| 1 | import { AgentScope } from "@t3tools/shared/AgentScope"; |
| 2 | import * as KeyedLock from "@t3tools/shared/KeyedLock"; |
| 3 | import { resolveProjectSettings } from "@t3tools/shared/projectSettings"; |
| 4 | import { |
| 5 | ModelSelection, |
| 6 | OrchestrationV2DomainEvent, |
| 7 | OrchestrationV2ProviderSession, |
| 8 | type OrchestrationV2ProviderThread, |
| 9 | OrchestrationV2RuntimeRequest, |
| 10 | ProviderInstanceId, |
| 11 | ProviderSessionId, |
| 12 | ThreadId, |
| 13 | type ProviderThreadId, |
| 14 | } from "@t3tools/contracts"; |
| 15 | import * as Cause from "effect/Cause"; |
| 16 | import * as Clock from "effect/Clock"; |
| 17 | import * as Context from "effect/Context"; |
| 18 | import * as DateTime from "effect/DateTime"; |
| 19 | import * as Duration from "effect/Duration"; |
| 20 | import * as Effect from "effect/Effect"; |
| 21 | import * as Exit from "effect/Exit"; |
| 22 | import * as Fiber from "effect/Fiber"; |
| 23 | import * as FiberSet from "effect/FiberSet"; |
| 24 | import * as FileSystem from "effect/FileSystem"; |
| 25 | import * as Layer from "effect/Layer"; |
| 26 | import * as Option from "effect/Option"; |
| 27 | import * as Queue from "effect/Queue"; |
| 28 | import * as Ref from "effect/Ref"; |
| 29 | import * as Schedule from "effect/Schedule"; |
| 30 | import * as Schema from "effect/Schema"; |
| 31 | import * as Scope from "effect/Scope"; |
| 32 | import * as Semaphore from "effect/Semaphore"; |
| 33 | import * as Stream from "effect/Stream"; |
| 34 | |
| 35 | import { normalizeModelMetricLabel } from "../observability/Attributes.ts"; |
| 36 | import { |
| 37 | providerSessionsTotal, |
| 38 | providerTurnDuration, |
| 39 | providerTurnsTotal, |
| 40 | withMetrics, |
| 41 | } from "../observability/Metrics.ts"; |
| 42 | import { ProviderWorkspaceMissingError } from "../provider/Errors.ts"; |
| 43 | import * as ProjectService from "../project/ProjectService.ts"; |
| 44 | import * as McpProviderSessions from "@t3tools/provider-core/server/McpProviderSessions"; |
| 45 | import * as ServerSettings from "../serverSettings.ts"; |
| 46 | import * as McpSessionRegistry from "../mcp/McpSessionRegistry.ts"; |
| 47 | import * as EventSink from "./EventSink.ts"; |
| 48 | import * as IdAllocator from "@t3tools/provider-core/server/IdAllocator"; |
| 49 | import * as ProviderEventIngestor from "./ProviderEventIngestor.ts"; |
| 50 | import * as ProviderAdapter from "@t3tools/provider-core/server/ProviderAdapter"; |
| 51 | import * as ProviderAdapterRegistry from "./ProviderAdapterRegistry.ts"; |
| 52 | import * as ProjectionStore from "./ProjectionStore.ts"; |
| 53 | |
| 54 | const DEFAULT_IDLE_TIMEOUT_MS = 30 * 60 * 1000; |
| 55 | const DEFAULT_MAX_IDLE_PIN_MS = 4 * 60 * 60 * 1000; |
| 56 | const RELEASE_SCOPE_CLOSE_TIMEOUT_MS = 30 * 1000; |
| 57 | |
| 58 | const busyTurnPrefix = (providerThreadId: ProviderThreadId) => `${providerThreadId}#`; |
| 59 | /** The identity a turn's start and its `turn.terminal` share. */ |
| 60 | const busyTurnKey = (providerThreadId: ProviderThreadId, runOrdinal: number) => |
| 61 | `${busyTurnPrefix(providerThreadId)}${runOrdinal}`; |
| 62 | const UNLOAD_THREAD_TIMEOUT_MS = 10 * 1000; |
| 63 | |
| 64 | export const ProviderSessionReleaseReason = Schema.Literals([ |
| 65 | "idle_timeout", |
| 66 | "runtime_error", |
| 67 | "manual_shutdown", |
| 68 | "server_shutdown", |
| 69 | ]); |
| 70 | export type ProviderSessionReleaseReason = typeof ProviderSessionReleaseReason.Type; |
| 71 | |
| 72 | /** |
| 73 | * ProviderSessionManager owns live session residency: open sessions, idle release, |
| 74 | * explicit shutdown, and release-on-runtime-failure. |
| 75 | * |
| 76 | * It intentionally does not resurrect persisted sessions. Process-loss recovery |
| 77 | * terminalizes provider-bound work and retires non-replayable effects; a later |
| 78 | * user command or durable replay-safe operation opens a session lazily. |
| 79 | */ |
| 80 | export class ProviderSessionOpenError extends Schema.TaggedError<ProviderSessionOpenError>()( |
| 81 | "ProviderSessionOpenError", |
| 82 | { |
| 83 | instanceId: ProviderInstanceId, |
| 84 | providerSessionId: ProviderSessionId, |
| 85 | cause: Schema.optional(Schema.Defect()), |
| 86 | }, |
| 87 | ) { |
| 88 | override get message(): string { |
| 89 | return `Failed to open provider instance ${this.instanceId} session ${this.providerSessionId}.`; |
| 90 | } |
| 91 | } |
| 92 | |
| 93 | export class ProviderSessionLookupError extends Schema.TaggedError<ProviderSessionLookupError>()( |
| 94 | "ProviderSessionLookupError", |
| 95 | { |
| 96 | providerSessionId: ProviderSessionId, |
| 97 | cause: Schema.optional(Schema.Defect()), |
| 98 | }, |
| 99 | ) { |
| 100 | override get message(): string { |
| 101 | return `Failed to look up provider session ${this.providerSessionId}.`; |
| 102 | } |
| 103 | } |
| 104 | |
| 105 | export class ProviderSessionCloseError extends Schema.TaggedError<ProviderSessionCloseError>()( |
| 106 | "ProviderSessionCloseError", |
| 107 | { |
| 108 | providerSessionId: ProviderSessionId, |
| 109 | cause: Schema.optional(Schema.Defect()), |
| 110 | }, |
| 111 | ) { |
| 112 | override get message(): string { |
| 113 | return `Failed to close provider session ${this.providerSessionId}.`; |
| 114 | } |
| 115 | } |
| 116 | |
| 117 | export class ProviderSessionReleaseError extends Schema.TaggedError<ProviderSessionReleaseError>()( |
| 118 | "ProviderSessionReleaseError", |
| 119 | { |
| 120 | providerSessionId: ProviderSessionId, |
| 121 | reason: ProviderSessionReleaseReason, |
| 122 | cause: Schema.optional(Schema.Defect()), |
| 123 | }, |
| 124 | ) { |
| 125 | override get message(): string { |
| 126 | return `Failed to release provider session ${this.providerSessionId}.`; |
| 127 | } |
| 128 | } |
| 129 | |
| 130 | export class ProviderSessionActivityError extends Schema.TaggedError<ProviderSessionActivityError>()( |
| 131 | "ProviderSessionActivityError", |
| 132 | { |
| 133 | providerSessionId: ProviderSessionId, |
| 134 | cause: Schema.optional(Schema.Defect()), |
| 135 | }, |
| 136 | ) { |
| 137 | override get message(): string { |
| 138 | return `Failed to update provider session activity for ${this.providerSessionId}.`; |
| 139 | } |
| 140 | } |
| 141 | |
| 142 | export const ProviderSessionManagerV2Error = Schema.Union([ |
| 143 | ProviderSessionOpenError, |
| 144 | ProviderWorkspaceMissingError, |
