| 1 | package ops |
| 2 | |
| 3 | import ( |
| 4 | "database/sql" |
| 5 | "os" |
| 6 | "path/filepath" |
| 7 | "slices" |
| 8 | "strings" |
| 9 | "testing" |
| 10 | "time" |
| 11 | |
| 12 | "github.com/carlosframework/platform/internal/provision" |
| 13 | "github.com/carlosframework/platform/internal/registry" |
| 14 | |
| 15 | _ "modernc.org/sqlite" |
| 16 | ) |
| 17 | |
| 18 | // TestConfigGenSkipsHibernatingRoutes pins Eleven's two-replicators lesson in |
| 19 | // executable form: a route with Hibernate=true must never appear in the fleet |
| 20 | // litestream config, because the activator's embedded Litestream owns that |
| 21 | // database's replication — a second replicator on the same WAL would fight it |
| 22 | // for checkpointing. |
| 23 | func TestConfigGenSkipsHibernatingRoutes(t *testing.T) { |
| 24 | dir := t.TempDir() |
| 25 | reg, err := registry.Open(filepath.Join(dir, "registry.db")) |
| 26 | if err != nil { |
| 27 | t.Fatal(err) |
| 28 | } |
| 29 | |
| 30 | // A unit tenant's data lives OUTSIDE dataDir entirely (its systemd |
| 31 | // StateDirectory, e.g. /var/lib/carlos-app/<host>) — a separate temp dir |
| 32 | // here stands in for that, deliberately not under dir, so this test |
| 33 | // actually exercises FleetDBs' DBPath-based discovery instead of |
| 34 | // accidentally passing because the file happened to sit where the glob |
| 35 | // already looks. |
| 36 | unitDir := t.TempDir() |
| 37 | |
| 38 | awake := filepath.Join(dir, "a.example.com.db") |
| 39 | asleep := filepath.Join(dir, "b.example.com.db") |
| 40 | unitTenant := filepath.Join(unitDir, "state.db") |
| 41 | for _, p := range []string{awake, asleep, unitTenant} { |
| 42 | if err := os.WriteFile(p, []byte("x"), 0o644); err != nil { |
| 43 | t.Fatal(err) |
| 44 | } |
| 45 | } |
| 46 | if err := reg.Upsert(registry.Route{Host: "a.example.com", DBPath: awake}); err != nil { |
| 47 | t.Fatal(err) |
| 48 | } |
| 49 | if err := reg.Upsert(registry.Route{Host: "b.example.com", DBPath: asleep, Hibernate: true}); err != nil { |
| 50 | t.Fatal(err) |
| 51 | } |
| 52 | // A unit-backed hibernating tenant (carlos-app@) with an operator-set |
| 53 | // -db: its wake/sleep is systemctl and the activator never replicates |
| 54 | // it, so this box-wide replicator must own its database — found via |
| 55 | // DBPath, not the dataDir glob (its file isn't under dataDir at all). |
| 56 | if err := reg.Upsert(registry.Route{Host: "c.example.com", DBPath: unitTenant, |
| 57 | Hibernate: true, Backing: registry.BackingUnit}); err != nil { |
| 58 | t.Fatal(err) |
| 59 | } |
| 60 | |
| 61 | dbs, err := FleetDBs(reg, dir) |
| 62 | if err != nil { |
| 63 | t.Fatal(err) |
| 64 | } |
| 65 | cfg := LitestreamConfig(dbs, "bucket-x", "fleet", "eu-west-1") |
| 66 | if !strings.Contains(cfg, awake) { |
| 67 | t.Fatalf("awake route's DB missing from config:\n%s", cfg) |
| 68 | } |
| 69 | if strings.Contains(cfg, asleep) { |
| 70 | t.Fatalf("hibernating route's DB must be excluded (activator owns its replication):\n%s", cfg) |
| 71 | } |
| 72 | if !strings.Contains(cfg, unitTenant) { |
| 73 | t.Fatalf("unit-backed hibernating route's DB (outside dataDir) missing (nobody else replicates it):\n%s", cfg) |
| 74 | } |
| 75 | // The registry itself lives in the data dir and is not anyone's hibernating |
| 76 | // instance — it must be replicated. |
| 77 | if !strings.Contains(cfg, filepath.Join(dir, "registry.db")) { |
| 78 | t.Fatalf("registry.db missing from config:\n%s", cfg) |
| 79 | } |
| 80 | if !strings.Contains(cfg, "GENERATED") { |
| 81 | t.Fatalf("generated-file warning missing:\n%s", cfg) |
| 82 | } |
| 83 | if !strings.Contains(cfg, "region: eu-west-1") { |
| 84 | t.Fatalf("replica region missing:\n%s", cfg) |
| 85 | } |
| 86 | } |
| 87 | |
| 88 | // TestFleetDBsDiscoversNestedDatabases pins the isolated-exec companion to |
| 89 | // the flat glob above (design §7): a route's DBPath being "claimed" only |
| 90 | // FILTERS what the glob already found. For a non-unit route it never adds |
| 91 | // a path by itself, so a nested row's database, under |
| 92 | // <dataDir>/instances/<id>/<host>.db, would stay invisible to replication |
| 93 | // forever without the nested glob to find it in the first place. |
| 94 | func TestFleetDBsDiscoversNestedDatabases(t *testing.T) { |
| 95 | dir := t.TempDir() |
| 96 | reg, err := registry.Open(filepath.Join(dir, "registry.db")) |
| 97 | if err != nil { |
| 98 | t.Fatal(err) |
| 99 | } |
| 100 | host := "nested.example.com" |
| 101 | nested := provision.NestedDBPath(dir, host) |
| 102 | if err := os.MkdirAll(provision.InstanceDir(dir, host), 0o755); err != nil { |
| 103 | t.Fatal(err) |
| 104 | } |
| 105 | if err := os.WriteFile(nested, []byte("x"), 0o644); err != nil { |
| 106 | t.Fatal(err) |
| 107 | } |
| 108 | // Not hibernating: a resident, activator-managed nested row is exactly |
| 109 | // as replicable by this box-wide pass as a flat one. |
