.Equal(gotSock, wantSock) { |
| 1581 | t.Fatalf("socket drop-in = %q, want v1's %q — a rolled-back version must not leave :465 bound", gotSock, wantSock) |
| 1582 | } |
| 1583 | // Bytes are not a binding: restoring the socket half has to re-run the |
| 1584 | // socket sequence, or systemd keeps the listener the rolled-back version |
| 1585 | // asked for. |
| 1586 | if !rec.called("restart " + caps.SocketUnit(host)) { |
| 1587 | t.Fatalf("calls = %v, want the socket unit re-bound during the restore", rec.calls) |
| 1588 | } |
| 1589 | // The ordering inside the rollback is load-bearing and easy to get |
| 1590 | // backwards: the symlink goes back FIRST, so the service the restore |
| 1591 | // starts comes up on the OLD binary. Restoring first would start v2's |
| 1592 | // binary under v1's unit text — the failure this whole branch exists to |
| 1593 | // undo. lastLiveAt, not liveAt: the same verb already ran during the |
| 1594 | // forward converge. |
| 1595 | if v, seen := rec.lastLiveAt("start " + caps.ServiceUnit(host)); !seen || v != "v1" { |
| 1596 | t.Fatalf("the restore's service start observed live=%q (seen=%v), want v1 — the symlink must be rolled back BEFORE the drop-ins are", v, seen) |
| 1597 | } |
| 1598 | if v, _ := LiveVersion(opt, "b.example.com"); v != "v1" { |
| 1599 | t.Fatalf("b live = %q, want v1 — the failed version must be poisoned for the rest of the pass", v) |
| 1600 | } |
| 1601 | line := out.String() |
| 1602 | if !strings.Contains(line, "rolled the symlink back") { |
| 1603 | t.Fatalf("printed %q, want a line saying the symlink was rolled back", line) |
| 1604 | } |
| 1605 | if !strings.Contains(line, "capability drop-ins") { |
| 1606 | t.Fatalf("printed %q, want a line saying the drop-ins were restored too", line) |
| 1607 | } |
| 1608 | } |
| 1609 | |
| 1610 | // TestExecBackedRouteWithCapabilitiesMaterializesNothing: Backing == "" is |
| 1611 | // the whole existing fleet. A release declaring the full relay set must |
| 1612 | // change no unit text on such a route (its process is an exec child, not a |
| 1613 | // carlos-app@ unit) while adoption itself proceeds exactly as before. This |
| 1614 | // is also why no existing test in this package can reach the converge path. |
| 1615 | func TestExecBackedRouteWithCapabilitiesMaterializesNothing(t *testing.T) { |
| 1616 | s := NewDirStore(t.TempDir()) |
| 1617 | ctx := context.Background() |
| 1618 | priv, pub := testKeys(t) |
| 1619 | opt := t.TempDir() |
| 1620 | reg := testRegistry(t) |
| 1621 | host := "exec.example.com" |
| 1622 | |
| 1623 | shipCapRelease(t, s, "bab", "keymail", "v1", relaySet(), []byte("relay")) |
| 1624 | if err := Promote(ctx, s, priv, "bab", "keymail", "canary/x", "v1"); err != nil { |
| 1625 | t.Fatalf("Promote: %v", err) |
| 1626 | } |
| 1627 | addCapRoute(t, reg, host, "keymail", "canary/x", "", "port:25", false, false) |
| 1628 | |
| 1629 | rec := newCapsRecorder(opt, host) |
| 1630 | o, etc := capsTestOps(t, rec) |
| 1631 | var restarted []string |
| 1632 | if err := Pass(ctx, reg, s, pub, opt, func(h string) error { |
| 1633 | restarted = append(restarted, h) |
| 1634 | return nil |
| 1635 | }, WithCapabilities(o)); err != nil { |
| 1636 | t.Fatalf("Pass: %v", err) |
| 1637 | } |
| 1638 | |
| 1639 | assertEmptyDir(t, etc, "an exec-backed route must materialize no unit text") |
| 1640 | if len(rec.calls) != 0 { |
| 1641 | t.Fatalf("systemctl calls = %v, want none", rec.calls) |
| 1642 | } |
| 1643 | if v, _ := LiveVersion(opt, host); v != "v1" { |
| 1644 | t.Fatalf("live = %q, want v1 — adoption itself must proceed", v) |
| 1645 | } |
| 1646 | if len(restarted) != 1 || restarted[0] != host { |
| 1647 | t.Fatalf("restarted = %v, want the ordinary callback for %s", restarted, host) |
| 1648 | } |
| 1649 | } |
| 1650 | |
| 1651 | // TestPlatformAliasRouteMaterializesNothing: a platform alias is a second |
| 1652 | // name for one process. A second carlos-app@ unit claiming :25 would break |
| 1653 | // the real one, so an alias materializes nothing — and, being outside |
| 1654 | // Adoptable(), is not adopted at all. |
| 1655 | func TestPlatformAliasRouteMaterializesNothing(t *testing.T) { |
| 1656 | s := NewDirStore(t.TempDir()) |
| 1657 | ctx := context.Background() |
| 1658 | priv, pub := testKeys(t) |
| 1659 | opt := t.TempDir() |
| 1660 | reg := testRegistry(t) |
| 1661 | host := "keymail.bab.apps.example.com" |
| 1662 | |
| 1663 | shipCapRelease(t, s, "bab", "keymail", "v1", relaySet(), []byte("relay")) |
| 1664 | if err := Promote(ctx, s, priv, "bab", "keymail", "canary/x", "v1"); err != nil { |
| 1665 | t.Fatalf("Promote: %v", err) |
| 1666 | } |
| 1667 | addCapRoute(t, reg, host, "keymail", "canary/x", registry.BackingUnit, "port:25", false, true) |
| 1668 | |
| 1669 | rec := newCapsRecorder(opt, host) |
| 1670 | o, etc := capsTestOps(t, rec) |
| 1671 | if err := Pass(ctx, reg, s, pub, opt, func(string) error { return nil }, WithCapabilities(o)); err != nil { |
| 1672 | t.Fatalf("Pass: %v", err) |
| 1673 | } |
| 1674 | |
| 1675 | assertEmptyDir(t, etc, "a platform alias must materialize no unit text") |
| 1676 | if len(rec.calls) != 0 { |
| 1677 | t.Fatalf("systemctl calls = %v, want none", rec.calls) |
| 1678 | } |
| 1679 | if v, _ := LiveVersion(opt, host); v != "" { |
| 1680 | t.Fatalf("live = %q, want unset — a platform alias is not adopted at all", v) |
| 1681 | } |
| 1682 | } |
| 1683 | |
| 1684 | // TestPassWithoutWithCapabilitiesIsUnchanged is the byte-identical-to-before |
| 1685 | // assertion, and the reason this task could land on its own: a unit-backed |
| 1686 | // route with a full capability manifest, adopted by a Pass that was never |
| 1687 | // given the option, writes no unit text anywhere and adopts exactly as it |
| 1688 | // always did. |
| 1689 | func TestPassWithoutWithCapabilitiesIsUnchanged(t *testing.T) { |
| 1690 | s := NewDirStore(t.TempDir()) |
| 1691 | ctx := context.Background() |
| 1692 | priv, pub := testKeys(t) |
| 1693 | opt := t.TempDir() |
| 1694 | reg := testRegistry(t) |
| 1695 | host := "unit.example.com" |
| 1696 | etc := t.TempDir() // stands in for /etc/systemd/system; nothing may reach it |
| 1697 | |
| 1698 | shipCapRelease(t, s, "bab", "keymail", "v1", relaySet(), []byte("relay")) |
| 1699 | if err := Promote(ctx, s, priv, "bab", "keymail", "canary/x", "v1"); err != nil { |
| 1700 | t.Fatalf("Promote: %v", err) |
| 1701 | } |
| 1702 | addCapRoute(t, reg, host, "keymail", "canary/x", registry.BackingUnit, "port:25", false, false) |
| 1703 | |
| 1704 | var restarted []string |
| 1705 | if err := Pass(ctx, reg, s, pub, opt, func(h string) error { |
| 1706 | restarted = append(restarted, h) |
| 1707 | return nil |
| 1708 | }); err != nil { |
| 1709 | t.Fatalf("Pass: %v", err) |
| 1710 | } |
| 1711 | |
| 1712 | assertEmptyDir(t, etc, "a Pass without WithCapabilities must write no unit text") |
| 1713 | if v, _ := LiveVersion(opt, host); v != "v1" { |
| 1714 | t.Fatalf("live = %q, want v1", v) |
| 1715 | } |
| 1716 | if len(restarted) != 1 || restarted[0] != host { |
| 1717 | t.Fatalf("restarted = %v, want the ordinary callback for %s", restarted, host) |
| 1718 | } |
| 1719 | } |
| 1720 | |
| 1721 | // TestUnconvergedRepointOfACapabilityRouteLogsTheWindow pins the 2026-08-03 |
| 1722 | // ruling on the canary fast loop. That loop runs inside carlos-edge.service |
| 1723 | // (ProtectSystem=strict) and CANNOT write /etc/systemd/system, so it repoints |
| 1724 | // a capability-bearing canary and leaves the unit shape to the hourly root |
| 1725 | // pass — a window of up to an hour where a new binary runs under the previous |
| 1726 | // version's drop-ins. The window is accepted; being silent about it is not. |
| 1727 | // |
| 1728 | // The signal is deliberately the cheapest honest one: "the new version's |
| 1729 | // manifest declares capabilities at all", not "the new set differs from what |
| 1730 | // is materialized" — the difference test needs the grants and a render of |
| 1731 | // both sides, and this line exists to mark the window, not to diff it. |
| 1732 | func TestUnconvergedRepointOfACapabilityRouteLogsTheWindow(t *testing.T) { |
| 1733 | s := NewDirStore(t.TempDir()) |
| 1734 | ctx := context.Background() |
| 1735 | priv, pub := testKeys(t) |
| 1736 | opt := t.TempDir() |
| 1737 | reg := testRegistry(t) |
| 1738 | |
| 1739 | shipCapRelease(t, s, "bab", "keymail", "v1", relaySet(), []byte("relay")) |
| 1740 | shipRelease(t, s, "bab", "plain", "v1", adoptPlatform, []byte("plain")) |
| 1741 | for _, app := range []string{"keymail", "plain"} { |
| 1742 | if err := Promote(ctx, s, priv, "bab", app, "canary/x", "v1"); err != nil { |
| 1743 | t.Fatalf("Promote %s: %v", app, err) |
| 1744 | } |
| 1745 | } |
| 1746 | addCapRoute(t, reg, "caps.example.com", "keymail", "canary/x", registry.BackingUnit, "port:25", false, false) |
| 1747 | // A unit-backed route whose release declares NOTHING must stay quiet — |
| 1748 | // otherwise the line fires on every ordinary tenant and nobody reads it. |
| 1749 | addCapRoute(t, reg, "plain.example.com", "plain", "canary/x", registry.BackingUnit, "", false, false) |
| 1750 | |
| 1751 | var out bytes.Buffer |
| 1752 | withCapturedStdout(t, &out, func() { |
| 1753 | if err := PassFilter(ctx, reg, s, pub, opt, func(string) error { return nil }, isCanaryTestRoute); err != nil { |
| 1754 | t.Fatalf("PassFilter: %v", err) |
| 1755 | } |
| 1756 | }) |
| 1757 | |
| 1758 | line := out.String() |
| 1759 | if !strings.Contains(line, "caps.example.com: v1 declares capabilities") { |
| 1760 | t.Fatalf("printed %q, want one line naming the unconverged capability repoint", line) |
| 1761 | } |
| 1762 | if !strings.Contains(line, "carlos-adopt.service") { |
| 1763 | t.Fatalf("printed %q, want the line to name what closes the window", line) |
| 1764 | } |
| 1765 | if strings.Contains(line, "plain.example.com: v1 declares") { |
| 1766 | t.Fatalf("printed %q, want silence for a route whose release declares nothing", line) |
| 1767 | } |
| 1768 | } |
| 1769 | |
| 1770 | // TestHibernatingUnitRouteConvergesWithoutStarting: a hibernating route |
| 1771 | // reaches adoptHost with restart == nil. Its drop-ins are unit TEXT, not a |
| 1772 | // wake, so they converge — but nothing may start the service, or an hourly |
| 1773 | // timer would defeat hibernation. |
| 1774 | func TestHibernatingUnitRouteConvergesWithoutStarting(t *testing.T) { |
| 1775 | s := NewDirStore(t.TempDir()) |
| 1776 | ctx := context.Background() |
| 1777 | priv, pub := testKeys(t) |
| 1778 | opt := t.TempDir() |
| 1779 | reg := testRegistry(t) |
| 1780 | host := "sleepy.example.com" |
| 1781 | |
| 1782 | // No daemon: daemon + hibernate is a contradiction converge refuses. |
| 1783 | set := &caps.Set{V: 1, Command: "web", Ports: []caps.Port{{Port: 25}}} |
| 1784 | shipCapRelease(t, s, "bab", "keymail", "v1", set, []byte("relay")) |
| 1785 | if err := Promote(ctx, s, priv, "bab", "keymail", "canary/x", "v1"); err != nil { |
| 1786 | t.Fatalf("Promote: %v", err) |
| 1787 | } |
| 1788 | addCapRoute(t, reg, host, "keymail", "canary/x", registry.BackingUnit, "port:25", true, false) |
| 1789 | |
| 1790 | rec := newCapsRecorder(opt, host) |
| 1791 | o, etc := capsTestOps(t, rec) |
| 1792 | if err := Pass(ctx, reg, s, pub, opt, func(string) error { |
| 1793 | t.Fatal("a hibernating host must never reach the restart callback") |
| 1794 | return nil |
| 1795 | }, WithCapabilities(o)); err != nil { |
| 1796 | t.Fatalf("Pass: %v", err) |
| 1797 | } |
| 1798 | |
| 1799 | svcPath, _ := caps.DropInPaths(etc, host) |
| 1800 | if _, err := os.Stat(svcPath); err != nil { |
| 1801 | t.Fatalf("stat %s: %v — a hibernating route's drop-ins still converge", svcPath, err) |
| 1802 | } |
| 1803 | if v, _ := LiveVersion(opt, host); v != "v1" { |
| 1804 | t.Fatalf("live = %q, want v1", v) |
| 1805 | } |
| 1806 | for _, c := range rec.calls { |
| 1807 | if strings.HasSuffix(c, caps.ServiceUnit(host)) { |
| 1808 | t.Fatalf("calls = %v, want no verb against the SERVICE unit — a drop-in is not a wake", rec.calls) |
| 1809 | } |
| 1810 | } |
| 1811 | } |
| 1812 | |
| 1813 | // TestConvergeRefusalRefusesTheVersion: a refusal from converge falls exactly |
| 1814 | // where an ensureStaged failure falls — nothing pointed live, the box keeps |
| 1815 | // serving what it has, one line naming the operator's fix. Pointing live |
| 1816 | // anyway would run a new binary against a unit shape the platform has just |
| 1817 | // declined to write. |
| 1818 | // |
| 1819 | // The refusal used here is spec §4e's daemon + hibernate contradiction, |
| 1820 | // because it is refused with nothing written at all. |
| 1821 | func TestConvergeRefusalRefusesTheVersion(t *testing.T) { |
| 1822 | s := NewDirStore(t.TempDir()) |
| 1823 | ctx := context.Background() |
| 1824 | priv, pub := testKeys(t) |
| 1825 | opt := t.TempDir() |
| 1826 | reg := testRegistry(t) |
| 1827 | host := "sleepy.example.com" |
| 1828 | |
| 1829 | shipRelease(t, s, "bab", "keymail", "v1", adoptPlatform, []byte("old")) |
| 1830 | if err := StageArtifact(opt, "bab", "v1", "keymail", []byte("old")); err != nil { |
| 1831 | t.Fatal(err) |
| 1832 | } |
| 1833 | addCapRoute(t, reg, host, "keymail", "canary/x", registry.BackingUnit, "port:25", true, false) |
| 1834 | if err := PointLive(opt, host, "bab", "v1", "keymail"); err != nil { |
| 1835 | t.Fatal(err) |
| 1836 | } |
| 1837 | shipCapRelease(t, s, "bab", "keymail", "v2", relaySet(), []byte("new")) |
| 1838 | if err := Promote(ctx, s, priv, "bab", "keymail", "canary/x", "v2"); err != nil { |
| 1839 | t.Fatalf("Promote v2: %v", err) |
| 1840 | } |
| 1841 | |
| 1842 | rec := newCapsRecorder(opt, host) |
| 1843 | o, etc := capsTestOps(t, rec) |
| 1844 | var out bytes.Buffer |
| 1845 | withCapturedStdout(t, &out, func() { |
| 1846 | if err := Pass(ctx, reg, s, pub, opt, func(string) error { return nil }, WithCapabilities(o)); err != nil { |
| 1847 | t.Fatalf("Pass: %v", err) |
| 1848 | } |
| 1849 | }) |
| 1850 | |
| 1851 | if v, _ := LiveVersion(opt, host); v != "v1" { |
| 1852 | t.Fatalf("live = %q, want v1 — a refused converge must not point live", v) |
| 1853 | } |
| 1854 | assertEmptyDir(t, etc, "a refusal writes nothing") |
| 1855 | if len(rec.calls) != 0 { |
| 1856 | t.Fatalf("systemctl calls = %v, want none", rec.calls) |
| 1857 | } |
| 1858 | line := out.String() |
| 1859 | if !strings.Contains(line, "not pointed live") { |
| 1860 | t.Fatalf("printed %q, want a line saying the version was not pointed live", line) |
| 1861 | } |
| 1862 | if !strings.Contains(line, "-hibernate=false") { |
| 1863 | t.Fatalf("printed %q, want converge's own line naming the operator's fix", line) |
| 1864 | } |
| 1865 | } |
| 1866 | |
| 1867 | // TestArmedPassRefusesAVersionWhoseManifestItCouldNotObtain is the second of |
| 1868 | // Task 7's two pins, in the one place it can bite: the binary is staged (so |
| 1869 | // ensureStaged succeeds) but its manifest is obtainable neither from the |
| 1870 | // sibling nor from the bucket. caps.Plan's nil Set means "leave the drop-ins |
| 1871 | // alone" — correct for a pass surveying the box, and exactly wrong here, |
| 1872 | // where it would point a new binary at the previous version's unit shape and |
| 1873 | // say nothing. So an armed pass refuses the version. |
| 1874 | // |
| 1875 | // The second half is the compatibility half: the SAME situation on a pass |
| 1876 | // with no WithCapabilities adopts normally, which is what keeps every |
| 1877 | // existing caller byte-identical. |
| 1878 | func TestArmedPassRefusesAVersionWhoseManifestItCouldNotObtain(t *testing.T) { |
| 1879 | dir := t.TempDir() |
| 1880 | s := NewDirStore(dir) |
| 1881 | ctx := context.Background() |
| 1882 | priv, pub := testKeys(t) |
| 1883 | opt := t.TempDir() |
| 1884 | reg := testRegistry(t) |
| 1885 | host := "keymail.dev" |
| 1886 | |
| 1887 | shipRelease(t, s, "bab", "keymail", "v1", adoptPlatform, []byte("old")) |
| 1888 | if err := StageArtifact(opt, "bab", "v1", "keymail", []byte("old")); err != nil { |
| 1889 | t.Fatal(err) |
| 1890 | } |
