| 1 | package main |
| 2 | |
| 3 | // 🤖 The live two-box mesh tests ("the kill-a-box demo", #550 Part 4 v2 |
| 4 | // Phase 3): TWO in-process boxes — two activator+router pairs on different |
| 5 | // loopback ports, each spawning REAL instance processes — sharing one REAL S3 |
| 6 | // bucket for leases, replicas, and mesh heartbeats. They prove the mesh's |
| 7 | // whole story: |
| 8 | // |
| 9 | // - a request landing on the NON-owner box is reverse-proxied to the HRW |
| 10 | // owner, which wakes the instance and serves; |
| 11 | // - killing the owner box moves its instance to the survivor with data |
| 12 | // intact — in BOTH kill flavors: |
| 13 | // drain (TestMeshLiveProxyAndDrainFailover): heartbeat deleted |
| 14 | // first, checkpoint + lease release — the survivor takes |
| 15 | // over on its next membership poll; |
| 16 | // hard kill (TestMeshLiveHardKillFailover): the box just dies — its |
| 17 | // heartbeat rots to staleness, its lease expires on TTL, and |
| 18 | // the survivor's wake is gated by the lease until then. |
| 19 | // |
| 20 | // Gated on LCHAT_TEST_S3_BUCKET like the other live tests. Run with: |
| 21 | // |
| 22 | // eval "$(aws configure export-credentials --profile eleven-dev --format env)" |
| 23 | // LCHAT_TEST_S3_BUCKET=<bucket> go test -run TestMeshLive -v . |
| 24 | |
| 25 | import ( |
| 26 | "bytes" |
| 27 | "context" |
| 28 | "crypto/rand" |
| 29 | "encoding/hex" |
| 30 | "encoding/json" |
| 31 | "fmt" |
| 32 | "io" |
| 33 | "net/http" |
| 34 | "net/http/httptest" |
| 35 | "os" |
| 36 | "os/exec" |
| 37 | "path/filepath" |
| 38 | "strings" |
| 39 | "sync" |
| 40 | "testing" |
| 41 | "time" |
| 42 | |
| 43 | "lchat/internal/activator" |
| 44 | ) |
| 45 | |
| 46 | // liveMeshBox is one box of the pair: its own activator (real s3Backend, real |
| 47 | // child processes), registry, mesh, and front door. |
| 48 | type liveMeshBox struct { |
| 49 | name string |
| 50 | self string // the front URL — this box's mesh identity |
| 51 | act *activatorSvc |
| 52 | mesh *meshSvc |
| 53 | front *httptest.Server |
| 54 | reg *Registry |
| 55 | route Route |
| 56 | |
| 57 | shutdownOnce sync.Once |
| 58 | } |
| 59 | |
| 60 | // shutdown drains the box exactly once — tests that dismantle a box by hand |
| 61 | // (the hard kill) consume the Once so cleanup can't double-drain it. |
| 62 | func (b *liveMeshBox) shutdown(ctx context.Context) { |
| 63 | b.shutdownOnce.Do(func() { b.act.shutdown(ctx) }) |
| 64 | } |
| 65 | |
| 66 | // setupMeshLive builds the shared world (S3 prefix, staged binary, env file) |
| 67 | // and the two boxes. Both boxes register the same host, each pointing at its |
| 68 | // OWN socket and DB path — a box must never see its peer's local files, so a |
| 69 | // failover's restore provably comes from S3. |
| 70 | func setupMeshLive(t *testing.T, host string) (boxA, boxB *liveMeshBox, adminToken string) { |
| 71 | t.Helper() |
| 72 | bucket := os.Getenv("LCHAT_TEST_S3_BUCKET") |
| 73 | if bucket == "" { |
| 74 | t.Skip("LCHAT_TEST_S3_BUCKET not set; skipping live two-box mesh test") |
| 75 | } |
| 76 | region := os.Getenv("LCHAT_TEST_S3_REGION") |
| 77 | if region == "" { |
| 78 | region = "eu-west-1" |
| 79 | } |
| 80 | rb := make([]byte, 4) |
| 81 | _, _ = rand.Read(rb) |
| 82 | prefix := fmt.Sprintf("mesh-live/%d-%s", time.Now().Unix(), hex.EncodeToString(rb)) |