| 110 | if err := reg.Upsert(registry.Route{Host: host, DBPath: nested, Socket: provision.NestedSocket(host)}); err != nil { |
| 111 | t.Fatal(err) |
| 112 | } |
| 113 | |
| 114 | dbs, err := FleetDBs(reg, dir) |
| 115 | if err != nil { |
| 116 | t.Fatal(err) |
| 117 | } |
| 118 | if !slices.Contains(dbs, nested) { |
| 119 | t.Fatalf("nested database missing from FleetDBs: %v, want %s", dbs, nested) |
| 120 | } |
| 121 | } |
| 122 | |
| 123 | // TestFleetDBsUnitTenantWithoutDBPathIsInvisible pins the honest half of the |
| 124 | // I1 fix: a BackingUnit route added without -db has no discoverable database |
| 125 | // path (the platform enforces no filename convention inside a unit tenant's |
| 126 | // state dir), so FleetDBs must not guess one — it must simply not appear in |
| 127 | // the generated config. This is the documented gap operators must close by |
| 128 | // hand (or cli.go's -hibernate requirement closes it for them); it must |
| 129 | // never masquerade as covered. |
| 130 | func TestFleetDBsUnitTenantWithoutDBPathIsInvisible(t *testing.T) { |
| 131 | dir := t.TempDir() |
| 132 | reg, err := registry.Open(filepath.Join(dir, "registry.db")) |
| 133 | if err != nil { |
| 134 | t.Fatal(err) |
| 135 | } |
| 136 | if err := reg.Upsert(registry.Route{Host: "no-db.example.com", Backing: registry.BackingUnit}); err != nil { |
| 137 | t.Fatal(err) |
| 138 | } |
| 139 | |
| 140 | dbs, err := FleetDBs(reg, dir) |
| 141 | if err != nil { |
| 142 | t.Fatal(err) |
| 143 | } |
| 144 | for _, p := range dbs { |
| 145 | if strings.Contains(p, "no-db") { |
| 146 | t.Fatalf("FleetDBs invented a path for a unit tenant with no --db: %v", dbs) |
| 147 | } |
| 148 | } |
| 149 | } |
| 150 | |
| 151 | // TestFleetDBsDedupesUnitPathUnderDataDir: a unit tenant's DBPath can, in |
| 152 | // principle, sit inside dataDir (nothing forbids it) — FleetDBs must not |
| 153 | // list it twice just because both the glob and the DBPath pass found it. |
| 154 | func TestFleetDBsDedupesUnitPathUnderDataDir(t *testing.T) { |
| 155 | dir := t.TempDir() |
| 156 | reg, err := registry.Open(filepath.Join(dir, "registry.db")) |
| 157 | if err != nil { |
| 158 | t.Fatal(err) |
| 159 | } |
| 160 | p := filepath.Join(dir, "d.example.com.db") |
| 161 | if err := os.WriteFile(p, []byte("x"), 0o644); err != nil { |
| 162 | t.Fatal(err) |
| 163 | } |
| 164 | if err := reg.Upsert(registry.Route{Host: "d.example.com", DBPath: p, Backing: registry.BackingUnit}); err != nil { |
| 165 | t.Fatal(err) |
| 166 | } |
| 167 | |
| 168 | dbs, err := FleetDBs(reg, dir) |
| 169 | if err != nil { |
| 170 | t.Fatal(err) |
| 171 | } |
| 172 | count := 0 |
| 173 | for _, got := range dbs { |
| 174 | if got == p { |
| 175 | count++ |
| 176 | } |
| 177 | } |
| 178 | if count != 1 { |
| 179 | t.Fatalf("FleetDBs listed %s %d times, want exactly once: %v", p, count, dbs) |
| 180 | } |
| 181 | } |
| 182 | |
| 183 | // The round-trip is the contract that lets restore-verify read the generated |
| 184 | // config without a YAML parser: whatever LitestreamConfig writes, |
| 185 | // ParseLitestreamDBs must recover exactly. |
| 186 | func TestConfigRoundTrip(t *testing.T) { |
| 187 | in := []string{"/data/registry.db", "/data/a.example.com.db", "/data/b.example.com.db"} |
| 188 | got := ParseLitestreamDBs(LitestreamConfig(in, "b", "p", "eu-west-1")) |
| 189 | if len(got) != len(in) { |
| 190 | t.Fatalf("round-trip lost databases: in %d, out %d (%v)", len(in), len(got), got) |
| 191 | } |
| 192 | want := map[string]bool{} |
| 193 | for _, p := range in { |
| 194 | want[p] = true |
| 195 | } |
| 196 | for _, p := range got { |
| 197 | if !want[p] { |
| 198 | t.Fatalf("round-trip invented path %q", p) |
| 199 | } |
| 200 | } |
| 201 | } |
| 202 | |
| 203 | func TestParseLitestreamDBsIgnoresNoise(t *testing.T) { |
| 204 | got := ParseLitestreamDBs("# comment\ndbs:\n - path: /data/x.db\n replicas:\n - url: s3://b/p/x.db\n region: eu-west-1\n\n") |
| 205 | if len(got) != 1 || got[0] != "/data/x.db" { |
| 206 | t.Fatalf("ParseLitestreamDBs = %#v, want [/data/x.db]", got) |
| 207 | } |
| 208 | if got := ParseLitestreamDBs(""); len(got) != 0 { |
| 209 | t.Fatalf("ParseLitestreamDBs(\"\") = %#v, want none", got) |
| 210 | } |
| 211 | } |
| 212 | |
| 213 | // Re-running the generator over an unchanged fleet must produce a |
| 214 | // byte-identical file — that is what lets the `litestream` verb report |
| 215 | // "unchanged" and skip a restart-worthy config write. |
| 216 | func TestConfigChangeDetection(t *testing.T) { |
| 217 | dbs := []string{"/data/b.db", "/data/a.db", "/data/registry.db"} |
| 218 | first := LitestreamConfig(dbs, "bucket", "prefix", "eu-west-1") |
| 219 | second := LitestreamConfig([]string{"/data/registry.db", "/data/a.db", "/data/b.db"}, "bucket", "prefix", "eu-west-1") |
| 220 | if first != second { |
| 221 | t.Fatalf("same fleet produced different configs:\n--- first\n%s\n--- second\n%s", first, second) |
| 222 | } |
