| 1 | package main |
| 2 | |
| 3 | // Unit + in-process integration tests for the activator mesh (mesh.go): HRW |
| 4 | // determinism and minimal redistribution, membership staleness, the |
| 5 | // owner-vs-proxy routing decision, proxy header/Host semantics, drain |
| 6 | // ordering (heartbeat first), the due-sweep ownership gate, and the mesh-off |
| 7 | // tripwire. Two "boxes" here are two activator+router pairs sharing an |
| 8 | // in-memory heartbeat store — the store IS the bucket, which is the whole |
| 9 | // membership design. The real-S3, real-process version (proxy-to-owner and |
| 10 | // the kill-a-box failover, drain and hard-kill variants) is in |
| 11 | // mesh_live_test.go. |
| 12 | |
| 13 | import ( |
| 14 | "context" |
| 15 | "fmt" |
| 16 | "io" |
| 17 | "net" |
| 18 | "net/http" |
| 19 | "net/http/httptest" |
| 20 | "net/http/httputil" |
| 21 | "net/url" |
| 22 | "path/filepath" |
| 23 | "sync" |
| 24 | "sync/atomic" |
| 25 | "testing" |
| 26 | "time" |
| 27 | |
| 28 | "lchat/internal/activator" |
| 29 | ) |
| 30 | |
| 31 | // --- fakes --------------------------------------------------------------------- |
| 32 | |
| 33 | // fakeMeshStore is an in-memory stand-in for the S3 heartbeat prefix. Shared |
| 34 | // between boxes it plays the bucket; rec (optional) feeds a fakeBackend's |
| 35 | // order log so drain tests can pin event ordering. |
| 36 | type fakeMeshStore struct { |
| 37 | mu sync.Mutex |
| 38 | beats map[string]time.Time |
| 39 | regions map[string]string // peer -> advertised region (like the real body's "region") |
| 40 | regionsCall int // how many times Regions was invoked — proves the beat gate |
| 41 | now func() time.Time |
| 42 | rec func(ev string) |
| 43 | } |
| 44 | |
| 45 | func newFakeMeshStore(now func() time.Time) *fakeMeshStore { |
| 46 | return &fakeMeshStore{beats: make(map[string]time.Time), regions: make(map[string]string), now: now} |
| 47 | } |
| 48 | |
| 49 | func (s *fakeMeshStore) Put(_ context.Context, peer, region string) error { |
| 50 | s.mu.Lock() |
| 51 | s.beats[peer] = s.now() |
| 52 | if region != "" { |
| 53 | s.regions[peer] = region // mirror the real store: region omitted when empty |
| 54 | } |
| 55 | rec := s.rec |
| 56 | s.mu.Unlock() |
| 57 | if rec != nil { |
| 58 | rec("hb-put") |
| 59 | } |
| 60 | return nil |
| 61 | } |
| 62 | |
| 63 | func (s *fakeMeshStore) List(context.Context) ([]activator.Heartbeat, error) { |
| 64 | s.mu.Lock() |
| 65 | defer s.mu.Unlock() |
| 66 | out := make([]activator.Heartbeat, 0, len(s.beats)) |
| 67 | for peer, seen := range s.beats { |
| 68 | out = append(out, activator.Heartbeat{Peer: peer, LastSeen: seen}) |
| 69 | } |
| 70 | return out, nil |
| 71 | } |
| 72 | |
| 73 | func (s *fakeMeshStore) Regions(context.Context) (map[string]string, error) { |
| 74 | s.mu.Lock() |
| 75 | defer s.mu.Unlock() |
| 76 | s.regionsCall++ |
| 77 | out := make(map[string]string, len(s.regions)) |
| 78 | for peer, region := range s.regions { |
| 79 | out[peer] = region |
| 80 | } |
| 81 | return out, nil |
| 82 | } |
| 83 | |
| 84 | func (s *fakeMeshStore) regionsCalls() int { |
| 85 | s.mu.Lock() |
| 86 | defer s.mu.Unlock() |
| 87 | return s.regionsCall |
| 88 | } |
| 89 | |
| 90 | func (s *fakeMeshStore) Delete(_ context.Context, peer string) error { |
| 91 | s.mu.Lock() |
| 92 | delete(s.beats, peer) |
| 93 | delete(s.regions, peer) |
| 94 | rec := s.rec |
| 95 | s.mu.Unlock() |
| 96 | if rec != nil { |
| 97 | rec("hb-delete") |
| 98 | } |
| 99 | return nil |
| 100 | } |
| 101 | |
| 102 | func (s *fakeMeshStore) setBeat(peer string, at time.Time) { |
| 103 | s.mu.Lock() |
| 104 | defer s.mu.Unlock() |
| 105 | s.beats[peer] = at |
| 106 | } |
| 107 | |
| 108 | // setRegion advertises a peer's placement region, as if its heartbeat body |
| 109 | // carried it — the fake counterpart of the real box PUTting with a region. |
| 110 | func (s *fakeMeshStore) setRegion(peer, region string) { |
| 111 | s.mu.Lock() |
| 112 | defer s.mu.Unlock() |
| 113 | s.regions[peer] = region |
| 114 | } |
| 115 | |
| 116 | func (s *fakeMeshStore) hasBeat(peer string) bool { |
| 117 | s.mu.Lock() |
| 118 | defer s.mu.Unlock() |
| 119 | _, ok := s.beats[peer] |
| 120 | return ok |
| 121 | } |
| 122 | |
| 123 | // hostOwnedBy generates deterministic host names until one's HRW owner is the |
| 124 | // wanted box — tests can't hardcode hosts because box identities (httptest |
| 125 | // ports) vary per run. |
| 126 | func hostOwnedBy(boxes []string, owner string) string { |
| 127 | for i := 0; ; i++ { |
| 128 | h := fmt.Sprintf("space-%d.test", i) |
| 129 | if hrwOwner(h, boxes) == owner { |
| 130 | return h |
| 131 | } |
| 132 | } |
| 133 | } |
| 134 | |
| 135 | // meshTestBox is one in-process "box": activator (fake backend) + router + |
| 136 | // front door + a live unix-socket listener standing in for any woken instance |
| 137 | // (it echoes the box's name and records what the backend saw). |
| 138 | type meshTestBox struct { |
| 139 | name string |
| 140 | act *activatorSvc |
| 141 | b *fakeBackend |
| 142 | reg *Registry |
| 143 | rt *router |
| 144 | front *httptest.Server |
| 145 | mesh *meshSvc |
| 146 | sock string |
| 147 | |
| 148 | seenHost atomic.Value // last Host the backend instance saw |
| 149 | fwdLeaked atomic.Bool // the forwarded marker reached the backend |
| 150 | } |
| 151 | |
| 152 | func (box *meshTestBox) self() string { return box.front.URL } |
| 153 | |