| 83 | t.Cleanup(func() { cleanupS3Prefix(t, bucket, prefix, region) }) |
| 84 | |
| 85 | // Sandbox the box-layout globals (shared by both in-process boxes: the |
| 86 | // staged binary, live symlink, and the instance env file are box-agnostic). |
| 87 | dataDir := t.TempDir() |
| 88 | sockDir := fmt.Sprintf("/tmp/lchat-mesh-%d", os.Getpid()) |
| 89 | if err := os.MkdirAll(sockDir, 0o755); err != nil { |
| 90 | t.Fatal(err) |
| 91 | } |
| 92 | t.Cleanup(func() { os.RemoveAll(sockDir) }) |
| 93 | optTmp := t.TempDir() |
| 94 | oldData, oldSock, oldOpt := instanceDataDir, instanceSocketDir, optDir |
| 95 | instanceDataDir, instanceSocketDir, optDir = dataDir, sockDir, optTmp |
| 96 | t.Cleanup(func() { instanceDataDir, instanceSocketDir, optDir = oldData, oldSock, oldOpt }) |
| 97 | |
| 98 | bin := filepath.Join(optTmp, "lchat-built") |
| 99 | build := exec.Command("go", "build", "-o", bin, ".") |
| 100 | build.Env = append(os.Environ(), "GOTOOLCHAIN=auto") |
| 101 | if out, err := build.CombinedOutput(); err != nil { |
| 102 | t.Fatalf("go build: %v\n%s", err, out) |
| 103 | } |
| 104 | if err := stageVersion(bin, "meshlive"); err != nil { |
| 105 | t.Fatalf("stage: %v", err) |
| 106 | } |
| 107 | if err := pointLive(host, "meshlive"); err != nil { |
| 108 | t.Fatalf("point live: %v", err) |
| 109 | } |
| 110 | |
| 111 | adminToken = "meshlive-" + hex.EncodeToString(rb) |
| 112 | if err := writeEnvFile(filepath.Join(dataDir, host+".env"), |
| 113 | map[string]string{"LCHAT_ADMIN_TOKEN": adminToken}); err != nil { |
| 114 | t.Fatalf("env file: %v", err) |
| 115 | } |
| 116 | |
| 117 | cfg := hibernateConfig{Bucket: bucket, Region: region, Prefix: prefix, Idle: 2 * time.Minute} |
| 118 | newBox := func(name string) *liveMeshBox { |
| 119 | box := &liveMeshBox{name: name} |
| 120 | var err error |
| 121 | box.reg, err = openRegistry(filepath.Join(t.TempDir(), "registry.db")) |
| 122 | if err != nil { |
| 123 | t.Fatalf("open registry: %v", err) |
| 124 | } |
| 125 | dbDir := filepath.Join(dataDir, name) |
| 126 | if err := os.MkdirAll(dbDir, 0o755); err != nil { |
| 127 | t.Fatal(err) |
| 128 | } |
| 129 | box.route = Route{Host: host, Kind: "instance", |
| 130 | Socket: filepath.Join(sockDir, name+".sock"), |
| 131 | DBPath: filepath.Join(dbDir, host+".db"), |
| 132 | Hibernate: true} |
| 133 | if err := box.reg.upsert(box.route); err != nil { |
| 134 | t.Fatalf("upsert: %v", err) |
| 135 | } |
| 136 | box.act = newActivator(cfg, &s3Backend{cfg: cfg}, box.reg) |
| 137 | rt, err := newRouter(box.reg, box.act) |
| 138 | if err != nil { |
| 139 | t.Fatalf("new router: %v", err) |
| 140 | } |
| 141 | box.front = httptest.NewServer(rt) |
| 142 | t.Cleanup(box.front.Close) |
| 143 | box.self = box.front.URL |
| 144 | |
| 145 | hb, err := activator.NewHeartbeatStore(context.Background(), bucket, cfg.meshPrefix(), region) |
| 146 | if err != nil { |
| 147 | t.Fatalf("heartbeat store: %v", err) |
| 148 | } |
| 149 | box.mesh = newMesh(box.self, hb, nil) |
| 150 | box.mesh.interval = time.Second |
| 151 | box.mesh.staleAfter = 3 * time.Second |
| 152 | box.act.mesh = box.mesh |
| 153 | box.mesh.start() |
| 154 | box.act.start() |
| 155 | t.Cleanup(func() { |
| 156 | ctx, cancel := context.WithTimeout(context.Background(), 60*time.Second) |
| 157 | defer cancel() |
| 158 | box.shutdown(ctx) |
| 159 | }) |
| 160 | return box |
| 161 | } |
| 162 | boxA, boxB = newBox("box-a"), newBox("box-b") |
