| 1 | package session |
| 2 | |
| 3 | import ( |
| 4 | "context" |
| 5 | "errors" |
| 6 | "fmt" |
| 7 | "sync" |
| 8 | "time" |
| 9 | ) |
| 10 | |
| 11 | // HostID returns the immutable host namespace used to validate SessionRef. |
| 12 | func (s *Service) HostID() string { |
| 13 | if s == nil { |
| 14 | return "" |
| 15 | } |
| 16 | return s.hostID |
| 17 | } |
| 18 | |
| 19 | type prepareRuntime struct{ done chan struct{} } |
| 20 | |
| 21 | // ClientBinding grants a caller access to one exact published runtime without |
| 22 | // granting authority to close the host-owned writer. Release only detaches this |
| 23 | // client; the host retires an idle runtime after its last binding disappears. |
| 24 | type ClientBinding struct { |
| 25 | service *Service |
| 26 | runtime *Runtime |
| 27 | once sync.Once |
| 28 | err error |
| 29 | } |
| 30 | |
| 31 | func (b *ClientBinding) Runtime() *Runtime { |
| 32 | if b == nil { |
| 33 | return nil |
| 34 | } |
| 35 | return b.runtime |
| 36 | } |
| 37 | |
| 38 | func (b *ClientBinding) Release(ctx context.Context) error { |
| 39 | if b == nil || b.service == nil || b.runtime == nil { |
| 40 | return nil |
| 41 | } |
| 42 | b.once.Do(func() { b.err = b.service.releaseBinding(ctx, b.runtime) }) |
| 43 | return b.err |
| 44 | } |
| 45 | |
| 46 | // PreparedRuntime owns a write handle that has been fully opened but is not |
| 47 | // yet visible through the host registry. Callers may build projections, seed |
| 48 | // initial events, and flush before atomically publishing the exact instance. |
| 49 | // A candidate must be either published or discarded. |
| 50 | type PreparedRuntime struct { |
| 51 | service *Service |
| 52 | runtime *Runtime |
| 53 | instance string |
| 54 | |
| 55 | mu sync.Mutex |
| 56 | published bool |
| 57 | discarded bool |
| 58 | } |
| 59 | |
| 60 | func (p *PreparedRuntime) Runtime() *Runtime { |
| 61 | if p == nil { |
| 62 | return nil |
| 63 | } |
| 64 | return p.runtime |
| 65 | } |
| 66 | |
| 67 | func NewService(hostID string, persistence SessionPersistence) (*Service, error) { |
| 68 | if hostID == "" || persistence == nil { |
| 69 | return nil, errors.New("session: host id and persistence are required") |
| 70 | } |
| 71 | service := &Service{ |
| 72 | hostID: hostID, persistence: persistence, |
| 73 | active: map[SessionRef]*Runtime{}, closed: map[SessionRef]error{}, preparing: map[SessionRef]*prepareRuntime{}, |
| 74 | bindings: map[*Runtime]int{}, retiring: map[*Runtime]chan struct{}{}, retireIdle: map[*Runtime]bool{}, |
| 75 | idleTimers: map[*Runtime]*time.Timer{}, idleWeight: map[*Runtime]int64{}, idleOrder: map[*Runtime]uint64{}, |
| 76 | idleBudget: 256 << 20, idleTTL: 60 * time.Second, |
| 77 | } |
| 78 | service.query = newQuery(hostID, persistence, service) |
| 79 | return service, nil |
| 80 | } |
| 81 | |
| 82 | func (s *Service) Create(ctx context.Context, options CreateOptions) (*Runtime, error) { |
| 83 | prepared, err := s.PrepareCreate(ctx, options) |
| 84 | if err != nil { |
| 85 | return nil, err |
| 86 | } |
| 87 | owner, err := s.Publish(prepared) |
| 88 | if err != nil { |
| 89 | _ = s.Discard(context.Background(), prepared) |
| 90 | return nil, err |
| 91 | } |
| 92 | return owner.Runtime(), nil |
| 93 | } |
| 94 | |
| 95 | // PrepareCreate reserves the immutable session identity and its writer lease |
| 96 | // without publishing an attachable runtime. This is the DSH prepare phase: |
| 97 | // host/controller state remains untouched until Publish succeeds. |
| 98 | func (s *Service) PrepareCreate(ctx context.Context, options CreateOptions) (*PreparedRuntime, error) { |
| 99 | if err := ctx.Err(); err != nil { |