| 223 | } |
| 224 | |
| 225 | func TestCheckRestoredDBRejectsEmptySchema(t *testing.T) { |
| 226 | dir := t.TempDir() |
| 227 | |
| 228 | good := filepath.Join(dir, "good.db") |
| 229 | db, err := sql.Open("sqlite", good) |
| 230 | if err != nil { |
| 231 | t.Fatal(err) |
| 232 | } |
| 233 | if _, err := db.Exec("CREATE TABLE t (id INTEGER PRIMARY KEY); INSERT INTO t VALUES (1)"); err != nil { |
| 234 | t.Fatal(err) |
| 235 | } |
| 236 | db.Close() |
| 237 | if err := CheckRestoredDB(good); err != nil { |
| 238 | t.Fatalf("CheckRestoredDB(good) = %v, want nil", err) |
| 239 | } |
| 240 | |
| 241 | // A schema-less database restores "successfully" but would be useless in a |
| 242 | // disaster — the check must refuse it. |
| 243 | empty := filepath.Join(dir, "empty.db") |
| 244 | db, err = sql.Open("sqlite", empty) |
| 245 | if err != nil { |
| 246 | t.Fatal(err) |
| 247 | } |
| 248 | if _, err := db.Exec("PRAGMA user_version=1"); err != nil { // force file creation |
| 249 | t.Fatal(err) |
| 250 | } |
| 251 | db.Close() |
| 252 | if err := CheckRestoredDB(empty); err == nil { |
| 253 | t.Fatal("CheckRestoredDB accepted a database with no tables") |
| 254 | } |
| 255 | |
| 256 | // Garbage bytes must fail loudly, not pass as an open-able file. |
| 257 | junk := filepath.Join(dir, "junk.db") |
| 258 | if err := os.WriteFile(junk, []byte("this is not a sqlite database at all"), 0o644); err != nil { |
| 259 | t.Fatal(err) |
| 260 | } |
| 261 | if err := CheckRestoredDB(junk); err == nil { |
| 262 | t.Fatal("CheckRestoredDB accepted a non-SQLite file") |
| 263 | } |
| 264 | } |
| 265 | |
| 266 | // The console's own SQLite — accounts, ownership, sessions, the deployment's |
| 267 | // Keymail credentials — lives one directory BELOW dataDir, where FleetDBs' |
| 268 | // flat <data>/*.db glob cannot see it. Losing it loses who everyone is, so it |
| 269 | // is included explicitly. |
| 270 | func TestFleetDBsIncludesConsoleDB(t *testing.T) { |
| 271 | dir := t.TempDir() |
| 272 | reg, err := registry.Open(filepath.Join(dir, "registry.db")) |
| 273 | if err != nil { |
| 274 | t.Fatal(err) |
| 275 | } |
| 276 | |
| 277 | // Absent console: nothing invented, and no error. |
| 278 | dbs, err := FleetDBs(reg, dir) |
| 279 | if err != nil { |
| 280 | t.Fatal(err) |
| 281 | } |
| 282 | for _, p := range dbs { |
| 283 | if strings.Contains(p, "console") { |
| 284 | t.Fatalf("FleetDBs listed a console DB on a box that has none: %v", dbs) |
| 285 | } |
| 286 | } |
| 287 | |
| 288 | consoleDB := filepath.Join(dir, "console", "console.db") |
| 289 | if err := os.MkdirAll(filepath.Dir(consoleDB), 0o750); err != nil { |
| 290 | t.Fatal(err) |
| 291 | } |
| 292 | if err := os.WriteFile(consoleDB, []byte("x"), 0o600); err != nil { |
| 293 | t.Fatal(err) |
| 294 | } |
| 295 | |
| 296 | dbs, err = FleetDBs(reg, dir) |
| 297 | if err != nil { |
| 298 | t.Fatal(err) |
| 299 | } |
| 300 | if !slices.Contains(dbs, consoleDB) { |
| 301 | t.Fatalf("console DB missing from the replication set: %v", dbs) |
| 302 | } |
| 303 | } |
| 304 | |
| 305 | // CONSOLE_DB is the console's own env var for its database; a box that moved |
| 306 | // it must not silently drop out of replication. |
| 307 | func TestFleetDBsHonoursConsoleDBEnv(t *testing.T) { |
| 308 | dir := t.TempDir() |
| 309 | reg, err := registry.Open(filepath.Join(dir, "registry.db")) |
| 310 | if err != nil { |
| 311 | t.Fatal(err) |
| 312 | } |
| 313 | elsewhere := filepath.Join(t.TempDir(), "moved-console.db") |
| 314 | if err := os.WriteFile(elsewhere, []byte("x"), 0o600); err != nil { |
| 315 | t.Fatal(err) |
| 316 | } |
| 317 | t.Setenv("CONSOLE_DB", elsewhere) |
| 318 | |
| 319 | dbs, err := FleetDBs(reg, dir) |
| 320 | if err != nil { |
| 321 | t.Fatal(err) |
| 322 | } |
| 323 | if !slices.Contains(dbs, elsewhere) { |
| 324 | t.Fatalf("relocated console DB missing from the replication set: %v", dbs) |
| 325 | } |
| 326 | } |
| 327 | |
| 328 | // The generated config must state the recovery window rather than inherit |
| 329 | // litestream's 24h snapshot retention by silence. A day is the shortest |
| 330 | // useful window there is: a bad migration noticed on Tuesday is unrecoverable |
| 331 | // if it landed on Sunday. |
| 332 | func TestConfigPinsSnapshotRetention(t *testing.T) { |
| 333 | cfg := LitestreamConfig([]string{"/data/registry.db"}, "b", "p", "eu-west-1") |
| 334 | if !strings.Contains(cfg, "snapshot:") { |
| 335 | t.Fatalf("generated config states no snapshot policy — it inherits a 24h window silently:\n%s", cfg) |
| 336 | } |
| 337 | if !strings.Contains(cfg, "retention: "+SnapshotRetention.String()) { |
| 338 | t.Fatalf("generated config does not pin retention to %s:\n%s", SnapshotRetention, cfg) |
| 339 | } |
| 340 | if SnapshotRetention <= 24*time.Hour { |
| 341 | t.Fatalf("SnapshotRetention = %s, which is no better than litestream's default", SnapshotRetention) |
| 342 | } |