| 163 | |
| 164 | // Both boxes must see the full membership before the test routes anything. |
| 165 | waitForLive(t, "both boxes to see each other", 15*time.Second, func() bool { |
| 166 | return len(boxA.mesh.liveBoxes()) == 2 && len(boxB.mesh.liveBoxes()) == 2 |
| 167 | }) |
| 168 | return boxA, boxB, adminToken |
| 169 | } |
| 170 | |
| 171 | // pickOwner orders the pair as (owner, survivor) for host by the same HRW |
| 172 | // hash the boxes themselves use. |
| 173 | func pickOwner(host string, a, b *liveMeshBox) (owner, survivor *liveMeshBox) { |
| 174 | if hrwOwner(host, []string{a.self, b.self}) == b.self { |
| 175 | return b, a |
| 176 | } |
| 177 | return a, b |
| 178 | } |
| 179 | |
| 180 | func meshGet(t *testing.T, front, host, path string) (int, string) { |
| 181 | t.Helper() |
| 182 | req, _ := http.NewRequest("GET", front+path, nil) |
| 183 | req.Host = host |
| 184 | resp, err := http.DefaultClient.Do(req) |
| 185 | if err != nil { |
| 186 | t.Fatalf("GET %s: %v", path, err) |
| 187 | } |
| 188 | defer resp.Body.Close() |
| 189 | body, _ := io.ReadAll(resp.Body) |
| 190 | return resp.StatusCode, string(body) |
| 191 | } |
| 192 | |
| 193 | func meshSetGroupName(t *testing.T, front, host, token, name string) { |
| 194 | t.Helper() |
| 195 | body, _ := json.Marshal(map[string]string{"name": name}) |
| 196 | req, _ := http.NewRequest("POST", front+"/api/admin/group", bytes.NewReader(body)) |
| 197 | req.Host = host |
| 198 | req.Header.Set("Authorization", "Bearer "+token) |
| 199 | req.Header.Set("Content-Type", "application/json") |
| 200 | resp, err := http.DefaultClient.Do(req) |
| 201 | if err != nil { |
| 202 | t.Fatalf("set group name: %v", err) |
| 203 | } |
| 204 | defer resp.Body.Close() |
| 205 | if resp.StatusCode != 200 { |
| 206 | b, _ := io.ReadAll(resp.Body) |
| 207 | t.Fatalf("set group name: %d %s", resp.StatusCode, b) |
| 208 | } |
| 209 | } |
| 210 | |
| 211 | // pollUntilServed hammers the survivor's front door until the instance |
| 212 | // answers 200, logging every distinct (status, body) transition — the test's |
| 213 | // narrative of the failover: proxy-to-dead-peer 503s, then lease-held 503s, |
| 214 | // then the wake. Returns the winning body. |
| 215 | func pollUntilServed(t *testing.T, front, host string, deadline time.Duration) string { |
| 216 | t.Helper() |
| 217 | start := time.Now() |
| 218 | last := "" |
| 219 | for time.Now().Before(start.Add(deadline)) { |
| 220 | code, body := meshGet(t, front, host, "/api/group") |
| 221 | if code == 200 { |
| 222 | t.Logf(" +%s: 200 — served", time.Since(start).Truncate(time.Millisecond)) |
| 223 | return body |
| 224 | } |
| 225 | if cur := fmt.Sprintf("%d %s", code, strings.TrimSpace(body)); cur != last { |
| 226 | t.Logf(" +%s: %s", time.Since(start).Truncate(time.Millisecond), cur) |
| 227 | last = cur |
| 228 | } |
| 229 | time.Sleep(time.Second) |
| 230 | } |
| 231 | t.Fatalf("instance never served on the survivor within %s", deadline) |
| 232 | return "" |
| 233 | } |
| 234 | |
| 235 | // The drain variant: proxy-to-owner, then an orderly box exit (heartbeat |
| 236 | // deleted FIRST, checkpoint, lease release) and the survivor taking over on |
| 237 | // its next membership poll — the same code path a deploy or scale-down runs. |
| 238 | func TestMeshLiveProxyAndDrainFailover(t *testing.T) { |
| 239 | const host = "mesh-drain.test" |