| 1891 | if err := StageManifest(opt, "bab", "v1", "keymail", []byte(`{"schema":1,"app":"keymail","version":"v1","kind":"binary"}`)); err != nil { |
| 1892 | t.Fatal(err) |
| 1893 | } |
| 1894 | addCapRoute(t, reg, host, "keymail", "canary/x", registry.BackingUnit, "port:25", false, false) |
| 1895 | if err := PointLive(opt, host, "bab", "v1", "keymail"); err != nil { |
| 1896 | t.Fatal(err) |
| 1897 | } |
| 1898 | |
| 1899 | // v2's binary is on disk with no sibling, and its manifest object is then |
| 1900 | // removed from the store: the back-fill fetch cannot answer either. |
| 1901 | shipCapRelease(t, s, "bab", "keymail", "v2", relaySet(), []byte("new")) |
| 1902 | if err := Promote(ctx, s, priv, "bab", "keymail", "canary/x", "v2"); err != nil { |
| 1903 | t.Fatalf("Promote v2: %v", err) |
| 1904 | } |
| 1905 | if err := StageArtifact(opt, "bab", "v2", "keymail", []byte("new")); err != nil { |
| 1906 | t.Fatal(err) |
| 1907 | } |
| 1908 | if err := os.Remove(filepath.Join(dir, filepath.FromSlash(ManifestKey("bab", "keymail", "v2")))); err != nil { |
| 1909 | t.Fatalf("remove v2 manifest: %v", err) |
| 1910 | } |
| 1911 | |
| 1912 | rec := newCapsRecorder(opt, host) |
| 1913 | o, etc := capsTestOps(t, rec) |
| 1914 | var out bytes.Buffer |
| 1915 | withCapturedStdout(t, &out, func() { |
| 1916 | if err := Pass(ctx, reg, s, pub, opt, func(string) error { return nil }, WithCapabilities(o)); err != nil { |
| 1917 | t.Fatalf("armed Pass: %v", err) |
| 1918 | } |
| 1919 | }) |
| 1920 | if v, _ := LiveVersion(opt, host); v != "v1" { |
| 1921 | t.Fatalf("live = %q, want v1 — an unknown unit shape must refuse the version", v) |
| 1922 | } |
| 1923 | assertEmptyDir(t, etc, "a refusal writes nothing") |
| 1924 | if line := out.String(); !strings.Contains(line, "not pointed live") { |
| 1925 | t.Fatalf("printed %q, want a line saying the version was not pointed live", line) |
| 1926 | } |
| 1927 | |
| 1928 | // Unarmed, the identical situation adopts exactly as it always did. |
| 1929 | if err := Pass(ctx, reg, s, pub, opt, func(string) error { return nil }); err != nil { |
| 1930 | t.Fatalf("unarmed Pass: %v", err) |
| 1931 | } |
| 1932 | if v, _ := LiveVersion(opt, host); v != "v2" { |
| 1933 | t.Fatalf("live = %q after an unarmed pass, want v2 — capabilities off means behaviour unchanged", v) |
| 1934 | } |
| 1935 | } |
| 1936 | |
| 1937 | // assertEmptyDir fails unless dir has no entries at all. |
| 1938 | func assertEmptyDir(t *testing.T, dir, why string) { |
| 1939 | t.Helper() |
| 1940 | ents, err := os.ReadDir(dir) |
| 1941 | if err != nil { |
| 1942 | t.Fatalf("read %s: %v", dir, err) |
| 1943 | } |
| 1944 | if len(ents) != 0 { |
| 1945 | var names []string |
| 1946 | for _, e := range ents { |
| 1947 | names = append(names, e.Name()) |
| 1948 | } |
| 1949 | t.Fatalf("%s contains %v — %s", dir, names, why) |
| 1950 | } |
| 1951 | } |
| 1952 | |
| 1953 | // TestAdoptRemovesDropInsWhenTheNewVersionDeclaresNothing is Task 7's second |
| 1954 | // pin, and the mutant it kills is a one-word one: Plan{Set: m.Capabilities}. |
| 1955 | // Manifest.Capabilities is a nil pointer for BOTH a legacy manifest and a |
| 1956 | // fetch that never happened, and caps.Plan reads a nil Set as "leave the |
| 1957 | // drop-ins exactly as they are". Passing it through verbatim would mean a |
| 1958 | // capability DROPPED from a manifest is never removed — a relay that stopped |
| 1959 | // asking for :25 would keep binding it, which is the exact opposite of spec |
| 1960 | // §4d's removal semantics. A manifest that RESOLVED always carries a Set, |
| 1961 | // empty when it declares nothing. |
| 1962 | func TestAdoptRemovesDropInsWhenTheNewVersionDeclaresNothing(t *testing.T) { |
| 1963 | s := NewDirStore(t.TempDir()) |
| 1964 | ctx := context.Background() |
| 1965 | priv, pub := testKeys(t) |
| 1966 | opt := t.TempDir() |
| 1967 | reg := testRegistry(t) |
| 1968 | host := "keymail.dev" |
| 1969 | |
| 1970 | shipCapRelease(t, s, "bab", "keymail", "v1", relaySet(), []byte("relay")) |
| 1971 | if err := Promote(ctx, s, priv, "bab", "keymail", "canary/x", "v1"); err != nil { |
| 1972 | t.Fatalf("Promote v1: %v", err) |
| 1973 | } |
| 1974 | addCapRoute(t, reg, host, "keymail", "canary/x", registry.BackingUnit, "port:25", false, false) |
| 1975 | |
| 1976 | rec := newCapsRecorder(opt, host) |
| 1977 | o, etc := capsTestOps(t, rec) |
| 1978 | if err := Pass(ctx, reg, s, pub, opt, func(string) error { return nil }, WithCapabilities(o)); err != nil { |
| 1979 | t.Fatalf("Pass v1: %v", err) |
| 1980 | } |
| 1981 | svcPath, sockPath := caps.DropInPaths(etc, host) |
| 1982 | for _, p := range []string{svcPath, sockPath} { |
| 1983 | if _, err := os.Stat(p); err != nil { |
| 1984 | t.Fatalf("stat %s after v1: %v", p, err) |
| 1985 | } |
| 1986 | } |
| 1987 | |
| 1988 | // v2 declares nothing at all — a plain manifest, exactly like every |
| 1989 | // release in every production bucket today. |
| 1990 | shipRelease(t, s, "bab", "keymail", "v2", adoptPlatform, []byte("plain relay")) |
| 1991 | if err := Promote(ctx, s, priv, "bab", "keymail", "canary/x", "v2"); err != nil { |
| 1992 | t.Fatalf("Promote v2: %v", err) |
| 1993 | } |
| 1994 | if err := Pass(ctx, reg, s, pub, opt, func(string) error { return nil }, WithCapabilities(o)); err != nil { |
| 1995 | t.Fatalf("Pass v2: %v", err) |
| 1996 | } |
| 1997 | |
| 1998 | if v, _ := LiveVersion(opt, host); v != "v2" { |
| 1999 | t.Fatalf("live = %q, want v2", v) |
| 2000 | } |
| 2001 | for _, p := range []string{svcPath, sockPath} { |
| 2002 | if _, err := os.Stat(p); !os.IsNotExist(err) { |
| 2003 | b, _ := os.ReadFile(p) |
| 2004 | t.Fatalf("%s still exists after a version that declares nothing (%q) — a dropped capability must be removed", p, b) |
| 2005 | } |
| 2006 | } |
| 2007 | // Removing the socket half has to reach systemd, or the listener the |
| 2008 | // removed file asked for is still bound. |
| 2009 | if !rec.called("restart " + caps.SocketUnit(host)) { |
| 2010 | t.Fatalf("calls = %v, want the socket unit re-bound when its drop-in was removed", rec.calls) |
| 2011 | } |
| 2012 | } |
| 2013 | |
| 2014 | // TestDaemonVersionChangeStartsTheServiceExactlyOnce covers the other half of |
| 2015 | // the deferred restart: a version change whose SOCKET shape is unchanged, so |
| 2016 | // nothing in converge's socket ordering runs and step 9's service restart is |
| 2017 | // the only thing that would have cycled the unit. Deferred, the app is |
| 2018 | // started exactly once, after the symlink moves — one cycle, on the new |
| 2019 | // binary, instead of two with the first one running the old. |
| 2020 | func TestDaemonVersionChangeStartsTheServiceExactlyOnce(t *testing.T) { |
| 2021 | s := NewDirStore(t.TempDir()) |
| 2022 | ctx := context.Background() |
| 2023 | priv, pub := testKeys(t) |
| 2024 | opt := t.TempDir() |
| 2025 | reg := testRegistry(t) |
| 2026 | host := "keymail.dev" |
| 2027 | |
| 2028 | v1 := &caps.Set{V: 1, Command: "relay", Daemon: true, Ports: []caps.Port{{Port: 25}}} |
| 2029 | v2 := &caps.Set{V: 1, Command: "relay-two", Daemon: true, Ports: []caps.Port{{Port: 25}}} |
| 2030 | shipCapRelease(t, s, "bab", "keymail", "v1", v1, []byte("relay one")) |
| 2031 | shipCapRelease(t, s, "bab", "keymail", "v2", v2, []byte("relay two")) |
| 2032 | if err := Promote(ctx, s, priv, "bab", "keymail", "canary/x", "v1"); err != nil { |
| 2033 | t.Fatalf("Promote v1: %v", err) |
| 2034 | } |
| 2035 | addCapRoute(t, reg, host, "keymail", "canary/x", registry.BackingUnit, "port:25", false, false) |
| 2036 | |
| 2037 | rec := newCapsRecorder(opt, host) |
| 2038 | o, _ := capsTestOps(t, rec) |
| 2039 | restart := func(string) error { return nil } |
| 2040 | if err := Pass(ctx, reg, s, pub, opt, restart, WithCapabilities(o)); err != nil { |
| 2041 | t.Fatalf("Pass v1: %v", err) |
| 2042 | } |
| 2043 | if n, _ := rec.serviceCycles(host); n != 1 { |
| 2044 | t.Fatalf("first adoption started the service %d time(s), want 1 (calls: %v)", n, rec.calls) |
| 2045 | } |
| 2046 | rec.reset() |
| 2047 | |
| 2048 | if err := Promote(ctx, s, priv, "bab", "keymail", "canary/x", "v2"); err != nil { |
| 2049 | t.Fatalf("Promote v2: %v", err) |
| 2050 | } |
| 2051 | if err := Pass(ctx, reg, s, pub, opt, restart, WithCapabilities(o)); err != nil { |
| 2052 | t.Fatalf("Pass v2: %v", err) |
| 2053 | } |
| 2054 | |
| 2055 | if v, _ := LiveVersion(opt, host); v != "v2" { |
| 2056 | t.Fatalf("live = %q, want v2", v) |
| 2057 | } |
| 2058 | if n, last := rec.serviceCycles(host); n != 1 || last != "v2" { |
| 2059 | t.Fatalf("service started/restarted %d time(s), last at live=%q; want exactly 1, at v2 (calls: %v)", n, last, rec.calls) |
| 2060 | } |
| 2061 | } |
| 2062 | |
| 2063 | // TestConvergeFailureAfterItsReloadRestoresTheDropIns is the failure the |
| 2064 | // converge contract hands to this caller by name: a systemd verb that fails |
| 2065 | // AFTER the daemon-reload leaves the new version's unit text on disk, loaded, |
| 2066 | // and deliberately un-rolled-back (Converge reports Changed with the error |
| 2067 | // and says the version-change caller owns it). Nothing points live, so those |
| 2068 | // bytes describe a version this box is not running — and they are permanent |
| 2069 | // if nobody undoes them: the next pass finds no diff, snapshots them as its |
| 2070 | // own Prev, and every later "restore" restores the wrong shape. |
| 2071 | // |
| 2072 | // The failing verb here is `enable` on the socket unit, which is post-reload |
| 2073 | // and NOT one of the diverging ones Converge rolls back itself. |
| 2074 | func TestConvergeFailureAfterItsReloadRestoresTheDropIns(t *testing.T) { |
| 2075 | s := NewDirStore(t.TempDir()) |
| 2076 | ctx := context.Background() |
| 2077 | priv, pub := testKeys(t) |
| 2078 | opt := t.TempDir() |
| 2079 | reg := testRegistry(t) |
| 2080 | host := "keymail.dev" |
| 2081 | |
| 2082 | // The box is on v1, which declares nothing, so there are no drop-ins yet. |
| 2083 | shipRelease(t, s, "bab", "keymail", "v1", adoptPlatform, []byte("old")) |
| 2084 | if err := StageArtifact(opt, "bab", "v1", "keymail", []byte("old")); err != nil { |
| 2085 | t.Fatal(err) |
| 2086 | } |
| 2087 | addCapRoute(t, reg, host, "keymail", "canary/x", registry.BackingUnit, "port:25", false, false) |
| 2088 | if err := PointLive(opt, host, "bab", "v1", "keymail"); err != nil { |
| 2089 | t.Fatal(err) |
| 2090 | } |
| 2091 | shipCapRelease(t, s, "bab", "keymail", "v2", relaySet(), []byte("new")) |
| 2092 | if err := Promote(ctx, s, priv, "bab", "keymail", "canary/x", "v2"); err != nil { |
| 2093 | t.Fatalf("Promote v2: %v", err) |
| 2094 | } |
| 2095 | |
| 2096 | rec := newCapsRecorder(opt, host) |
| 2097 | o, etc := capsTestOps(t, rec) |
| 2098 | rec.fail["enable "+caps.SocketUnit(host)] = errors.New("unit masked") |
| 2099 | |
| 2100 | if err := Pass(ctx, reg, s, pub, opt, func(string) error { return nil }, WithCapabilities(o)); err != nil { |
| 2101 | t.Fatalf("Pass: %v", err) |
| 2102 | } |
| 2103 | |
| 2104 | if v, _ := LiveVersion(opt, host); v != "v1" { |
| 2105 | t.Fatalf("live = %q, want v1", v) |
| 2106 | } |
| 2107 | svcPath, sockPath := caps.DropInPaths(etc, host) |
| 2108 | for _, p := range []string{svcPath, sockPath} { |
| 2109 | if b, err := os.ReadFile(p); err == nil { |
| 2110 | t.Fatalf("%s survives a converge that failed after its reload (%q) — v2's unit shape must not be left on disk under v1's binary", p, b) |
| 2111 | } |
| 2112 | } |
| 2113 | } |
| 2114 | |
| 2115 | // TestConvergeRefusalLeavesAnEarlierVersionsDropInsAlone is the other side of |
| 2116 | // that coin, and the reason the restore above is conditional. An ORDINARY |
| 2117 | // refusal writes nothing — and the refusals that fire earliest fire before |
| 2118 | // the drop-ins are even read, so the snapshot is EMPTY. caps.Restore reads an |
| 2119 | // empty snapshot as "this route had no generated drop-ins" and honours it by |
| 2120 | // REMOVING the ones it finds, so restoring unconditionally on a refusal would |
| 2121 | // delete the shape the running version is serving under. |
| 2122 | // |
| 2123 | // The refusal here is spec §4e's daemon + hibernate contradiction, which |
| 2124 | // Converge checks before it reads anything. |
| 2125 | func TestConvergeRefusalLeavesAnEarlierVersionsDropInsAlone(t *testing.T) { |
| 2126 | s := NewDirStore(t.TempDir()) |
| 2127 | ctx := context.Background() |
| 2128 | priv, pub := testKeys(t) |
| 2129 | opt := t.TempDir() |
| 2130 | reg := testRegistry(t) |
| 2131 | host := "sleepy.example.com" |
| 2132 | |
| 2133 | // v1 is a hibernating tenant with a granted port and no daemon: a |
| 2134 | // perfectly legal shape, and it converges. |
| 2135 | v1 := &caps.Set{V: 1, Command: "web", Ports: []caps.Port{{Port: 25}}} |
| 2136 | shipCapRelease(t, s, "bab", "keymail", "v1", v1, []byte("web one")) |
| 2137 | if err := Promote(ctx, s, priv, "bab", "keymail", "canary/x", "v1"); err != nil { |
| 2138 | t.Fatalf("Promote v1: %v", err) |
| 2139 | } |
| 2140 | addCapRoute(t, reg, host, "keymail", "canary/x", registry.BackingUnit, "port:25", true, false) |
| 2141 | |
| 2142 | rec := newCapsRecorder(opt, host) |
| 2143 | o, etc := capsTestOps(t, rec) |
| 2144 | if err := Pass(ctx, reg, s, pub, opt, func(string) error { return nil }, WithCapabilities(o)); err != nil { |
| 2145 | t.Fatalf("Pass v1: %v", err) |
| 2146 | } |
| 2147 | svcPath, sockPath := caps.DropInPaths(etc, host) |
| 2148 | wantSvc, err := os.ReadFile(svcPath) |
| 2149 | if err != nil { |
| 2150 | t.Fatalf("read v1 service drop-in: %v", err) |
| 2151 | } |
| 2152 | wantSock, err := os.ReadFile(sockPath) |
| 2153 | if err != nil { |
| 2154 | t.Fatalf("read v1 socket drop-in: %v", err) |
| 2155 | } |
| 2156 | |
| 2157 | // v2 adds `daemon` to a hibernating route — refused before anything is |
| 2158 | // read or written. |
| 2159 | shipCapRelease(t, s, "bab", "keymail", "v2", relaySet(), []byte("web two")) |
| 2160 | if err := Promote(ctx, s, priv, "bab", "keymail", "canary/x", "v2"); err != nil { |
| 2161 | t.Fatalf("Promote v2: %v", err) |
| 2162 | } |
| 2163 | if err := Pass(ctx, reg, s, pub, opt, func(string) error { return nil }, WithCapabilities(o)); err != nil { |
| 2164 | t.Fatalf("Pass v2: %v", err) |
| 2165 | } |
| 2166 | |
| 2167 | if v, _ := LiveVersion(opt, host); v != "v1" { |
| 2168 | t.Fatalf("live = %q, want v1 — the refusal must not point live", v) |
| 2169 | } |
| 2170 | gotSvc, _ := os.ReadFile(svcPath) |
| 2171 | gotSock, _ := os.ReadFile(sockPath) |
| 2172 | if !bytes.Equal(gotSvc, wantSvc) || !bytes.Equal(gotSock, wantSock) { |
| 2173 | t.Fatalf("drop-ins after a refusal = %q / %q, want v1's %q / %q left exactly as they were", |
| 2174 | gotSvc, gotSock, wantSvc, wantSock) |
| 2175 | } |
| 2176 | } |
| 2177 | |
| 2178 | // TestConvergeRefusalDoesNotPoisonTheVersionForOtherHosts is the 2026-08-03 |
| 2179 | // ruling on the poison. A refusal bounces nothing — no symlink moved, no unit |
| 2180 | // cycled — so there is no bad build to keep off the next host, which is the |
| 2181 | // only thing the poison exists for. And the refusal is often HOST-shaped: the |
| 2182 | // one here is a port a third route's drop-in already holds on this box. |
| 2183 | // Poisoning on that would let one misconfigured route freeze every other host |
| 2184 | // out of a version that is fine for them, every pass, until an operator |
| 2185 | // noticed. |
| 2186 | // |
| 2187 | // The ordering is the point: a.example.com is refused FIRST (routes are swept |
| 2188 | // in host order), so a poison set there is what b.example.com would trip over. |
| 2189 | func TestConvergeRefusalDoesNotPoisonTheVersionForOtherHosts(t *testing.T) { |
| 2190 | s := NewDirStore(t.TempDir()) |
| 2191 | ctx := context.Background() |
| 2192 | priv, pub := testKeys(t) |
| 2193 | opt := t.TempDir() |
| 2194 | reg := testRegistry(t) |
| 2195 | |
| 2196 | shipCapRelease(t, s, "bab", "keymail", "v1", relaySet(), []byte("relay")) |
| 2197 | if err := Promote(ctx, s, priv, "bab", "keymail", "canary/x", "v1"); err != nil { |
| 2198 | t.Fatalf("Promote: %v", err) |
| 2199 | } |
| 2200 | // a holds the grant for :25 and is refused, because something else on |