| 145 | ProviderSessionLookupError, |
| 146 | ProviderSessionCloseError, |
| 147 | ProviderSessionReleaseError, |
| 148 | ProviderSessionActivityError, |
| 149 | ]); |
| 150 | export type ProviderSessionManagerV2Error = typeof ProviderSessionManagerV2Error.Type; |
| 151 | |
| 152 | export interface ProviderSessionManagerV2Shape { |
| 153 | readonly shutdown: Effect.Effect<void>; |
| 154 | readonly open: (input: { |
| 155 | readonly threadId: ThreadId; |
| 156 | readonly providerSessionId: ProviderSessionId; |
| 157 | readonly modelSelection: ModelSelection; |
| 158 | readonly runtimePolicy: ProviderAdapter.ProviderAdapterV2RuntimePolicy; |
| 159 | readonly resumeFromSession?: OrchestrationV2ProviderSession; |
| 160 | readonly initialNativeThreadId?: string; |
| 161 | readonly initialProviderItemIdentityVersion?: 2; |
| 162 | }) => Effect.Effect< |
| 163 | ProviderAdapter.ProviderAdapterV2SessionRuntime, |
| 164 | ProviderSessionManagerV2Error |
| 165 | >; |
| 166 | readonly get: ( |
| 167 | providerSessionId: ProviderSessionId, |
| 168 | ) => Effect.Effect< |
| 169 | Option.Option<ProviderAdapter.ProviderAdapterV2SessionRuntime>, |
| 170 | ProviderSessionManagerV2Error |
| 171 | >; |
| 172 | /** Whether the session's runtime is running. Unlike `get`, not activity. */ |
| 173 | readonly isLive: (providerSessionId: ProviderSessionId) => Effect.Effect<boolean>; |
| 174 | readonly close: ( |
| 175 | providerSessionId: ProviderSessionId, |
| 176 | ) => Effect.Effect<void, ProviderSessionManagerV2Error>; |
| 177 | /** Closes every live runtime owned by one provider instance. */ |
| 178 | readonly closeInstance: ( |
| 179 | instanceId: ProviderInstanceId, |
| 180 | ) => Effect.Effect<void, ProviderSessionManagerV2Error>; |
| 181 | readonly release: (input: { |
| 182 | readonly providerSessionId: ProviderSessionId; |
| 183 | readonly reason: ProviderSessionReleaseReason; |
| 184 | readonly detail?: string; |
| 185 | }) => Effect.Effect<void, ProviderSessionManagerV2Error>; |
| 186 | readonly detach: (input: { |
| 187 | readonly providerSessionId: ProviderSessionId; |
| 188 | readonly threadId: ThreadId; |
| 189 | readonly detail?: string; |
| 190 | /** |
| 191 | * True for terminal detaches (thread archived or deleted): the thread's |
| 192 | * MCP credentials are revoked immediately instead of surviving for a |
| 193 | * potential re-attach. |
| 194 | */ |
| 195 | readonly revokeMcpCredential?: boolean; |
| 196 | }) => Effect.Effect<void, ProviderSessionManagerV2Error>; |
| 197 | } |
| 198 | |
| 199 | export class ProviderSessionManagerV2 extends Context.Service< |
| 200 | ProviderSessionManagerV2, |
| 201 | ProviderSessionManagerV2Shape |
| 202 | >()("t3/orchestration-v2/ProviderSessionManager/ProviderSessionManagerV2") {} |
| 203 | |
| 204 | interface LiveSessionEntry { |
| 205 | readonly attachedThreadIds: ReadonlySet<ThreadId>; |
| 206 | readonly loadedProviderThreadKeyByThread: ReadonlyMap<ThreadId, string>; |
| 207 | /** |
| 208 | * MCP credential session id issued for each attached thread. Revocation on |
| 209 | * detach/release is scoped to these ids so tearing down a superseded |
| 210 | * session cannot revoke a replacement session's credential for the same |
| 211 | * thread (the workspace-handoff sequence opens the replacement before the |
| 212 | * outbox executes the old session's detach). |
| 213 | */ |
| 214 | readonly mcpCredentialIdByThread: ReadonlyMap<ThreadId, string>; |
| 215 | readonly supportsMultipleProviderThreads: boolean; |
| 216 | readonly runtime: ProviderAdapter.ProviderAdapterV2SessionRuntime; |
| 217 | readonly exposedRuntime: ProviderAdapter.ProviderAdapterV2SessionRuntime; |
| 218 | readonly eventSubscribers: Ref.Ref< |
| 219 | ReadonlyMap<number, Queue.Queue<ProviderSessionEventSignal, Cause.Done>> |
| 220 | >; |
| 221 | readonly requestEventPermit: Semaphore.Semaphore; |
| 222 | readonly scope: Scope.Closeable; |
| 223 | readonly idleGeneration: number; |
| 224 | /** |
| 225 | * Turns this session is running, keyed by `busyTurnKey`. A turn's start adds |
| 226 | * it and its `turn.terminal` (or a failed start) removes it, so a turn can |
| 227 | * only clear itself and the session is idle when the set is empty. |
| 228 | */ |
| 229 | readonly busyTurns: ReadonlySet<string>; |
| 230 | readonly lastActivityAtMs: number; |
| 231 | readonly idleFiber: Fiber.Fiber<void, never> | null; |
| 232 | /** Set when idle release is deferred for pending background work; bounds total deferral. */ |
| 233 | readonly pinnedSinceMs: number | null; |
| 234 | /** |
| 235 | * Shared runtimes only: the provider thread each attached app thread last |
| 236 | * started a turn on, and the timer that unloads it once it has been idle |
| 237 | * for `idleTimeoutMs`. |
| 238 | */ |
| 239 | readonly idleThreadUnloads: ReadonlyMap<ThreadId, IdleThreadUnload>; |
| 240 | } |
| 241 | |
| 242 | interface IdleThreadUnload { |
| 243 | readonly providerThread: OrchestrationV2ProviderThread; |
| 244 | readonly generation: number; |
| 245 | readonly fiber: Fiber.Fiber<void, never> | null; |
| 246 | } |
| 247 | |
| 248 | type ProviderSessionEventSignal = |
| 249 | | { readonly type: "event"; readonly event: ProviderAdapter.ProviderAdapterV2Event } |
| 250 | | { |
| 251 | readonly type: "failure"; |
| 252 | readonly cause: Cause.Cause<ProviderAdapter.ProviderAdapterV2Error>; |
| 253 | }; |
| 254 | |
| 255 | export interface ProviderSessionManagerV2LayerOptions { |
| 256 | readonly idleTimeoutMs?: number; |
| 257 | /** Cap on how long idle release may be deferred for pending background work. */ |
| 258 | readonly maxIdlePinMs?: number; |
| 259 | /** Test replay harnesses can omit T3's MCP server from provider protocol fixtures. */ |
| 260 | readonly configureMcp?: boolean; |
| 261 | } |
| 262 | |
| 263 | function releaseStatusFor( |
| 264 | reason: ProviderSessionReleaseReason, |
| 265 | ): OrchestrationV2ProviderSession["status"] { |
| 266 | return reason === "runtime_error" ? "error" : "stopped"; |
| 267 | } |
| 268 | |
| 269 | function releasedRuntimeRequestStatusFor( |
| 270 | reason: ProviderSessionReleaseReason, |
| 271 | ): OrchestrationV2RuntimeRequest["status"] { |
| 272 | return reason === "manual_shutdown" || reason === "server_shutdown" ? "cancelled" : "expired"; |
| 273 | } |
| 274 | |
| 275 | function sessionKey(providerSessionId: ProviderSessionId): string { |
| 276 | return String(providerSessionId); |
| 277 | } |
| 278 | |
| 279 | /** |