| 343 | // And the round-trip the restore-verify verb depends on must survive it. |
| 344 | if got := ParseLitestreamDBs(cfg); len(got) != 1 || got[0] != "/data/registry.db" { |
| 345 | t.Fatalf("ParseLitestreamDBs broke on the retention stanza: %v", got) |
| 346 | } |
| 347 | } |
| 348 | |
| 349 | // A config that STATES a snapshot policy the installed binary silently |
| 350 | // ignores is worse than one that states nothing: it reads as a decision |
| 351 | // someone made. litestream 0.3.13 (still on the flagship, verified 2026-08-15) |
| 352 | // parses the stanza and drops it on the floor, and 0.5 accepts unknown keys |
| 353 | // just as quietly — so parsing proves nothing and the VERSION is the only |
| 354 | // signal there is. |
| 355 | func TestSnapshotPolicySupported(t *testing.T) { |
| 356 | cases := []struct { |
| 357 | out string |
| 358 | want bool |
| 359 | }{ |
| 360 | {"v0.3.13", false}, // the flagship's binary, dated 2023 |
| 361 | {"0.3.13", false}, // some builds print no leading v |
| 362 | {"v0.4.9", false}, // still pre-snapshot-policy |
| 363 | {"0.5.15", true}, // tito's binary, and the repo's pin |
| 364 | {"v0.5.14", true}, // the vendored library version |
| 365 | {"v1.0.0", true}, // a future major must not regress to "unsupported" |
| 366 | {"litestream v0.5.15\n", true}, |
| 367 | {"", false}, // unknown is treated as unsupported: warn, never assume |
| 368 | {"not-a-version", false}, |
| 369 | } |
| 370 | for _, c := range cases { |
| 371 | if got := SnapshotPolicySupported(c.out); got != c.want { |
| 372 | t.Errorf("SnapshotPolicySupported(%q) = %v, want %v", c.out, got, c.want) |
| 373 | } |
| 374 | } |
| 375 | } |
| 376 | |
| 377 | // The drill must hold ONE restored database at a time, not all of them. |
| 378 | // |
| 379 | // This is not tidiness. carlos-restore-verify.service runs PrivateTmp=yes, so |
| 380 | // its temp dir is a tmpfs whose pages are charged to the unit's own cgroup — |
| 381 | // accumulating every restored copy there spends MemoryMax on files. Measured |
| 382 | // on carlos-tito-ie 2026-08-15: 54 databases totalling 247M against |
| 383 | // MemoryMax=256M, killed by the OOM killer at "256.2M memory peak" after ~40 |
| 384 | // clean restores. Deleting each copy once it has been checked bounds the peak |
| 385 | // at the LARGEST SINGLE database instead of their sum. |
| 386 | func TestRestoreVerifyHoldsOneRestoreAtATime(t *testing.T) { |
| 387 | tmp := t.TempDir() |
| 388 | srcDir := t.TempDir() |
| 389 | cfgPath := filepath.Join(t.TempDir(), "litestream.yml") |
| 390 | // Real source databases: the freshness check reads their headers, and a |
| 391 | // config naming files that do not exist is itself a finding now. |
| 392 | var dbs []string |
| 393 | for _, name := range []string{"a.db", "b.db", "c.db"} { |
| 394 | p := filepath.Join(srcDir, name) |
| 395 | writeCounterDB(t, p, 3) |
| 396 | dbs = append(dbs, p) |
| 397 | } |
| 398 | if err := os.WriteFile(cfgPath, []byte(LitestreamConfig(dbs, "b", "p", "eu-west-1")), 0o644); err != nil { |
| 399 | t.Fatal(err) |
| 400 | } |
| 401 | |
| 402 | maxSeen := 0 |
| 403 | prev := restoreAndCheckFn |
| 404 | restoreAndCheckFn = func(_, _, dbPath, tmpDir string) error { |
| 405 | // Stand in for a restore of a CURRENT replica: a byte copy of the |
| 406 | // source, so its change counter matches and freshness passes. |
| 407 | body, err := os.ReadFile(dbPath) |
| 408 | if err != nil { |
| 409 | return err |
| 410 | } |
| 411 | if err := os.WriteFile(filepath.Join(tmpDir, filepath.Base(dbPath)), body, 0o600); err != nil { |
| 412 | return err |
| 413 | } |
| 414 | ents, err := os.ReadDir(tmpDir) |
| 415 | if err != nil { |
| 416 | return err |
| 417 | } |
| 418 | if len(ents) > maxSeen { |
| 419 | maxSeen = len(ents) |
| 420 | } |
| 421 | return nil |
| 422 | } |
| 423 | t.Cleanup(func() { restoreAndCheckFn = prev }) |
| 424 | |
| 425 | if err := RestoreVerify("litestream", cfgPath, tmp); err != nil { |
| 426 | t.Fatalf("RestoreVerify: %v", err) |
| 427 | } |
| 428 | if maxSeen > 1 { |
| 429 | t.Fatalf("as many as %d restored copies were held at once — the drill spends MemoryMax on tmpfs files", maxSeen) |
| 430 | } |
| 431 | left, err := os.ReadDir(tmp) |
| 432 | if err != nil { |
| 433 | t.Fatal(err) |
| 434 | } |
| 435 | if len(left) != 0 { |
| 436 | t.Fatalf("RestoreVerify left %d file(s) behind", len(left)) |
| 437 | } |
| 438 | } |
| 439 | |
| 440 | // --- replica freshness --------------------------------------------------- |
| 441 | // |
| 442 | // The drill proves a replica RESTORES. It has never proved the replica is |
| 443 | // CURRENT — and those come apart exactly when it matters. On 2026-08-11 the |
| 444 | // flagship's litestream died with "permission denied" writing its shadow WAL |
| 445 | // into a DynamicUser tenant's state dir; had nobody noticed by hand, the daily |
| 446 | // drill would have gone on passing indefinitely, restoring a replica frozen at |