| 2201 | // this box already claims it. b declares the same port with NO grant, so |
| 2202 | // it claims nothing, materializes only its service half, and is fine. |
| 2203 | addCapRoute(t, reg, "a.example.com", "keymail", "canary/x", registry.BackingUnit, "port:25", false, false) |
| 2204 | addCapRoute(t, reg, "b.example.com", "keymail", "canary/x", registry.BackingUnit, "", false, false) |
| 2205 | |
| 2206 | rec := newCapsRecorder(opt, "a.example.com") |
| 2207 | o, etc := capsTestOps(t, rec) |
| 2208 | // A drop-in belonging to a route this pass never sees, holding :25. |
| 2209 | incumbent := filepath.Join(etc, caps.SocketUnit("zz.example.com")+".d") |
| 2210 | if err := os.MkdirAll(incumbent, 0o755); err != nil { |
| 2211 | t.Fatal(err) |
| 2212 | } |
| 2213 | if err := os.WriteFile(filepath.Join(incumbent, caps.DropInName), |
| 2214 | []byte(caps.Marker+"\n[Socket]\nListenStream=25\n"), 0o644); err != nil { |
| 2215 | t.Fatal(err) |
| 2216 | } |
| 2217 | |
| 2218 | var out bytes.Buffer |
| 2219 | withCapturedStdout(t, &out, func() { |
| 2220 | if err := Pass(ctx, reg, s, pub, opt, func(string) error { return nil }, WithCapabilities(o)); err != nil { |
| 2221 | t.Fatalf("Pass: %v", err) |
| 2222 | } |
| 2223 | }) |
| 2224 | |
| 2225 | if v, _ := LiveVersion(opt, "a.example.com"); v != "" { |
| 2226 | t.Fatalf("a live = %q, want unset — its own refusal keeps it off the version", v) |
| 2227 | } |
| 2228 | if v, _ := LiveVersion(opt, "b.example.com"); v != "v1" { |
| 2229 | t.Fatalf("b live = %q, want v1 — another host's refusal must not freeze this one out of the version", v) |
| 2230 | } |
| 2231 | line := out.String() |
| 2232 | if strings.Contains(line, "failed on another host this pass") { |
| 2233 | t.Fatalf("printed %q, want no poison — a refusal is not a bad build", line) |
| 2234 | } |
| 2235 | if !strings.Contains(line, "a.example.com: capabilities for v1 were refused") { |
| 2236 | t.Fatalf("printed %q, want a's own refusal named", line) |
| 2237 | } |
| 2238 | } |
| 2239 | |
| 2240 | // TestPointLiveFailureRestoresTheDropIns: converge succeeded, the symlink did |
| 2241 | // not move. The box is still serving the previous version, and the unit shape |
| 2242 | // on disk is the new one — a port the running binary never asked for, bound |
| 2243 | // in public. The drop-ins have to follow the symlink, which stayed put. |
| 2244 | // |
| 2245 | // PointLive is failed by parking a non-empty directory at the temp-link path |
| 2246 | // it renames through, which is as close to a disk-level swap failure as this |
| 2247 | // package can get without a fake filesystem. |
| 2248 | func TestPointLiveFailureRestoresTheDropIns(t *testing.T) { |
| 2249 | s := NewDirStore(t.TempDir()) |
| 2250 | ctx := context.Background() |
| 2251 | priv, pub := testKeys(t) |
| 2252 | opt := t.TempDir() |
| 2253 | reg := testRegistry(t) |
| 2254 | host := "keymail.dev" |
| 2255 | |
| 2256 | v1 := &caps.Set{V: 1, Command: "relay", Daemon: true, Ports: []caps.Port{{Port: 25}}} |
| 2257 | v2 := &caps.Set{V: 1, Command: "relay", Daemon: true, Ports: []caps.Port{{Port: 25}, {Port: 465}}} |
| 2258 | shipCapRelease(t, s, "bab", "keymail", "v1", v1, []byte("relay one")) |
| 2259 | shipCapRelease(t, s, "bab", "keymail", "v2", v2, []byte("relay two")) |
| 2260 | if err := Promote(ctx, s, priv, "bab", "keymail", "canary/x", "v1"); err != nil { |
| 2261 | t.Fatalf("Promote v1: %v", err) |
| 2262 | } |
| 2263 | addCapRoute(t, reg, host, "keymail", "canary/x", registry.BackingUnit, "port:25,port:465", false, false) |
| 2264 | |
| 2265 | rec := newCapsRecorder(opt, host) |
| 2266 | o, etc := capsTestOps(t, rec) |
| 2267 | if err := Pass(ctx, reg, s, pub, opt, func(string) error { return nil }, WithCapabilities(o)); err != nil { |
| 2268 | t.Fatalf("Pass v1: %v", err) |
| 2269 | } |
| 2270 | svcPath, sockPath := caps.DropInPaths(etc, host) |
| 2271 | wantSvc, _ := os.ReadFile(svcPath) |
| 2272 | wantSock, _ := os.ReadFile(sockPath) |
| 2273 | if !strings.Contains(string(wantSock), "ListenStream=25") || strings.Contains(string(wantSock), "465") { |
| 2274 | t.Fatalf("v1 socket drop-in = %q, want :25 only", wantSock) |
| 2275 | } |
| 2276 | |
| 2277 | // Block the symlink swap. |
| 2278 | blocker := LiveLink(opt, host) + ".tmp" |
| 2279 | if err := os.MkdirAll(filepath.Join(blocker, "occupied"), 0o755); err != nil { |
| 2280 | t.Fatalf("park a blocker at %s: %v", blocker, err) |
| 2281 | } |
| 2282 | if err := Promote(ctx, s, priv, "bab", "keymail", "canary/x", "v2"); err != nil { |
| 2283 | t.Fatalf("Promote v2: %v", err) |
| 2284 | } |
| 2285 | |
| 2286 | var out bytes.Buffer |
| 2287 | withCapturedStdout(t, &out, func() { |
| 2288 | if err := Pass(ctx, reg, s, pub, opt, func(string) error { return nil }, WithCapabilities(o)); err != nil { |
| 2289 | t.Fatalf("Pass v2: %v", err) |
| 2290 | } |
| 2291 | }) |
| 2292 | |
| 2293 | if v, _ := LiveVersion(opt, host); v != "v1" { |
| 2294 | t.Fatalf("live = %q, want v1 — the repoint was supposed to fail", v) |
| 2295 | } |
| 2296 | gotSvc, _ := os.ReadFile(svcPath) |
| 2297 | gotSock, _ := os.ReadFile(sockPath) |
| 2298 | if !bytes.Equal(gotSvc, wantSvc) || !bytes.Equal(gotSock, wantSock) { |
| 2299 | t.Fatalf("drop-ins = %q / %q, want v1's %q / %q — :465 must not stay bound under a binary that never asked for it", |
| 2300 | gotSvc, gotSock, wantSvc, wantSock) |
| 2301 | } |
| 2302 | if line := out.String(); !strings.Contains(line, "repoint failed") || !strings.Contains(line, "capability drop-ins") { |
| 2303 | t.Fatalf("printed %q, want a line naming both the failed repoint and the restored drop-ins", line) |
| 2304 | } |
| 2305 | } |
| 2306 | |
| 2307 | // TestArmedPassConvergesFromTheStagedSiblingWhenTheBucketCannotAnswer pins |
| 2308 | // the read ensureStaged does on its already-staged path. The binary is on |
| 2309 | // disk with its sibling beside it and the bucket object is gone: the sibling |
| 2310 | // IS the manifest in hand there, and dropping that read would send this |
| 2311 | // straight to a bucket fetch that cannot answer, then to a refusal of a |
| 2312 | // version whose capabilities are sitting on disk, readable, all along. |
| 2313 | func TestArmedPassConvergesFromTheStagedSiblingWhenTheBucketCannotAnswer(t *testing.T) { |
| 2314 | dir := t.TempDir() |
| 2315 | s := NewDirStore(dir) |
| 2316 | ctx := context.Background() |
| 2317 | priv, pub := testKeys(t) |
| 2318 | opt := t.TempDir() |
| 2319 | reg := testRegistry(t) |
| 2320 | host := "keymail.dev" |
| 2321 | |
| 2322 | shipRelease(t, s, "bab", "keymail", "v1", adoptPlatform, []byte("old")) |
| 2323 | if err := StageArtifact(opt, "bab", "v1", "keymail", []byte("old")); err != nil { |
| 2324 | t.Fatal(err) |
| 2325 | } |
| 2326 | addCapRoute(t, reg, host, "keymail", "canary/x", registry.BackingUnit, "port:25", false, false) |
| 2327 | if err := PointLive(opt, host, "bab", "v1", "keymail"); err != nil { |
| 2328 | t.Fatal(err) |
| 2329 | } |
| 2330 | |
| 2331 | set := relaySet() |
| 2332 | shipCapRelease(t, s, "bab", "keymail", "v2", set, []byte("new")) |
| 2333 | if err := Promote(ctx, s, priv, "bab", "keymail", "canary/x", "v2"); err != nil { |
| 2334 | t.Fatalf("Promote v2: %v", err) |
| 2335 | } |
| 2336 | // v2 is staged the way a box that adopted it once already would have it: |
| 2337 | // binary AND sibling on disk. Then the bucket object goes away. |
| 2338 | if err := StageArtifact(opt, "bab", "v2", "keymail", []byte("new")); err != nil { |
| 2339 | t.Fatal(err) |
| 2340 | } |
| 2341 | raw, _, err := s.Get(ctx, ManifestKey("bab", "keymail", "v2")) |
| 2342 | if err != nil { |
| 2343 | t.Fatalf("read v2 manifest: %v", err) |
| 2344 | } |
| 2345 | if err := StageManifest(opt, "bab", "v2", "keymail", raw); err != nil { |
| 2346 | t.Fatalf("StageManifest: %v", err) |
| 2347 | } |
| 2348 | if err := os.Remove(filepath.Join(dir, filepath.FromSlash(ManifestKey("bab", "keymail", "v2")))); err != nil { |
| 2349 | t.Fatalf("remove v2 manifest: %v", err) |
| 2350 | } |
| 2351 | |
| 2352 | rec := newCapsRecorder(opt, host) |
| 2353 | o, etc := capsTestOps(t, rec) |
| 2354 | if err := Pass(ctx, reg, s, pub, opt, func(string) error { return nil }, WithCapabilities(o)); err != nil { |
| 2355 | t.Fatalf("Pass: %v", err) |
| 2356 | } |
| 2357 | |
| 2358 | if v, _ := LiveVersion(opt, host); v != "v2" { |
| 2359 | t.Fatalf("live = %q, want v2 — the sibling answered, so nothing needed the bucket", v) |
| 2360 | } |
| 2361 | g := caps.Grants{Ports: map[int]bool{25: true}} |
| 2362 | svcPath, _ := caps.DropInPaths(etc, host) |
| 2363 | wantSvc, err := caps.RenderService("keymail", set, g) |
| 2364 | if err != nil { |
| 2365 | t.Fatalf("RenderService: %v", err) |
| 2366 | } |
| 2367 | gotSvc, err := os.ReadFile(svcPath) |
| 2368 | if err != nil || !bytes.Equal(gotSvc, wantSvc) { |
| 2369 | t.Fatalf("service drop-in = %q (%v), want the sibling's capabilities materialized: %q", gotSvc, err, wantSvc) |
| 2370 | } |
| 2371 | } |
| 2372 | |
| 2373 | // TestRestartFailureSaysSoWhenTheRestoreAlsoFails: the summary line is the |
| 2374 | // only thing most operators read, and it must not claim a restore that did |
| 2375 | // not happen. A rollback that could not put the drop-ins back is the worst |
| 2376 | // state this path can end in — the symlink says v1, the unit text says v2 — |
| 2377 | // so it is exactly the case that must not be reported as a tidy rollback. |
| 2378 | // |
| 2379 | // The restore is failed by making the socket drop-in's directory unwritable |
| 2380 | // at the moment the daemon restart fails, i.e. after the forward converge has |
| 2381 | // written it and before the rollback tries to. |
| 2382 | func TestRestartFailureSaysSoWhenTheRestoreAlsoFails(t *testing.T) { |
| 2383 | if os.Geteuid() == 0 { |
| 2384 | t.Skip("running as root: directory permissions do not refuse a write") |
| 2385 | } |
| 2386 | s := NewDirStore(t.TempDir()) |
| 2387 | ctx := context.Background() |
| 2388 | priv, pub := testKeys(t) |
| 2389 | opt := t.TempDir() |
| 2390 | reg := testRegistry(t) |
| 2391 | host := "keymail.dev" |
| 2392 | |
| 2393 | v1 := &caps.Set{V: 1, Command: "relay", Daemon: true, Ports: []caps.Port{{Port: 25}}} |
| 2394 | v2 := &caps.Set{V: 1, Command: "relay", Daemon: true, Ports: []caps.Port{{Port: 25}, {Port: 465}}} |
| 2395 | shipCapRelease(t, s, "bab", "keymail", "v1", v1, []byte("relay one")) |
| 2396 | shipCapRelease(t, s, "bab", "keymail", "v2", v2, []byte("relay two")) |
| 2397 | if err := Promote(ctx, s, priv, "bab", "keymail", "canary/x", "v1"); err != nil { |
| 2398 | t.Fatalf("Promote v1: %v", err) |
| 2399 | } |
| 2400 | addCapRoute(t, reg, host, "keymail", "canary/x", registry.BackingUnit, "port:25,port:465", false, false) |
| 2401 | |
| 2402 | rec := newCapsRecorder(opt, host) |
| 2403 | o, etc := capsTestOps(t, rec) |
| 2404 | if err := Pass(ctx, reg, s, pub, opt, func(string) error { return nil }, WithCapabilities(o)); err != nil { |
| 2405 | t.Fatalf("Pass v1: %v", err) |
| 2406 | } |
| 2407 | |
| 2408 | _, sockPath := caps.DropInPaths(etc, host) |
| 2409 | sockDir := filepath.Dir(sockPath) |
| 2410 | t.Cleanup(func() { os.Chmod(sockDir, 0o755) }) |
| 2411 | rec.fail["restart "+caps.ServiceUnit(host)] = errors.New("relay failed to start") |
| 2412 | rec.onCall = func(call string) { |
| 2413 | if call == "restart "+caps.ServiceUnit(host) { |
| 2414 | if err := os.Chmod(sockDir, 0o555); err != nil { |
| 2415 | t.Errorf("chmod %s: %v", sockDir, err) |
| 2416 | } |
| 2417 | } |
| 2418 | } |
| 2419 | |
| 2420 | if err := Promote(ctx, s, priv, "bab", "keymail", "canary/x", "v2"); err != nil { |
| 2421 | t.Fatalf("Promote v2: %v", err) |
| 2422 | } |
| 2423 | var out bytes.Buffer |
| 2424 | withCapturedStdout(t, &out, func() { |
| 2425 | if err := Pass(ctx, reg, s, pub, opt, func(string) error { return nil }, WithCapabilities(o)); err != nil { |
| 2426 | t.Fatalf("Pass v2: %v", err) |
| 2427 | } |
| 2428 | }) |
| 2429 | |
| 2430 | line := out.String() |
| 2431 | if !strings.Contains(line, "could not restore") { |
| 2432 | t.Fatalf("printed %q, want the summary to say the restore failed", line) |
| 2433 | } |
| 2434 | if strings.Contains(line, "and restored the previous version's capability drop-ins") { |
| 2435 | t.Fatalf("printed %q, want no claim of a restore that did not happen", line) |
| 2436 | } |
| 2437 | if v, _ := LiveVersion(opt, host); v != "v1" { |
| 2438 | t.Fatalf("live = %q, want v1 — the symlink half of the rollback still has to happen", v) |
| 2439 | } |
| 2440 | } |
| 2441 | |
| 2442 | // TestFirstAdoptionRestartFailureLeavesTheConvergedDropIns: with no previous |
| 2443 | // version there is no symlink to roll back — the box stays live on the |
| 2444 | // version whose restart just failed. Handing caps.Restore the empty snapshot |
| 2445 | // that describes "this route had no drop-ins" would then REMOVE the shape the |
| 2446 | // live version needs, leaving it unconfigured until the hourly grant-drift |
| 2447 | // pass rebuilds it. The honest state is the one the symlink names. |
| 2448 | func TestFirstAdoptionRestartFailureLeavesTheConvergedDropIns(t *testing.T) { |
| 2449 | s := NewDirStore(t.TempDir()) |
| 2450 | ctx := context.Background() |
| 2451 | priv, pub := testKeys(t) |
| 2452 | opt := t.TempDir() |
| 2453 | reg := testRegistry(t) |
| 2454 | host := "keymail.dev" |
| 2455 | |
| 2456 | set := relaySet() |
| 2457 | shipCapRelease(t, s, "bab", "keymail", "v1", set, []byte("relay")) |
| 2458 | if err := Promote(ctx, s, priv, "bab", "keymail", "canary/x", "v1"); err != nil { |
| 2459 | t.Fatalf("Promote: %v", err) |
| 2460 | } |
| 2461 | addCapRoute(t, reg, host, "keymail", "canary/x", registry.BackingUnit, "port:25", false, false) |
| 2462 | |
| 2463 | rec := newCapsRecorder(opt, host) |
| 2464 | o, etc := capsTestOps(t, rec) |
| 2465 | rec.fail["restart "+caps.ServiceUnit(host)] = errors.New("relay failed to start") |
| 2466 | |
| 2467 | var out bytes.Buffer |
| 2468 | withCapturedStdout(t, &out, func() { |
| 2469 | if err := Pass(ctx, reg, s, pub, opt, func(string) error { return nil }, WithCapabilities(o)); err != nil { |
| 2470 | t.Fatalf("Pass: %v", err) |
| 2471 | } |
| 2472 | }) |
| 2473 | |
| 2474 | if v, _ := LiveVersion(opt, host); v != "v1" { |
| 2475 | t.Fatalf("live = %q, want v1 — there is no previous version to roll back to", v) |
| 2476 | } |
| 2477 | g := caps.Grants{Ports: map[int]bool{25: true}} |
| 2478 | svcPath, sockPath := caps.DropInPaths(etc, host) |
| 2479 | wantSvc, _ := caps.RenderService("keymail", set, g) |
| 2480 | wantSock, _ := caps.RenderSocket(set, g) |
| 2481 | gotSvc, err := os.ReadFile(svcPath) |
| 2482 | if err != nil || !bytes.Equal(gotSvc, wantSvc) { |
| 2483 | t.Fatalf("service drop-in = %q (%v), want the converged shape the live version needs: %q", gotSvc, err, wantSvc) |
| 2484 | } |
| 2485 | gotSock, err := os.ReadFile(sockPath) |
| 2486 | if err != nil || !bytes.Equal(gotSock, wantSock) { |
| 2487 | t.Fatalf("socket drop-in = %q (%v), want %q", gotSock, err, wantSock) |
| 2488 | } |
| 2489 | if line := out.String(); !strings.Contains(line, "no previous version to roll back to") { |
| 2490 | t.Fatalf("printed %q, want the line to say why nothing was rolled back", line) |
| 2491 | } |
| 2492 | } |
| 2493 | |
| 2494 | // ---- the production flag on the adopt path (2026-08-04 spec §6) ------------- |
| 2495 | |
| 2496 | // The default: no flag, no window (2026-08-04 spec §2). The whole fleet |
| 2497 | // runs on canary/* today, so nothing live loses protection — and a |
| 2498 | // rapid-iteration app's stable is just another channel. |
| 2499 | func TestPassNoProductionFlagInstallsImmediately(t *testing.T) { |
| 2500 | s := NewDirStore(t.TempDir()) |
| 2501 | ctx := context.Background() |
| 2502 | priv, pub := testKeys(t) |
| 2503 | opt := t.TempDir() |
| 2504 | reg := testRegistry(t) |
| 2505 | |
| 2506 | shipRelease(t, s, "bab", "myapp", "v1", adoptPlatform, []byte("bin")) |
| 2507 | if err := Promote(ctx, s, priv, "bab", "myapp", "stable", "v1"); err != nil { |
| 2508 | t.Fatal(err) |
| 2509 | } |
| 2510 | addRoute(t, reg, "h.example.com", "myapp", "", false) |
| 2511 | |