| 240 | boxA, boxB, adminToken := setupMeshLive(t, host) |
| 241 | owner, survivor := pickOwner(host, boxA, boxB) |
| 242 | t.Logf("HRW owner of %s is %s (%s); survivor is %s (%s)", host, owner.name, owner.self, survivor.name, survivor.self) |
| 243 | |
| 244 | // --- Act 1: a request to the NON-owner proxies to the owner and serves. --- |
| 245 | start := time.Now() |
| 246 | code, _ := meshGet(t, survivor.front.URL, host, "/api/group") |
| 247 | if code != 200 { |
| 248 | t.Fatalf("request via non-owner: %d, want 200", code) |
| 249 | } |
| 250 | t.Logf("request to the non-owner proxied + woke the instance on %s in %s", owner.name, time.Since(start).Truncate(time.Millisecond)) |
| 251 | if owner.act.entryCount() != 1 || !owner.act.entryAwake(host) { |
| 252 | t.Fatalf("instance is not awake on the owner (entries=%d)", owner.act.entryCount()) |
| 253 | } |
| 254 | if survivor.act.entryCount() != 0 { |
| 255 | t.Fatal("the non-owner woke a local copy instead of proxying") |
| 256 | } |
| 257 | |
| 258 | // Write real data through the proxied path. |
| 259 | groupName := "Mesh failover " + hex.EncodeToString([]byte(host)) |
| 260 | meshSetGroupName(t, survivor.front.URL, host, adminToken, groupName) |
| 261 | t.Logf("wrote group name %q through the non-owner's front door", groupName) |
| 262 | |
| 263 | // --- Act 2: drain the owner. Front door dark first (runRouter shuts the |
| 264 | // HTTP servers down before the activator — a draining box accepts no new |
| 265 | // work), then the drain proper: heartbeat delete, checkpoint, release. --- |
| 266 | start = time.Now() |
| 267 | ctx, cancel := context.WithTimeout(context.Background(), 60*time.Second) |
| 268 | defer cancel() |
| 269 | owner.front.Close() |
| 270 | owner.shutdown(ctx) |
| 271 | t.Logf("owner %s drained in %s (heartbeat deleted, instance checkpointed, lease released)", owner.name, time.Since(start).Truncate(time.Millisecond)) |
| 272 | |
| 273 | // The heartbeat really is gone from the bucket — only the survivor beats. |
| 274 | hbs, err := survivor.mesh.store.List(context.Background()) |
| 275 | if err != nil { |
| 276 | t.Fatalf("list heartbeats: %v", err) |
| 277 | } |
| 278 | for _, hb := range hbs { |
| 279 | if hb.Peer == owner.self { |
| 280 | t.Fatal("drained owner's heartbeat is still in the bucket") |
| 281 | } |
| 282 | } |
| 283 | |
| 284 | // --- Act 3: the survivor takes over — wake local, restore from S3. --- |
| 285 | body := pollUntilServed(t, survivor.front.URL, host, 60*time.Second) |
| 286 | if !strings.Contains(body, groupName) { |
| 287 | t.Fatalf("survivor lost the data: %s does not contain %q", body, groupName) |
| 288 | } |
| 289 | if !survivor.act.entryAwake(host) { |
| 290 | t.Fatal("instance is not awake on the survivor") |
| 291 | } |
| 292 | t.Logf("survivor %s restored the instance from S3 with data intact: %s", survivor.name, body) |
| 293 | } |
| 294 | |
| 295 | // The hard-kill variant: the owner box dies with no goodbye — its heartbeat |
| 296 | // is left to rot (stale in staleAfter) and its lease to expire (TTL). The |
| 297 | // survivor's requests walk the whole failure narrative: proxy to the corpse |
| 298 | // (503), local wake gated by the still-held lease (503 + Retry-After), then |
| 299 | // the lease expires and the instance wakes on the survivor with every byte |
| 300 | // that had replicated. |