| 447 | // the moment of death. A valid backup of last week is not a backup. |
| 448 | |
| 449 | func writeCounterDB(t *testing.T, path string, writes int) { |
| 450 | t.Helper() |
| 451 | db, err := sql.Open("sqlite", path) |
| 452 | if err != nil { |
| 453 | t.Fatal(err) |
| 454 | } |
| 455 | defer db.Close() |
| 456 | // DELETE journalling: the header change counter advances per commit |
| 457 | // without needing a checkpoint, which keeps the fixture honest about what |
| 458 | // it is testing. |
| 459 | if _, err := db.Exec("PRAGMA journal_mode=DELETE"); err != nil { |
| 460 | t.Fatal(err) |
| 461 | } |
| 462 | if _, err := db.Exec("CREATE TABLE IF NOT EXISTS t (id INTEGER PRIMARY KEY)"); err != nil { |
| 463 | t.Fatal(err) |
| 464 | } |
| 465 | for i := 0; i < writes; i++ { |
| 466 | // A fresh rowid each call, so re-seeding an existing file adds |
| 467 | // transactions rather than colliding on the primary key. |
| 468 | if _, err := db.Exec("INSERT INTO t (id) VALUES (NULL)"); err != nil { |
| 469 | t.Fatal(err) |
| 470 | } |
| 471 | } |
| 472 | } |
| 473 | |
| 474 | func TestChangeCounterAdvancesWithWrites(t *testing.T) { |
| 475 | dir := t.TempDir() |
| 476 | a := filepath.Join(dir, "a.db") |
| 477 | writeCounterDB(t, a, 1) |
| 478 | first, err := ChangeCounter(a) |
| 479 | if err != nil { |
| 480 | t.Fatal(err) |
| 481 | } |
| 482 | writeCounterDB(t, a, 5) |
| 483 | second, err := ChangeCounter(a) |
| 484 | if err != nil { |
| 485 | t.Fatal(err) |
| 486 | } |
| 487 | if second <= first { |
| 488 | t.Fatalf("change counter did not advance across writes: %d then %d", first, second) |
| 489 | } |
| 490 | |
| 491 | junk := filepath.Join(dir, "junk.db") |
| 492 | if err := os.WriteFile(junk, []byte("nope"), 0o600); err != nil { |
| 493 | t.Fatal(err) |
| 494 | } |
| 495 | if _, err := ChangeCounter(junk); err == nil { |
| 496 | t.Fatal("ChangeCounter accepted a file that is not a SQLite database") |
| 497 | } |
| 498 | } |
| 499 | |
| 500 | func TestReplicaFreshness(t *testing.T) { |
| 501 | dir := t.TempDir() |
| 502 | local := filepath.Join(dir, "local.db") |
| 503 | behind := filepath.Join(dir, "behind.db") |
| 504 | writeCounterDB(t, behind, 1) // the replica, frozen early |
| 505 | writeCounterDB(t, local, 9) // the source, written since |
| 506 | |
| 507 | long := time.Now().Add(-2 * time.Hour) |
| 508 | recent := time.Now() |
| 509 | |
| 510 | // Quiet since well before the drill, replica still behind: the replicator |
| 511 | // has had every chance to catch up and has not. That is the alarm. |
| 512 | if v, detail := replicaFreshness(local, behind, long); v != freshnessStale { |
| 513 | t.Fatalf("a replica behind a long-quiet source = %v (%s), want stale", v, detail) |
| 514 | } |
| 515 | // Written moments ago: the replica being a transaction or two behind is |
| 516 | // ordinary lag, not a fault. Never cry wolf here — a drill that goes red |
| 517 | // on healthy churn is a drill people stop reading. |
| 518 | if v, _ := replicaFreshness(local, behind, recent); v != freshnessOK { |
| 519 | t.Fatalf("a replica behind a just-written source = %v, want ok (still catching up)", v) |
| 520 | } |
| 521 | // Caught up. |
| 522 | if v, _ := replicaFreshness(local, local, long); v != freshnessOK { |
| 523 | t.Fatalf("an up-to-date replica = %v, want ok", v) |
| 524 | } |
| 525 | // The source is gone: a config naming a database that no longer exists is |
| 526 | // its own finding — the replica can never advance again. |
| 527 | if v, _ := replicaFreshness(filepath.Join(dir, "absent.db"), behind, long); v != freshnessNoSource { |
| 528 | t.Fatalf("a missing source = %v, want no-source", v) |
| 529 | } |
| 530 | // Unreadable header: say nothing rather than invent a verdict. |
| 531 | junk := filepath.Join(dir, "junk.db") |
| 532 | if err := os.WriteFile(junk, []byte("nope"), 0o600); err != nil { |
| 533 | t.Fatal(err) |
| 534 | } |
| 535 | if v, _ := replicaFreshness(junk, behind, long); v != freshnessUnknown { |
| 536 | t.Fatalf("an unreadable source = %v, want unknown", v) |
| 537 | } |
| 538 | } |
| 539 | |
| 540 | // The whole point, end to end: a replica that restores perfectly but has |
| 541 | // stopped tracking its source must FAIL the run. Before this, such a replica |
| 542 | // was reported "ok" every night for as long as the replicator stayed dead. |
| 543 | func TestRestoreVerifyFailsOnAStaleReplica(t *testing.T) { |
| 544 | srcDir, replicaDir := t.TempDir(), t.TempDir() |
| 545 | cfgPath := filepath.Join(t.TempDir(), "litestream.yml") |
| 546 | |
| 547 | live := filepath.Join(srcDir, "live.db") |
| 548 | writeCounterDB(t, live, 2) |
| 549 | // Snapshot the replica HERE — this stands for the moment the replicator |
| 550 | // died — then let the source move on without it. |
| 551 | frozen, err := os.ReadFile(live) |
| 552 | if err != nil { |
| 553 | t.Fatal(err) |
| 554 | } |
| 555 | writeCounterDB(t, live, 20) |