| 2512 | if err := Pass(ctx, reg, s, pub, opt, func(string) error { return nil }); err != nil { |
| 2513 | t.Fatalf("Pass: %v", err) |
| 2514 | } |
| 2515 | if v, _ := LiveVersion(opt, "h.example.com"); v != "v1" { |
| 2516 | t.Fatalf("live = %q, want v1 — no production flag means no window", v) |
| 2517 | } |
| 2518 | } |
| 2519 | |
| 2520 | // Fast-track: a production app's held head installs the moment its pointer |
| 2521 | // carries the root-signed waiver (spec §5). |
| 2522 | func TestPassFastTrackSkipsTheWindow(t *testing.T) { |
| 2523 | s := NewDirStore(t.TempDir()) |
| 2524 | ctx := context.Background() |
| 2525 | priv, pub := testKeys(t) |
| 2526 | opt := t.TempDir() |
| 2527 | reg := testRegistry(t) |
| 2528 | |
| 2529 | shipRelease(t, s, "bab", "myapp", "v1", adoptPlatform, []byte("bin")) |
| 2530 | if err := Promote(ctx, s, priv, "bab", "myapp", "stable", "v1"); err != nil { |
| 2531 | t.Fatal(err) |
| 2532 | } |
| 2533 | if err := SetProduction(ctx, s, priv, "bab", "myapp", true, "test"); err != nil { |
| 2534 | t.Fatal(err) |
| 2535 | } |
| 2536 | addRoute(t, reg, "h.example.com", "myapp", "", false) |
| 2537 | |
| 2538 | if err := Pass(ctx, reg, s, pub, opt, func(string) error { return nil }); err != nil { |
| 2539 | t.Fatal(err) |
| 2540 | } |
| 2541 | if v, _ := LiveVersion(opt, "h.example.com"); v != "" { |
| 2542 | t.Fatalf("live = %q, want held (production, no waiver yet)", v) |
| 2543 | } |
| 2544 | if _, err := FastTrack(ctx, s, priv, "bab", "myapp", "stable", "test"); err != nil { |
| 2545 | t.Fatal(err) |
| 2546 | } |
| 2547 | if err := Pass(ctx, reg, s, pub, opt, func(string) error { return nil }); err != nil { |
| 2548 | t.Fatal(err) |
| 2549 | } |
| 2550 | if v, _ := LiveVersion(opt, "h.example.com"); v != "v1" { |
| 2551 | t.Fatalf("live = %q, want v1 — the waiver did not skip the window", v) |
| 2552 | } |
| 2553 | } |
| 2554 | |
| 2555 | // A production app whose flag OBJECT rots (exists, won't verify) keeps |
| 2556 | // baking — an attacker who can corrupt but not sign shortens nothing. |
| 2557 | func TestPassCorruptProductionFlagFailsTowardBaking(t *testing.T) { |
| 2558 | s := NewDirStore(t.TempDir()) |
| 2559 | ctx := context.Background() |
| 2560 | priv, pub := testKeys(t) |
| 2561 | opt := t.TempDir() |
| 2562 | reg := testRegistry(t) |
| 2563 | |
| 2564 | shipRelease(t, s, "bab", "myapp", "v1", adoptPlatform, []byte("bin")) |
| 2565 | if err := Promote(ctx, s, priv, "bab", "myapp", "stable", "v1"); err != nil { |
| 2566 | t.Fatal(err) |
| 2567 | } |
| 2568 | if _, err := s.Put(ctx, ProductionKey("bab", "myapp"), []byte("not a wrapper"), ""); err != nil { |
| 2569 | t.Fatal(err) |
| 2570 | } |
| 2571 | addRoute(t, reg, "h.example.com", "myapp", "", false) |
| 2572 | |
| 2573 | if err := Pass(ctx, reg, s, pub, opt, func(string) error { return nil }); err != nil { |
| 2574 | t.Fatalf("Pass: %v", err) |
| 2575 | } |
| 2576 | if v, _ := LiveVersion(opt, "h.example.com"); v != "" { |
| 2577 | t.Fatalf("live = %q, want held — a corrupt flag must read as production", v) |
| 2578 | } |
| 2579 | } |
| 2580 | |
| 2581 | // ORDERING, pinned: ObserveProduction runs BEFORE this pass's bake verdicts |
| 2582 | // are taken, so the bucket flag re-latches a wiped memory within ONE pass — |
| 2583 | // the whole basis of prodMemory's "a mixed-version rollout costs at most one |
| 2584 | // pass of this memory" comment. Simulates the old binary that re-marshals a |
| 2585 | // struct with no Production field and thereby deletes the latch. |
| 2586 | func TestPassReLatchesProductionMemoryInTheSamePass(t *testing.T) { |
| 2587 | s := NewDirStore(t.TempDir()) |
| 2588 | ctx := context.Background() |
| 2589 | priv, pub := testKeys(t) |
| 2590 | opt := t.TempDir() |
| 2591 | reg := testRegistry(t) |
| 2592 | |
| 2593 | shipRelease(t, s, "bab", "myapp", "v1", adoptPlatform, []byte("bin")) |
| 2594 | if err := Promote(ctx, s, priv, "bab", "myapp", "stable", "v1"); err != nil { |
| 2595 | t.Fatal(err) |
| 2596 | } |
| 2597 | if err := SetProduction(ctx, s, priv, "bab", "myapp", true, "test"); err != nil { |
| 2598 | t.Fatal(err) |
| 2599 | } |
| 2600 | addRoute(t, reg, "h.example.com", "myapp", "", false) |
| 2601 | |
| 2602 | if err := Pass(ctx, reg, s, pub, opt, func(string) error { return nil }); err != nil { |
| 2603 | t.Fatal(err) |
| 2604 | } |
| 2605 | st := LoadFirstSeen(FirstSeenPath(opt)) |
| 2606 | if len(st.Production) == 0 { |
| 2607 | t.Fatalf("first pass did not latch the production memory") |
| 2608 | } |
| 2609 | |
| 2610 | // The old binary's write: every channel observation survives, the |
| 2611 | // production memory (and with it the replay floor) is gone. |
| 2612 | st.Production = map[string]*prodMemory{} |
| 2613 | if err := st.Save(); err != nil { |
| 2614 | t.Fatalf("Save wiped clock: %v", err) |
| 2615 | } |
| 2616 | |
| 2617 | if err := Pass(ctx, reg, s, pub, opt, func(string) error { return nil }); err != nil { |
| 2618 | t.Fatal(err) |
| 2619 | } |
| 2620 | if v, _ := LiveVersion(opt, "h.example.com"); v != "" { |
| 2621 | t.Fatalf("live = %q, want still held — the bucket flag must re-latch BEFORE this pass's bake verdict, not after it", v) |
| 2622 | } |
| 2623 | if st := LoadFirstSeen(FirstSeenPath(opt)); len(st.Production) == 0 { |
| 2624 | t.Fatalf("the re-latch was not persisted") |
| 2625 | } |
| 2626 | } |
| 2627 | |
| 2628 | // A clock file that exists but does not parse is rewritten within ONE pass, |
| 2629 | // even when the pass observes nothing new — the taint-forced Save. Without |
| 2630 | // it a corrupt file would sit there forever, and every held head on the box |
| 2631 | // would fail toward baking pass after pass (ProductionBake's st.tainted). |
| 2632 | func TestPassRewritesATaintedClockInOnePass(t *testing.T) { |
| 2633 | s := NewDirStore(t.TempDir()) |
| 2634 | ctx := context.Background() |
| 2635 | _, pub := testKeys(t) |
| 2636 | opt := t.TempDir() |
| 2637 | reg := testRegistry(t) // no routes: nothing new is observed this pass |
| 2638 | |
| 2639 | if err := os.WriteFile(FirstSeenPath(opt), []byte("{not json"), 0o644); err != nil { |
| 2640 | t.Fatal(err) |
| 2641 | } |
| 2642 | if st := LoadFirstSeen(FirstSeenPath(opt)); !st.tainted { |
| 2643 | t.Fatalf("setup: the corrupt clock did not read as tainted") |
| 2644 | } |
| 2645 | |
| 2646 | if err := Pass(ctx, reg, s, pub, opt, func(string) error { return nil }); err != nil { |
| 2647 | t.Fatalf("Pass: %v", err) |
| 2648 | } |
| 2649 | if st := LoadFirstSeen(FirstSeenPath(opt)); st.tainted { |
| 2650 | t.Fatalf("the pass left the clock file corrupt — a tainted clock must force its own Save") |
| 2651 | } |
| 2652 | } |
| 2653 | |
| 2654 | // prodReadStore counts Get calls for one exact key. |
| 2655 | type prodReadStore struct { |
| 2656 | Store |
| 2657 | key string |
| 2658 | n int |
| 2659 | } |
| 2660 | |
| 2661 | func (p *prodReadStore) Get(ctx context.Context, key string) ([]byte, string, error) { |
| 2662 | if key == p.key { |
| 2663 | p.n++ |
| 2664 | } |
| 2665 | return p.Store.Get(ctx, key) |
| 2666 | } |
| 2667 | |
| 2668 | // The 2-3 SECOND canary fast loop must never pay a production-flag bucket |
| 2669 | // read: canary/* has no window (DefaultHolds), so no canary-only pass ever |
| 2670 | // has a HELD head to ask about. The hourly pass pays exactly one read per |
| 2671 | // (account, app) with a held head — not one per channel, not one per route. |
| 2672 | func TestCanaryFastLoopPaysNoProductionRead(t *testing.T) { |
| 2673 | dir := NewDirStore(t.TempDir()) |
| 2674 | ctx := context.Background() |
| 2675 | priv, pub := testKeys(t) |
| 2676 | opt := t.TempDir() |
| 2677 | reg := testRegistry(t) |
| 2678 | |
| 2679 | shipRelease(t, dir, "bab", "myapp", "v1", adoptPlatform, []byte("bin")) |
| 2680 | for _, ch := range []string{"canary/x", "canary/y", "stable"} { |
| 2681 | if err := Promote(ctx, dir, priv, "bab", "myapp", ch, "v1"); err != nil { |
| 2682 | t.Fatalf("Promote %s: %v", ch, err) |
| 2683 | } |
| 2684 | } |
| 2685 | if err := SetProduction(ctx, dir, priv, "bab", "myapp", true, "test"); err != nil { |
| 2686 | t.Fatal(err) |
| 2687 | } |
| 2688 | addRoute(t, reg, "cx.example.com", "myapp", "canary/x", false) |
| 2689 | addRoute(t, reg, "cy.example.com", "myapp", "canary/y", false) |
| 2690 | addRoute(t, reg, "s.example.com", "myapp", "stable", false) |
| 2691 | |
| 2692 | s := &prodReadStore{Store: dir, key: ProductionKey("bab", "myapp")} |
| 2693 | if err := PassFilter(ctx, reg, s, pub, opt, func(string) error { return nil }, isCanaryTestRoute); err != nil { |
| 2694 | t.Fatalf("PassFilter: %v", err) |
| 2695 | } |
| 2696 | if s.n != 0 { |
| 2697 | t.Fatalf("canary-only pass read the production flag %d times, want 0 — the fast loop must not pay bucket reads for a window canary/* does not have", s.n) |
| 2698 | } |
| 2699 | |
| 2700 | // Not vacuous: the full pass, which DOES resolve a held head, reads it — |
| 2701 | // once, for two canary channels, a stable one and three routes. |
| 2702 | if err := Pass(ctx, reg, s, pub, opt, func(string) error { return nil }); err != nil { |
| 2703 | t.Fatalf("Pass: %v", err) |
| 2704 | } |
| 2705 | if s.n != 1 { |
| 2706 | t.Fatalf("full pass read the production flag %d times, want exactly 1 per (account, app)", s.n) |
| 2707 | } |
| 2708 | } |
| 2709 | |
| 2710 | // A NARROWED pass (the canary fast loop) must never repair a tainted clock: |
| 2711 | // it fetched no production flags, so its Production map is empty, and writing |
| 2712 | // it would replace the corrupt file with that emptiness AND clear the taint |
| 2713 | // that is currently failing closed — before anything re-observed the flags. |
| 2714 | // The full pass, which holds the flags, does the repair. |
| 2715 | func TestNarrowedPassNeverRepairsATaintedClock(t *testing.T) { |
| 2716 | s := NewDirStore(t.TempDir()) |
| 2717 | ctx := context.Background() |
| 2718 | priv, pub := testKeys(t) |
| 2719 | opt := t.TempDir() |
| 2720 | reg := testRegistry(t) |
| 2721 | |
| 2722 | shipRelease(t, s, "bab", "myapp", "v1", adoptPlatform, []byte("bin")) |
| 2723 | for _, ch := range []string{"canary/x", "stable"} { |
| 2724 | if err := Promote(ctx, s, priv, "bab", "myapp", ch, "v1"); err != nil { |
| 2725 | t.Fatalf("Promote %s: %v", ch, err) |
| 2726 | } |
| 2727 | } |
| 2728 | if err := SetProduction(ctx, s, priv, "bab", "myapp", true, "test"); err != nil { |
| 2729 | t.Fatal(err) |
| 2730 | } |
| 2731 | addRoute(t, reg, "cx.example.com", "myapp", "canary/x", false) |
| 2732 | addRoute(t, reg, "s.example.com", "myapp", "stable", false) |
| 2733 | |
| 2734 | if err := os.WriteFile(FirstSeenPath(opt), []byte("{not json"), 0o644); err != nil { |
| 2735 | t.Fatal(err) |
| 2736 | } |
| 2737 | |
| 2738 | if err := PassFilter(ctx, reg, s, pub, opt, func(string) error { return nil }, isCanaryTestRoute); err != nil { |
| 2739 | t.Fatalf("PassFilter: %v", err) |
| 2740 | } |
| 2741 | if raw, err := os.ReadFile(FirstSeenPath(opt)); err != nil { |
| 2742 | t.Fatal(err) |
| 2743 | } else if json.Valid(raw) { |
| 2744 | t.Fatalf("the canary fast loop overwrote a tainted clock — a pass that fetched no flags cannot repair what it would overwrite; file is now %q", raw) |
| 2745 | } |
| 2746 | |
| 2747 | // The hourly pass has the flags in hand, so it repairs — and the app's |
| 2748 | // production state survives, re-latched from the bucket in that same pass. |
| 2749 | if err := Pass(ctx, reg, s, pub, opt, func(string) error { return nil }); err != nil { |
| 2750 | t.Fatalf("Pass: %v", err) |
| 2751 | } |
| 2752 | st := LoadFirstSeen(FirstSeenPath(opt)) |
| 2753 | if st.tainted { |
| 2754 | t.Fatalf("the full pass did not repair the tainted clock") |
| 2755 | } |
| 2756 | if mem := st.Production[prodKey("bab", "myapp")]; mem == nil || !mem.WasProduction { |
| 2757 | t.Fatalf("production memory = %+v, want the latch re-established by the repairing pass", mem) |
| 2758 | } |
| 2759 | } |
| 2760 | |
| 2761 | // The off-ratchet, end to end on the adopt path: flipping production OFF does |
| 2762 | // not release a held head immediately — the off is honored only after the |
| 2763 | // stable window has elapsed on this box's own sighting of it (spec §4). This |
| 2764 | // is the dangerous direction, so it is pinned here and not only in bake_test: |
| 2765 | // it is also what fails if ProductionBake's hasBakeableHead carve-out is ever |
| 2766 | // passed a constant false from this call site. |
| 2767 | func TestPassProductionOffStillRatchets(t *testing.T) { |
| 2768 | s := NewDirStore(t.TempDir()) |
| 2769 | ctx := context.Background() |
| 2770 | priv, pub := testKeys(t) |
| 2771 | opt := t.TempDir() |
| 2772 | reg := testRegistry(t) |
| 2773 | |
| 2774 | shipRelease(t, s, "bab", "myapp", "v1", adoptPlatform, []byte("bin")) |
| 2775 | if err := Promote(ctx, s, priv, "bab", "myapp", "stable", "v1"); err != nil { |
| 2776 | t.Fatal(err) |
| 2777 | } |
| 2778 | if err := SetProduction(ctx, s, priv, "bab", "myapp", true, "test"); err != nil { |
| 2779 | t.Fatal(err) |
| 2780 | } |
| 2781 | addRoute(t, reg, "h.example.com", "myapp", "", false) |
| 2782 | |
| 2783 | if err := Pass(ctx, reg, s, pub, opt, func(string) error { return nil }); err != nil { |
| 2784 | t.Fatal(err) |
| 2785 | } |
| 2786 | if v, _ := LiveVersion(opt, "h.example.com"); v != "" { |
| 2787 | t.Fatalf("live = %q, want held (declared production)", v) |
| 2788 | } |
| 2789 | |
| 2790 | // Flip it off. An operator who can flip the flag must not thereby have a |
| 2791 | // same-minute bypass of the window their held head is sitting in. |
| 2792 | if err := SetProduction(ctx, s, priv, "bab", "myapp", false, "test"); err != nil { |
| 2793 | t.Fatal(err) |
| 2794 | } |
| 2795 | if err := Pass(ctx, reg, s, pub, opt, func(string) error { return nil }); err != nil { |
| 2796 | t.Fatal(err) |
| 2797 | } |
| 2798 | if v, _ := LiveVersion(opt, "h.example.com"); v != "" { |
| 2799 | t.Fatalf("live = %q, want STILL held — production: false must ratchet through the stable window, never release a held head on sight", v) |
| 2800 | } |
| 2801 | // And the box recorded the off-sighting, which is what the ratchet is |
| 2802 | // measured from — its own clock, never the flag's timestamp. |
| 2803 | mem := LoadFirstSeen(FirstSeenPath(opt)).Production[prodKey("bab", "myapp")] |
| 2804 | if mem == nil || mem.OffFirstSeen.IsZero() || !mem.WasProduction { |
| 2805 | t.Fatalf("production memory = %+v, want a latched app with an off-sighting recorded", mem) |
| 2806 | } |
| 2807 | } |
| 2808 | |
| 2809 | // ---- pipeline windows on the adopt path (release pipelines v2 spec §3) ------ |
| 2810 | |
| 2811 | // TestPassEnforcesPipelineWindows: an app with pipeline {edge(0), |
| 2812 | // production(24h)}. Four behaviours in one lifecycle, in order: a release at |
| 2813 | // the head of "production" is HELD even though the app's production flag is |
| 2814 | // absent (pipelined apps never consult the flag); the same release at the |
| 2815 | // head of "edge" adopts immediately; the declared 24h — narrower than the |
| 2816 | // 72h a brand-new channel baselines at — ratchets in only after the OLD |
| 2817 | // window has elapsed on this box's own clock, and then governs; and a |
| 2818 | // FastTrack-re-signed pointer adopts immediately while held (spec §3 |
| 2819 | // amendment — ratcheting the waiver would make it wait out the very window |
| 2820 | // it exists to skip). |
| 2821 | func TestPassEnforcesPipelineWindows(t *testing.T) { |
| 2822 | s := NewDirStore(t.TempDir()) |
| 2823 | ctx := context.Background() |
| 2824 | priv, pub := testKeys(t) |
| 2825 | opt := t.TempDir() |
| 2826 | reg := testRegistry(t) |
| 2827 | |
| 2828 | // NO production flag is ever set in this test: a pipelined app's windows |
| 2829 | // come from the pipeline alone. (Legacy behaviour — no flag, no window — |
| 2830 | // would adopt the production head immediately.) |
| 2831 | if _, err := SetPipeline(ctx, s, priv, "bab", "myapp", []PipelineChannel{ |
| 2832 | {Name: "edge"}, {Name: "production", Bake: "24h"}, |
| 2833 | }, "test"); err != nil { |
| 2834 | t.Fatalf("SetPipeline: %v", err) |
| 2835 | } |