| 154 | func newMeshTestBox(t *testing.T, name string, store meshStore, clk *fakeClock, domains []string) *meshTestBox { |
| 155 | t.Helper() |
| 156 | box := &meshTestBox{name: name, b: &fakeBackend{}} |
| 157 | |
| 158 | reg, err := openRegistry(filepath.Join(t.TempDir(), "registry.db")) |
| 159 | if err != nil { |
| 160 | t.Fatalf("open registry: %v", err) |
| 161 | } |
| 162 | box.reg = reg |
| 163 | box.act = newActivator(hibernateConfig{Bucket: "test-bucket", Idle: time.Minute}, box.b, reg) |
| 164 | box.act.now = clk.now |
| 165 | |
| 166 | box.sock = tmpSock(t) |
| 167 | ln, err := net.Listen("unix", box.sock) |
| 168 | if err != nil { |
| 169 | t.Fatalf("listen unix: %v", err) |
| 170 | } |
| 171 | srv := &http.Server{Handler: http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) { |
| 172 | box.seenHost.Store(r.Host) |
| 173 | if r.Header.Get(meshForwardedHeader) != "" { |
| 174 | box.fwdLeaked.Store(true) |
| 175 | } |
| 176 | io.WriteString(w, name) |
| 177 | })} |
| 178 | go srv.Serve(ln) |
| 179 | t.Cleanup(func() { srv.Close() }) |
| 180 | |
| 181 | box.rt, err = newRouter(reg, box.act) |
| 182 | if err != nil { |
| 183 | t.Fatalf("new router: %v", err) |
| 184 | } |
| 185 | box.front = httptest.NewServer(box.rt) |
| 186 | t.Cleanup(box.front.Close) |
| 187 | |
| 188 | box.mesh = newMesh(box.front.URL, store, domains) |
| 189 | box.mesh.now = clk.now |
| 190 | box.mesh.reg = reg // region-scoped placement reads home_region from the registry |
| 191 | box.act.mesh = box.mesh |
| 192 | return box |
| 193 | } |
| 194 | |
| 195 | // addHost registers a hibernating route for host on this box (pointing at its |
| 196 | // stand-in instance socket) and reloads the router. |
| 197 | func (box *meshTestBox) addHost(t *testing.T, host string) { |
| 198 | t.Helper() |
| 199 | if err := box.reg.upsert(Route{Host: host, Kind: "instance", Socket: box.sock, Hibernate: true}); err != nil { |
| 200 | t.Fatal(err) |
| 201 | } |
| 202 | if err := box.rt.reload(); err != nil { |
| 203 | t.Fatalf("reload: %v", err) |
| 204 | } |
| 205 | } |
| 206 | |
| 207 | // getVia is proxyGet plus arbitrary headers. |
| 208 | func getVia(t *testing.T, base, host string, hdr map[string]string) (string, int) { |
| 209 | t.Helper() |
| 210 | req, err := http.NewRequest(http.MethodGet, base+"/", nil) |
| 211 | if err != nil { |
| 212 | t.Fatalf("new request: %v", err) |
| 213 | } |
| 214 | req.Host = host |
| 215 | for k, v := range hdr { |
| 216 | req.Header.Set(k, v) |
| 217 | } |
| 218 | resp, err := http.DefaultClient.Do(req) |
| 219 | if err != nil { |
| 220 | t.Fatalf("request: %v", err) |
| 221 | } |
| 222 | defer resp.Body.Close() |
| 223 | b, _ := io.ReadAll(resp.Body) |
| 224 | return string(b), resp.StatusCode |
| 225 | } |
| 226 | |
| 227 | // --- rendezvous hashing ---------------------------------------------------------- |
| 228 | |
| 229 | // hrwOwner is a pure function: same answer for any ordering of the member |
| 230 | // set, on every box, forever — that's what lets placement need no directory. |
| 231 | func TestHRWOwnerDeterministic(t *testing.T) { |
| 232 | boxes := []string{"http://a", "http://b", "http://c"} |
| 233 | perms := [][]string{ |
| 234 | {"http://a", "http://b", "http://c"}, |
| 235 | {"http://c", "http://a", "http://b"}, |
| 236 | {"http://b", "http://c", "http://a"}, |
| 237 | } |
| 238 | owned := map[string]int{} |
| 239 | for i := 0; i < 300; i++ { |
| 240 | h := fmt.Sprintf("host-%d.test", i) |
| 241 | want := hrwOwner(h, boxes) |
| 242 | for _, p := range perms { |
| 243 | if got := hrwOwner(h, p); got != want { |
| 244 | t.Fatalf("owner of %s depends on member order: %q vs %q", h, got, want) |
| 245 | } |
| 246 | } |
| 247 | owned[want]++ |
| 248 | } |
| 249 | // The hash must actually spread load — every box owns a healthy share. |
| 250 | for _, b := range boxes { |
| 251 | if owned[b] < 50 { |
| 252 | t.Fatalf("box %s owns only %d/300 hosts — distribution is broken: %v", b, owned[b], owned) |
| 253 | } |
| 254 | } |
| 255 | if hrwOwner("x.test", []string{"http://solo"}) != "http://solo" { |
| 256 | t.Fatal("single box does not own everything") |
| 257 | } |
| 258 | if hrwOwner("x.test", nil) != "" { |
| 259 | t.Fatal("empty member set should own nothing") |
| 260 | } |
| 261 | } |
| 262 | |
| 263 | // The property that makes membership churn cheap: adding a box moves only the |
| 264 | // ~1/N of hosts it steals (all TO the joiner), removing one moves only the |
| 265 | // leaver's hosts (scattered over survivors) — everyone else stays put, so a |
| 266 | // scale event never stampedes the fleet. |
| 267 | func TestHRWRedistributionMinimal(t *testing.T) { |
| 268 | four := []string{"http://a", "http://b", "http://c", "http://d"} |
| 269 | five := append(append([]string{}, four...), "http://e") |
| 270 | const n = 1000 |
| 271 | |
| 272 | before := make(map[string]string, n) |
| 273 | for i := 0; i < n; i++ { |
| 274 | h := fmt.Sprintf("host-%d.test", i) |
| 275 | before[h] = hrwOwner(h, four) |
| 276 | } |
| 277 | |
| 278 | // Join: every moved host must have moved onto the new box. |
| 279 | moved := 0 |
| 280 | for h, was := range before { |
| 281 | now := hrwOwner(h, five) |
| 282 | if now != was { |
| 283 | moved++ |
| 284 | if now != "http://e" { |
| 285 | t.Fatalf("join moved %s from %s to %s — only the joiner may steal", h, was, now) |
| 286 | } |
| 287 | } |
| 288 | } |
| 289 | if moved < n/10 || moved > n/2 { |