| 556 | // And make the source quiet for longer than the grace window, so this is |
| 557 | // unambiguously "replication stopped" rather than "still catching up". |
| 558 | old := time.Now().Add(-2 * time.Hour) |
| 559 | if err := os.Chtimes(live, old, old); err != nil { |
| 560 | t.Fatal(err) |
| 561 | } |
| 562 | |
| 563 | if err := os.WriteFile(cfgPath, []byte(LitestreamConfig([]string{live}, "b", "p", "eu-west-1")), 0o644); err != nil { |
| 564 | t.Fatal(err) |
| 565 | } |
| 566 | prev := restoreAndCheckFn |
| 567 | restoreAndCheckFn = func(_, _, dbPath, tmpDir string) error { |
| 568 | return os.WriteFile(filepath.Join(tmpDir, filepath.Base(dbPath)), frozen, 0o600) |
| 569 | } |
| 570 | t.Cleanup(func() { restoreAndCheckFn = prev }) |
| 571 | |
| 572 | err = RestoreVerify("litestream", cfgPath, replicaDir) |
| 573 | if err == nil { |
| 574 | t.Fatal("a replica frozen at the moment its replicator died passed the drill") |
| 575 | } |
| 576 | if !strings.Contains(err.Error(), "STOPPED TRACKING") { |
| 577 | t.Fatalf("error does not name the failure: %v", err) |
| 578 | } |
| 579 | } |
| 580 | |
| 581 | // A unit tenant whose app manages its OWN sub-instances keeps a TREE of |
| 582 | // databases, not one file: kass on the flagship has registry.db plus an |
| 583 | // instances/ directory that grows as the tenant creates instances. Naming a |
| 584 | // single -db path cannot express that, which is why the flagship's config is |
| 585 | // still hand-maintained ("carlos ops litestream takes over once nested layouts |
| 586 | // are supported") — and why it had already drifted, missing one instance |
| 587 | // database that the tenant created after the list was written. |
| 588 | // |
| 589 | // So: a BackingUnit route whose -db names a DIRECTORY replicates every *.db |
| 590 | // beneath it. Exec routes are untouched by this — the activator consumes their |
| 591 | // DBPath verbatim (Restore, Seed, the --db flag), so a directory there would |
| 592 | // be meaningless. |
| 593 | func TestFleetDBsWalksAUnitTenantDirectory(t *testing.T) { |
| 594 | dir := t.TempDir() |
| 595 | reg, err := registry.Open(filepath.Join(dir, "registry.db")) |
| 596 | if err != nil { |
| 597 | t.Fatal(err) |
| 598 | } |
| 599 | state := t.TempDir() |
| 600 | nested := filepath.Join(state, "instances") |
| 601 | if err := os.MkdirAll(nested, 0o750); err != nil { |
| 602 | t.Fatal(err) |
| 603 | } |
| 604 | top := filepath.Join(state, "registry.db") |
| 605 | inst1 := filepath.Join(nested, "5vepbveytb8y.db") |
| 606 | inst2 := filepath.Join(nested, "eafu5brhjs26.db") |
| 607 | for _, p := range []string{top, inst1, inst2} { |
| 608 | if err := os.WriteFile(p, []byte("x"), 0o600); err != nil { |
| 609 | t.Fatal(err) |
| 610 | } |
| 611 | } |
| 612 | // Litestream's own shadow-WAL directory sits beside each database and is |
| 613 | // full of files that are not databases. Replicating a replicator's |
| 614 | // bookkeeping would be absurd, and dot-directories are its convention. |
| 615 | meta := filepath.Join(state, ".registry.db-litestream", "generations", "abc", "wal") |
| 616 | if err := os.MkdirAll(meta, 0o750); err != nil { |
| 617 | t.Fatal(err) |
| 618 | } |
| 619 | if err := os.WriteFile(filepath.Join(meta, "00000000.db"), []byte("x"), 0o600); err != nil { |
| 620 | t.Fatal(err) |
| 621 | } |
| 622 | // Sidecars and empty placeholders are not databases either. |
| 623 | for _, p := range []string{top + "-wal", top + "-shm", filepath.Join(state, "empty.db")} { |
| 624 | if err := os.WriteFile(p, nil, 0o600); err != nil { |
| 625 | t.Fatal(err) |
| 626 | } |
| 627 | } |
| 628 | |
| 629 | if err := reg.Upsert(registry.Route{Host: "home.kass.training", Backing: registry.BackingUnit, DBPath: state}); err != nil { |
| 630 | t.Fatal(err) |
| 631 | } |
| 632 | |
| 633 | dbs, err := FleetDBs(reg, dir) |
| 634 | if err != nil { |
| 635 | t.Fatal(err) |
| 636 | } |
| 637 | for _, want := range []string{top, inst1, inst2} { |
| 638 | if !slices.Contains(dbs, want) { |
| 639 | t.Fatalf("tree database %s missing: %v", want, dbs) |
| 640 | } |
| 641 | } |
| 642 | for _, unwanted := range []string{meta, top + "-wal", top + "-shm", filepath.Join(state, "empty.db")} { |
| 643 | if slices.Contains(dbs, unwanted) { |
| 644 | t.Fatalf("FleetDBs swept up %s: %v", unwanted, dbs) |
| 645 | } |
| 646 | } |
| 647 | for _, got := range dbs { |
| 648 | if strings.Contains(got, "-litestream") { |
| 649 | t.Fatalf("FleetDBs descended into a litestream meta directory: %s", got) |
| 650 | } |
| 651 | } |
| 652 | } |
| 653 | |
| 654 | // The tree rule must not leak to exec routes: their DBPath is consumed |
| 655 | // verbatim by the activator, so a directory there is a mistake, not a tree. |
| 656 | func TestFleetDBsDoesNotWalkExecRouteDBPath(t *testing.T) { |
| 657 | dir := t.TempDir() |
| 658 | reg, err := registry.Open(filepath.Join(dir, "registry.db")) |
| 659 | if err != nil { |
| 660 | t.Fatal(err) |
| 661 | } |
| 662 | state := t.TempDir() |