| 2836 | shipRelease(t, s, "bab", "myapp", "v1", adoptPlatform, []byte("bin one")) |
| 2837 | for _, ch := range []string{"edge", "production"} { |
| 2838 | if err := Promote(ctx, s, priv, "bab", "myapp", ch, "v1"); err != nil { |
| 2839 | t.Fatalf("Promote %s: %v", ch, err) |
| 2840 | } |
| 2841 | } |
| 2842 | addRoute(t, reg, "e.example.com", "myapp", "edge", false) |
| 2843 | addRoute(t, reg, "p.example.com", "myapp", "production", false) |
| 2844 | |
| 2845 | // Sub-step 1: the production head is held with no flag in sight. |
| 2846 | if err := Pass(ctx, reg, s, pub, opt, func(string) error { return nil }); err != nil { |
| 2847 | t.Fatalf("Pass: %v", err) |
| 2848 | } |
| 2849 | if v, _ := LiveVersion(opt, "p.example.com"); v != "" { |
| 2850 | t.Fatalf("production live = %q, want held — a pipelined app's bake window must apply with no production flag at all", v) |
| 2851 | } |
| 2852 | // Sub-step 2: the zero-bake entry channel adopted in that same pass. |
| 2853 | if v, _ := LiveVersion(opt, "e.example.com"); v != "v1" { |
| 2854 | t.Fatalf("edge live = %q, want v1 — the pipeline's zero-bake channel adopts immediately", v) |
| 2855 | } |
| 2856 | |
| 2857 | // Sub-step 3: the ratchet lands. The pass above latched the 72h |
| 2858 | // fail-closed baseline and parked the declared 24h as pending. Backdate |
| 2859 | // the box's own clocks: the pending sighting past the old window (72h |
| 2860 | // jittered < 90h < 91h), and v1's first sighting past 24h jittered |
| 2861 | // (< 30h < 31h) but NOT past 72h — adoption then proves the enforced |
| 2862 | // window really is the ratcheted-in 24h, not an expired baseline. |
| 2863 | st := LoadFirstSeen(FirstSeenPath(opt)) |
| 2864 | key := channelClockKey("bab", "myapp", "production") |
| 2865 | mem := st.Holds[key] |
| 2866 | if mem == nil || mem.Window != 72*time.Hour || mem.Pending != 24*time.Hour || mem.PendingAt.IsZero() { |
| 2867 | t.Fatalf("hold memory = %+v, want the 72h baseline with 24h pending", mem) |
| 2868 | } |
| 2869 | mem.PendingAt = mem.PendingAt.Add(-91 * time.Hour) |
| 2870 | if seen, ok := st.Channels[key]["v1"]; !ok { |
| 2871 | t.Fatalf("no first sighting recorded for v1 at the production head") |
| 2872 | } else { |
| 2873 | st.Channels[key]["v1"] = seen.Add(-31 * time.Hour) |
| 2874 | } |
| 2875 | if err := st.Save(); err != nil { |
| 2876 | t.Fatalf("Save backdated clock: %v", err) |
| 2877 | } |
| 2878 | if err := Pass(ctx, reg, s, pub, opt, func(string) error { return nil }); err != nil { |
| 2879 | t.Fatalf("Pass: %v", err) |
| 2880 | } |
| 2881 | if v, _ := LiveVersion(opt, "p.example.com"); v != "v1" { |
| 2882 | t.Fatalf("production live = %q, want v1 — the ratcheted-in 24h window elapsed on the box's own clock", v) |
| 2883 | } |
| 2884 | // And the landing was PERSISTED by that pass: ObservePipeline's changed |
| 2885 | // flag must reach the Save path, or the promotion would re-run (and the |
| 2886 | // window claim silently regress) every pass until some other change |
| 2887 | // happened to force a write. |
| 2888 | if mem := LoadFirstSeen(FirstSeenPath(opt)).Holds[key]; mem == nil || mem.Window != 24*time.Hour || !mem.PendingAt.IsZero() { |
| 2889 | t.Fatalf("persisted hold memory = %+v, want the landed 24h window", mem) |
| 2890 | } |
| 2891 | |
| 2892 | // Sub-step 4: fast-track. v2 at the production head is inside the (now |
| 2893 | // 24h) window; the root-re-signed waiver releases it, unchanged by the |
| 2894 | // pipeline. |
| 2895 | shipRelease(t, s, "bab", "myapp", "v2", adoptPlatform, []byte("bin two")) |
| 2896 | if err := Promote(ctx, s, priv, "bab", "myapp", "production", "v2"); err != nil { |
| 2897 | t.Fatalf("Promote v2: %v", err) |
| 2898 | } |
| 2899 | if err := Pass(ctx, reg, s, pub, opt, func(string) error { return nil }); err != nil { |
| 2900 | t.Fatalf("Pass: %v", err) |
| 2901 | } |
| 2902 | if v, _ := LiveVersion(opt, "p.example.com"); v != "v1" { |
| 2903 | t.Fatalf("production live = %q, want still v1 — v2 is inside the window until the waiver", v) |
| 2904 | } |
| 2905 | if _, err := FastTrack(ctx, s, priv, "bab", "myapp", "production", "test"); err != nil { |
| 2906 | t.Fatalf("FastTrack: %v", err) |
| 2907 | } |
| 2908 | if err := Pass(ctx, reg, s, pub, opt, func(string) error { return nil }); err != nil { |
| 2909 | t.Fatalf("Pass: %v", err) |
| 2910 | } |
| 2911 | if v, _ := LiveVersion(opt, "p.example.com"); v != "v2" { |
| 2912 | t.Fatalf("production live = %q, want v2 — FastTrack must keep bypassing a pipelined hold", v) |
| 2913 | } |
| 2914 | } |
| 2915 | |
| 2916 | // TestPassPipelinedAppCanaryKeepsNoWindow: canary/* lives OUTSIDE the |
| 2917 | // pipeline (spec §2), so a pipelined app's canary routes must keep their |
| 2918 | // zero window. The trap is structural: HoldsForApp's map carries only the |
| 2919 | // pipeline's channels, and HoldFor's fallback for a name missing from the |
| 2920 | // map is the 72h fail-closed default — without canary/* stitched back in at |
| 2921 | // zero, declaring a pipeline would freeze every canary deploy of the app. |
| 2922 | func TestPassPipelinedAppCanaryKeepsNoWindow(t *testing.T) { |
| 2923 | s := NewDirStore(t.TempDir()) |
| 2924 | ctx := context.Background() |
| 2925 | priv, pub := testKeys(t) |
| 2926 | opt := t.TempDir() |
| 2927 | reg := testRegistry(t) |
| 2928 | |
| 2929 | if _, err := SetPipeline(ctx, s, priv, "bab", "myapp", []PipelineChannel{ |
| 2930 | {Name: "edge"}, {Name: "production", Bake: "24h"}, |
| 2931 | }, "test"); err != nil { |
| 2932 | t.Fatalf("SetPipeline: %v", err) |
| 2933 | } |
| 2934 | shipRelease(t, s, "bab", "myapp", "v1", adoptPlatform, []byte("bin")) |
| 2935 | for _, ch := range []string{"canary/x", "production"} { |
| 2936 | if err := Promote(ctx, s, priv, "bab", "myapp", ch, "v1"); err != nil { |
| 2937 | t.Fatalf("Promote %s: %v", ch, err) |
| 2938 | } |
| 2939 | } |
| 2940 | addRoute(t, reg, "c.example.com", "myapp", "canary/x", false) |
| 2941 | addRoute(t, reg, "p.example.com", "myapp", "production", false) |
| 2942 | |
| 2943 | if err := Pass(ctx, reg, s, pub, opt, func(string) error { return nil }); err != nil { |
| 2944 | t.Fatalf("Pass: %v", err) |
| 2945 | } |
| 2946 | if v, _ := LiveVersion(opt, "c.example.com"); v != "v1" { |
| 2947 | t.Fatalf("canary live = %q, want v1 — canary/* is outside the pipeline and must keep its zero window", v) |
| 2948 | } |
| 2949 | // Not vacuous: the held production head in the SAME pass proves the |
| 2950 | // pipelined holds map (not DefaultHolds-with-no-flag, which would have |
| 2951 | // adopted it) is what governed — including the canary route above. |
| 2952 | if v, _ := LiveVersion(opt, "p.example.com"); v != "" { |
| 2953 | t.Fatalf("production live = %q, want held — without the hold this test proves nothing about the map the canary resolved through", v) |
| 2954 | } |
| 2955 | } |
| 2956 | |
| 2957 | // TestPassCorruptPipelineFailsTowardLegacyHolds: a pipeline object that |
| 2958 | // EXISTS but will not read or verify is not a pipeline and not an absence — |
| 2959 | // the app falls back to the legacy path (DefaultHolds + the production |
| 2960 | // flag), the same fail-closed posture an old binary that cannot parse |
| 2961 | // pipeline.json takes, whose mitigation is that enabling a nonzero-bake |
| 2962 | // pipeline also sets production: true (spec §3). And the errored read must |
| 2963 | // NEVER reach the ratchet memory as a nil: nil means verified-absent, and a |
| 2964 | // nil fold would land pending promotions — a corrupt object masquerading as |
| 2965 | // a deletion. |
| 2966 | func TestPassCorruptPipelineFailsTowardLegacyHolds(t *testing.T) { |
| 2967 | s := NewDirStore(t.TempDir()) |
| 2968 | ctx := context.Background() |
| 2969 | priv, pub := testKeys(t) |
| 2970 | opt := t.TempDir() |
| 2971 | reg := testRegistry(t) |
| 2972 | |
| 2973 | shipRelease(t, s, "bab", "myapp", "v1", adoptPlatform, []byte("bin")) |
| 2974 | if err := Promote(ctx, s, priv, "bab", "myapp", "production", "v1"); err != nil { |
| 2975 | t.Fatalf("Promote: %v", err) |
| 2976 | } |
| 2977 | // The flag the console sets alongside any nonzero-bake pipeline. |
| 2978 | if err := SetProduction(ctx, s, priv, "bab", "myapp", true, "test"); err != nil { |
| 2979 | t.Fatalf("SetProduction: %v", err) |
| 2980 | } |
| 2981 | if _, err := s.Put(ctx, PipelineKey("bab", "myapp"), []byte("not a wrapper"), ""); err != nil { |
| 2982 | t.Fatalf("corrupt the pipeline object: %v", err) |
| 2983 | } |
| 2984 | addRoute(t, reg, "p.example.com", "myapp", "production", false) |
| 2985 | |
| 2986 | // Pre-seed ratchet memory with a pending loosening whose old window has |
| 2987 | // long elapsed, and a v1 sighting past the pending 24h but inside the |
| 2988 | // enforced 72h. If the errored read were wrongly folded as nil, the |
| 2989 | // promotion would land and the memory would read 24h below. |
| 2990 | now := time.Now() |
| 2991 | st := LoadFirstSeen(FirstSeenPath(opt)) |
| 2992 | key := channelClockKey("bab", "myapp", "production") |
| 2993 | st.Holds[key] = &holdMemory{ |
| 2994 | Window: 72 * time.Hour, SetAt: now.Add(-100 * time.Hour), |
| 2995 | Pending: 24 * time.Hour, PendingAt: now.Add(-91 * time.Hour), |
| 2996 | } |
| 2997 | st.Channels[key] = map[string]time.Time{"v1": now.Add(-31 * time.Hour)} |
| 2998 | if err := st.Save(); err != nil { |
| 2999 | t.Fatalf("Save pre-seeded clock: %v", err) |
| 3000 | } |
| 3001 | |
| 3002 | if err := Pass(ctx, reg, s, pub, opt, func(string) error { return nil }); err != nil { |
| 3003 | t.Fatalf("Pass: %v", err) |
| 3004 | } |
| 3005 | if v, _ := LiveVersion(opt, "p.example.com"); v != "" { |
| 3006 | t.Fatalf("live = %q, want held — an unreadable pipeline must fall to legacy holds with the flag applying, never adopt free", v) |
| 3007 | } |
| 3008 | mem := LoadFirstSeen(FirstSeenPath(opt)).Holds[key] |
| 3009 | if mem == nil || mem.Window != 72*time.Hour || mem.PendingAt.IsZero() { |
| 3010 | t.Fatalf("hold memory after the errored read = %+v, want untouched (72h enforced, promotion still pending) — an errored read must never tick the ratchet", mem) |
| 3011 | } |
| 3012 | } |
| 3013 | |
| 3014 | // TestPassErroredPipelineReadKeepsTightenedWindows: the memory floor. A |
| 3015 | // verified pipeline tightened edge's window to 24h — and DefaultHolds gives |
| 3016 | // edge ZERO, so a plain fall-back to DefaultHolds on a later errored read |
| 3017 | // would adopt instantly: one transient GET or verify failure silently |
| 3018 | // defeating the exact window the bake exists to provide. Spec §3 is |
| 3019 | // binding: corruption must never loosen — on an errored read, every channel |
| 3020 | // the box's own ratchet remembers is enforced at max(its DefaultHolds |
| 3021 | // resolution, the remembered window). |
| 3022 | func TestPassErroredPipelineReadKeepsTightenedWindows(t *testing.T) { |
| 3023 | s := NewDirStore(t.TempDir()) |
| 3024 | ctx := context.Background() |
| 3025 | priv, pub := testKeys(t) |
| 3026 | opt := t.TempDir() |
| 3027 | reg := testRegistry(t) |
| 3028 | |
| 3029 | if _, err := SetPipeline(ctx, s, priv, "bab", "myapp", []PipelineChannel{ |
| 3030 | {Name: "edge", Bake: "24h"}, |
| 3031 | }, "test"); err != nil { |
| 3032 | t.Fatalf("SetPipeline: %v", err) |
| 3033 | } |
| 3034 | // The flag the console co-writes with any nonzero-bake pipeline. |
| 3035 | if err := SetProduction(ctx, s, priv, "bab", "myapp", true, "test"); err != nil { |
| 3036 | t.Fatalf("SetProduction: %v", err) |
| 3037 | } |
| 3038 | shipRelease(t, s, "bab", "myapp", "v1", adoptPlatform, []byte("bin")) |
| 3039 | if err := Promote(ctx, s, priv, "bab", "myapp", "edge", "v1"); err != nil { |
| 3040 | t.Fatalf("Promote: %v", err) |
| 3041 | } |
| 3042 | addRoute(t, reg, "e.example.com", "myapp", "edge", false) |
| 3043 | |
| 3044 | // Pass 1, pipeline verified: edge's zero baseline tightens to 24h |
| 3045 | // immediately (a tighten is never pending), and v1's clock starts — held. |
| 3046 | if err := Pass(ctx, reg, s, pub, opt, func(string) error { return nil }); err != nil { |
| 3047 | t.Fatalf("Pass: %v", err) |
| 3048 | } |
| 3049 | if v, _ := LiveVersion(opt, "e.example.com"); v != "" { |
| 3050 | t.Fatalf("live = %q, want held — the verified pipeline tightened edge to 24h", v) |
| 3051 | } |
| 3052 | if mem := LoadFirstSeen(FirstSeenPath(opt)).Holds[channelClockKey("bab", "myapp", "edge")]; mem == nil || mem.Window != 24*time.Hour { |
| 3053 | t.Fatalf("hold memory = %+v, want edge tightened to 24h", mem) |
| 3054 | } |
| 3055 | |
| 3056 | // Corrupt the object in place. The next read ERRORS (exists, won't |
| 3057 | // verify) — and must not hand edge back its zero legacy window. |
| 3058 | _, etag, err := s.Get(ctx, PipelineKey("bab", "myapp")) |
| 3059 | if err != nil { |
| 3060 | t.Fatalf("read pipeline etag: %v", err) |
| 3061 | } |
| 3062 | if _, err := s.Put(ctx, PipelineKey("bab", "myapp"), []byte("not a wrapper"), etag); err != nil { |
| 3063 | t.Fatalf("corrupt the pipeline object: %v", err) |
| 3064 | } |
| 3065 | if err := Pass(ctx, reg, s, pub, opt, func(string) error { return nil }); err != nil { |
| 3066 | t.Fatalf("Pass: %v", err) |
| 3067 | } |
| 3068 | if v, _ := LiveVersion(opt, "e.example.com"); v != "" { |
| 3069 | t.Fatalf("live = %q, want STILL held — an errored read must keep the ratcheted 24h floor on edge, never fall to its zero legacy window", v) |
| 3070 | } |
| 3071 | } |
| 3072 | |
| 3073 | // TestPassDeletedPipelineRatchetsDown: deleting pipeline.json is a plain |
| 3074 | // bucket write — no signing key — so verified absence must never collapse a |
| 3075 | // ratcheted window instantly. Deletion ratchets DOWN to the legacy defaults |
| 3076 | // instead: each remembered channel keeps its enforced window for one full |
| 3077 | // old-window cycle, then lands at HoldFor(DefaultHolds, name) and the app |
| 3078 | // resumes a legacy-shaped pass. |
| 3079 | func TestPassDeletedPipelineRatchetsDown(t *testing.T) { |
| 3080 | s := NewDirStore(t.TempDir()) |
| 3081 | ctx := context.Background() |
| 3082 | priv, pub := testKeys(t) |
| 3083 | opt := t.TempDir() |
| 3084 | reg := testRegistry(t) |
| 3085 | |
| 3086 | if _, err := SetPipeline(ctx, s, priv, "bab", "myapp", []PipelineChannel{ |
| 3087 | {Name: "edge", Bake: "24h"}, |
| 3088 | }, "test"); err != nil { |
| 3089 | t.Fatalf("SetPipeline: %v", err) |
| 3090 | } |
| 3091 | // The flag the console co-writes with any nonzero-bake pipeline. |
| 3092 | if err := SetProduction(ctx, s, priv, "bab", "myapp", true, "test"); err != nil { |
| 3093 | t.Fatalf("SetProduction: %v", err) |
| 3094 | } |
| 3095 | shipRelease(t, s, "bab", "myapp", "v1", adoptPlatform, []byte("bin")) |
| 3096 | if err := Promote(ctx, s, priv, "bab", "myapp", "edge", "v1"); err != nil { |
| 3097 | t.Fatalf("Promote: %v", err) |
| 3098 | } |
| 3099 | addRoute(t, reg, "e.example.com", "myapp", "edge", false) |
| 3100 | |
| 3101 | // Pass 1, pipeline verified: edge tightens to 24h immediately — held. |
| 3102 | if err := Pass(ctx, reg, s, pub, opt, func(string) error { return nil }); err != nil { |
| 3103 | t.Fatalf("Pass: %v", err) |
| 3104 | } |
| 3105 | if v, _ := LiveVersion(opt, "e.example.com"); v != "" { |
| 3106 | t.Fatalf("live = %q, want held — the verified pipeline tightened edge to 24h", v) |
| 3107 | } |
| 3108 | |
| 3109 | // The attack: delete the object outright. |
| 3110 | if err := s.Delete(ctx, PipelineKey("bab", "myapp")); err != nil { |
| 3111 | t.Fatalf("delete pipeline: %v", err) |
| 3112 | } |
| 3113 | |
| 3114 | // Pass 2: STILL held. Bare DefaultHolds would hand edge its zero legacy |
| 3115 | // window back and adopt v1 in this very pass. |
| 3116 | if err := Pass(ctx, reg, s, pub, opt, func(string) error { return nil }); err != nil { |
| 3117 | t.Fatalf("Pass: %v", err) |
| 3118 | } |
| 3119 | if v, _ := LiveVersion(opt, "e.example.com"); v != "" { |
| 3120 | t.Fatalf("live = %q, want STILL held — deletion must ratchet the 24h window down, never collapse it", v) |
| 3121 | } |
| 3122 | key := channelClockKey("bab", "myapp", "edge") |
| 3123 | st := LoadFirstSeen(FirstSeenPath(opt)) |
| 3124 | mem := st.Holds[key] |