| 280 | * Runtime requests with no provider turn belong to the live session itself. |
| 281 | * Their node and transcript item are runless too, so they bypass the normal |
| 282 | * per-run subscriber and are persisted by the session event pump. |
| 283 | */ |
| 284 | function sessionScopedRuntimeRequestThreadId( |
| 285 | event: ProviderAdapter.ProviderAdapterV2Event, |
| 286 | ): ThreadId | undefined { |
| 287 | switch (event.type) { |
| 288 | case "runtime_request.updated": |
| 289 | return event.runtimeRequest.providerTurnId === null ? event.threadId : undefined; |
| 290 | case "node.updated": |
| 291 | return event.node.runId === null && event.node.runtimeRequestId !== null |
| 292 | ? event.node.threadId |
| 293 | : undefined; |
| 294 | case "turn_item.updated": |
| 295 | return event.turnItem.runId === null && |
| 296 | (event.turnItem.type === "approval_request" || event.turnItem.type === "user_input_request") |
| 297 | ? event.turnItem.threadId |
| 298 | : undefined; |
| 299 | default: |
| 300 | return undefined; |
| 301 | } |
| 302 | } |
| 303 | |
| 304 | function providerThreadRuntimeKey( |
| 305 | providerThread: Parameters< |
| 306 | ProviderAdapter.ProviderAdapterV2SessionRuntime["resumeThread"] |
| 307 | >[0]["providerThread"], |
| 308 | ): string { |
| 309 | const nativeThreadRef = providerThread.nativeThreadRef; |
| 310 | return nativeThreadRef === null |
| 311 | ? String(providerThread.id) |
| 312 | : `${nativeThreadRef.driver}:${nativeThreadRef.nativeId}`; |
| 313 | } |
| 314 | |
| 315 | function providerThreadLoadKey(input: { |
| 316 | readonly providerThread: Parameters< |
| 317 | ProviderAdapter.ProviderAdapterV2SessionRuntime["resumeThread"] |
| 318 | >[0]["providerThread"]; |
| 319 | readonly modelSelection?: ModelSelection; |
| 320 | readonly runtimePolicy?: ProviderAdapter.ProviderAdapterV2RuntimePolicy; |
| 321 | }): string { |
| 322 | return JSON.stringify({ |
| 323 | providerThread: providerThreadRuntimeKey(input.providerThread), |
| 324 | modelSelection: input.modelSelection ?? null, |
| 325 | runtimePolicy: input.runtimePolicy ?? null, |
| 326 | }); |
| 327 | } |
| 328 | |
| 329 | export const layerWithOptions = ( |
| 330 | options: ProviderSessionManagerV2LayerOptions = {}, |
| 331 | ): Layer.Layer< |
| 332 | ProviderSessionManagerV2, |
| 333 | never, |
| 334 | | EventSink.EventSinkV2 |
| 335 | | FileSystem.FileSystem |
| 336 | | IdAllocator.IdAllocatorV2 |
| 337 | | McpProviderSessions.McpProviderSessions |
| 338 | | McpSessionRegistry.McpSessionRegistry |
| 339 | | ProjectionStore.ProjectionStoreV2 |
| 340 | | ProviderEventIngestor.ProviderEventIngestorV2 |
| 341 | | ProviderAdapterRegistry.ProviderAdapterRegistryV2 |
| 342 | > => |
| 343 | Layer.effect( |
| 344 | ProviderSessionManagerV2, |
| 345 | Effect.gen(function* () { |
| 346 | const registry = yield* ProviderAdapterRegistry.ProviderAdapterRegistryV2; |
| 347 | const fileSystem = yield* FileSystem.FileSystem; |
| 348 | const mcpSessionRegistry = yield* McpSessionRegistry.McpSessionRegistry; |
| 349 | const mcpSessions = yield* McpProviderSessions.McpProviderSessions; |
| 350 | /** |
| 351 | * Optional so the many focused tests that assemble this layer by hand do |
| 352 | * not each need a settings stub; the production composition always |
| 353 | * provides it. When present, an unreadable settings file withholds |
| 354 | * browser access rather than granting it — an explicit "off" silently |
| 355 | * becoming "on" would violate the user's stated choice, whereas the |
| 356 | * reverse costs an agent one toolset and is visible immediately (#7083). |
| 357 | */ |
| 358 | const serverSettings = yield* Effect.serviceOption(ServerSettings.ServerSettingsService); |
| 359 | const projectService = yield* Effect.serviceOption(ProjectService.ProjectService); |
| 360 | const eventSink = yield* EventSink.EventSinkV2; |
| 361 | const idAllocator = yield* IdAllocator.IdAllocatorV2; |
| 362 | const providerEventIngestor = yield* ProviderEventIngestor.ProviderEventIngestorV2; |
| 363 | const projectionStore = yield* ProjectionStore.ProjectionStoreV2; |
| 364 | const agentScope = yield* AgentScope; |
| 365 | const agentAccessSettings = Effect.fn("ProviderSessionManagerV2.agentAccessSettings")( |
| 366 | function* (threadId: ThreadId) { |
| 367 | if (Option.isNone(serverSettings)) return { browser: true, device: false }; |
| 368 | return yield* Effect.gen(function* () { |
| 369 | const settings = yield* serverSettings.value.getSettings; |
| 370 | const thread = yield* projectionStore.getThread(threadId); |
| 371 | const entries = Object.values(settings.projectSettingsOverrides); |
| 372 | const browserOverridden = entries.some( |
| 373 | (entry) => entry.enableAgentBrowserAccess !== undefined, |
| 374 | ); |
| 375 | const deviceOverridden = entries.some( |
| 376 | (entry) => entry.enableAgentDeviceAccess !== undefined, |
| 377 | ); |
| 378 | if (browserOverridden || deviceOverridden) { |
| 379 | const project = Option.isSome(projectService) |
| 380 | ? yield* projectService.value.getById(thread.projectId) |
| 381 | : Option.none(); |
| 382 | if (Option.isNone(project)) |
| 383 | return { |
| 384 | browser: browserOverridden ? false : settings.enableAgentBrowserAccess, |
| 385 | device: deviceOverridden ? false : settings.enableAgentDeviceAccess, |
| 386 | }; |
| 387 | } |
| 388 | const effective = resolveProjectSettings(settings, thread.projectId).settings; |
| 389 | return { |
| 390 | browser: effective.enableAgentBrowserAccess, |
| 391 | device: effective.enableAgentDeviceAccess, |
| 392 | }; |
| 393 | }).pipe( |
| 394 | Effect.catch((cause) => |
| 395 | Effect.logWarning( |
| 396 | "Could not resolve agent access; withholding browser and device tools.", |
| 397 | { threadId, cause }, |
| 398 | ).pipe(Effect.as({ browser: false, device: false })), |
| 399 | ), |
| 400 | ); |
| 401 | }, |
| 402 | ); |
| 403 | const layerScope = yield* Effect.scope; |
| 404 | // Ctrl+C, or a stop that signals the whole process group, reaches the |
| 405 | // provider CLIs with the server. They report their own background work |
| 406 | // stopped before shutdown captures restart continuations, so provider |
| 407 | // events after the signal are dropped; restart recovery owns that state. |
| 408 | const shutdownSignal = { received: false }; |
| 409 | const onShutdownSignal = () => { |
| 410 | shutdownSignal.received = true; |
| 411 | }; |
| 412 | yield* Effect.acquireRelease( |