| 100 | return nil, err |
| 101 | } |
| 102 | session, err := s.persistence.Create(options) |
| 103 | if err != nil { |
| 104 | return nil, err |
| 105 | } |
| 106 | session.externalizeDurableHistory() |
| 107 | ref := SessionRef{HostID: s.hostID, SessionID: session.ID()} |
| 108 | candidate, err := newRuntime(ref, session) |
| 109 | if err != nil { |
| 110 | return nil, errors.Join(err, session.Close(context.Background())) |
| 111 | } |
| 112 | candidate.owner = s |
| 113 | return &PreparedRuntime{service: s, runtime: candidate, instance: randomID()}, nil |
| 114 | } |
| 115 | |
| 116 | // Publish makes the exact prepared runtime visible and returns the host |
| 117 | // authority over that instance. It never replaces an existing instance with the |
| 118 | // same identity; the caller must resolve that ownership conflict explicitly. |
| 119 | // |
| 120 | // Only a RuntimeOwner can terminate a published runtime. Clients attach through |
| 121 | // ClientBinding and can only detach themselves. |
| 122 | func (s *Service) Publish(prepared *PreparedRuntime) (*RuntimeOwner, error) { |
| 123 | if prepared == nil || prepared.service != s || prepared.runtime == nil { |
| 124 | return nil, errors.New("session: invalid prepared runtime") |
| 125 | } |
| 126 | prepared.mu.Lock() |
| 127 | defer prepared.mu.Unlock() |
| 128 | if prepared.discarded { |
| 129 | return nil, errors.New("session: prepared runtime was discarded") |
| 130 | } |
| 131 | candidate := prepared.runtime |
| 132 | if prepared.published { |
| 133 | return &RuntimeOwner{service: s, runtime: candidate, instance: prepared.instance}, nil |
| 134 | } |
| 135 | ref := candidate.ref |
| 136 | s.mu.Lock() |
| 137 | if current := s.active[ref]; current != nil { |
| 138 | s.mu.Unlock() |
| 139 | return nil, fmt.Errorf("%w: %s", ErrSessionExists, ref.SessionID) |
| 140 | } |
| 141 | delete(s.closed, ref) |
| 142 | candidate.instance = prepared.instance |
| 143 | s.active[ref] = candidate |
| 144 | s.revision.Add(1) |
| 145 | s.mu.Unlock() |
| 146 | prepared.published = true |
| 147 | return &RuntimeOwner{service: s, runtime: candidate, instance: prepared.instance}, nil |
| 148 | } |
| 149 | |
| 150 | // RuntimeOwner is the host-side authority over one exact published instance. |
| 151 | // Controller and client code must never hold one: they use ClientBinding so a |
| 152 | // failed attach can only undo its own bind, never dispose a shared runtime. |
| 153 | type RuntimeOwner struct { |
| 154 | service *Service |
| 155 | runtime *Runtime |
| 156 | instance string |
| 157 | } |
| 158 | |
| 159 | // Runtime exposes the owned instance for host preparation work. |
| 160 | func (o *RuntimeOwner) Runtime() *Runtime { |
| 161 | if o == nil { |
| 162 | return nil |
| 163 | } |
| 164 | return o.runtime |
| 165 | } |
| 166 | |
| 167 | // Bind attaches a client to this instance without transferring ownership. |
| 168 | func (o *RuntimeOwner) Bind() (*ClientBinding, error) { |
| 169 | if o == nil || o.service == nil || o.runtime == nil { |
| 170 | return nil, ErrSessionNotRunning |
| 171 | } |
| 172 | return o.service.Bind(o.runtime) |
| 173 | } |
| 174 | |
| 175 | // Close terminates this exact instance. It refuses while any client is still |
| 176 | // bound, and it can never affect a same-ID successor published later. |
| 177 | func (o *RuntimeOwner) Close(ctx context.Context) error { |
| 178 | if o == nil || o.service == nil || o.runtime == nil { |
| 179 | return ErrSessionNotRunning |
| 180 | } |
| 181 | return o.service.closeOwned(ctx, o.runtime, o.instance) |
| 182 | } |
| 183 | |
| 184 | // Owner returns the host authority for the exact active instance. It fails for |