| 3125 | if mem == nil || mem.Window != 24*time.Hour || mem.Pending != 0 || mem.PendingAt.IsZero() { |
| 3126 | t.Fatalf("hold memory = %+v, want 24h enforced with the legacy 0 pending", mem) |
| 3127 | } |
| 3128 | pendingAt := mem.PendingAt |
| 3129 | |
| 3130 | // Pass 3, absence re-observed: the pending clock must NOT restart. |
| 3131 | if err := Pass(ctx, reg, s, pub, opt, func(string) error { return nil }); err != nil { |
| 3132 | t.Fatalf("Pass: %v", err) |
| 3133 | } |
| 3134 | if mem := LoadFirstSeen(FirstSeenPath(opt)).Holds[key]; mem == nil || !mem.PendingAt.Equal(pendingAt) { |
| 3135 | t.Fatalf("re-observed absence restarted the pending clock: %+v, want PendingAt %v", mem, pendingAt) |
| 3136 | } |
| 3137 | |
| 3138 | // Elapse the OLD window on the box's own clock (24h jittered < 30h < |
| 3139 | // 31h). Next pass: the legacy default lands, edge is back at zero, and a |
| 3140 | // legacy-shaped pass adopts v1. |
| 3141 | st = LoadFirstSeen(FirstSeenPath(opt)) |
| 3142 | st.Holds[key].PendingAt = st.Holds[key].PendingAt.Add(-31 * time.Hour) |
| 3143 | if seen, ok := st.Channels[key]["v1"]; ok { |
| 3144 | st.Channels[key]["v1"] = seen.Add(-31 * time.Hour) |
| 3145 | } |
| 3146 | if err := st.Save(); err != nil { |
| 3147 | t.Fatalf("Save backdated clock: %v", err) |
| 3148 | } |
| 3149 | if err := Pass(ctx, reg, s, pub, opt, func(string) error { return nil }); err != nil { |
| 3150 | t.Fatalf("Pass: %v", err) |
| 3151 | } |
| 3152 | if v, _ := LiveVersion(opt, "e.example.com"); v != "v1" { |
| 3153 | t.Fatalf("live = %q, want v1 — the old window elapsed, so legacy holds (edge 0) resume", v) |
| 3154 | } |
| 3155 | if mem := LoadFirstSeen(FirstSeenPath(opt)).Holds[key]; mem == nil || mem.Window != 0 || !mem.PendingAt.IsZero() { |
| 3156 | t.Fatalf("persisted hold memory = %+v, want the landed legacy 0", mem) |
| 3157 | } |
| 3158 | } |
| 3159 | |
| 3160 | // ratchetedProductionApp arranges the case the whole app-defined-channel |
| 3161 | // posture turns on, and the one the edge-named tests below cannot reach: a |
| 3162 | // FLAGLESS app whose pipeline names a channel OUTSIDE the frozen vocabulary |
| 3163 | // ("production"), with a bake SHORTER than that name's legacy resolution. |
| 3164 | // HoldFor's fallback for an unrecognised name is the 72h fail-closed default, |
| 3165 | // so the declared 24h is a LOOSENING that ratchets in over the old window — |
| 3166 | // and once it has, the box enforces 24h while DefaultHolds still resolves the |
| 3167 | // same name at 72h. Any "is this app pipelined" test phrased as "does the |
| 3168 | // memory RAISE a window above legacy" answers NO here, which is exactly the |
| 3169 | // shape that made an app-defined channel the one place the verdict collapsed. |
| 3170 | // |
| 3171 | // Leaves: v1 adopted, the memory ratcheted in at 24h with nothing pending, |
| 3172 | // and v2 at the production head INSIDE that 24h window — held, on the |
| 3173 | // pipeline's word alone, with no production flag anywhere. |
| 3174 | func ratchetedProductionApp(t *testing.T, s Store, ctx context.Context, priv ed25519.PrivateKey, pub ed25519.PublicKey, opt string, reg *registry.Registry) { |
| 3175 | t.Helper() |
| 3176 | if _, err := SetPipeline(ctx, s, priv, "bab", "myapp", []PipelineChannel{ |
| 3177 | {Name: "edge"}, {Name: "production", Bake: "24h"}, |
| 3178 | }, "test"); err != nil { |
| 3179 | t.Fatalf("SetPipeline: %v", err) |
| 3180 | } |
| 3181 | shipRelease(t, s, "bab", "myapp", "v1", adoptPlatform, []byte("bin one")) |
| 3182 | if err := Promote(ctx, s, priv, "bab", "myapp", "edge", "v1"); err != nil { |
| 3183 | t.Fatalf("Promote edge: %v", err) |
| 3184 | } |
| 3185 | if err := Promote(ctx, s, priv, "bab", "myapp", "production", "v1"); err != nil { |
| 3186 | t.Fatalf("Promote production: %v", err) |
| 3187 | } |
| 3188 | addRoute(t, reg, "p.example.com", "myapp", "production", false) |
| 3189 | |
| 3190 | // Pass 1 latches the 72h fail-closed baseline and parks the declared 24h |
| 3191 | // as pending — held. |
| 3192 | if err := Pass(ctx, reg, s, pub, opt, func(string) error { return nil }); err != nil { |
| 3193 | t.Fatalf("Pass 1: %v", err) |
| 3194 | } |
| 3195 | if v, _ := LiveVersion(opt, "p.example.com"); v != "" { |
| 3196 | t.Fatalf("live = %q, want held at the 72h baseline", v) |
| 3197 | } |
| 3198 | // Elapse the OLD window on the box's own clock so the 24h lands, and age |
| 3199 | // v1 past 24h (but not past 72h) so adoption proves WHICH window governs. |
| 3200 | key := channelClockKey("bab", "myapp", "production") |
| 3201 | st := LoadFirstSeen(FirstSeenPath(opt)) |
| 3202 | mem := st.Holds[key] |
| 3203 | if mem == nil || mem.Window != 72*time.Hour || mem.Pending != 24*time.Hour { |
| 3204 | t.Fatalf("hold memory = %+v, want the 72h baseline with 24h pending", mem) |
| 3205 | } |
| 3206 | mem.PendingAt = mem.PendingAt.Add(-91 * time.Hour) |
| 3207 | if seen, ok := st.Channels[key]["v1"]; !ok { |
| 3208 | t.Fatalf("no first sighting recorded for v1") |
| 3209 | } else { |
| 3210 | st.Channels[key]["v1"] = seen.Add(-31 * time.Hour) |
| 3211 | } |
| 3212 | if err := st.Save(); err != nil { |
| 3213 | t.Fatalf("Save backdated clock: %v", err) |
| 3214 | } |
| 3215 | if err := Pass(ctx, reg, s, pub, opt, func(string) error { return nil }); err != nil { |
| 3216 | t.Fatalf("Pass 2: %v", err) |
| 3217 | } |
| 3218 | if v, _ := LiveVersion(opt, "p.example.com"); v != "v1" { |
| 3219 | t.Fatalf("live = %q, want v1 — the ratcheted-in 24h elapsed", v) |
| 3220 | } |
| 3221 | if mem := LoadFirstSeen(FirstSeenPath(opt)).Holds[key]; mem == nil || mem.Window != 24*time.Hour || !mem.PendingAt.IsZero() { |
| 3222 | t.Fatalf("hold memory = %+v, want the landed 24h with nothing pending", mem) |
| 3223 | } |
| 3224 | // A fresh head inside the ratcheted window: held while the pipeline reads. |
| 3225 | shipRelease(t, s, "bab", "myapp", "v2", adoptPlatform, []byte("bin two")) |
| 3226 | if err := Promote(ctx, s, priv, "bab", "myapp", "production", "v2"); err != nil { |
| 3227 | t.Fatalf("Promote v2: %v", err) |
| 3228 | } |
| 3229 | if err := Pass(ctx, reg, s, pub, opt, func(string) error { return nil }); err != nil { |
| 3230 | t.Fatalf("Pass 3: %v", err) |
| 3231 | } |
| 3232 | if v, _ := LiveVersion(opt, "p.example.com"); v != "v1" { |
| 3233 | t.Fatalf("live = %q, want still v1 — v2 is inside the ratcheted 24h window", v) |
| 3234 | } |
| 3235 | } |
| 3236 | |
| 3237 | // TestPassErroredReadOnAppDefinedChannelHoldsWithoutFlag is the C1 probe: the |
| 3238 | // same collapse the edge-named test below covers, on the naming that actually |
| 3239 | // hides it. With the memory ratcheted DOWN to 24h against a 72h legacy |
| 3240 | // resolution, no window in the map is raised above legacy — so a verdict |
| 3241 | // derived from the OVERLAY finds nothing, falls to the production flag this |
| 3242 | // app has never had, and adopts v2 the moment one GET of pipeline.json errors. |
| 3243 | // The evidence has to be the ratchet's own liveness, not the overlay. |
| 3244 | func TestPassErroredReadOnAppDefinedChannelHoldsWithoutFlag(t *testing.T) { |
| 3245 | s := NewDirStore(t.TempDir()) |
| 3246 | ctx := context.Background() |
| 3247 | priv, pub := testKeys(t) |
| 3248 | opt := t.TempDir() |
| 3249 | reg := testRegistry(t) |
| 3250 | ratchetedProductionApp(t, s, ctx, priv, pub, opt, reg) |
| 3251 | |
| 3252 | _, etag, err := s.Get(ctx, PipelineKey("bab", "myapp")) |
| 3253 | if err != nil { |
| 3254 | t.Fatalf("read pipeline etag: %v", err) |
| 3255 | } |
| 3256 | if _, err := s.Put(ctx, PipelineKey("bab", "myapp"), []byte("not a wrapper"), etag); err != nil { |
| 3257 | t.Fatalf("corrupt the pipeline object: %v", err) |
| 3258 | } |
| 3259 | if err := Pass(ctx, reg, s, pub, opt, func(string) error { return nil }); err != nil { |
| 3260 | t.Fatalf("Pass: %v", err) |
| 3261 | } |
| 3262 | if v, _ := LiveVersion(opt, "p.example.com"); v != "v1" { |
| 3263 | t.Fatalf("live = %q, want still v1 — an errored read on an app-defined channel must keep the hold, with no production flag in sight", v) |
| 3264 | } |
| 3265 | } |
| 3266 | |
| 3267 | // TestPassDeletedPipelineOnAppDefinedChannelHoldsWithoutFlag: the same probe |
| 3268 | // on the other keyless bucket write. Deleting pipeline.json hands the channel |
| 3269 | // its legacy 72h back — LONGER than the 24h the box was enforcing, so the |
| 3270 | // window itself never loosens — but the app must not fall to the flagless |
| 3271 | // legacy VERDICT in the same pass either, or one DELETE adopts a build that |
| 3272 | // was minutes old. It returns to the legacy path only once the deletion |
| 3273 | // ratchet has run the old window out on the box's own clock. |
| 3274 | func TestPassDeletedPipelineOnAppDefinedChannelHoldsWithoutFlag(t *testing.T) { |
| 3275 | s := NewDirStore(t.TempDir()) |
| 3276 | ctx := context.Background() |
| 3277 | priv, pub := testKeys(t) |
| 3278 | opt := t.TempDir() |
| 3279 | reg := testRegistry(t) |
| 3280 | ratchetedProductionApp(t, s, ctx, priv, pub, opt, reg) |
| 3281 | |
| 3282 | if err := s.Delete(ctx, PipelineKey("bab", "myapp")); err != nil { |
| 3283 | t.Fatalf("delete pipeline: %v", err) |
| 3284 | } |
| 3285 | if err := Pass(ctx, reg, s, pub, opt, func(string) error { return nil }); err != nil { |
| 3286 | t.Fatalf("Pass: %v", err) |
| 3287 | } |
| 3288 | if v, _ := LiveVersion(opt, "p.example.com"); v != "v1" { |
| 3289 | t.Fatalf("live = %q, want still v1 — deleting the pipeline must not adopt a held head in the same pass", v) |
| 3290 | } |
| 3291 | |
| 3292 | // And it does return to legacy: elapse the deletion ratchet's own cycle on |
| 3293 | // the box's clock, and the app is a legacy app again — no floor, no |
| 3294 | // pipeline evidence, the flagless legacy verdict, so v2 adopts. |
| 3295 | key := channelClockKey("bab", "myapp", "production") |
| 3296 | st := LoadFirstSeen(FirstSeenPath(opt)) |
| 3297 | mem := st.Holds[key] |
| 3298 | if mem == nil { |
| 3299 | t.Fatalf("no hold memory for the deleted pipeline's channel") |
| 3300 | } |
| 3301 | if mem.DeletedAt.IsZero() { |
| 3302 | t.Fatalf("hold memory = %+v, want the deletion stamped so its ratchet can run out", mem) |
| 3303 | } |
| 3304 | mem.DeletedAt = mem.DeletedAt.Add(-91 * time.Hour) |
| 3305 | st.Holds[key] = mem |
| 3306 | if err := st.Save(); err != nil { |
| 3307 | t.Fatalf("Save backdated clock: %v", err) |
| 3308 | } |
| 3309 | if err := Pass(ctx, reg, s, pub, opt, func(string) error { return nil }); err != nil { |
| 3310 | t.Fatalf("Pass: %v", err) |
| 3311 | } |
| 3312 | if v, _ := LiveVersion(opt, "p.example.com"); v != "v2" { |
| 3313 | t.Fatalf("live = %q, want v2 — once the deletion ratchet has run out the app is legacy again, and a flagless legacy app holds nothing", v) |
| 3314 | } |
| 3315 | } |
| 3316 | |
| 3317 | // TestPassReplayedSubsetPipelineCannotPoisonTheDeletionStamp: the re-review |
| 3318 | // probe (PR #258, IMPORTANT-1). The deletion stamp is what carries the bake |
| 3319 | // verdict across a DELETE for an app-defined channel, so anything that can |
| 3320 | // leave a STALE stamp standing — one that expires unnoticed and then blocks |
| 3321 | // its own re-arming — is a way to strip the app of every ratchet it has. |
| 3322 | // |
| 3323 | // The chain, all of it keyless bucket writes plus one historical SIGNED |
| 3324 | // object anybody who ever saw the bucket can keep a copy of: |
| 3325 | // |
| 3326 | // 1. DELETE pipeline.json — every remembered channel is stamped. |
| 3327 | // 2. PUT BACK an OLD, still-validly-signed pipeline that names only a SUBSET |
| 3328 | // of the app's channels (here: edge, not production). There is no |
| 3329 | // object-level replay floor — only a per-channel SetAt one — so this is |
| 3330 | // accepted as a verified read. |
| 3331 | // 3. Dwell until the un-named channel's stale stamp has run out. |
| 3332 | // 4. DELETE again. A stamp is never restamped while non-zero, so production |
| 3333 | // is left with window == legacy, nothing pending and a dead stamp: no |
| 3334 | // live ratchet, no evidence, and the flagless legacy verdict adopts a |
| 3335 | // build that is minutes old. |
| 3336 | // |
| 3337 | // The fix is that a verified read clears the stamp for EVERY remembered |
| 3338 | // channel of the app, not only the ones the object names — so step 2 can |
| 3339 | // only ever DISARM the machinery, never leave a fuse burning behind it, and |
| 3340 | // step 4 stamps afresh. |
| 3341 | func TestPassReplayedSubsetPipelineCannotPoisonTheDeletionStamp(t *testing.T) { |
| 3342 | s := NewDirStore(t.TempDir()) |
| 3343 | ctx := context.Background() |
| 3344 | priv, pub := testKeys(t) |
| 3345 | opt := t.TempDir() |
| 3346 | reg := testRegistry(t) |
| 3347 | |
| 3348 | // The historical object: a legitimately signed, edge-only pipeline from |
| 3349 | // before production existed. Kept aside, exactly as an attacker with a |
| 3350 | // copy of yesterday's bucket would. |
| 3351 | if _, err := SetPipeline(ctx, s, priv, "bab", "myapp", []PipelineChannel{{Name: "edge"}}, "test"); err != nil { |
| 3352 | t.Fatalf("SetPipeline (historical): %v", err) |
| 3353 | } |
| 3354 | historical, _, err := s.Get(ctx, PipelineKey("bab", "myapp")) |
| 3355 | if err != nil { |
| 3356 | t.Fatalf("read the historical pipeline: %v", err) |
| 3357 | } |
| 3358 | ratchetedProductionApp(t, s, ctx, priv, pub, opt, reg) |
| 3359 | |
| 3360 | // 1. Delete: production is stamped, and its held head stays held. |
| 3361 | if err := s.Delete(ctx, PipelineKey("bab", "myapp")); err != nil { |
| 3362 | t.Fatalf("delete pipeline: %v", err) |
| 3363 | } |
| 3364 | if err := Pass(ctx, reg, s, pub, opt, func(string) error { return nil }); err != nil { |
| 3365 | t.Fatalf("Pass after delete: %v", err) |
| 3366 | } |
| 3367 | if v, _ := LiveVersion(opt, "p.example.com"); v != "v1" { |
| 3368 | t.Fatalf("live = %q, want still v1 after the delete", v) |
| 3369 | } |
| 3370 | |
| 3371 | // 2. Replay the historical edge-only object (create-only PUT: the key is |
| 3372 | // absent after the delete, which is precisely the opening). |
| 3373 | if _, err := s.Put(ctx, PipelineKey("bab", "myapp"), historical, ""); err != nil { |
| 3374 | t.Fatalf("replay the historical pipeline: %v", err) |
| 3375 | } |
| 3376 | if err := Pass(ctx, reg, s, pub, opt, func(string) error { return nil }); err != nil { |
| 3377 | t.Fatalf("Pass after the replay: %v", err) |
| 3378 | } |
| 3379 | if v, _ := LiveVersion(opt, "p.example.com"); v != "v1" { |
| 3380 | t.Fatalf("live = %q, want still v1 — a verified pipeline is the app being pipelined, not a reason to adopt", v) |
| 3381 | } |
| 3382 | // The property that makes step 4 safe: the verified read left NO stamp |
| 3383 | // standing anywhere under the app, including on the channel it never |
| 3384 | // named. |
| 3385 | st := LoadFirstSeen(FirstSeenPath(opt)) |
| 3386 | for _, name := range []string{"edge", "production"} { |
| 3387 | if mem := st.Holds[channelClockKey("bab", "myapp", name)]; mem != nil && !mem.DeletedAt.IsZero() { |
| 3388 | t.Fatalf("%s keeps a deletion stamp (%v) after a verified read — a stale stamp can expire unnoticed and then refuse to re-arm", name, mem.DeletedAt) |
| 3389 | } |
| 3390 | } |
| 3391 | |
| 3392 | // 3. Dwell. Anything still stamped ages out; v2's own first sighting is |
| 3393 | // deliberately NOT aged, so the only thing that can adopt it below is the |
| 3394 | // verdict collapsing. |
| 3395 | st = LoadFirstSeen(FirstSeenPath(opt)) |
| 3396 | for _, name := range []string{"edge", "production"} { |
| 3397 | if mem := st.Holds[channelClockKey("bab", "myapp", name)]; mem != nil && !mem.DeletedAt.IsZero() { |
| 3398 | mem.DeletedAt = mem.DeletedAt.Add(-91 * time.Hour) |
| 3399 | } |
| 3400 | } |
| 3401 | if err := st.Save(); err != nil { |
| 3402 | t.Fatalf("Save dwelt clock: %v", err) |
| 3403 | } |
| 3404 | |
| 3405 | // 4. Delete again. The stamp re-arms, so the head is still held. |
| 3406 | if err := s.Delete(ctx, PipelineKey("bab", "myapp")); err != nil { |
| 3407 | t.Fatalf("second delete: %v", err) |
| 3408 | } |