| 301 | func TestMeshLiveHardKillFailover(t *testing.T) { |
| 302 | const host = "mesh-kill.test" |
| 303 | boxA, boxB, adminToken := setupMeshLive(t, host) |
| 304 | owner, survivor := pickOwner(host, boxA, boxB) |
| 305 | t.Logf("HRW owner of %s is %s (%s); survivor is %s (%s)", host, owner.name, owner.self, survivor.name, survivor.self) |
| 306 | |
| 307 | // --- Act 1: wake on the owner and write data. --- |
| 308 | if code, _ := meshGet(t, owner.front.URL, host, "/api/group"); code != 200 { |
| 309 | t.Fatalf("wake on owner: %d, want 200", code) |
| 310 | } |
| 311 | groupName := "Hard kill survivor " + hex.EncodeToString([]byte(host)) |
| 312 | meshSetGroupName(t, owner.front.URL, host, adminToken, groupName) |
| 313 | |
| 314 | // Make the write durable in S3 before the "crash" — a real hard kill |
| 315 | // loses only the last seconds of unshipped WAL (that's the design's RPO); |
| 316 | // the test pins the data so the assertion is about failover, not timing. |
| 317 | owner.act.mu.Lock() |
| 318 | entry := owner.act.entries[host] |
| 319 | owner.act.mu.Unlock() |
| 320 | if entry == nil { |
| 321 | t.Fatal("no live entry on the owner") |
| 322 | } |
| 323 | rep := entry.rep.(*activator.LeasedReplicator) |
| 324 | syncCtx, syncCancel := context.WithTimeout(context.Background(), 30*time.Second) |
| 325 | if err := rep.SyncAndWait(syncCtx); err != nil { |
| 326 | syncCancel() |
| 327 | t.Fatalf("sync before kill: %v", err) |
| 328 | } |
| 329 | syncCancel() |
| 330 | t.Logf("wrote group name %q on the owner and confirmed it replicated to S3", groupName) |
| 331 | |
| 332 | // --- Act 2: hard-kill the owner box. Everything at once, nothing orderly: |
| 333 | // front door gone, sweeper gone, heartbeat loop gone (object NOT deleted), |
| 334 | // instance child SIGKILLed, lease renewal abandoned (NOT released). --- |
| 335 | owner.front.Close() |
| 336 | close(owner.act.sweepStop) |
| 337 | <-owner.act.sweepDone |
| 338 | owner.mesh.halt() // the heartbeat object stays — it must rot to staleness |
| 339 | owner.act.mu.Lock() |
| 340 | delete(owner.act.entries, host) |
| 341 | owner.act.mu.Unlock() |
| 342 | proc := entry.proc.(*childProc) |
| 343 | if err := proc.cmd.Process.Kill(); err != nil { |
| 344 | t.Fatalf("SIGKILL instance child: %v", err) |
| 345 | } |
| 346 | <-proc.done |
| 347 | abandonCtx, abandonCancel := context.WithTimeout(context.Background(), 30*time.Second) |
| 348 | rep.Abandon(abandonCtx) |
| 349 | abandonCancel() |
| 350 | owner.shutdownOnce.Do(func() {}) // consume: cleanup must not re-drain a corpse |
| 351 | t.Logf("owner %s hard-killed: heartbeat left in S3 (stale after %s), lease left to expire (TTL %s)", |
| 352 | owner.name, survivor.mesh.staleAfter, activator.LeaseTTL) |
| 353 | |
| 354 | // --- Act 3: the survivor takes over once staleness + lease expiry pass. --- |
| 355 | start := time.Now() |
| 356 | body := pollUntilServed(t, survivor.front.URL, host, 120*time.Second) |
| 357 | if !strings.Contains(body, groupName) { |
| 358 | t.Fatalf("survivor lost the data: %s does not contain %q", body, groupName) |
| 359 | } |
| 360 | if !survivor.act.entryAwake(host) { |
| 361 | t.Fatal("instance is not awake on the survivor") |
| 362 | } |
| 363 | t.Logf("survivor %s took over %s after the owner's death with data intact: %s", |
| 364 | survivor.name, time.Since(start).Truncate(time.Millisecond), body) |
| 365 | } |
| 366 | |
amadan