| 290 | t.Fatalf("join moved %d/%d hosts, want ~1/5 (a fifth box should steal about its share)", moved, n) |
| 291 | } |
| 292 | |
| 293 | // Leave: only the leaver's hosts move. |
| 294 | three := []string{"http://a", "http://c", "http://d"} |
| 295 | for h, was := range before { |
| 296 | now := hrwOwner(h, three) |
| 297 | if was == "http://b" { |
| 298 | if now == "http://b" { |
| 299 | t.Fatalf("%s still owned by the removed box", h) |
| 300 | } |
| 301 | } else if now != was { |
| 302 | t.Fatalf("leave moved %s from %s to %s — hosts not owned by the leaver must not move", h, was, now) |
| 303 | } |
| 304 | } |
| 305 | } |
| 306 | |
| 307 | // --- membership -------------------------------------------------------------- |
| 308 | |
| 309 | // Staleness: a heartbeat older than staleAfter means a dead box — judged at |
| 310 | // read time, so a box that dies between polls drops out on the clock, not on |
| 311 | // the next LIST. Self is always a member (a box counts itself even before its |
| 312 | // first PUT lands), and garbage records never enter the hash. |
| 313 | func TestMeshMembershipStaleness(t *testing.T) { |
| 314 | clk := &fakeClock{t: time.Unix(1_700_000_000, 0)} |
| 315 | store := newFakeMeshStore(clk.now) |
| 316 | m := newMesh("http://a", store, nil) |
| 317 | m.now = clk.now |
| 318 | |
| 319 | // Before any beat: membership of one — the degenerate single box. |
| 320 | if got := m.liveBoxes(); len(got) != 1 || got[0] != "http://a" { |
| 321 | t.Fatalf("liveBoxes before any beat = %v, want just self", got) |
| 322 | } |
| 323 | |
| 324 | store.setBeat("http://b", clk.now()) |
| 325 | store.setBeat("this is not a url", clk.now()) // a foreign/garbage record |
| 326 | m.beat() |
| 327 | if !store.hasBeat("http://a") { |
| 328 | t.Fatal("beat did not PUT our own heartbeat") |
| 329 | } |
| 330 | live := m.liveBoxes() |
| 331 | if len(live) != 2 { |
| 332 | t.Fatalf("liveBoxes = %v, want self+b (garbage excluded)", live) |
| 333 | } |
| 334 | |
| 335 | // b goes quiet: past staleAfter it is dead — with NO new list needed. |
| 336 | clk.advance(m.staleAfter + time.Second) |
| 337 | if got := m.liveBoxes(); len(got) != 1 || got[0] != "http://a" { |
| 338 | t.Fatalf("liveBoxes with a stale peer = %v, want just self", got) |
| 339 | } |
| 340 | } |
| 341 | |
| 342 | // The routing decision: proxy only when the HRW owner is another LIVE box; |
| 343 | // own it (wake locally) when the owner is us — or dead, because a stale owner |
| 344 | // is out of the member set and the hash has already reassigned. The lease |
| 345 | // arbitrates if this view is ever wrong; the decision only places. |
| 346 | func TestMeshRoutingDecision(t *testing.T) { |
| 347 | clk := &fakeClock{t: time.Unix(1_700_000_000, 0)} |
| 348 | store := newFakeMeshStore(clk.now) |
| 349 | m := newMesh("http://a", store, nil) |
| 350 | m.now = clk.now |
| 351 | store.setBeat("http://b", clk.now()) |
| 352 | m.beat() |
| 353 | |
| 354 | boxes := []string{"http://a", "http://b"} |
| 355 | theirs := hostOwnedBy(boxes, "http://b") |
| 356 | ours := hostOwnedBy(boxes, "http://a") |
| 357 | |
| 358 | if peer, remote := m.remoteOwner(theirs); !remote || peer != "http://b" { |
| 359 | t.Fatalf("remoteOwner(%s) = %q, %v; want b, true", theirs, peer, remote) |
| 360 | } |
| 361 | if _, remote := m.remoteOwner(ours); remote { |
| 362 | t.Fatalf("remoteOwner(%s) says proxy a host we own", ours) |
| 363 | } |
| 364 | if m.owns(theirs) || !m.owns(ours) { |
| 365 | t.Fatal("owns() disagrees with remoteOwner()") |
| 366 | } |
| 367 | |
| 368 | // The owner dies: its hosts fall through to a local wake. |
| 369 | clk.advance(m.staleAfter + time.Second) |
| 370 | if _, remote := m.remoteOwner(theirs); remote { |
| 371 | t.Fatal("remoteOwner still proxies to a stale (dead) box — its hosts must wake locally") |
| 372 | } |
| 373 | if !m.owns(theirs) { |
| 374 | t.Fatal("survivor does not own the dead box's host") |
| 375 | } |
| 376 | } |
| 377 | |
| 378 | // --- domain scoping ------------------------------------------------------------ |
| 379 | |
| 380 | // inScope's matching rule: exact domain or a proper subdomain of one, case- |
| 381 | // insensitive (hosts and configured domains alike); nothing else. |
| 382 | func TestMeshInScope(t *testing.T) { |
| 383 | m := newMesh("http://a", newFakeMeshStore(time.Now), []string{"11ch.at", "11relay.com"}) |
| 384 | cases := []struct { |
| 385 | host string |
| 386 | want bool |
| 387 | }{ |
| 388 | {"11ch.at", true}, // the bare domain itself |
| 389 | {"abc.11ch.at", true}, // a subdomain |
| 390 | {"ABC.11CH.AT", true}, // case-insensitive |
| 391 | {"11relay.com", true}, |
| 392 | {"abc.11relay.com", true}, |
| 393 | {"family.luan.chat", false}, // a different domain entirely |
| 394 | {"evil11ch.at", false}, // must not match on a bare suffix |
| 395 | {"11ch.at.evil.com", false}, // must not match as a prefix |
| 396 | } |
| 397 | for _, c := range cases { |
| 398 | if got := m.inScope(c.host); got != c.want { |
| 399 | t.Errorf("inScope(%q) = %v, want %v", c.host, got, c.want) |
| 400 | } |
| 401 | } |
| 402 | } |
| 403 | |
| 404 | // An out-of-scope host always resolves locally, regardless of what HRW would |
| 405 | // otherwise say — it's never proxied away, and this box always reports owning |
| 406 | // it (the sweep's due-wake gate must never skip it). Domain scoping exists |
| 407 | // because a registry row for an out-of-scope host is only ever replicated to |