| 413 | Effect.sync(() => { |
| 414 | process.on("SIGINT", onShutdownSignal); |
| 415 | process.on("SIGTERM", onShutdownSignal); |
| 416 | }), |
| 417 | () => |
| 418 | Effect.sync(() => { |
| 419 | process.off("SIGINT", onShutdownSignal); |
| 420 | process.off("SIGTERM", onShutdownSignal); |
| 421 | }), |
| 422 | ); |
| 423 | const sessions = yield* Ref.make(new Map<string, LiveSessionEntry>()); |
| 424 | // One retry per released entry, so a later release with the same id |
| 425 | // cannot drop cleanup for threads only the earlier session served. |
| 426 | const releaseRecordRetries = yield* FiberSet.make(); |
| 427 | const nextSubscriberId = yield* Ref.make(0); |
| 428 | const sessionOpen = yield* KeyedLock.make<ProviderSessionId>(); |
| 429 | // Orders a thread's attach against a detach unloading it on the same session. |
| 430 | const threadAttachment = yield* KeyedLock.make<string>(); |
| 431 | const threadAttachmentKey = (input: { |
| 432 | readonly providerSessionId: ProviderSessionId; |
| 433 | readonly threadId: ThreadId; |
| 434 | }) => `${input.providerSessionId}\u0000${input.threadId}`; |
| 435 | const idleTimeoutMs = Math.max(1, options.idleTimeoutMs ?? DEFAULT_IDLE_TIMEOUT_MS); |
| 436 | const maxIdlePinMs = Math.max(0, options.maxIdlePinMs ?? DEFAULT_MAX_IDLE_PIN_MS); |
| 437 | interface PreparedMcpCredential { |
| 438 | readonly mcpCredentialId: string | undefined; |
| 439 | /** True when this call minted the credential (vs reusing a live one). */ |
| 440 | readonly issued: boolean; |
| 441 | } |
| 442 | /** |
| 443 | * Reservations protect a credential between prepareMcpSession handing it |
| 444 | * out and the owning session entry becoming visible in `sessions`. |
| 445 | * Adapters like ACP and OpenCode consume the credential eagerly during |
| 446 | * openSession, so a racing release must not revoke it in that window |
| 447 | * (rotating afterwards cannot repair an already-configured process). |
| 448 | * The holder MUST drop the reservation once the entry is recorded or the |
| 449 | * open fails. |
| 450 | */ |
| 451 | const mcpCredentialReservations = new Map<string, number>(); |
| 452 | const mcpReservationKey = (threadId: ThreadId, mcpCredentialId: string) => |
| 453 | `${threadId}\0${mcpCredentialId}`; |
| 454 | const reserveMcpCredential = (threadId: ThreadId, mcpCredentialId: string) => { |
| 455 | const key = mcpReservationKey(threadId, mcpCredentialId); |
| 456 | mcpCredentialReservations.set(key, (mcpCredentialReservations.get(key) ?? 0) + 1); |
| 457 | }; |
| 458 | const dropMcpCredentialReservation = (threadId: ThreadId, mcpCredentialId: string) => { |
| 459 | const key = mcpReservationKey(threadId, mcpCredentialId); |
| 460 | const count = mcpCredentialReservations.get(key) ?? 0; |
| 461 | if (count <= 1) { |
| 462 | mcpCredentialReservations.delete(key); |
| 463 | } else { |
| 464 | mcpCredentialReservations.set(key, count - 1); |
| 465 | } |
| 466 | }; |
| 467 | const isMcpCredentialReserved = (threadId: ThreadId, mcpCredentialId: string) => |
| 468 | (mcpCredentialReservations.get(mcpReservationKey(threadId, mcpCredentialId)) ?? 0) > 0; |
| 469 | const mcpPrepareLock = yield* KeyedLock.make<ThreadId>(); |
| 470 | /** |
| 471 | * Resolves (or mints) the thread's MCP credential and returns it with a |
| 472 | * reservation held; the caller must drop the reservation exactly once. |
| 473 | * Serialized per thread so two concurrent prepares cannot interleave |
| 474 | * their rotate steps and revoke each other's freshly minted credential. |
| 475 | */ |
| 476 | const prepareMcpSession = ( |
| 477 | threadId: ThreadId, |
| 478 | providerInstanceId: ProviderInstanceId, |
| 479 | ): Effect.Effect<PreparedMcpCredential> => |
| 480 | options.configureMcp === false |
| 481 | ? mcpSessions |
| 482 | .clear(threadId) |
| 483 | .pipe(Effect.as<PreparedMcpCredential>({ mcpCredentialId: undefined, issued: false })) |
| 484 | : mcpPrepareLock.withLock( |
| 485 | threadId, |
| 486 | Effect.gen(function* () { |
| 487 | // Reuse a still-valid credential for this thread instead of |
| 488 | // rotating: long-lived provider processes (codex app-server) |
| 489 | // build their MCP client once per conversation and keep using |
| 490 | // the credential it started with, so a thread that detaches and |
| 491 | // re-attaches across a workspace handoff must come back to the |
| 492 | // same token or the process's tool calls fail auth. |
| 493 | const { browser: browserToolsAvailable, device: deviceToolsAvailable } = |
| 494 | yield* agentAccessSettings(threadId); |
| 495 | const capabilities = new Set< |
| 496 | import("../mcp/McpInvocationContext.ts").McpCapability |
| 497 | >(["orchestration", "worktree", "pull-requests"]); |
| 498 | if (browserToolsAvailable) capabilities.add("preview"); |
| 499 | if (deviceToolsAvailable) capabilities.add("device"); |
| 500 | const existing = yield* mcpSessions.read(threadId); |
| 501 | if (existing !== undefined) { |
| 502 | // Reserve before the async resolve so a release cannot |
| 503 | // revoke the credential between validation and reservation. |
| 504 | reserveMcpCredential(threadId, existing.providerSessionId); |
| 505 | const rawToken = existing.authorizationHeader.replace(/^Bearer\s+/, ""); |
| 506 | // The caller only learns of the reservation once this returns, |
| 507 | // so a stop while resolving must drop it here. |
| 508 | const resolved = yield* mcpSessionRegistry |
| 509 | .resolve(rawToken) |
| 510 | .pipe( |
| 511 | Effect.onInterrupt(() => |
| 512 | Effect.sync(() => |
| 513 | dropMcpCredentialReservation(threadId, existing.providerSessionId), |
| 514 | ), |
| 515 | ), |
| 516 | ); |
| 517 | if ( |
| 518 | resolved !== undefined && |
| 519 | resolved.thread.threadId === threadId && |
| 520 | resolved.thread.providerInstanceId === providerInstanceId && |
| 521 | // A flipped browser-access setting must not survive through |
| 522 | // credential reuse: rotate so the new scope reflects it. |
| 523 | resolved.capabilities.has("preview") === browserToolsAvailable && |
| 524 | resolved.capabilities.has("device") === deviceToolsAvailable |
| 525 | ) { |
| 526 | return { mcpCredentialId: existing.providerSessionId, issued: false }; |
| 527 | } |
| 528 | dropMcpCredentialReservation(threadId, existing.providerSessionId); |
| 529 | } |
| 530 | yield* mcpSessionRegistry.revokeThread(threadId); |
| 531 | const credential = yield* mcpSessionRegistry.issue({ |