| 185 | // an unknown or superseded instance, so a delayed caller can never acquire |
| 186 | // authority over a same-ID successor. It is used by the flows that publish a |
| 187 | // brand-new identity in the same call; attaching to an existing session must go |
| 188 | // through Open and its ClientBinding instead. |
| 189 | func (s *Service) Owner(runtime *Runtime) (*RuntimeOwner, error) { |
| 190 | if runtime == nil || runtime.owner != s { |
| 191 | return nil, ErrSessionNotRunning |
| 192 | } |
| 193 | s.mu.Lock() |
| 194 | defer s.mu.Unlock() |
| 195 | if runtime.instance == "" || s.active[runtime.ref] != runtime { |
| 196 | return nil, ErrSessionNotRunning |
| 197 | } |
| 198 | return &RuntimeOwner{service: s, runtime: runtime, instance: runtime.instance}, nil |
| 199 | } |
| 200 | |
| 201 | // Discard closes an unpublished candidate and releases its writer lease. |
| 202 | // Published runtimes must be closed through Service.Close so exact-instance |
| 203 | // unregistering cannot be bypassed. |
| 204 | func (s *Service) Discard(ctx context.Context, prepared *PreparedRuntime) error { |
| 205 | if prepared == nil || prepared.service != s || prepared.runtime == nil { |
| 206 | return nil |
| 207 | } |
| 208 | prepared.mu.Lock() |
| 209 | defer prepared.mu.Unlock() |
| 210 | if prepared.published { |
| 211 | return errors.New("session: published runtime cannot be discarded") |
| 212 | } |
| 213 | if prepared.discarded { |
| 214 | return prepared.runtime.close(ctx) |
| 215 | } |
| 216 | prepared.discarded = true |
| 217 | return prepared.runtime.close(ctx) |
| 218 | } |
| 219 | |
| 220 | // Open attaches a client to an existing session and returns a binding that can |
| 221 | // only detach this client. It never hands out authority to close a runtime that |
| 222 | // another client may already be using. |
| 223 | func (s *Service) Open(ctx context.Context, ref SessionRef) (*ClientBinding, error) { |
| 224 | return s.openBinding(ctx, ref) |
| 225 | } |
| 226 | |
| 227 | // openRuntime resolves the exact published instance. It is internal so only the |
| 228 | // service's own prepare/publish flows can reach a runtime without a grant. |
| 229 | func (s *Service) openRuntime(ctx context.Context, ref SessionRef) (*Runtime, error) { |
| 230 | if err := ref.validate(s.hostID); err != nil { |
| 231 | return nil, err |
| 232 | } |
| 233 | for { |
| 234 | s.mu.Lock() |
| 235 | if current := s.active[ref]; current != nil { |
| 236 | s.mu.Unlock() |
| 237 | return current, nil |
| 238 | } |
| 239 | if pending := s.preparing[ref]; pending != nil { |
| 240 | done := pending.done |
| 241 | s.mu.Unlock() |
| 242 | select { |
| 243 | case <-done: |
| 244 | continue |
| 245 | case <-ctx.Done(): |
| 246 | return nil, ctx.Err() |
| 247 | } |
| 248 | } |
| 249 | pending := &prepareRuntime{done: make(chan struct{})} |
| 250 | s.preparing[ref] = pending |
| 251 | s.mu.Unlock() |
| 252 | break |
| 253 | } |
| 254 | |
| 255 | session, err := s.persistence.Open(ref.SessionID, ReadWrite) |
| 256 | if err != nil { |
| 257 | s.finishPrepare(ref) |
| 258 | return nil, err |
| 259 | } |
| 260 | if _, _, recoverErr := session.RecoverInterrupted(ctx); recoverErr != nil { |
| 261 | _ = session.Close(context.Background()) |
| 262 | s.finishPrepare(ref) |
| 263 | return nil, fmt.Errorf("session: close interrupted runtime: %w", recoverErr) |
| 264 | } |
| 265 | session.externalizeDurableHistory() |
| 266 | candidate, err := newRuntime(ref, session) |
| 267 | if err != nil { |
| 268 | closeErr := session.Close(context.Background()) |
| 269 | s.finishPrepare(ref) |