| 408 | // the box(es) it already lives on (control.go's registerInstance): a host in |
| 409 | // scope legitimately has rows on every mesh member and needs real HRW |
| 410 | // placement; a host out of scope does not, and hashing it elsewhere would |
| 411 | // just 404 there. |
| 412 | func TestMeshDomainScopingBypassesHRW(t *testing.T) { |
| 413 | clk := &fakeClock{t: time.Unix(1_700_000_000, 0)} |
| 414 | store := newFakeMeshStore(clk.now) |
| 415 | m := newMesh("http://a", store, []string{"11ch.at"}) |
| 416 | m.now = clk.now |
| 417 | store.setBeat("http://b", clk.now()) |
| 418 | m.beat() |
| 419 | |
| 420 | boxes := []string{"http://a", "http://b"} |
| 421 | // A host under the in-scope domain that genuinely hashes to the peer. |
| 422 | // hostOwnedBy's "space-N.test" hosts fall under no configured domain, so |
| 423 | // build one under 11ch.at directly. |
| 424 | var inScopeTheirs string |
| 425 | for i := 0; ; i++ { |
| 426 | h := fmt.Sprintf("space-%d.11ch.at", i) |
| 427 | if hrwOwner(h, boxes) == "http://b" { |
| 428 | inScopeTheirs = h |
| 429 | break |
| 430 | } |
| 431 | } |
| 432 | if peer, remote := m.remoteOwner(inScopeTheirs); !remote || peer != "http://b" { |
| 433 | t.Fatalf("in-scope host: remoteOwner = %q, %v; want b, true", peer, remote) |
| 434 | } |
| 435 | |
| 436 | // The same host name, but under a domain the mesh isn't scoped to. Find |
| 437 | // one that HRW would (incorrectly, if honored) assign to the peer. |
| 438 | var outOfScope string |
| 439 | for i := 0; ; i++ { |
| 440 | h := fmt.Sprintf("family-%d.luan.chat", i) |
| 441 | if hrwOwner(h, boxes) == "http://b" { |
| 442 | outOfScope = h |
| 443 | break |
| 444 | } |
| 445 | } |
| 446 | if _, remote := m.remoteOwner(outOfScope); remote { |
| 447 | t.Fatalf("out-of-scope host %q was proxied away — domain scoping did not apply", outOfScope) |
| 448 | } |
| 449 | if !m.owns(outOfScope) { |
| 450 | t.Fatalf("out-of-scope host %q: owns() = false — the due-wake gate would wrongly skip it", outOfScope) |
| 451 | } |
| 452 | } |
| 453 | |
| 454 | // --- region placement (docs/multi-edge.md, Phase 1) ----------------------------- |
| 455 | |
| 456 | // candidatesFor scopes HRW to the live boxes whose region matches a space's |
| 457 | // home_region, with an all-boxes fallback that keeps pre-region behaviour |
| 458 | // byte-identical. One setup, four rules the task pins: |
| 459 | // (a) a space pinned to a region routes to that region's box; |
| 460 | // (b) a space with empty home_region falls back to all-boxes HRW (== today); |
| 461 | // (c) a home_region no live box advertises falls back to all-boxes; |
| 462 | // (d) a peer that advertises no region is treated as agnostic (only ever a |
| 463 | // fallback candidate, never a region match). |
| 464 | func TestMeshRegionPlacement(t *testing.T) { |
| 465 | clk := &fakeClock{t: time.Unix(1_700_000_000, 0)} |
| 466 | store := newFakeMeshStore(clk.now) |
| 467 | reg, err := openRegistry(filepath.Join(t.TempDir(), "registry.db")) |
| 468 | if err != nil { |
| 469 | t.Fatalf("open registry: %v", err) |
| 470 | } |
| 471 | |
| 472 | // Three live boxes: self in "ie", a peer in "fi", and an agnostic peer that |
| 473 | // advertises no region at all (rule d). |
| 474 | m := newMesh("http://a", store, []string{"11ch.at"}) |
| 475 | m.now = clk.now |
| 476 | m.reg = reg |
| 477 | m.region = "ie" |
| 478 | store.setBeat("http://b", clk.now()) |
| 479 | store.setRegion("http://b", "fi") |
| 480 | store.setBeat("http://c", clk.now()) // agnostic peer — no setRegion |
| 481 | m.beat() // our PUT records self's region; List+Regions load the view |
| 482 | |
| 483 | all := []string{"http://a", "http://b", "http://c"} |
| 484 | if got := m.liveBoxes(); !sameSet(got, all) { |
| 485 | t.Fatalf("liveBoxes = %v, want %v", got, all) |
| 486 | } |
| 487 | |
| 488 | // A helper to register an in-scope space with a given home_region. |
| 489 | pin := func(host, region string) { |
| 490 | if err := reg.upsert(Route{Host: host, Kind: "instance", Socket: "/x.sock", Hibernate: true, HomeRegion: region}); err != nil { |
| 491 | t.Fatalf("upsert %s: %v", host, err) |
| 492 | } |
| 493 | } |
| 494 | |
| 495 | // (a) Pinned to "fi" → only the fi box is a candidate, so it owns and every |
| 496 | // other box proxies to it, regardless of what bare HRW over all three says. |
| 497 | pin("fi-space.11ch.at", "fi") |
| 498 | if got := m.candidatesFor("fi-space.11ch.at"); !sameSet(got, []string{"http://b"}) { |
| 499 | t.Fatalf("home_region fi: candidates = %v, want [http://b]", got) |
| 500 | } |
| 501 | if m.owns("fi-space.11ch.at") { |
| 502 | t.Fatal("box a wrongly owns an fi-pinned space") |
| 503 | } |
| 504 | if peer, remote := m.remoteOwner("fi-space.11ch.at"); !remote || peer != "http://b" { |
| 505 | t.Fatalf("remoteOwner(fi-space) = %q,%v; want http://b,true", peer, remote) |
| 506 | } |
| 507 | |
| 508 | // Self-region match: a space pinned to our own "ie" is owned locally. |
| 509 | pin("ie-space.11ch.at", "ie") |
| 510 | if got := m.candidatesFor("ie-space.11ch.at"); !sameSet(got, []string{"http://a"}) { |
| 511 | t.Fatalf("home_region ie: candidates = %v, want [http://a]", got) |
| 512 | } |
| 513 | if !m.owns("ie-space.11ch.at") { |
| 514 | t.Fatal("box a does not own an ie-pinned space it is the region for") |
| 515 | } |
| 516 | |
| 517 | // (b) Empty home_region (a host with no row at all is empty too) → all boxes, |
| 518 | // and the chosen owner is identical to bare HRW over the whole live set. |