| 532 | threadId, |
| 533 | providerInstanceId, |
| 534 | browserToolsAvailable, |
| 535 | capabilities, |
| 536 | }); |
| 537 | yield* mcpSessions.set(credential.config); |
| 538 | reserveMcpCredential(threadId, credential.config.providerSessionId); |
| 539 | return { mcpCredentialId: credential.config.providerSessionId, issued: true }; |
| 540 | }), |
| 541 | ); |
| 542 | /** |
| 543 | * With a credential id, revocation is scoped to that credential and the |
| 544 | * config slot is cleared only while it still holds it; a replacement |
| 545 | * session's newer credential survives. Without one (attach failed before |
| 546 | * a credential was recorded), fall back to thread-wide revocation. |
| 547 | */ |
| 548 | const clearMcpSession = (threadId: ThreadId, mcpCredentialId?: string) => |
| 549 | mcpCredentialId === undefined |
| 550 | ? mcpSessionRegistry |
| 551 | .revokeThread(threadId) |
| 552 | .pipe(Effect.tap(() => mcpSessions.clear(threadId))) |
| 553 | : mcpSessionRegistry.revokeProviderSession(mcpCredentialId).pipe( |
| 554 | Effect.tap(() => |
| 555 | Effect.gen(function* () { |
| 556 | if ((yield* mcpSessions.read(threadId))?.providerSessionId === mcpCredentialId) { |
| 557 | yield* mcpSessions.clear(threadId); |
| 558 | } |
| 559 | }), |
| 560 | ), |
| 561 | ); |
| 562 | |
| 563 | const publishToSubscribers = ( |
| 564 | subscribers: Ref.Ref< |
| 565 | ReadonlyMap<number, Queue.Queue<ProviderSessionEventSignal, Cause.Done>> |
| 566 | >, |
| 567 | signal: ProviderSessionEventSignal, |
| 568 | ) => |
| 569 | Ref.get(subscribers).pipe( |
| 570 | Effect.flatMap((current) => |
| 571 | Effect.forEach(current.values(), (queue) => Queue.offer(queue, signal), { |
| 572 | discard: true, |
| 573 | }), |
| 574 | ), |
| 575 | ); |
| 576 | |
| 577 | const failSubscribers = (entry: LiveSessionEntry, detail: string) => |
| 578 | Effect.gen(function* () { |
| 579 | const error = new ProviderAdapter.ProviderAdapterEventStreamError({ |
| 580 | driver: entry.runtime.driver, |
| 581 | providerSessionId: entry.runtime.providerSessionId, |
| 582 | cause: detail, |
| 583 | }); |
| 584 | const subscribers = yield* Ref.getAndSet(entry.eventSubscribers, new Map()); |
| 585 | yield* Effect.forEach( |
| 586 | subscribers.values(), |
| 587 | (queue) => |
| 588 | Queue.offer(queue, { |
| 589 | type: "failure", |
| 590 | cause: Cause.fail(error), |
| 591 | }), |
| 592 | { discard: true }, |
| 593 | ); |
| 594 | }); |
| 595 | |
| 596 | const closeSubscribers = (entry: LiveSessionEntry) => |
| 597 | Effect.gen(function* () { |
| 598 | const subscribers = yield* Ref.getAndSet(entry.eventSubscribers, new Map()); |
| 599 | yield* Effect.forEach( |
| 600 | subscribers.values(), |
| 601 | (queue) => Queue.clear(queue).pipe(Effect.andThen(Queue.end(queue))), |
| 602 | { discard: true }, |
| 603 | ); |
| 604 | }); |
| 605 | |
| 606 | // Preserve already-published terminal events while ending subscriptions. |
| 607 | // Server shutdown intentionally clears them; a provider-announced Stop |
| 608 | // must let consumers drain them before the stream completes. |
| 609 | const endSubscribers = (entry: LiveSessionEntry) => |
| 610 | Effect.gen(function* () { |
| 611 | const subscribers = yield* Ref.getAndSet(entry.eventSubscribers, new Map()); |
| 612 | yield* Effect.forEach(subscribers.values(), (queue) => Queue.end(queue), { |
| 613 | discard: true, |
| 614 | }); |
| 615 | }); |
| 616 | |
| 617 | const cancelIdleFiber = (fiber: Fiber.Fiber<void, never> | null) => |
| 618 | fiber === null ? Effect.void : Fiber.interrupt(fiber).pipe(Effect.ignore); |
| 619 | |
| 620 | const writeProviderSessionEvents = (input: { |
| 621 | readonly runtime: ProviderAdapter.ProviderAdapterV2SessionRuntime; |
| 622 | readonly threadIds: Iterable<ThreadId>; |
| 623 | readonly type: "provider-session.attached" | "provider-session.updated"; |
| 624 | readonly payload: OrchestrationV2ProviderSession; |
| 625 | }) => |
| 626 | Effect.gen(function* () { |
| 627 | const now = yield* DateTime.now; |
| 628 | const events = yield* Effect.forEach(input.threadIds, (threadId) => |
| 629 | Effect.gen(function* () { |
| 630 | return { |
| 631 | id: yield* idAllocator.allocate.event({ |
| 632 | threadId, |
| 633 | providerSessionId: input.runtime.providerSessionId, |
| 634 | }), |
| 635 | type: input.type, |
| 636 | threadId, |
| 637 | driver: input.runtime.driver, |
| 638 | providerInstanceId: input.runtime.instanceId, |
| 639 | occurredAt: now, |
| 640 | payload: input.payload, |
| 641 | } satisfies OrchestrationV2DomainEvent; |
| 642 | }), |
| 643 | ); |
| 644 | if (events.length > 0) { |
| 645 | yield* eventSink.write({ events }); |
| 646 | } |
| 647 | }); |
| 648 | |
| 649 | const writeReleasedSessionEvents = (input: { |
| 650 | readonly entry: LiveSessionEntry; |
| 651 | readonly reason: ProviderSessionReleaseReason; |
| 652 | readonly detail?: string; |
| 653 | }) => |
| 654 | Effect.gen(function* () { |
| 655 | const now = yield* DateTime.now; |
| 656 | const payload: OrchestrationV2ProviderSession = { |
| 657 | ...input.entry.runtime.providerSession, |
| 658 | status: releaseStatusFor(input.reason), |
| 659 | updatedAt: now, |
| 660 | lastError: |
| 661 | input.reason === "runtime_error" |
| 662 | ? (input.detail ?? "Provider runtime failed.") |
| 663 | : null, |
| 664 | }; |
| 665 | yield* writeProviderSessionEvents({ |
| 666 | runtime: input.entry.runtime, |
| 667 | threadIds: input.entry.attachedThreadIds, |
| 668 | type: "provider-session.updated", |
| 669 | payload, |
| 670 | }); |
| 671 | }); |
| 672 | |
| 673 | const writeReleasedRuntimeRequestEvents = (input: { |
| 674 | readonly entry: LiveSessionEntry; |
| 675 | readonly reason: ProviderSessionReleaseReason; |
| 676 | /** Requests created later belong to a replacement session with the same id. */ |
| 677 | readonly releasedAt: DateTime.Utc; |
| 678 | }) => |
| 679 | Effect.gen(function* () { |
| 680 | const providerSessionId = input.entry.runtime.providerSessionId; |
| 681 | const now = yield* DateTime.now; |
| 682 | const status = releasedRuntimeRequestStatusFor(input.reason); |
| 683 | const reason = |
| 684 | input.reason === "runtime_error" |
| 685 | ? "Provider session failed before this runtime request was resolved." |
| 686 | : "Provider session was closed before this runtime request was resolved."; |
| 687 | |
| 688 | const events: Array<OrchestrationV2DomainEvent> = []; |