| 3409 | if err := Pass(ctx, reg, s, pub, opt, func(string) error { return nil }); err != nil { |
| 3410 | t.Fatalf("Pass after the second delete: %v", err) |
| 3411 | } |
| 3412 | if v, _ := LiveVersion(opt, "p.example.com"); v != "v1" { |
| 3413 | t.Fatalf("live = %q, want STILL v1 — a replayed subset pipeline must not be able to strip the app of its deletion ratchet", v) |
| 3414 | } |
| 3415 | } |
| 3416 | |
| 3417 | // TestPassErroredPipelineHoldsWithoutProductionFlag: the memory floor is only |
| 3418 | // half a defence if the app's held head is then ADOPTED anyway. The bake |
| 3419 | // VERDICT ("does the window apply at all?") must ride on the same evidence |
| 3420 | // the WINDOWS do — the box's own ratchet memory — not on the production flag |
| 3421 | // the console co-writes. A co-write is a bucket write the box cannot verify |
| 3422 | // happened: an app pipelined before that co-write existed, an app whose flag |
| 3423 | // was deleted, a direct-bucket pipeline write — all leave a flagless app with |
| 3424 | // a live ratchet memory, and a legacy verdict would adopt its held head the |
| 3425 | // moment one GET of pipeline.json errored. NO production flag is set here. |
| 3426 | func TestPassErroredPipelineHoldsWithoutProductionFlag(t *testing.T) { |
| 3427 | s := NewDirStore(t.TempDir()) |
| 3428 | ctx := context.Background() |
| 3429 | priv, pub := testKeys(t) |
| 3430 | opt := t.TempDir() |
| 3431 | reg := testRegistry(t) |
| 3432 | |
| 3433 | if _, err := SetPipeline(ctx, s, priv, "bab", "myapp", []PipelineChannel{ |
| 3434 | {Name: "edge", Bake: "24h"}, |
| 3435 | }, "test"); err != nil { |
| 3436 | t.Fatalf("SetPipeline: %v", err) |
| 3437 | } |
| 3438 | shipRelease(t, s, "bab", "myapp", "v1", adoptPlatform, []byte("bin")) |
| 3439 | if err := Promote(ctx, s, priv, "bab", "myapp", "edge", "v1"); err != nil { |
| 3440 | t.Fatalf("Promote: %v", err) |
| 3441 | } |
| 3442 | addRoute(t, reg, "e.example.com", "myapp", "edge", false) |
| 3443 | |
| 3444 | // Pass 1, pipeline verified: edge's zero baseline tightens to 24h |
| 3445 | // immediately (a tighten is never pending) — held, flag or no flag. |
| 3446 | if err := Pass(ctx, reg, s, pub, opt, func(string) error { return nil }); err != nil { |
| 3447 | t.Fatalf("Pass: %v", err) |
| 3448 | } |
| 3449 | if v, _ := LiveVersion(opt, "e.example.com"); v != "" { |
| 3450 | t.Fatalf("live = %q, want held — a pipelined app's window applies with no production flag at all", v) |
| 3451 | } |
| 3452 | |
| 3453 | // Corrupt the object in place: the next read ERRORS. |
| 3454 | _, etag, err := s.Get(ctx, PipelineKey("bab", "myapp")) |
| 3455 | if err != nil { |
| 3456 | t.Fatalf("read pipeline etag: %v", err) |
| 3457 | } |
| 3458 | if _, err := s.Put(ctx, PipelineKey("bab", "myapp"), []byte("not a wrapper"), etag); err != nil { |
| 3459 | t.Fatalf("corrupt the pipeline object: %v", err) |
| 3460 | } |
| 3461 | if err := Pass(ctx, reg, s, pub, opt, func(string) error { return nil }); err != nil { |
| 3462 | t.Fatalf("Pass: %v", err) |
| 3463 | } |
| 3464 | if v, _ := LiveVersion(opt, "e.example.com"); v != "" { |
| 3465 | t.Fatalf("live = %q, want STILL held — the ratchet memory IS the evidence this app is pipelined; the verdict must not fall back to the production flag", v) |
| 3466 | } |
| 3467 | if mem := LoadFirstSeen(FirstSeenPath(opt)).Holds[channelClockKey("bab", "myapp", "edge")]; mem == nil || mem.Window != 24*time.Hour { |
| 3468 | t.Fatalf("hold memory = %+v, want the 24h floor still remembered — otherwise this test proves nothing", mem) |
| 3469 | } |
| 3470 | } |
| 3471 | |
| 3472 | // TestPassDeletedPipelineHoldsWithoutProductionFlag: same law on the |
| 3473 | // verified-absent path. Deleting pipeline.json is a plain bucket write; the |
| 3474 | // deletion ratchet keeps the 24h window enforced for one old-window cycle, |
| 3475 | // and the VERDICT must honour it without any production flag — otherwise |
| 3476 | // deleting one object adopts every held head of a flagless pipelined app in |
| 3477 | // the very next pass, which is precisely the collapse the ratchet exists to |
| 3478 | // prevent. |
| 3479 | func TestPassDeletedPipelineHoldsWithoutProductionFlag(t *testing.T) { |
| 3480 | s := NewDirStore(t.TempDir()) |
| 3481 | ctx := context.Background() |
| 3482 | priv, pub := testKeys(t) |
| 3483 | opt := t.TempDir() |
| 3484 | reg := testRegistry(t) |
| 3485 | |
| 3486 | if _, err := SetPipeline(ctx, s, priv, "bab", "myapp", []PipelineChannel{ |
| 3487 | {Name: "edge", Bake: "24h"}, |
| 3488 | }, "test"); err != nil { |
| 3489 | t.Fatalf("SetPipeline: %v", err) |
| 3490 | } |
| 3491 | shipRelease(t, s, "bab", "myapp", "v1", adoptPlatform, []byte("bin")) |
| 3492 | if err := Promote(ctx, s, priv, "bab", "myapp", "edge", "v1"); err != nil { |
| 3493 | t.Fatalf("Promote: %v", err) |
| 3494 | } |
| 3495 | addRoute(t, reg, "e.example.com", "myapp", "edge", false) |
| 3496 | |
| 3497 | if err := Pass(ctx, reg, s, pub, opt, func(string) error { return nil }); err != nil { |
| 3498 | t.Fatalf("Pass: %v", err) |
| 3499 | } |
| 3500 | if v, _ := LiveVersion(opt, "e.example.com"); v != "" { |
| 3501 | t.Fatalf("live = %q, want held — the verified pipeline tightened edge to 24h", v) |
| 3502 | } |
| 3503 | |
| 3504 | if err := s.Delete(ctx, PipelineKey("bab", "myapp")); err != nil { |
| 3505 | t.Fatalf("delete pipeline: %v", err) |
| 3506 | } |
| 3507 | if err := Pass(ctx, reg, s, pub, opt, func(string) error { return nil }); err != nil { |
| 3508 | t.Fatalf("Pass: %v", err) |
| 3509 | } |
| 3510 | if v, _ := LiveVersion(opt, "e.example.com"); v != "" { |
| 3511 | t.Fatalf("live = %q, want STILL held — deletion ratchets the window down over a cycle, and the verdict must ride on that memory, not on a production flag nobody set", v) |
| 3512 | } |
| 3513 | key := channelClockKey("bab", "myapp", "edge") |
| 3514 | if mem := LoadFirstSeen(FirstSeenPath(opt)).Holds[key]; mem == nil || mem.Window != 24*time.Hour || mem.PendingAt.IsZero() { |
| 3515 | t.Fatalf("hold memory = %+v, want 24h enforced with the legacy 0 pending", mem) |
| 3516 | } |
| 3517 | |
| 3518 | // And the ratchet still LANDS: once the old window has elapsed on the |
| 3519 | // box's own clock the app is legacy again — memory at the legacy default, |
| 3520 | // no floor map, and the bare-DefaultHolds verdict adopts. The fix must not |
| 3521 | // pin a once-pipelined app to a hold forever. |
| 3522 | st := LoadFirstSeen(FirstSeenPath(opt)) |
| 3523 | st.Holds[key].PendingAt = st.Holds[key].PendingAt.Add(-31 * time.Hour) |
| 3524 | if seen, ok := st.Channels[key]["v1"]; ok { |
| 3525 | st.Channels[key]["v1"] = seen.Add(-31 * time.Hour) |
| 3526 | } |
| 3527 | if err := st.Save(); err != nil { |
| 3528 | t.Fatalf("Save backdated clock: %v", err) |
| 3529 | } |
| 3530 | if err := Pass(ctx, reg, s, pub, opt, func(string) error { return nil }); err != nil { |
| 3531 | t.Fatalf("Pass: %v", err) |
| 3532 | } |
| 3533 | if v, _ := LiveVersion(opt, "e.example.com"); v != "v1" { |
| 3534 | t.Fatalf("live = %q, want v1 — the old window elapsed, so legacy holds (edge 0) resume and the legacy verdict adopts", v) |
| 3535 | } |
| 3536 | } |
| 3537 | |
| 3538 | // TestPassNoPipelineNoMemoryStaysLegacy pins the other half of the deletion |
| 3539 | // ratchet's contract: an app with NO pipeline and NO ratchet memory — every |
| 3540 | // legacy app on every pass — keeps the bare DefaultHolds path byte-identically. |
| 3541 | // Edge adopts in the first pass, and no holds memory is minted from nothing. |
| 3542 | func TestPassNoPipelineNoMemoryStaysLegacy(t *testing.T) { |
| 3543 | s := NewDirStore(t.TempDir()) |
| 3544 | ctx := context.Background() |
| 3545 | priv, pub := testKeys(t) |
| 3546 | opt := t.TempDir() |
| 3547 | reg := testRegistry(t) |
| 3548 | |
| 3549 | shipRelease(t, s, "bab", "myapp", "v1", adoptPlatform, []byte("bin")) |
| 3550 | if err := Promote(ctx, s, priv, "bab", "myapp", "edge", "v1"); err != nil { |
| 3551 | t.Fatalf("Promote: %v", err) |
| 3552 | } |
| 3553 | addRoute(t, reg, "e.example.com", "myapp", "edge", false) |
| 3554 | if err := Pass(ctx, reg, s, pub, opt, func(string) error { return nil }); err != nil { |
| 3555 | t.Fatalf("Pass: %v", err) |
| 3556 | } |
| 3557 | if v, _ := LiveVersion(opt, "e.example.com"); v != "v1" { |
| 3558 | t.Fatalf("live = %q, want v1 — a legacy app's edge adopts immediately, untouched", v) |
| 3559 | } |
| 3560 | if holds := LoadFirstSeen(FirstSeenPath(opt)).Holds; len(holds) != 0 { |
| 3561 | t.Fatalf("holds memory = %v, want none — no pipeline was ever observed", holds) |
| 3562 | } |
| 3563 | } |
| 3564 | |
| 3565 | // TestPassNullHoldsEntryDoesNotPanic: {"holds":{"x":null}} parses cleanly |
| 3566 | // into a nil *holdMemory, and the adopt pass's floor loops iterate the map. |
| 3567 | // An unguarded mem.Window dereference panics the whole pass — fleet blast |
| 3568 | // radius — on both the errored-read and the verified-absent path. |
| 3569 | func TestPassNullHoldsEntryDoesNotPanic(t *testing.T) { |
| 3570 | s := NewDirStore(t.TempDir()) |
| 3571 | ctx := context.Background() |
| 3572 | priv, pub := testKeys(t) |
| 3573 | opt := t.TempDir() |
| 3574 | reg := testRegistry(t) |
| 3575 | |
| 3576 | if _, err := SetPipeline(ctx, s, priv, "bab", "myapp", []PipelineChannel{ |
| 3577 | {Name: "edge", Bake: "24h"}, |
| 3578 | }, "test"); err != nil { |
| 3579 | t.Fatalf("SetPipeline: %v", err) |
| 3580 | } |
| 3581 | if err := SetProduction(ctx, s, priv, "bab", "myapp", true, "test"); err != nil { |
| 3582 | t.Fatalf("SetProduction: %v", err) |
| 3583 | } |
| 3584 | shipRelease(t, s, "bab", "myapp", "v1", adoptPlatform, []byte("bin")) |
| 3585 | if err := Promote(ctx, s, priv, "bab", "myapp", "edge", "v1"); err != nil { |
| 3586 | t.Fatalf("Promote: %v", err) |
| 3587 | } |
| 3588 | addRoute(t, reg, "e.example.com", "myapp", "edge", false) |
| 3589 | |
| 3590 | // A clock file whose holds map carries a null entry — legal JSON, loads |
| 3591 | // clean, nil pointer in the map. |
| 3592 | blob := `{"jitter":0,"channels":{},"holds":{"` + channelClockKey("bab", "myapp", "edge") + `":null}}` |
| 3593 | if err := os.WriteFile(FirstSeenPath(opt), []byte(blob), 0o644); err != nil { |
| 3594 | t.Fatal(err) |
| 3595 | } |
| 3596 | |
| 3597 | // Errored read: corrupt the pipeline object so the memory-floor loop runs. |
| 3598 | _, etag, err := s.Get(ctx, PipelineKey("bab", "myapp")) |
| 3599 | if err != nil { |
| 3600 | t.Fatalf("read pipeline etag: %v", err) |
| 3601 | } |
| 3602 | if _, err := s.Put(ctx, PipelineKey("bab", "myapp"), []byte("not a wrapper"), etag); err != nil { |
| 3603 | t.Fatalf("corrupt the pipeline object: %v", err) |
| 3604 | } |
| 3605 | if err := Pass(ctx, reg, s, pub, opt, func(string) error { return nil }); err != nil { |
| 3606 | t.Fatalf("Pass over a null holds entry (errored read): %v", err) |
| 3607 | } |
| 3608 | |
| 3609 | // Verified absence: same clock shape, object deleted outright. |
| 3610 | if err := os.WriteFile(FirstSeenPath(opt), []byte(blob), 0o644); err != nil { |
| 3611 | t.Fatal(err) |
| 3612 | } |
| 3613 | if err := s.Delete(ctx, PipelineKey("bab", "myapp")); err != nil { |
| 3614 | t.Fatalf("delete pipeline: %v", err) |
| 3615 | } |
| 3616 | if err := Pass(ctx, reg, s, pub, opt, func(string) error { return nil }); err != nil { |
| 3617 | t.Fatalf("Pass over a null holds entry (verified absent): %v", err) |
| 3618 | } |
| 3619 | } |
| 3620 | |
| 3621 | // The 2-3 SECOND canary fast loop must never pay a pipeline bucket read |
| 3622 | // either: canary/* is outside every pipeline, so no canary-only pass has a |
| 3623 | // verdict the object could change — the mirror of |
| 3624 | // TestCanaryFastLoopPaysNoProductionRead. The hourly pass pays exactly one |
| 3625 | // read per (account, app) with a non-canary head — not one per channel, not |
| 3626 | // one per route. |
| 3627 | func TestCanaryFastLoopPaysNoPipelineRead(t *testing.T) { |
| 3628 | dir := NewDirStore(t.TempDir()) |
| 3629 | ctx := context.Background() |
| 3630 | priv, pub := testKeys(t) |
| 3631 | opt := t.TempDir() |
| 3632 | reg := testRegistry(t) |
| 3633 | |
| 3634 | shipRelease(t, dir, "bab", "myapp", "v1", adoptPlatform, []byte("bin")) |
| 3635 | for _, ch := range []string{"canary/x", "canary/y", "edge", "stable"} { |
| 3636 | if err := Promote(ctx, dir, priv, "bab", "myapp", ch, "v1"); err != nil { |
| 3637 | t.Fatalf("Promote %s: %v", ch, err) |
| 3638 | } |
| 3639 | } |
| 3640 | addRoute(t, reg, "cx.example.com", "myapp", "canary/x", false) |
| 3641 | addRoute(t, reg, "cy.example.com", "myapp", "canary/y", false) |
| 3642 | addRoute(t, reg, "e.example.com", "myapp", "edge", false) |
| 3643 | addRoute(t, reg, "s.example.com", "myapp", "stable", false) |
| 3644 | |
| 3645 | s := &prodReadStore{Store: dir, key: PipelineKey("bab", "myapp")} |
| 3646 | if err := PassFilter(ctx, reg, s, pub, opt, func(string) error { return nil }, isCanaryTestRoute); err != nil { |
| 3647 | t.Fatalf("PassFilter: %v", err) |
| 3648 | } |
| 3649 | if s.n != 0 { |
| 3650 | t.Fatalf("canary-only pass read the pipeline %d times, want 0 — canary/* is outside every pipeline", s.n) |
| 3651 | } |
| 3652 | |
| 3653 | // Not vacuous: the full pass, with an edge and a stable head resolved, |
| 3654 | // reads it — once for the whole app. |
| 3655 | if err := Pass(ctx, reg, s, pub, opt, func(string) error { return nil }); err != nil { |
| 3656 | t.Fatalf("Pass: %v", err) |
| 3657 | } |
| 3658 | if s.n != 1 { |
| 3659 | t.Fatalf("full pass read the pipeline %d times, want exactly 1 per (account, app)", s.n) |
| 3660 | } |
| 3661 | } |
| 3662 | |
| 3663 | // ---- WithStartUnit: the rollback must not leave the service down ---- |
| 3664 | |
| 3665 | // startUnitFixture is the shared setup of the WithStartUnit tests: one |
| 3666 | // plain instance route live on "old" (staged locally, symlink pointed), with |
| 3667 | // "new" promoted at the channel head; canary/* so no bake window applies. |
| 3668 | // The route carries no Unit column — this package never reads it (the |
| 3669 | // named-unit shape lives in internal/host's dispatch). What these tests |
| 3670 | // model is the CALLBACK CONTRACT: a restart that fails the way a named |
| 3671 | // unit's does (returns the error), and a start capability handed in through |
| 3672 | // WithStartUnit. |
| 3673 | func startUnitFixture(t *testing.T, host string) (context.Context, *registry.Registry, Store, ed25519.PublicKey, string) { |
| 3674 | t.Helper() |
| 3675 | s := NewDirStore(t.TempDir()) |
| 3676 | ctx := context.Background() |
| 3677 | priv, pub := testKeys(t) |
| 3678 | opt := t.TempDir() |
| 3679 | reg := testRegistry(t) |
| 3680 | |
| 3681 | shipRelease(t, s, "bab", "console", "old", adoptPlatform, []byte("good build")) |
| 3682 | shipRelease(t, s, "bab", "console", "new", adoptPlatform, []byte("bad build")) |
| 3683 | if err := StageArtifact(opt, "bab", "old", "console", []byte("good build")); err != nil { |
| 3684 | t.Fatal(err) |
| 3685 | } |
| 3686 | addRoute(t, reg, host, "console", "canary/x", false) |
| 3687 | if err := PointLive(opt, host, "bab", "old", "console"); err != nil { |
| 3688 | t.Fatal(err) |
| 3689 | } |
| 3690 | if err := Promote(ctx, s, priv, "bab", "console", "canary/x", "new"); err != nil { |
| 3691 | t.Fatalf("Promote new: %v", err) |
| 3692 | } |
| 3693 | return ctx, reg, s, pub, opt |
| 3694 | } |
| 3695 | |
| 3696 | // TestFailedRestartLeavesTheUnitRUNNINGOnTheOldVersion. |
| 3697 | // |
| 3698 | // The rollback repointed the symlink and stopped there, which left the box |
| 3699 | // with a correct symlink and no process. Worse, try-restart is a NO-OP on a |
| 3700 | // unit that is not running — so the NEXT pass exited 0, stamped SetVersions, |
| 3701 | // and produced exactly the lying header the whole mechanism exists to |
| 3702 | // prevent, over a service that was not running at all. |
| 3703 | func TestFailedRestartLeavesTheUnitRUNNINGOnTheOldVersion(t *testing.T) { |
| 3704 | host := "console.example.com" |
| 3705 | ctx, reg, s, pub, opt := startUnitFixture(t, host) |
| 3706 | |
| 3707 | var started []string |