| 270 | return nil, errors.Join(err, closeErr) |
| 271 | } |
| 272 | candidate.owner = s |
| 273 | s.mu.Lock() |
| 274 | pending := s.preparing[ref] |
| 275 | delete(s.preparing, ref) |
| 276 | if current := s.active[ref]; current != nil { |
| 277 | if pending != nil { |
| 278 | close(pending.done) |
| 279 | } |
| 280 | s.mu.Unlock() |
| 281 | _ = candidate.close(context.Background()) |
| 282 | return current, nil |
| 283 | } |
| 284 | delete(s.closed, ref) |
| 285 | // Stamp the publish grant so Owner can hand the host authority over exactly |
| 286 | // this instance and no same-ID successor. |
| 287 | candidate.instance = randomID() |
| 288 | s.active[ref] = candidate |
| 289 | s.revision.Add(1) |
| 290 | if pending != nil { |
| 291 | close(pending.done) |
| 292 | } |
| 293 | s.mu.Unlock() |
| 294 | return candidate, nil |
| 295 | } |
| 296 | |
| 297 | func (s *Service) finishPrepare(ref SessionRef) { |
| 298 | s.mu.Lock() |
| 299 | if pending := s.preparing[ref]; pending != nil { |
| 300 | delete(s.preparing, ref) |
| 301 | close(pending.done) |
| 302 | } |
| 303 | s.mu.Unlock() |
| 304 | } |
| 305 | |
| 306 | func (s *Service) Runtime(ref SessionRef) (*Runtime, bool) { |
| 307 | if ref.validate(s.hostID) != nil { |
| 308 | return nil, false |
| 309 | } |
| 310 | s.mu.Lock() |
| 311 | runtime := s.active[ref] |
| 312 | s.mu.Unlock() |
| 313 | return runtime, runtime != nil |
| 314 | } |
| 315 | |
| 316 | // Bind attaches a client to an exact published runtime. The returned binding |
| 317 | // can only release its own reference; it cannot dispose the shared runtime. |
| 318 | func (s *Service) Bind(runtime *Runtime) (*ClientBinding, error) { |
| 319 | if runtime == nil || runtime.owner != s { |
| 320 | return nil, ErrSessionNotRunning |
| 321 | } |
| 322 | s.mu.Lock() |
| 323 | defer s.mu.Unlock() |
| 324 | if s.retiring[runtime] != nil { |
| 325 | return nil, ErrRuntimeRetiring |
| 326 | } |
| 327 | if s.active[runtime.ref] != runtime { |
| 328 | return nil, ErrSessionNotRunning |
| 329 | } |
| 330 | s.bindings[runtime]++ |
| 331 | delete(s.retireIdle, runtime) |
| 332 | if timer := s.idleTimers[runtime]; timer != nil { |
| 333 | s.removeIdleCacheLocked(runtime, true) |
| 334 | } |
| 335 | return &ClientBinding{service: s, runtime: runtime}, nil |
| 336 | } |
| 337 | |
| 338 | // OpenBinding opens or reuses a runtime and attaches a client capability. |
| 339 | func (s *Service) openBinding(ctx context.Context, ref SessionRef) (*ClientBinding, error) { |
| 340 | for { |
| 341 | runtime, err := s.openRuntime(ctx, ref) |
| 342 | if err != nil { |
| 343 | return nil, err |
| 344 | } |
| 345 | binding, err := s.Bind(runtime) |
| 346 | if err == nil { |
| 347 | return binding, err |
| 348 | } |
| 349 | if errors.Is(err, ErrRuntimeRetiring) { |
| 350 | s.mu.Lock() |
| 351 | done := s.retiring[runtime] |
| 352 | s.mu.Unlock() |
| 353 | if done != nil { |
| 354 | select { |
| 355 | case <-done: |
| 356 | continue |
| 357 | case <-ctx.Done(): |
| 358 | return nil, ctx.Err() |
| 359 | } |
| 360 | } |
| 361 | continue |
| 362 | } |
| 363 | return nil, err |
| 364 | } |
| 365 | } |
| 366 | |
| 367 | func (s *Service) releaseBinding(ctx context.Context, runtime *Runtime) error { |
| 368 | s.mu.Lock() |
| 369 | count := s.bindings[runtime] |
| 370 | if count <= 1 { |
| 371 | delete(s.bindings, runtime) |
| 372 | s.retireIdle[runtime] = true |
| 373 | } else { |
| 374 | s.bindings[runtime] = count - 1 |
| 375 | } |
| 376 | s.mu.Unlock() |
| 377 | if count <= 1 { |
| 378 | var flushErr error |
| 379 | if !runtime.executionBusy() { |
| 380 | _, flushErr = runtime.session.Flush(ctx) |
| 381 | } |
| 382 | s.scheduleIdleRetirement(runtime) |