| 689 | for (const threadId of input.entry.attachedThreadIds) { |
| 690 | const projection = yield* projectionStore.getThreadRecords( |
| 691 | threadId, |
| 692 | ["runtimeRequests", "nodes", "turnItems"], |
| 693 | { turnItemTypes: ["approval_request", "user_input_request"] }, |
| 694 | ); |
| 695 | const releasedRequests = projection.runtimeRequests.filter( |
| 696 | (request) => |
| 697 | request.status === "pending" && |
| 698 | request.responseCapability.type === "live" && |
| 699 | request.responseCapability.providerSessionId === providerSessionId && |
| 700 | DateTime.isLessThanOrEqualTo(request.createdAt, input.releasedAt), |
| 701 | ); |
| 702 | |
| 703 | for (const request of releasedRequests) { |
| 704 | events.push({ |
| 705 | id: yield* idAllocator.allocate.event({ |
| 706 | threadId, |
| 707 | providerSessionId, |
| 708 | }), |
| 709 | type: "runtime-request.updated", |
| 710 | threadId, |
| 711 | nodeId: request.nodeId, |
| 712 | driver: input.entry.runtime.driver, |
| 713 | occurredAt: now, |
| 714 | payload: { |
| 715 | ...request, |
| 716 | status, |
| 717 | responseCapability: { |
| 718 | type: "not_resumable", |
| 719 | reason, |
| 720 | }, |
| 721 | resolvedAt: now, |
| 722 | }, |
| 723 | }); |
| 724 | |
| 725 | const requestNode = projection.nodes.find((node) => node.id === request.nodeId); |
| 726 | if (requestNode !== undefined) { |
| 727 | events.push({ |
| 728 | id: yield* idAllocator.allocate.event({ |
| 729 | threadId, |
| 730 | providerSessionId, |
| 731 | }), |
| 732 | type: "node.updated", |
| 733 | threadId, |
| 734 | ...(requestNode.runId === null ? {} : { runId: requestNode.runId }), |
| 735 | nodeId: requestNode.id, |
| 736 | driver: input.entry.runtime.driver, |
| 737 | occurredAt: now, |
| 738 | payload: { |
| 739 | ...requestNode, |
| 740 | status: input.reason === "runtime_error" ? "failed" : "cancelled", |
| 741 | completedAt: now, |
| 742 | }, |
| 743 | }); |
| 744 | } |
| 745 | |
| 746 | const turnItem = projection.turnItems.find( |
| 747 | (item) => |
| 748 | (item.type === "approval_request" || item.type === "user_input_request") && |
| 749 | item.requestId === request.id, |
| 750 | ); |
| 751 | if (turnItem !== undefined) { |
| 752 | events.push({ |
| 753 | id: yield* idAllocator.allocate.event({ |
| 754 | threadId, |
| 755 | providerSessionId, |
| 756 | }), |
| 757 | type: "turn-item.updated", |
| 758 | threadId, |
| 759 | ...(turnItem.runId === null ? {} : { runId: turnItem.runId }), |
| 760 | ...(turnItem.nodeId === null ? {} : { nodeId: turnItem.nodeId }), |
| 761 | driver: input.entry.runtime.driver, |
| 762 | occurredAt: now, |
| 763 | payload: { |
| 764 | ...turnItem, |
| 765 | status: input.reason === "runtime_error" ? "failed" : "cancelled", |
| 766 | completedAt: now, |
| 767 | updatedAt: now, |
| 768 | }, |
| 769 | }); |
| 770 | } |
| 771 | } |
| 772 | } |
| 773 | |
| 774 | if (events.length > 0) { |
| 775 | yield* eventSink.write({ events }); |
| 776 | } |
| 777 | }); |
| 778 | |
| 779 | // Reads the background tasks a released session's process left on its |
| 780 | // provider threads. Read as soon as the session leaves the live map: |
| 781 | // after that, a replacement session with the same id can open and report |
| 782 | // tasks of its own on the same provider threads. Server shutdown keeps |
| 783 | // rosters, since startup recovery clears them and records the cancelled |
| 784 | // work. Only runtimes whose tasks end with their process opt in (Claude, |
| 785 | // Muse). OpenCode's background subagents run on a server shared by its |
| 786 | // sessions and can outlive this one, so their roster stays. |
| 787 | const readReleasedBackgroundTasks = (input: { |
| 788 | readonly entry: LiveSessionEntry; |
| 789 | readonly reason: ProviderSessionReleaseReason; |
| 790 | readonly detachedThreadId?: ThreadId; |
| 791 | }) => |
| 792 | Effect.gen(function* () { |
| 793 | if ( |
| 794 | input.reason === "server_shutdown" || |
| 795 | input.entry.runtime.backgroundTasksEndWithProcess !== true |
| 796 | ) { |
| 797 | return []; |
| 798 | } |
| 799 | const providerSessionId = input.entry.runtime.providerSessionId; |
| 800 | // A detach that released the session already removed its thread |
| 801 | // from the attached set. |
| 802 | const threadIds = new Set(input.entry.attachedThreadIds); |
| 803 | if (input.detachedThreadId !== undefined) threadIds.add(input.detachedThreadId); |
| 804 | const released = []; |
| 805 | for (const threadId of threadIds) { |
| 806 | const projection = yield* projectionStore.getThreadRecords(threadId, [ |
| 807 | "providerThreads", |
| 808 | ]); |
| 809 | for (const providerThread of projection.providerThreads) { |
| 810 | const tasks = providerThread.pendingBackgroundTasks ?? []; |
| 811 | if (providerThread.providerSessionId !== providerSessionId || tasks.length === 0) { |
| 812 | continue; |
| 813 | } |
| 814 | released.push({ |
| 815 | threadId, |
| 816 | providerSessionId, |
| 817 | providerThreadId: providerThread.id, |
| 818 | taskIds: new Set(tasks.map((task) => task.taskId)), |
| 819 | }); |
| 820 | } |
| 821 | } |
| 822 | return released; |
| 823 | }).pipe( |
| 824 | Effect.catchCause((cause) => |
| 825 | Effect.logWarning("orchestration-v2.provider-session-release-roster-read-failed", { |
| 826 | providerSessionId: input.entry.runtime.providerSessionId, |
| 827 | cause, |
| 828 | }).pipe(Effect.as([])), |
| 829 | ), |
| 830 | ); |
| 831 | type ReleasedBackgroundTasks = Effect.Success< |
| 832 | ReturnType<typeof readReleasedBackgroundTasks> |
| 833 | >[number]; |
| 834 | |
| 835 | // Removes the tasks `readReleasedBackgroundTasks` found from their |
| 836 | // rosters. Their process is gone, so nothing will report them ending, |
| 837 | // and a leftover roster keeps the thread Waiting (#15239). Tasks a |
| 838 | // replacement session reported since are left in place. The write |
| 839 | // commits only if the stored row still equals what was read, and is |
| 840 | // retried a few times when a provider update lands in between. |
| 841 | const writeReleasedBackgroundWorkEvents = ( |
| 842 | released: ReadonlyArray<ReleasedBackgroundTasks>, |
| 843 | ) => |
| 844 | Effect.forEach( |
| 845 | released, |
| 846 | ({ threadId, providerSessionId, providerThreadId, taskIds }) => |
| 847 | Effect.gen(function* () { |
| 848 | for (let attempt = 0; attempt < 3; attempt++) { |
| 849 | const projection = yield* projectionStore.getThreadRecords(threadId, [ |