| 519 | noRegion := "plain.11ch.at" // no reg row → homeRegionOf "" |
| 520 | if got := m.candidatesFor(noRegion); !sameSet(got, all) { |
| 521 | t.Fatalf("empty home_region: candidates = %v, want all %v", got, all) |
| 522 | } |
| 523 | if hrwOwner(noRegion, m.candidatesFor(noRegion)) != hrwOwner(noRegion, m.liveBoxes()) { |
| 524 | t.Fatal("empty home_region must place identically to pre-region HRW") |
| 525 | } |
| 526 | |
| 527 | // (c) A home_region no live box advertises → fall back to all boxes. |
| 528 | pin("de-space.11ch.at", "de") |
| 529 | if got := m.candidatesFor("de-space.11ch.at"); !sameSet(got, all) { |
| 530 | t.Fatalf("unmatched home_region de: candidates = %v, want all %v", got, all) |
| 531 | } |
| 532 | |
| 533 | // (d) The agnostic peer "http://c" is never a region match (it advertised |
| 534 | // none), yet is present in every all-boxes fallback above — already asserted |
| 535 | // by the sameSet(all) checks that include it. |
| 536 | } |
| 537 | |
| 538 | // TestMeshRegionAgnosticUnchanged pins the core back-compat promise directly: a |
| 539 | // mesh with no region set and no registry (exactly how the existing unit tests |
| 540 | // and today's fleet run) has candidatesFor == liveBoxes for every host, so |
| 541 | // owns/remoteOwner cannot diverge from pre-region behaviour. |
| 542 | func TestMeshRegionAgnosticUnchanged(t *testing.T) { |
| 543 | clk := &fakeClock{t: time.Unix(1_700_000_000, 0)} |
| 544 | store := newFakeMeshStore(clk.now) |
| 545 | m := newMesh("http://a", store, []string{"11ch.at"}) |
| 546 | m.now = clk.now // no region, no reg — the pre-region shape |
| 547 | store.setBeat("http://b", clk.now()) |
| 548 | m.beat() |
| 549 | |
| 550 | for i := 0; i < 50; i++ { |
| 551 | host := fmt.Sprintf("space-%d.11ch.at", i) |
| 552 | if !sameSet(m.candidatesFor(host), m.liveBoxes()) { |
| 553 | t.Fatalf("region-agnostic mesh scoped %s away from the full live set", host) |
| 554 | } |
| 555 | } |
| 556 | } |
| 557 | |
| 558 | // TestMeshBeatRegionFetchGate proves the zero-cost-when-inert promise: a box |
| 559 | // with no region declared (prod today) fetches NO peer regions on a beat, so |
| 560 | // region placement adds zero S3 requests over pre-region behaviour; a box that |
| 561 | // declares a region does fetch them. |
| 562 | func TestMeshBeatRegionFetchGate(t *testing.T) { |
| 563 | clk := &fakeClock{t: time.Unix(1_700_000_000, 0)} |
| 564 | |
| 565 | // Agnostic box: beats never touch Regions. |
| 566 | store := newFakeMeshStore(clk.now) |
| 567 | agnostic := newMesh("http://a", store, []string{"11ch.at"}) |
| 568 | agnostic.now = clk.now // no region |
| 569 | agnostic.beat() |
| 570 | agnostic.beat() |
| 571 | if got := store.regionsCalls(); got != 0 { |
| 572 | t.Fatalf("agnostic box made %d Regions call(s) across two beats, want 0 — an inert feature must cost nothing", got) |
| 573 | } |
| 574 | |
| 575 | // Region-scoped box: each beat fetches peer regions. |
| 576 | store2 := newFakeMeshStore(clk.now) |
| 577 | scoped := newMesh("http://b", store2, []string{"11ch.at"}) |
| 578 | scoped.now = clk.now |
| 579 | scoped.region = "ie" |
| 580 | scoped.beat() |
| 581 | scoped.beat() |
| 582 | if got := store2.regionsCalls(); got != 2 { |
| 583 | t.Fatalf("region-scoped box made %d Regions call(s) across two beats, want 2", got) |
| 584 | } |
| 585 | } |
| 586 | |
| 587 | // sameSet compares two peer slices ignoring order (liveBoxes/candidatesFor draw |
| 588 | // from a map, so order is not defined). |
| 589 | func sameSet(a, b []string) bool { |
| 590 | if len(a) != len(b) { |
| 591 | return false |
| 592 | } |
| 593 | seen := make(map[string]int, len(a)) |
| 594 | for _, x := range a { |
| 595 | seen[x]++ |
| 596 | } |
| 597 | for _, y := range b { |
| 598 | seen[y]-- |
| 599 | } |
| 600 | for _, n := range seen { |
| 601 | if n != 0 { |
| 602 | return false |
| 603 | } |
| 604 | } |
| 605 | return true |
| 606 | } |
| 607 | |
| 608 | // meshFromEnv reads LCHAT_MESH_REGION (optional): set = this box's placement |
| 609 | // region, unset = region-agnostic (the empty string, today's behaviour). |
| 610 | func TestMeshFromEnvRegion(t *testing.T) { |
| 611 | t.Setenv("LCHAT_MESH_SELF", "https://box-a.internal") |
| 612 | t.Setenv("LCHAT_MESH_DOMAINS", "11ch.at") |
| 613 | t.Setenv("LCHAT_MESH_TOKEN", "shared-secret") |
| 614 | |
| 615 | t.Setenv("LCHAT_MESH_REGION", "") |
| 616 | m, err := meshFromEnv(&hibernateConfig{Bucket: "b", Region: "eu-west-1"}) |
| 617 | if err != nil || m == nil { |
| 618 | t.Fatalf("unset region: %v, %v", m, err) |
| 619 | } |
| 620 | if m.region != "" { |
| 621 | t.Fatalf("region = %q, want \"\" (agnostic) when LCHAT_MESH_REGION is unset", m.region) |
| 622 | } |
| 623 | |
| 624 | t.Setenv("LCHAT_MESH_REGION", " ie ") |
| 625 | m, err = meshFromEnv(&hibernateConfig{Bucket: "b", Region: "eu-west-1"}) |
| 626 | if err != nil || m == nil { |
| 627 | t.Fatalf("set region: %v, %v", m, err) |
| 628 | } |
| 629 | if m.region != "ie" { |
| 630 | t.Fatalf("region = %q, want the trimmed \"ie\"", m.region) |
| 631 | } |
| 632 | } |
| 633 | |
| 634 | // meshFromEnv refuses to start the mesh without an explicit domain allowlist — |
| 635 | // failing loudly beats silently letting HRW assign a host to a box that has |
| 636 | // no registry row for it (see the inScope doc comment). |
| 637 | func TestMeshFromEnvRequiresDomains(t *testing.T) { |
| 638 | t.Setenv("LCHAT_MESH_SELF", "https://box-a.internal") |