| 3708 | restart := func(string) error { return errors.New("Job for console.service failed") } |
| 3709 | start := func(h string) error { started = append(started, h); return nil } |
| 3710 | |
| 3711 | if err := Pass(ctx, reg, s, pub, opt, restart, WithStartUnit(start)); err != nil { |
| 3712 | t.Fatalf("Pass: %v", err) |
| 3713 | } |
| 3714 | |
| 3715 | if v, _ := LiveVersion(opt, host); v != "old" { |
| 3716 | t.Fatalf("symlink = %q, want it rolled back to old", v) |
| 3717 | } |
| 3718 | if len(started) != 1 || started[0] != host { |
| 3719 | t.Fatalf("started = %v, want the unit started once on the restored version — a rollback that leaves the service down is not a rollback", started) |
| 3720 | } |
| 3721 | rt, err := reg.Get(host) |
| 3722 | if err != nil || rt == nil { |
| 3723 | t.Fatalf("get %s: %+v, %v", host, rt, err) |
| 3724 | } |
| 3725 | if rt.Version == "new" { |
| 3726 | t.Fatal("SetVersions stamped the failed build") |
| 3727 | } |
| 3728 | } |
| 3729 | |
| 3730 | // TestSecondPassOverAStillBrokenBuildDoesNotStamp: the second-order failure. |
| 3731 | // With the unit running again after pass 1, pass 2's try-restart is a real |
| 3732 | // restart that fails again — so it rolls back again and still does not stamp. |
| 3733 | func TestSecondPassOverAStillBrokenBuildDoesNotStamp(t *testing.T) { |
| 3734 | host := "console.example.com" |
| 3735 | ctx, reg, s, pub, opt := startUnitFixture(t, host) |
| 3736 | |
| 3737 | // The fake is STATEFUL, and the state is the whole defect narrative: a |
| 3738 | // failed restart leaves the unit DOWN, and try-restart on a not-running |
| 3739 | // unit is an exit-0 NO-OP. With the fix reverted (no start on rollback) |
| 3740 | // pass 1 leaves running=false, pass 2's restart returns nil, and |
| 3741 | // SetVersions stamps "new" — this test's assertion fires. An |
| 3742 | // unconditionally-erroring fake could not fail on that revert. |
| 3743 | running := true |
| 3744 | restart := func(string) error { |
| 3745 | if !running { |
| 3746 | return nil // try-restart is a no-op on a stopped unit |
| 3747 | } |
| 3748 | running = false // the failed restart left it down |
| 3749 | return errors.New("still broken") |
| 3750 | } |
| 3751 | start := func(string) error { running = true; return nil } |
| 3752 | for i := 0; i < 2; i++ { |
| 3753 | if err := Pass(ctx, reg, s, pub, opt, restart, WithStartUnit(start)); err != nil { |
| 3754 | t.Fatalf("pass %d: %v", i, err) |
| 3755 | } |
| 3756 | } |
| 3757 | rt, err := reg.Get(host) |
| 3758 | if err != nil || rt == nil { |
| 3759 | t.Fatalf("get %s: %+v, %v", host, rt, err) |
| 3760 | } |
| 3761 | if rt.Version == "new" { |
| 3762 | t.Fatal("the second pass stamped a build that has never run") |
| 3763 | } |
| 3764 | if v, _ := LiveVersion(opt, host); v != "old" { |
| 3765 | t.Fatalf("symlink after two passes = %q, want old", v) |
| 3766 | } |
| 3767 | } |
| 3768 | |
| 3769 | // TestStartFailureStaysPoisonedAndSaysSo: when the start ALSO fails the |
| 3770 | // refusal stands exactly as before — poisoned, unstamped, rolled back — and |
| 3771 | // the failure is printed, because at that point the box is genuinely broken |
| 3772 | // and an operator reading only "rolled the symlink back" would take a host |
| 3773 | // that is serving nothing for a recovered one. |
| 3774 | func TestStartFailureStaysPoisonedAndSaysSo(t *testing.T) { |
| 3775 | host := "console.example.com" |
| 3776 | ctx, reg, s, pub, opt := startUnitFixture(t, host) |
| 3777 | |
| 3778 | restart := func(string) error { return errors.New("Job for console.service failed") } |
| 3779 | start := func(string) error { return errors.New("Job for console.service failed again") } |
| 3780 | |
| 3781 | var out bytes.Buffer |
| 3782 | withCapturedStdout(t, &out, func() { |
| 3783 | if err := Pass(ctx, reg, s, pub, opt, restart, WithStartUnit(start)); err != nil { |
| 3784 | t.Fatalf("Pass: %v", err) |
| 3785 | } |
| 3786 | }) |
| 3787 | |
| 3788 | if v, _ := LiveVersion(opt, host); v != "old" { |
| 3789 | t.Fatalf("symlink = %q, want old", v) |
| 3790 | } |
| 3791 | rt, err := reg.Get(host) |
| 3792 | if err != nil || rt == nil { |
| 3793 | t.Fatalf("get %s: %+v, %v", host, rt, err) |
| 3794 | } |
| 3795 | if rt.Version == "new" { |
| 3796 | t.Fatal("a build whose restart AND start both failed was stamped") |
| 3797 | } |
| 3798 | if !strings.Contains(out.String(), "could not start it") { |
| 3799 | t.Fatalf("printed %q, want the failed start reported", out.String()) |
| 3800 | } |
| 3801 | } |
| 3802 | |
| 3803 | // TestSuccessfulRestartNeverConsultsStart: the start capability belongs to |
| 3804 | // the ROLLBACK path alone. A restart that succeeds — the every-deploy case — |
| 3805 | // must not touch it, or every ordinary adoption would grow a systemctl start. |
| 3806 | func TestSuccessfulRestartNeverConsultsStart(t *testing.T) { |
| 3807 | host := "console.example.com" |
| 3808 | ctx, reg, s, pub, opt := startUnitFixture(t, host) |
| 3809 | |
| 3810 | restart := func(string) error { return nil } |
| 3811 | start := func(h string) error { |
| 3812 | t.Fatalf("start consulted for %q after a SUCCESSFUL restart", h) |
| 3813 | return nil |
| 3814 | } |
| 3815 | if err := Pass(ctx, reg, s, pub, opt, restart, WithStartUnit(start)); err != nil { |
| 3816 | t.Fatalf("Pass: %v", err) |
| 3817 | } |
| 3818 | if v, _ := LiveVersion(opt, host); v != "new" { |
| 3819 | t.Fatalf("live = %q, want new", v) |
| 3820 | } |
| 3821 | } |
| 3822 | |
| 3823 | // TestFailedRollbackStartsNothing: the start is gated on the rollback |
| 3824 | // having HAPPENED, not on a previous version existing. "Link present, |
| 3825 | // staged version gone" is a documented real state (materialize.go — an |
| 3826 | // operator's /opt cleanup, a partially restored volume): PointLive refuses |
| 3827 | // to repoint at a version directory that is not there, the symlink still |
| 3828 | // names the refused build, and a start would launch exactly that build. |
| 3829 | // The refusal is also no longer silent — the pass says the rollback failed. |
| 3830 | func TestFailedRollbackStartsNothing(t *testing.T) { |
| 3831 | host := "console.example.com" |
| 3832 | ctx, reg, s, pub, opt := startUnitFixture(t, host) |
| 3833 | |
| 3834 | // The operator's cleanup: old's staged directory is gone, so the |
| 3835 | // rollback's PointLive will refuse. The live symlink still dangles at |
| 3836 | // old, which is how adoptHost sees cur = "old" and attempts the update. |
| 3837 | if err := os.RemoveAll(filepath.Dir(VersionPath(opt, "bab", "old", "console"))); err != nil { |
| 3838 | t.Fatal(err) |
| 3839 | } |
| 3840 | |
| 3841 | restart := func(string) error { return errors.New("Job for console.service failed") } |
| 3842 | start := func(h string) error { |
| 3843 | t.Fatalf("start consulted for %q after a rollback that DID NOT HAPPEN — the symlink still names the refused build", h) |
| 3844 | return nil |
| 3845 | } |
| 3846 | var out bytes.Buffer |
| 3847 | withCapturedStdout(t, &out, func() { |
| 3848 | if err := Pass(ctx, reg, s, pub, opt, restart, WithStartUnit(start)); err != nil { |
| 3849 | t.Fatalf("Pass: %v", err) |
| 3850 | } |
| 3851 | }) |
| 3852 | if !strings.Contains(out.String(), "could not roll the symlink back") { |
| 3853 | t.Fatalf("printed %q, want the failed rollback reported instead of swallowed", out.String()) |
| 3854 | } |
| 3855 | rt, err := reg.Get(host) |
| 3856 | if err != nil || rt == nil { |
| 3857 | t.Fatalf("get %s: %+v, %v", host, rt, err) |
| 3858 | } |
| 3859 | if rt.Version == "new" { |
| 3860 | t.Fatal("a build whose restart failed was stamped") |
| 3861 | } |
| 3862 | } |
| 3863 | |
| 3864 | // TestFirstAdoptionRestartFailureStartsNothing: with no previous version |
| 3865 | // (cur == "") the symlink could not be rolled back — it still names the |
| 3866 | // version whose restart just failed — so there is no RESTORED version to |
| 3867 | // start. Starting there would run the exact build the failure refused. |
| 3868 | func TestFirstAdoptionRestartFailureStartsNothing(t *testing.T) { |
| 3869 | s := NewDirStore(t.TempDir()) |
| 3870 | ctx := context.Background() |
| 3871 | priv, pub := testKeys(t) |
| 3872 | opt := t.TempDir() |
| 3873 | reg := testRegistry(t) |
| 3874 | host := "console.example.com" |
| 3875 | |
| 3876 | shipRelease(t, s, "bab", "console", "new", adoptPlatform, []byte("bad build")) |
| 3877 | addRoute(t, reg, host, "console", "canary/x", false) |
| 3878 | if err := Promote(ctx, s, priv, "bab", "console", "canary/x", "new"); err != nil { |
| 3879 | t.Fatalf("Promote: %v", err) |
| 3880 | } |
| 3881 | |
| 3882 | restart := func(string) error { return errors.New("Job for console.service failed") } |
| 3883 | start := func(h string) error { |
| 3884 | t.Fatalf("start consulted for %q on a first adoption — nothing was rolled back, so there is nothing restored to start", h) |
| 3885 | return nil |
| 3886 | } |
| 3887 | if err := Pass(ctx, reg, s, pub, opt, restart, WithStartUnit(start)); err != nil { |
| 3888 | t.Fatalf("Pass: %v", err) |
| 3889 | } |
| 3890 | rt, err := reg.Get(host) |
| 3891 | if err != nil || rt == nil { |
| 3892 | t.Fatalf("get %s: %+v, %v", host, rt, err) |
| 3893 | } |
| 3894 | if rt.Version == "new" { |
| 3895 | t.Fatal("a build whose restart failed was stamped") |
| 3896 | } |
| 3897 | } |
| 3898 | |
| 3899 | // TestRefusedPromotionIsNeitherStagedNorRelayed: a refusal naming the current |
| 3900 | // pointer stops adoptHost before ensureStaged, so the agent never relays a |
| 3901 | // root pass for it and nothing restarts. |
| 3902 | func TestRefusedPromotionIsNeitherStagedNorRelayed(t *testing.T) { |
| 3903 | host := "console.example.com" |
| 3904 | ctx, reg, s, pub, opt := startUnitFixture(t, host) |
| 3905 | bake := t.TempDir() |
| 3906 | ptr, err := Channel(ctx, s, pub, "bab", "console", "canary/x") |
| 3907 | if err != nil { |
| 3908 | t.Fatal(err) |
| 3909 | } |
| 3910 | if err := WriteRefusal(bake, Refusal{Host: host, App: "console", Version: ptr.Version, |
| 3911 | PointerPromotedAt: ptr.PromotedAt, Reason: "exited: status=2"}); err != nil { |
| 3912 | t.Fatal(err) |
| 3913 | } |
| 3914 | restart := func(h string) error { t.Fatalf("restart consulted for refused %s", h); return nil } |
| 3915 | relayed := false |
| 3916 | var out bytes.Buffer |
| 3917 | withCapturedStdout(t, &out, func() { |
| 3918 | if err := Pass(ctx, reg, s, pub, opt, restart, BakeDir(bake), |
| 3919 | OnStageBlocked(func(string, string, error) { relayed = true })); err != nil { |
| 3920 | t.Fatalf("Pass: %v", err) |
| 3921 | } |
| 3922 | }) |
| 3923 | if v, _ := LiveVersion(opt, host); v != "old" { |
| 3924 | t.Fatalf("live = %q, want old", v) |
| 3925 | } |
| 3926 | if relayed { |
| 3927 | t.Fatal("a refused promotion was relayed to root") |
| 3928 | } |
| 3929 | // The ordering this gate exists for: the check sits BEFORE ensureStaged. |
| 3930 | // Staging succeeds in this fixture, so "relayed" alone cannot tell the |
| 3931 | // gate's position; the absence of the build and its manifest sibling can. |
| 3932 | for _, p := range []string{ |
| 3933 | VersionPath(opt, "bab", "new", "console"), |
| 3934 | StagedManifestPath(opt, "bab", "new", "console"), |
| 3935 | } { |
| 3936 | if _, err := os.Stat(p); !errors.Is(err, os.ErrNotExist) { |
| 3937 | t.Fatalf("refused version was staged: stat %s = %v, want not-exist", p, err) |
| 3938 | } |
| 3939 | } |
| 3940 | if !strings.Contains(out.String(), "promote --retry") { |
| 3941 | t.Fatalf("printed %q, want the retry hint", out.String()) |
| 3942 | } |
| 3943 | } |
| 3944 | |
| 3945 | // TestChannelBackOnTheRestoredBuildClearsTheRefusal (final review F8): v2 |
| 3946 | // was refused and rolled back to v1, then the channel was promoted back to |
| 3947 | // v1. The pass finds the box up to date, restarts nothing, and removes the |
| 3948 | // v2 record that no longer blocks anything. |
| 3949 | func TestChannelBackOnTheRestoredBuildClearsTheRefusal(t *testing.T) { |
| 3950 | host := "console.example.com" |
| 3951 | s := NewDirStore(t.TempDir()) |
| 3952 | ctx := context.Background() |
| 3953 | priv, pub := testKeys(t) |
| 3954 | opt := t.TempDir() |
| 3955 | reg := testRegistry(t) |
| 3956 | shipRelease(t, s, "bab", "console", "v1", adoptPlatform, []byte("good build")) |
| 3957 | shipRelease(t, s, "bab", "console", "v2", adoptPlatform, []byte("bad build")) |
| 3958 | if err := StageArtifact(opt, "bab", "v1", "console", []byte("good build")); err != nil { |
| 3959 | t.Fatal(err) |
| 3960 | } |
| 3961 | addRoute(t, reg, host, "console", "canary/x", false) |
| 3962 | if err := PointLive(opt, host, "bab", "v1", "console"); err != nil { |
| 3963 | t.Fatal(err) |
| 3964 | } |
| 3965 | if err := Promote(ctx, s, priv, "bab", "console", "canary/x", "v1"); err != nil { |
| 3966 | t.Fatal(err) |
| 3967 | } |
| 3968 | bake := t.TempDir() |
| 3969 | if err := WriteRefusal(bake, Refusal{Host: host, App: "console", Version: "v2", |
| 3970 | PointerPromotedAt: time.Unix(1, 0).UTC(), Reason: "exited: status=2", RestoredVersion: "v1", Recovered: true}); err != nil { |
| 3971 | t.Fatal(err) |
| 3972 | } |
| 3973 | restart := func(h string) error { t.Fatalf("restart consulted for up-to-date %s", h); return nil } |
| 3974 | if err := Pass(ctx, reg, s, pub, opt, restart, BakeDir(bake)); err != nil { |
| 3975 | t.Fatalf("Pass: %v", err) |
| 3976 | } |
| 3977 | if v, _ := LiveVersion(opt, host); v != "v1" { |
| 3978 | t.Fatalf("live = %q, want v1", v) |
| 3979 | } |
| 3980 | if r, _ := ReadRefusal(bake, host); r != nil { |
| 3981 | t.Fatalf("stale v2 refusal survived the channel's return to v1: %+v", r) |
| 3982 | } |
| 3983 | } |
| 3984 | |
| 3985 | // TestUpToDateKeepsARefusalThatStillBlocks: the link is on the refused build |
| 3986 | // only when its rollback failed; that record still names the pointer and |
| 3987 | // stays for the watcher. |
| 3988 | func TestUpToDateKeepsARefusalThatStillBlocks(t *testing.T) { |
| 3989 | host := "console.example.com" |
| 3990 | ctx, reg, s, pub, opt := startUnitFixture(t, host) |
| 3991 | if err := StageArtifact(opt, "bab", "new", "console", []byte("bad build")); err != nil { |
| 3992 | t.Fatal(err) |
| 3993 | } |
| 3994 | if err := PointLive(opt, host, "bab", "new", "console"); err != nil { |
| 3995 | t.Fatal(err) |
| 3996 | } |
| 3997 | ptr, err := Channel(ctx, s, pub, "bab", "console", "canary/x") |
| 3998 | if err != nil { |
| 3999 | t.Fatal(err) |
| 4000 | } |
| 4001 | bake := t.TempDir() |
| 4002 | if err := WriteRefusal(bake, Refusal{Host: host, App: "console", Version: ptr.Version, |
| 4003 | PointerPromotedAt: ptr.PromotedAt, Reason: "exited: status=2", RecoveryError: "could not roll the symlink back"}); err != nil { |
| 4004 | t.Fatal(err) |
| 4005 | } |
| 4006 | if err := Pass(ctx, reg, s, pub, opt, func(string) error { return nil }, BakeDir(bake)); err != nil { |
| 4007 | t.Fatalf("Pass: %v", err) |
| 4008 | } |
| 4009 | if r, _ := ReadRefusal(bake, host); r == nil { |
| 4010 | t.Fatal("a refusal that still blocks the pointer was removed") |
| 4011 | } |
| 4012 | } |
| 4013 | |
| 4014 | // TestRePromotionClearsTheBlock: a pointer re-signed for the same version |
| 4015 | // (what promote --retry does) is adopted, and the stale record is removed. |
| 4016 | func TestRePromotionClearsTheBlock(t *testing.T) { |
| 4017 | host := "console.example.com" |
| 4018 | ctx, reg, s, pub, opt := startUnitFixture(t, host) |
| 4019 | bake := t.TempDir() |
| 4020 | if err := WriteRefusal(bake, Refusal{Host: host, App: "console", Version: "new", |
| 4021 | PointerPromotedAt: time.Unix(1, 0).UTC(), Reason: "older promotion"}); err != nil { |
| 4022 | t.Fatal(err) |
| 4023 | } |
| 4024 | if err := Pass(ctx, reg, s, pub, opt, func(string) error { return nil }, BakeDir(bake)); err != nil { |
| 4025 | t.Fatalf("Pass: %v", err) |
| 4026 | } |
| 4027 | if v, _ := LiveVersion(opt, host); v != "new" { |
| 4028 | t.Fatalf("live = %q, want new (the record names an older promotion)", v) |
| 4029 | } |
| 4030 | if r, _ := ReadRefusal(bake, host); r != nil { |
| 4031 | t.Fatalf("stale refusal survived a successful adoption: %+v", r) |
| 4032 | } |
| 4033 | } |
| 4034 | |
amadan