| 383 | return flushErr |
| 384 | } |
| 385 | return nil |
| 386 | } |
| 387 | |
| 388 | func (s *Service) closeIfUnbound(ctx context.Context, runtime *Runtime) error { |
| 389 | if runtime == nil { |
| 390 | return nil |
| 391 | } |
| 392 | s.scheduleIdleRetirement(runtime) |
| 393 | return nil |
| 394 | } |
| 395 | |
| 396 | func (s *Service) scheduleIdleRetirement(runtime *Runtime) { |
| 397 | if runtime == nil { |
| 398 | return |
| 399 | } |
| 400 | weight := runtime.session.cacheWeight() |
| 401 | s.mu.Lock() |
| 402 | if s.active[runtime.ref] != runtime || s.bindings[runtime] != 0 || !s.retireIdle[runtime] { |
| 403 | s.mu.Unlock() |
| 404 | return |
| 405 | } |
| 406 | if runtime.executionBusy() { |
| 407 | s.mu.Unlock() |
| 408 | return |
| 409 | } |
| 410 | if s.idleTimers[runtime] != nil { |
| 411 | s.mu.Unlock() |
| 412 | return |
| 413 | } |
| 414 | ttl := s.idleTTL |
| 415 | s.idleClock++ |
| 416 | s.idleOrder[runtime] = s.idleClock |
| 417 | s.idleWeight[runtime] = weight |
| 418 | s.idleUsed += weight |
| 419 | if ttl > 0 { |
| 420 | s.idleTimers[runtime] = time.AfterFunc(ttl, func() { |
| 421 | _ = s.retireIfUnbound(context.Background(), runtime) |
| 422 | }) |
| 423 | } |
| 424 | var victims []*Runtime |
| 425 | for s.idleBudget >= 0 && s.idleUsed > s.idleBudget && len(s.idleOrder) > 0 { |
| 426 | var oldest *Runtime |
| 427 | var order uint64 |
| 428 | for candidate, candidateOrder := range s.idleOrder { |
| 429 | if oldest == nil || candidateOrder < order { |
| 430 | oldest, order = candidate, candidateOrder |
| 431 | } |
| 432 | } |
| 433 | if oldest == nil { |
| 434 | break |
| 435 | } |
| 436 | s.removeIdleCacheLocked(oldest, true) |
| 437 | victims = append(victims, oldest) |
| 438 | } |
| 439 | s.mu.Unlock() |
| 440 | for _, victim := range victims { |
| 441 | _ = s.retireIfUnbound(context.Background(), victim) |
| 442 | } |
| 443 | if ttl <= 0 && len(victims) == 0 { |
| 444 | _ = s.retireIfUnbound(context.Background(), runtime) |
| 445 | } |
| 446 | } |
| 447 | |
| 448 | func (s *Service) removeIdleCacheLocked(runtime *Runtime, stop bool) { |
| 449 | if timer := s.idleTimers[runtime]; timer != nil && stop { |
| 450 | timer.Stop() |
| 451 | } |
| 452 | delete(s.idleTimers, runtime) |
| 453 | s.idleUsed -= s.idleWeight[runtime] |
| 454 | if s.idleUsed < 0 { |
| 455 | s.idleUsed = 0 |
| 456 | } |
| 457 | delete(s.idleWeight, runtime) |
| 458 | delete(s.idleOrder, runtime) |
| 459 | } |
| 460 | |
| 461 | func (s *Service) retireIfUnbound(ctx context.Context, runtime *Runtime) error { |
| 462 | s.mu.Lock() |
| 463 | s.removeIdleCacheLocked(runtime, false) |
| 464 | if s.active[runtime.ref] != runtime || s.bindings[runtime] != 0 || !s.retireIdle[runtime] || runtime.executionBusy() { |
| 465 | s.mu.Unlock() |
| 466 | return nil |
| 467 | } |
| 468 | if done := s.retiring[runtime]; done != nil { |
| 469 | s.mu.Unlock() |
| 470 | select { |
| 471 | case <-done: |
| 472 | return nil |
| 473 | case <-ctx.Done(): |
| 474 | return ctx.Err() |
| 475 | } |
| 476 | } |
| 477 | done := make(chan struct{}) |
| 478 | s.retiring[runtime] = done |
| 479 | s.mu.Unlock() |
| 480 | |
| 481 | err := runtime.close(ctx) |
| 482 | s.mu.Lock() |
| 483 | delete(s.retiring, runtime) |
| 484 | if !errors.Is(err, ErrRuntimeBusy) && s.active[runtime.ref] == runtime && s.bindings[runtime] == 0 { |
| 485 | delete(s.active, runtime.ref) |
| 486 | delete(s.retireIdle, runtime) |
| 487 | s.removeIdleCacheLocked(runtime, true) |
| 488 | s.closed[runtime.ref] = err |
| 489 | s.revision.Add(1) |
| 490 | } |
| 491 | close(done) |
| 492 | s.mu.Unlock() |
| 493 | return err |
| 494 | } |
| 495 |