| 639 | t.Setenv("LCHAT_MESH_TOKEN", "shared-secret") |
| 640 | t.Setenv("LCHAT_MESH_DOMAINS", "") |
| 641 | if _, err := meshFromEnv(&hibernateConfig{Bucket: "b", Region: "eu-west-1"}); err == nil { |
| 642 | t.Fatal("mesh without LCHAT_MESH_DOMAINS should be a config error") |
| 643 | } |
| 644 | |
| 645 | t.Setenv("LCHAT_MESH_DOMAINS", "11ch.at, 11relay.com") |
| 646 | m, err := meshFromEnv(&hibernateConfig{Bucket: "b", Region: "eu-west-1"}) |
| 647 | if err != nil || m == nil { |
| 648 | t.Fatalf("valid domains: %v, %v", m, err) |
| 649 | } |
| 650 | if len(m.domains) != 2 || m.domains[0] != "11ch.at" || m.domains[1] != "11relay.com" { |
| 651 | t.Fatalf("domains = %v, want the parsed, trimmed, lowercased list", m.domains) |
| 652 | } |
| 653 | } |
| 654 | |
| 655 | // meshFromEnv refuses to start the mesh without a shared route-replication |
| 656 | // secret — an unauthenticated /api/mesh/route would let any caller hand this |
| 657 | // box arbitrary routes (see meshreplicate.go). |
| 658 | func TestMeshFromEnvRequiresToken(t *testing.T) { |
| 659 | t.Setenv("LCHAT_MESH_SELF", "https://box-a.internal") |
| 660 | t.Setenv("LCHAT_MESH_DOMAINS", "11ch.at") |
| 661 | t.Setenv("LCHAT_MESH_TOKEN", "") |
| 662 | if _, err := meshFromEnv(&hibernateConfig{Bucket: "b", Region: "eu-west-1"}); err == nil { |
| 663 | t.Fatal("mesh without LCHAT_MESH_TOKEN should be a config error") |
| 664 | } |
| 665 | |
| 666 | t.Setenv("LCHAT_MESH_TOKEN", "shared-secret") |
| 667 | m, err := meshFromEnv(&hibernateConfig{Bucket: "b", Region: "eu-west-1"}) |
| 668 | if err != nil || m == nil { |
| 669 | t.Fatalf("valid token: %v, %v", m, err) |
| 670 | } |
| 671 | if m.token != "shared-secret" { |
| 672 | t.Fatalf("token = %q, want it read through", m.token) |
| 673 | } |
| 674 | } |
| 675 | |
| 676 | // --- the proxy hop ------------------------------------------------------------- |
| 677 | |
| 678 | // Scheme/Host/SNI semantics of the peer hop: the instance hostname survives |
| 679 | // (the peer routes by Host); for https peers the outgoing URL keeps the |
| 680 | // instance host so SNI and cert verification run against a name the peer can |
| 681 | // actually prove (its wildcard/per-host cert), while the dialer pins TCP to |
| 682 | // the peer's address; for http peers the URL targets the peer directly. The |
| 683 | // forwarded marker rides every hop. InsecureSkipVerify appears nowhere. |
| 684 | func TestPeerProxyRewrite(t *testing.T) { |
| 685 | rewrite := func(p *httputil.ReverseProxy, host string) *http.Request { |
| 686 | in := httptest.NewRequest("GET", "http://placeholder/api/x", nil) |
| 687 | in.Host = host |
| 688 | in.RemoteAddr = "192.0.2.7:4711" |
| 689 | out := in.Clone(context.Background()) |
| 690 | p.Rewrite(&httputil.ProxyRequest{In: in, Out: out}) |
| 691 | return out |
| 692 | } |
| 693 | |
| 694 | httpsPeer, _ := url.Parse("https://box-b.internal") |
| 695 | out := rewrite(newPeerProxy("http://self", httpsPeer), "space.11ch.at") |
| 696 | if out.URL.Scheme != "https" || out.URL.Host != "space.11ch.at" || out.Host != "space.11ch.at" { |
| 697 | t.Fatalf("https hop: scheme=%s urlHost=%s host=%s — SNI/verification must target the instance host", out.URL.Scheme, out.URL.Host, out.Host) |
| 698 | } |
| 699 | if got := peerDialAddr(httpsPeer); got != "box-b.internal:443" { |
| 700 | t.Fatalf("https dial addr = %q, want the peer's address", got) |
| 701 | } |
| 702 | if out.Header.Get(meshForwardedHeader) != "http://self" { |
| 703 | t.Fatal("forwarded marker missing — the one-hop loop guard would not hold") |
| 704 | } |
| 705 | |
| 706 | httpPeer, _ := url.Parse("http://10.0.0.5:9000") |
| 707 | out = rewrite(newPeerProxy("http://self", httpPeer), "space.11ch.at") |
| 708 | if out.URL.Scheme != "http" || out.URL.Host != "10.0.0.5:9000" || out.Host != "space.11ch.at" { |
| 709 | t.Fatalf("http hop: scheme=%s urlHost=%s host=%s — dial the peer, Host stays the instance", out.URL.Scheme, out.URL.Host, out.Host) |
| 710 | } |
| 711 | if got := peerDialAddr(httpPeer); got != "10.0.0.5:9000" { |
| 712 | t.Fatalf("http dial addr = %q", got) |
| 713 | } |
| 714 | bare, _ := url.Parse("http://box-c.internal") |
| 715 | if got := peerDialAddr(bare); got != "box-c.internal:80" { |
| 716 | t.Fatalf("default http port: %q", got) |
| 717 | } |
| 718 | } |
| 719 | |
| 720 | // Two boxes, one shared "bucket": a request landing on the non-owner is |
| 721 | // reverse-proxied to the owner (which wakes the instance), the instance |
| 722 | // hostname survives the hop, the forwarded marker never reaches a backend, |
| 723 | // a request already carrying the marker is served where it stands (one hop |
| 724 | // max), and a host with a live local entry stays local even when the hash |
| 725 | // points elsewhere. |
| 726 | func TestMeshProxyToOwner(t *testing.T) { |
| 727 | clk := &fakeClock{t: time.Unix(1_700_000_000, 0)} |
| 728 | store := newFakeMeshStore(clk.now) |
| 729 | box1 := newMeshTestBox(t, "box1", store, clk, nil) |
| 730 | box2 := newMeshTestBox(t, "box2", store, clk, nil) |
| 731 | // Beat both (twice for box1, whose first beat predates box2's PUT) so |
| 732 | // each box sees the full membership. |
| 733 | box1.mesh.beat() |
| 734 | box2.mesh.beat() |
| 735 | box1.mesh.beat() |
| 736 | |
| 737 | boxes := []string{box1.self(), box2.self()} |
| 738 | host := hostOwnedBy(boxes, box2.self()) |
| 739 | box1.addHost(t, host) |
| 740 | box2.addHost(t, host) |