| 663 | if err := os.WriteFile(filepath.Join(state, "inner.db"), []byte("x"), 0o600); err != nil { |
| 664 | t.Fatal(err) |
| 665 | } |
| 666 | if err := reg.Upsert(registry.Route{Host: "exec.example.com", DBPath: state}); err != nil { |
| 667 | t.Fatal(err) |
| 668 | } |
| 669 | |
| 670 | dbs, err := FleetDBs(reg, dir) |
| 671 | if err != nil { |
| 672 | t.Fatal(err) |
| 673 | } |
| 674 | if slices.Contains(dbs, filepath.Join(state, "inner.db")) { |
| 675 | t.Fatalf("an exec route's directory DBPath was walked: %v", dbs) |
| 676 | } |
| 677 | } |
| 678 | |
| 679 | // Cutting a box over from a hand-maintained config to a generated one is the |
| 680 | // moment a database can silently stop being replicated: the new list is |
| 681 | // derived from the registry, and anything the registry does not know about |
| 682 | // simply is not in it. Nobody would notice — replication of a dropped database |
| 683 | // just stops, and the drill only checks what the config names. |
| 684 | // |
| 685 | // So the generator refuses to shrink the set unless told to. |
| 686 | func TestCoverageShortfall(t *testing.T) { |
| 687 | existing := LitestreamConfig([]string{"/data/a.db", "/data/b.db", "/data/registry.db"}, "b", "p", "eu-west-1") |
| 688 | |
| 689 | // A superset is fine — that is the ordinary case (console.db arriving). |
| 690 | if missing := CoverageShortfall(existing, []string{"/data/a.db", "/data/b.db", "/data/registry.db", "/data/console/console.db"}); len(missing) != 0 { |
| 691 | t.Fatalf("a superset was reported as a shortfall: %v", missing) |
| 692 | } |
| 693 | // Identical is fine. |
| 694 | if missing := CoverageShortfall(existing, []string{"/data/registry.db", "/data/b.db", "/data/a.db"}); len(missing) != 0 { |
| 695 | t.Fatalf("an identical set was reported as a shortfall: %v", missing) |
| 696 | } |
| 697 | // Dropping one is the thing to catch, and the caller must be told WHICH. |
| 698 | missing := CoverageShortfall(existing, []string{"/data/a.db", "/data/registry.db"}) |
| 699 | if len(missing) != 1 || missing[0] != "/data/b.db" { |
| 700 | t.Fatalf("CoverageShortfall = %v, want [/data/b.db]", missing) |
| 701 | } |
| 702 | // No existing config (a fresh box) can never be a shortfall. |
| 703 | if missing := CoverageShortfall("", []string{"/data/a.db"}); len(missing) != 0 { |
| 704 | t.Fatalf("a fresh box reported a shortfall: %v", missing) |
| 705 | } |
| 706 | } |
| 707 | |
| 708 | // Two databases with the same basename must not replicate to the same place. |
| 709 | // |
| 710 | // Every SQLite-backed tenant on the fleet names its database registry.db or |
| 711 | // data.db; the box's own registry is /data/registry.db. Deriving the replica |
| 712 | // URL from filepath.Base collapses all of them onto one S3 key, and two |
| 713 | // litestream replicators writing one prefix interleave their WAL into a |
| 714 | // replica that restores as neither database. Tree discovery makes the |
| 715 | // collision the normal case rather than a curiosity: every tenant tree has a |
| 716 | // registry.db in it. |
| 717 | func TestReplicaURLsAreUniquePerPath(t *testing.T) { |
| 718 | cfg := LitestreamConfig([]string{ |
| 719 | "/data/registry.db", |
| 720 | "/var/lib/carlos-app/home.kass.training/registry.db", |
| 721 | "/var/lib/carlos-app/other.example.com/registry.db", |
| 722 | }, "bucket-x", "fleet", "eu-west-1") |
| 723 | |
| 724 | var urls []string |
| 725 | for _, line := range strings.Split(cfg, "\n") { |
| 726 | if u, ok := strings.CutPrefix(strings.TrimSpace(line), "- url: "); ok { |
| 727 | urls = append(urls, u) |
| 728 | } |
| 729 | } |
| 730 | if len(urls) != 3 { |
| 731 | t.Fatalf("expected 3 replica urls, got %d:\n%s", len(urls), cfg) |
| 732 | } |
| 733 | seen := map[string]bool{} |
| 734 | for _, u := range urls { |
| 735 | if seen[u] { |
| 736 | t.Fatalf("two databases share the replica url %q — their WALs would interleave:\n%s", u, cfg) |
| 737 | } |
| 738 | seen[u] = true |
| 739 | } |
| 740 | // And the key should still be legible to whoever is doing the restore. |
| 741 | if !strings.Contains(cfg, "home.kass.training") { |
| 742 | t.Fatalf("replica keys lost the path that identifies them:\n%s", cfg) |
| 743 | } |
| 744 | } |
| 745 | |
| 746 | // The dataDir glob takes every *.db it finds, which means the box-wide |
| 747 | // replicator will happily stream anything that lands there — including things |
| 748 | // that are not tenant databases at all. The flagship has a 410M |
| 749 | // harbourmail-purge-backup.db sitting in /data: a leftover from a purge, owned |
| 750 | // by root, matching the glob exactly. Today it is not replicated only because |
| 751 | // that box's config is hand-written; the moment the generator takes over it |
| 752 | // would join the replication set and start costing storage forever. |
| 753 | // |
| 754 | // So the glob proposes and the REGISTRY disposes: a file under dataDir is |