| 850 | "providerThreads", |
| 851 | ]); |
| 852 | const providerThread = projection.providerThreads.find( |
| 853 | (candidate) => candidate.id === providerThreadId, |
| 854 | ); |
| 855 | const tasks = providerThread?.pendingBackgroundTasks ?? []; |
| 856 | const remaining = tasks.filter((task) => !taskIds.has(task.taskId)); |
| 857 | if (providerThread === undefined || remaining.length === tasks.length) return; |
| 858 | const now = yield* DateTime.now; |
| 859 | const { committed } = yield* eventSink.writeIfProviderThreadUnchanged({ |
| 860 | expected: providerThread, |
| 861 | events: [ |
| 862 | { |
| 863 | id: yield* idAllocator.allocate.event({ threadId, providerSessionId }), |
| 864 | type: "provider-thread.updated", |
| 865 | threadId, |
| 866 | driver: providerThread.driver, |
| 867 | providerInstanceId: providerThread.providerInstanceId, |
| 868 | occurredAt: now, |
| 869 | payload: { |
| 870 | ...providerThread, |
| 871 | pendingBackgroundTasks: remaining, |
| 872 | updatedAt: now, |
| 873 | }, |
| 874 | }, |
| 875 | ], |
| 876 | }); |
| 877 | if (committed) return; |
| 878 | } |
| 879 | yield* Effect.logWarning( |
| 880 | "orchestration-v2.provider-session-release-roster-conflict", |
| 881 | { |
| 882 | providerSessionId, |
| 883 | providerThreadId, |
| 884 | }, |
| 885 | ); |
| 886 | }), |
| 887 | { discard: true }, |
| 888 | ); |
| 889 | |
| 890 | // Records a released session as stopped and resolves the live runtime |
| 891 | // requests and background-task rosters it left. Each write runs even if |
| 892 | // another fails. Once a replacement session opens with the same id, it |
| 893 | // owns the session status, so only the requests are settled. |
| 894 | const writeReleaseRecords = (input: { |
| 895 | readonly entry: LiveSessionEntry; |
| 896 | readonly reason: ProviderSessionReleaseReason; |
| 897 | readonly detail?: string; |
| 898 | readonly releasedBackgroundTasks: ReadonlyArray<ReleasedBackgroundTasks>; |
| 899 | readonly releasedAt: DateTime.Utc; |
| 900 | readonly replaced: boolean; |
| 901 | }) => |
| 902 | Effect.all( |
| 903 | [ |
| 904 | input.replaced |
| 905 | ? Effect.succeed(Exit.void) |
| 906 | : Effect.exit(writeReleasedSessionEvents(input)), |
| 907 | Effect.exit( |
| 908 | writeReleasedRuntimeRequestEvents(input).pipe( |
| 909 | input.entry.requestEventPermit.withPermits(1), |
| 910 | ), |
| 911 | ), |
| 912 | Effect.exit(writeReleasedBackgroundWorkEvents(input.releasedBackgroundTasks)), |
| 913 | ], |
| 914 | { concurrency: 1 }, |
| 915 | ).pipe(Effect.flatMap(Exit.asVoidAll)); |
| 916 | |
| 917 | // The session already left the live map, so a later release finds |
| 918 | // nothing to do. Without a retry the UI would keep a ready session and |
| 919 | // answerable approvals until a server restart. Each attempt holds the |
| 920 | // session's open lock, so it sees a replacement that opened meanwhile. |
| 921 | const retryReleaseRecords = ( |
| 922 | input: Omit<Parameters<typeof writeReleaseRecords>[0], "replaced">, |
| 923 | ) => { |
| 924 | const providerSessionId = input.entry.runtime.providerSessionId; |
| 925 | const attempt = Effect.gen(function* () { |
| 926 | const exit = yield* Effect.exit( |
| 927 | sessionOpen.withLock( |
| 928 | providerSessionId, |
| 929 | Effect.gen(function* () { |
| 930 | const replaced = (yield* Ref.get(sessions)).has(sessionKey(providerSessionId)); |
| 931 | yield* writeReleaseRecords({ ...input, replaced }); |
| 932 | }), |
| 933 | ), |
| 934 | ); |
| 935 | if (Exit.isSuccess(exit) || Cause.hasInterruptsOnly(exit.cause)) return yield* exit; |
| 936 | yield* Effect.logWarning("orchestration-v2.provider-session-release-records-failed", { |
| 937 | providerSessionId, |
| 938 | cause: exit.cause, |
| 939 | }); |
| 940 | // A failed SQL commit is a defect, so every failure but interruption |
| 941 | // is retried. |
| 942 | return yield* Effect.fail(exit.cause); |
| 943 | }); |
| 944 | return attempt.pipe( |
| 945 | Effect.retry({ |
| 946 | schedule: Schedule.exponential("1 second").pipe( |
| 947 | Schedule.modifyDelay(({ duration }) => |
| 948 | Effect.succeed(Duration.min(duration, Duration.seconds(30))), |
| 949 | ), |
| 950 | ), |
| 951 | }), |
| 952 | Effect.delay("1 second"), |
| 953 | FiberSet.run(releaseRecordRetries), |
| 954 | ); |
| 955 | }; |
| 956 | |
| 957 | const logReleaseFailure = |
| 958 | (providerSessionId: ProviderSessionId) => |
| 959 | <E, R>(release: Effect.Effect<void, E, R>) => |
| 960 | release.pipe( |
| 961 | Effect.catchCause((cause) => |
| 962 | Effect.logWarning("orchestration-v2.provider-session-release-failed", { |
| 963 | providerSessionId, |
| 964 | cause, |
| 965 | }), |
| 966 | ), |
| 967 | ); |
| 968 | |
| 969 | // Removes the live entry and reads the request cleanup cutoff while |
| 970 | // holding the entry's request permit. A request the event pump is |
| 971 | // persisting for this runtime lands before the cutoff, and once the |
| 972 | // entry is gone the pump persists no more for it. A replacement's |
| 973 | // requests come after its own open. |
| 974 | const removeLiveEntry = (input: { |
| 975 | readonly providerSessionId: ProviderSessionId; |
| 976 | readonly onlyIfIdleGeneration?: number; |
| 977 | }): Effect.Effect<readonly [Option.Option<LiveSessionEntry>, DateTime.Utc]> => |
| 978 | Effect.gen(function* () { |
| 979 | const key = sessionKey(input.providerSessionId); |
| 980 | const candidate = (yield* Ref.get(sessions)).get(key); |
| 981 | if (candidate === undefined) { |
| 982 | return [Option.none<LiveSessionEntry>(), yield* DateTime.now] as const; |
| 983 | } |
| 984 | const removed = yield* Effect.zip( |
| 985 | Ref.modify(sessions, (current) => { |
| 986 | const existing = current.get(key); |
| 987 | if (existing !== candidate) { |
| 988 | return [existing === undefined ? "gone" : "changed", current] as const; |
| 989 | } |
| 990 | if ( |
| 991 | input.onlyIfIdleGeneration !== undefined && |
| 992 | (existing.busyTurns.size > 0 || |
| 993 | existing.idleGeneration !== input.onlyIfIdleGeneration) |
| 994 | ) { |
| 995 | return ["kept", current] as const; |
| 996 | } |
| 997 | const updated = new Map(current); |
| 998 | updated.delete(key); |
| 999 | return ["removed", updated] as const; |
| 1000 | }), |
| 1001 | DateTime.now, |
| 1002 | ).pipe(candidate.requestEventPermit.withPermits(1)); |
| 1003 | const [outcome, releasedAt] = removed; |
| 1004 | // Another entry took this id while the permit was held; release it instead. |