| 741 | |
| 742 | // (a) Request to the NON-owner: proxied to the owner and served there. |
| 743 | if body, code := proxyGet(t, box1.front.URL, host); code != 200 || body != "box2" { |
| 744 | t.Fatalf("non-owner request: got %d %q, want 200 box2 (proxied to owner)", code, body) |
| 745 | } |
| 746 | if acq, _, _ := box2.b.counts(); acq != 1 { |
| 747 | t.Fatalf("owner acquires = %d, want 1 (the wake happened on the owner)", acq) |
| 748 | } |
| 749 | if acq, _, starts := box1.b.counts(); acq != 0 || starts != 0 { |
| 750 | t.Fatalf("non-owner touched its own activator (acquires=%d starts=%d)", acq, starts) |
| 751 | } |
| 752 | if got, _ := box2.seenHost.Load().(string); got != host { |
| 753 | t.Fatalf("owner backend saw Host %q, want %q — the hostname must survive the hop", got, host) |
| 754 | } |
| 755 | if box2.fwdLeaked.Load() { |
| 756 | t.Fatal("the forwarded marker leaked through to the backend") |
| 757 | } |
| 758 | |
| 759 | // (b) Request straight to the owner: local, no extra wake. |
| 760 | if body, code := proxyGet(t, box2.front.URL, host); code != 200 || body != "box2" { |
| 761 | t.Fatalf("owner request: got %d %q", code, body) |
| 762 | } |
| 763 | if acq, _, _ := box2.b.counts(); acq != 1 { |
| 764 | t.Fatalf("owner re-acquired for a host it already has awake (acquires=%d)", acq) |
| 765 | } |
| 766 | |
| 767 | // (c) One hop max: a request already carrying the forwarded marker is |
| 768 | // served where it stands — here the non-owner, which wakes it locally |
| 769 | // (the lease would arbitrate a real conflict; the fake just observes). |
| 770 | var host2 string // a second box2-owned host, distinct from the first |
| 771 | for i := 0; ; i++ { |
| 772 | if h := fmt.Sprintf("space-b-%d.test", i); hrwOwner(h, boxes) == box2.self() { |
| 773 | host2 = h |
| 774 | break |
| 775 | } |
| 776 | } |
| 777 | box1.addHost(t, host2) |
| 778 | box2.addHost(t, host2) |
| 779 | if body, code := getVia(t, box1.front.URL, host2, map[string]string{meshForwardedHeader: "http://elsewhere"}); code != 200 || body != "box1" { |
| 780 | t.Fatalf("forwarded request: got %d %q, want 200 box1 (served in place, never re-proxied)", code, body) |
| 781 | } |
| 782 | if acq, _, _ := box1.b.counts(); acq != 1 { |
| 783 | t.Fatalf("non-owner acquires = %d after forwarded request, want 1", acq) |
| 784 | } |
| 785 | |
| 786 | // (d) Live-local-entry rule: host2 is now awake on box1, so an ordinary |
| 787 | // request to box1 stays local despite the hash pointing at box2 — an |
| 788 | // awake instance finishes its life where it is. |
| 789 | if body, code := proxyGet(t, box1.front.URL, host2); code != 200 || body != "box1" { |
| 790 | t.Fatalf("live-entry request: got %d %q, want 200 box1 (local entry wins over the hint)", code, body) |
| 791 | } |
| 792 | if acq, _, _ := box2.b.counts(); acq != 1 { |
| 793 | t.Fatalf("owner was woken for a host living on its peer (acquires=%d)", acq) |
| 794 | } |
| 795 | } |
| 796 | |
| 797 | // --- mesh off: the tripwire ------------------------------------------------------ |
| 798 | |
| 799 | // With the mesh off (LCHAT_MESH_SELF unset → act.mesh nil) the activator is |
| 800 | // byte-identical to the single-box shape: a host whose HRW owner WOULD be a |
| 801 | // peer is woken locally, nothing is proxied, and no heartbeat is ever |
| 802 | // written. The same box with only the mesh flipped on proxies the same-shaped |
| 803 | // request — proving the nil check is the single divergence point. |
| 804 | func TestMeshOffTripwire(t *testing.T) { |
| 805 | clk := &fakeClock{t: time.Unix(1_700_000_000, 0)} |
| 806 | store := newFakeMeshStore(clk.now) |
| 807 | box1 := newMeshTestBox(t, "box1", store, clk, nil) |
| 808 | box2 := newMeshTestBox(t, "box2", store, clk, nil) |
| 809 | box1.act.mesh = nil // the tripwire: box1 is mesh-off |
| 810 | box2.mesh.beat() // box2 participates; its heartbeat is in the store |
| 811 | |
| 812 | boxes := []string{box1.self(), box2.self()} |
| 813 | host := hostOwnedBy(boxes, box2.self()) |
| 814 | box1.addHost(t, host) |
| 815 | box2.addHost(t, host) |
| 816 | |
| 817 | if body, code := proxyGet(t, box1.front.URL, host); code != 200 || body != "box1" { |
| 818 | t.Fatalf("mesh-off request: got %d %q, want 200 box1 (woken locally, never proxied)", code, body) |
| 819 | } |
| 820 | if acq, _, _ := box1.b.counts(); acq != 1 { |
| 821 | t.Fatalf("mesh-off box acquires = %d, want 1", acq) |
| 822 | } |
| 823 | if acq, _, _ := box2.b.counts(); acq != 0 { |
| 824 | t.Fatalf("peer was touched by a mesh-off box (acquires=%d)", acq) |
| 825 | } |
| 826 | if store.hasBeat(box1.self()) { |
| 827 | t.Fatal("mesh-off box wrote a heartbeat") |
| 828 | } |
| 829 | |
| 830 | // Contrast: flip the mesh on — the ONLY change — and the same-shaped |
| 831 | // request (a fresh host, since the first now has a live local entry) |
| 832 | // proxies to the owner. |
| 833 | box1.act.mesh = box1.mesh |
| 834 | box1.mesh.beat() |
| 835 | var host2 string |
| 836 | for i := 0; ; i++ { |
| 837 | h := fmt.Sprintf("contrast-%d.test", i) |
| 838 | if hrwOwner(h, boxes) == box2.self() { |
| 839 | host2 = h |
| 840 | break |
| 841 | } |
| 842 | } |
| 843 | box1.addHost(t, host2) |
| 844 | box2.addHost(t, host2) |
| 845 | if body, code := proxyGet(t, box1.front.URL, host2); code != 200 || body != "box2" { |
| 846 | t.Fatalf("mesh-on contrast: got %d %q, want 200 box2 (proxied)", code, body) |
| 847 | } |
| 848 | } |
| 849 | |