| 755 | // replicated when some route claims it (or it is the box's own registry or |
| 756 | // console). Anything else is skipped and NAMED — never silently, because a |
| 757 | // skipped database that someone did want is exactly the failure this whole |
| 758 | // area keeps producing. |
| 759 | func TestFleetDBsSkipsUnclaimedFilesAndSaysSo(t *testing.T) { |
| 760 | dir := t.TempDir() |
| 761 | reg, err := registry.Open(filepath.Join(dir, "registry.db")) |
| 762 | if err != nil { |
| 763 | t.Fatal(err) |
| 764 | } |
| 765 | claimed := filepath.Join(dir, "app.example.com.db") |
| 766 | stray := filepath.Join(dir, "harbourmail-purge-backup.db") |
| 767 | for _, p := range []string{claimed, stray} { |
| 768 | if err := os.WriteFile(p, []byte("x"), 0o600); err != nil { |
| 769 | t.Fatal(err) |
| 770 | } |
| 771 | } |
| 772 | if err := reg.Upsert(registry.Route{Host: "app.example.com", DBPath: claimed}); err != nil { |
| 773 | t.Fatal(err) |
| 774 | } |
| 775 | |
| 776 | dbs, skipped, err := FleetDBsWithSkipped(reg, dir, nil) |
| 777 | if err != nil { |
| 778 | t.Fatal(err) |
| 779 | } |
| 780 | if !slices.Contains(dbs, claimed) { |
| 781 | t.Fatalf("a route's own database was skipped: %v", dbs) |
| 782 | } |
| 783 | if !slices.Contains(dbs, filepath.Join(dir, "registry.db")) { |
| 784 | t.Fatalf("the box registry was skipped: %v", dbs) |
| 785 | } |
| 786 | if slices.Contains(dbs, stray) { |
| 787 | t.Fatalf("an unclaimed file joined the replication set: %v", dbs) |
| 788 | } |
| 789 | if !slices.Contains(skipped, stray) { |
| 790 | t.Fatalf("the unclaimed file was dropped SILENTLY — skipped = %v", skipped) |
| 791 | } |
| 792 | } |
| 793 | |
| 794 | // A host with a layout migration journal is between two directories, so |
| 795 | // the generator claims neither of its database paths: its file is named |
| 796 | // among the skipped rather than replicated from a place it is leaving. |
| 797 | func TestFleetDBsSkipsAHostTheLayoutPassHolds(t *testing.T) { |
| 798 | dir := t.TempDir() |
| 799 | reg, err := registry.Open(filepath.Join(dir, "registry.db")) |
| 800 | if err != nil { |
| 801 | t.Fatal(err) |
| 802 | } |
| 803 | held := filepath.Join(dir, "held.example.com.db") |
| 804 | kept := filepath.Join(dir, "kept.example.com.db") |
| 805 | for _, p := range []string{held, kept} { |
| 806 | if err := os.WriteFile(p, []byte("x"), 0o600); err != nil { |
| 807 | t.Fatal(err) |
| 808 | } |
| 809 | } |
| 810 | for _, rt := range []registry.Route{ |
| 811 | {Host: "held.example.com", DBPath: held}, |
| 812 | {Host: "kept.example.com", DBPath: kept}, |
| 813 | } { |
| 814 | if err := reg.Upsert(rt); err != nil { |
| 815 | t.Fatal(err) |
| 816 | } |
| 817 | } |
| 818 | journal := provision.LayoutJournalPath(dir, "held.example.com") |
| 819 | if err := os.MkdirAll(filepath.Dir(journal), 0o700); err != nil { |
| 820 | t.Fatal(err) |
| 821 | } |
| 822 | if err := os.WriteFile(journal, []byte("{}"), 0o600); err != nil { |
| 823 | t.Fatal(err) |
| 824 | } |
| 825 | |
| 826 | dbs, skipped, err := FleetDBsWithSkipped(reg, dir, LayoutJournalPresent(dir)) |
| 827 | if err != nil { |
| 828 | t.Fatal(err) |
| 829 | } |
| 830 | if slices.Contains(dbs, held) { |
| 831 | t.Fatalf("a held host's database joined the replication set: %v", dbs) |
| 832 | } |
| 833 | if !slices.Contains(skipped, held) { |
| 834 | t.Fatalf("a held host's database was dropped silently: skipped = %v", skipped) |
| 835 | } |
| 836 | if !slices.Contains(dbs, kept) { |
| 837 | t.Fatalf("a host with no journal was skipped: %v", dbs) |
| 838 | } |
| 839 | |
| 840 | // No hook is the old behaviour: every route's database is claimed. |
| 841 | dbs, _, err = FleetDBsWithSkipped(reg, dir, nil) |
| 842 | if err != nil { |
| 843 | t.Fatal(err) |
| 844 | } |
| 845 | if !slices.Contains(dbs, held) { |
| 846 | t.Fatalf("with no hook the held host was skipped: %v", dbs) |
| 847 | } |
| 848 | } |
| 849 | |
| 850 | // LayoutJournalPresent answers true for anything at the journal path that |
| 851 | // is not definitely absent, a dangling symlink included: a journal that |
| 852 | // cannot be read is not proof there is none. |
| 853 | func TestLayoutJournalPresentIsLstatNotAbsent(t *testing.T) { |
| 854 | dir := t.TempDir() |
| 855 | present := LayoutJournalPresent(dir) |
| 856 | if present("a.example.com") { |
| 857 | t.Fatal("no journal directory, but a journal was reported") |
| 858 | } |
| 859 | link := provision.LayoutJournalPath(dir, "a.example.com") |
| 860 | if err := os.MkdirAll(filepath.Dir(link), 0o700); err != nil { |
| 861 | t.Fatal(err) |
| 862 | } |
| 863 | if err := os.Symlink(filepath.Join(dir, "nowhere"), link); err != nil { |
| 864 | t.Fatal(err) |
| 865 | } |
| 866 | if !present("a.example.com") { |
| 867 | t.Fatal("a dangling symlink at the journal path was read as no journal") |
| 868 | } |
| 869 | if present("b.example.com") { |
| 870 | t.Fatal("another host was reported held") |
| 871 | } |
| 872 | } |
| 873 | |
amadan