| 850 | // meshFromEnv: unset = off (nil, nil — the zero-change default); set without |
| 851 | // the hibernation bucket = a loud config error; malformed URLs rejected. |
| 852 | func TestMeshFromEnv(t *testing.T) { |
| 853 | // A valid domain allowlist and token throughout — |
| 854 | // TestMeshFromEnvRequiresDomains/TestMeshFromEnvRequiresToken cover their |
| 855 | // missing cases on their own. |
| 856 | t.Setenv("LCHAT_MESH_DOMAINS", "11ch.at") |
| 857 | t.Setenv("LCHAT_MESH_TOKEN", "shared-secret") |
| 858 | |
| 859 | t.Setenv("LCHAT_MESH_SELF", "") |
| 860 | if m, err := meshFromEnv(&hibernateConfig{Bucket: "b"}); err != nil || m != nil { |
| 861 | t.Fatalf("unset: got %v, %v; want nil, nil (mesh off)", m, err) |
| 862 | } |
| 863 | |
| 864 | t.Setenv("LCHAT_MESH_SELF", "https://box-a.internal") |
| 865 | if _, err := meshFromEnv(nil); err == nil { |
| 866 | t.Fatal("mesh without a hibernation bucket should be a config error") |
| 867 | } |
| 868 | |
| 869 | for _, bad := range []string{"box-a.internal", "ftp://box-a", "https://box-a/path"} { |
| 870 | t.Setenv("LCHAT_MESH_SELF", bad) |
| 871 | if _, err := meshFromEnv(&hibernateConfig{Bucket: "b", Region: "eu-west-1"}); err == nil { |
| 872 | t.Fatalf("LCHAT_MESH_SELF %q should be rejected", bad) |
| 873 | } |
| 874 | } |
| 875 | |
| 876 | t.Setenv("LCHAT_MESH_SELF", "https://box-a.internal/") |
| 877 | m, err := meshFromEnv(&hibernateConfig{Bucket: "b", Region: "eu-west-1"}) |
| 878 | if err != nil || m == nil { |
| 879 | t.Fatalf("valid self: %v, %v", m, err) |
| 880 | } |
| 881 | if m.self != "https://box-a.internal" { |
| 882 | t.Fatalf("self = %q, want the trailing slash trimmed", m.self) |
| 883 | } |
| 884 | } |
| 885 | |
| 886 | // --- drain ordering --------------------------------------------------------------- |
| 887 | |
| 888 | // The drain contract, order-pinned: the heartbeat is deleted FIRST — before |
| 889 | // the next-due probe, before the child stops, before the final checkpoint and |
| 890 | // lease release — so peers take over the moment the leases free up. Same code |
| 891 | // path for deploys, box failure drills, and scale-down. |
| 892 | func TestMeshDrainDeletesHeartbeatFirst(t *testing.T) { |
| 893 | a, b, clk, _ := testActivator(t, time.Hour) |
| 894 | store := newFakeMeshStore(clk.now) |
| 895 | store.rec = b.record // interleave heartbeat ops with the backend's order log |
| 896 | m := newMesh("http://box-a", store, nil) |
| 897 | m.now = clk.now |
| 898 | m.interval = time.Hour // the loop must never beat on its own mid-test |
| 899 | a.mesh = m |
| 900 | m.start() |
| 901 | |
| 902 | release, err := a.beginRequest(context.Background(), testRoute("drain.example.com")) |
| 903 | if err != nil { |
| 904 | t.Fatalf("beginRequest: %v", err) |
| 905 | } |
| 906 | release() |
| 907 | |
| 908 | ctx, cancel := context.WithTimeout(context.Background(), 5*time.Second) |
| 909 | defer cancel() |
| 910 | a.shutdown(ctx) |
| 911 | |
| 912 | if store.hasBeat("http://box-a") { |
| 913 | t.Fatal("drain left our heartbeat behind") |
| 914 | } |
| 915 | ev := b.events() |
| 916 | del := indexOf(ev, "hb-delete") |
| 917 | if del == -1 { |
| 918 | t.Fatalf("no heartbeat delete in %v", ev) |
| 919 | } |
| 920 | if put := indexOf(ev, "hb-put"); put == -1 || put > del { |
| 921 | t.Fatalf("heartbeat lifecycle out of order: %v", ev) |
| 922 | } |
| 923 | for _, later := range []string{"nextDue", "stop:instance"} { |
| 924 | if i := indexOf(ev, later); i == -1 || i < del { |
| 925 | t.Fatalf("drain order wrong: %q at %d, hb-delete at %d — the heartbeat must go first (%v)", later, i, del, ev) |
| 926 | } |
| 927 | } |
| 928 | } |
| 929 | |
| 930 | // --- due-time sweeps under the mesh ----------------------------------------------- |
| 931 | |
| 932 | // Only the HRW owner due-wakes a host; a non-owner skips it every tick. When |
| 933 | // the owner dies, staleness reassigns ownership and the next sweep fires the |
| 934 | // wake here — failover for schedules is the same eventual path as requests. |
| 935 | func TestMeshDueSweepOnlyOwner(t *testing.T) { |
| 936 | a, b, clk, reg := testActivator(t, 10*time.Second) |
| 937 | store := newFakeMeshStore(clk.now) |
| 938 | m := newMesh("http://box-a", store, nil) |
| 939 | m.now = clk.now |
| 940 | a.mesh = m |
| 941 | store.setBeat("http://box-b", clk.now()) |
| 942 | m.beat() |
| 943 | |
| 944 | host := hostOwnedBy([]string{"http://box-a", "http://box-b"}, "http://box-b") |
| 945 | route := testRoute(host) |
| 946 | if err := reg.upsert(route); err != nil { |
| 947 | t.Fatal(err) |
| 948 | } |
| 949 | if err := reg.setNextDue(host, clk.now().Add(-time.Second).Unix()); err != nil { |
| 950 | t.Fatal(err) |
| 951 | } |
| 952 | |
| 953 | // The peer owns it: our sweeps must not wake it, however overdue. |
| 954 | for i := 0; i < 3; i++ { |
| 955 | a.sweepOnce() |
| 956 | } |
| 957 | if _, _, starts := b.counts(); starts != 0 { |
| 958 | t.Fatalf("starts=%d — a non-owner due-woke a host", starts) |
| 959 | } |
| 960 | |
| 961 | // The owner dies (heartbeat goes stale): ownership reassigns to us on the |
| 962 | // next read, and the sweep fires the wake. |
| 963 | clk.advance(m.staleAfter + time.Second) |
| 964 | a.sweepOnce() |
| 965 | waitFor(t, "post-failover due wake", func() bool { return a.entryAwake(host) }) |
| 966 | if _, _, starts := b.counts(); starts != 1 { |
| 967 | t.Fatalf("starts=%d after owner death, want 1", starts) |
| 968 | } |
| 969 | } |
| 970 | |
amadan