| 1005 | if (outcome === "changed") return yield* removeLiveEntry(input); |
| 1006 | return [ |
| 1007 | outcome === "removed" ? Option.some(candidate) : Option.none<LiveSessionEntry>(), |
| 1008 | releasedAt, |
| 1009 | ] as const; |
| 1010 | }); |
| 1011 | |
| 1012 | // Scope close can wedge on a misbehaving adapter finalizer (e.g. a |
| 1013 | // provider process that never yields its message stream). Time-box it so |
| 1014 | // the caller, and any lock or worker it holds, moves on and leaves a |
| 1015 | // diagnosable trail. A close that finishes late is still logged. |
| 1016 | const closeScopeWithin = ( |
| 1017 | scope: Scope.Closeable, |
| 1018 | annotations: { readonly providerSessionId?: ProviderSessionId; readonly reason: string }, |
| 1019 | ) => |
| 1020 | Effect.gen(function* () { |
| 1021 | const closeFiber = yield* Scope.close(scope, Exit.void).pipe( |
| 1022 | Effect.exit, |
| 1023 | Effect.forkDetach({ startImmediately: true }), |
| 1024 | ); |
| 1025 | const closeExit = yield* Fiber.join(closeFiber).pipe( |
| 1026 | Effect.timeoutOption(RELEASE_SCOPE_CLOSE_TIMEOUT_MS), |
| 1027 | ); |
| 1028 | if (Option.isNone(closeExit)) { |
| 1029 | yield* Effect.logWarning("orchestration-v2.provider-session-scope-close-timeout", { |
| 1030 | ...annotations, |
| 1031 | timeoutMs: RELEASE_SCOPE_CLOSE_TIMEOUT_MS, |
| 1032 | }); |
| 1033 | yield* Fiber.join(closeFiber).pipe( |
| 1034 | Effect.flatMap((exit) => |
| 1035 | Exit.isFailure(exit) |
| 1036 | ? Effect.logWarning("orchestration-v2.provider-session-scope-close-failed", { |
| 1037 | ...annotations, |
| 1038 | cause: exit.cause, |
| 1039 | }) |
| 1040 | : Effect.logInfo( |
| 1041 | "orchestration-v2.provider-session-scope-close-completed-late", |
| 1042 | annotations, |
| 1043 | ), |
| 1044 | ), |
| 1045 | Effect.forkDetach, |
| 1046 | ); |
| 1047 | } |
| 1048 | return closeExit; |
| 1049 | }); |
| 1050 | |
| 1051 | const releaseEntry = (input: { |
| 1052 | readonly providerSessionId: ProviderSessionId; |
| 1053 | readonly reason: ProviderSessionReleaseReason; |
| 1054 | readonly detail?: string; |
| 1055 | /** The thread whose detach released the session; it is no longer attached. */ |
| 1056 | readonly detachedThreadId?: ThreadId; |
| 1057 | readonly cancelIdleFiber?: boolean; |
| 1058 | readonly onlyIfIdleGeneration?: number; |
| 1059 | readonly gracefulSubscribers?: boolean; |
| 1060 | }) => |
| 1061 | Effect.acquireUseRelease( |
| 1062 | removeLiveEntry(input), |
| 1063 | ([entry, releasedAt]) => |
| 1064 | Option.match(entry, { |
| 1065 | onNone: () => Effect.void, |
| 1066 | onSome: (entry) => |
| 1067 | Effect.gen(function* () { |
| 1068 | const releasedBackgroundTasks = yield* readReleasedBackgroundTasks({ |
| 1069 | entry, |
| 1070 | reason: input.reason, |
| 1071 | ...(input.detachedThreadId === undefined |
| 1072 | ? {} |
| 1073 | : { detachedThreadId: input.detachedThreadId }), |
| 1074 | }); |
| 1075 | if (input.cancelIdleFiber !== false) { |
| 1076 | yield* cancelIdleFiber(entry.idleFiber); |
| 1077 | } |
| 1078 | if (input.gracefulSubscribers === true) { |
| 1079 | yield* endSubscribers(entry); |
| 1080 | } else if (input.reason === "server_shutdown") { |
| 1081 | yield* closeSubscribers(entry); |
| 1082 | } else { |
| 1083 | yield* failSubscribers( |
| 1084 | entry, |
| 1085 | input.detail ?? `Provider session released: ${input.reason}.`, |
| 1086 | ); |
| 1087 | } |
| 1088 | const closeExit = yield* closeScopeWithin(entry.scope, { |
| 1089 | providerSessionId: input.providerSessionId, |
| 1090 | reason: input.reason, |
| 1091 | }); |
| 1092 | const records = { |
| 1093 | entry, |
| 1094 | reason: input.reason, |
| 1095 | ...(input.detail === undefined ? {} : { detail: input.detail }), |
| 1096 | releasedBackgroundTasks, |
| 1097 | releasedAt, |
| 1098 | }; |
| 1099 | const recorded = yield* Effect.exit( |
| 1100 | writeReleaseRecords({ ...records, replaced: false }), |
| 1101 | ); |
| 1102 | if (Exit.isFailure(recorded)) { |
| 1103 | yield* retryReleaseRecords(records); |
| 1104 | return yield* recorded; |
| 1105 | } |
| 1106 | if (Option.isSome(closeExit) && Exit.isFailure(closeExit.value)) { |
| 1107 | return yield* Effect.failCause(closeExit.value.cause); |
| 1108 | } |
| 1109 | }).pipe( |
| 1110 | withMetrics({ |
| 1111 | counter: providerSessionsTotal, |
| 1112 | attributes: { |
| 1113 | provider: entry.runtime.driver, |
| 1114 | operation: "release", |
| 1115 | reason: input.reason, |
| 1116 | }, |
| 1117 | }), |
| 1118 | ), |
| 1119 | }), |
| 1120 | ([entry]) => |
| 1121 | Option.match(entry, { |
| 1122 | onNone: () => Effect.void, |
| 1123 | onSome: (entry) => |
| 1124 | // Revoke every credential this session recorded, including for |
| 1125 | // threads that detached without re-attaching: the provider |
| 1126 | // process is gone, so nothing holds them anymore. Skip threads |
| 1127 | // a live replacement session took over, since credential reuse |
| 1128 | // means the replacement may hold this very credential. |
| 1129 | Ref.get(sessions).pipe( |
| 1130 | Effect.flatMap((current) => |
| 1131 | Effect.forEach( |
| 1132 | entry.mcpCredentialIdByThread, |
| 1133 | ([threadId, mcpCredentialId]) => { |
| 1134 | // Id-sensitive: a stale record for the same thread but |
| 1135 | // a DIFFERENT credential (left behind by an old session |
| 1136 | // the thread rotated away from) must not veto revoking |
| 1137 | // this session's own credential, or it leaks forever. |
| 1138 | // A reservation means an in-flight open is configuring |
| 1139 | // a provider process with this credential right now; |
| 1140 | // revoking it here would strand that process (eager |
| 1141 | // adapters cannot pick up a rotated token). |
| 1142 | const heldElsewhere = |
| 1143 | isMcpCredentialReserved(threadId, mcpCredentialId) || |
| 1144 | Array.from(current.values()).some( |
| 1145 | (other) => |
| 1146 | other !== entry && |
| 1147 | (other.attachedThreadIds.has(threadId) || |
| 1148 | other.mcpCredentialIdByThread.get(threadId) === mcpCredentialId), |
| 1149 | ); |
| 1150 | return heldElsewhere |
| 1151 | ? Effect.void |
| 1152 | : clearMcpSession(threadId, mcpCredentialId); |
| 1153 | }, |
| 1154 | { discard: true }, |
| 1155 | ), |
| 1156 | ), |
| 1157 | ), |
| 1158 | }), |
| 1159 | ).pipe( |
| 1160 | Effect.catchCause((cause) => |
| 1161 | Effect.fail( |
| 1162 | new ProviderSessionReleaseError({ |
| 1163 | providerSessionId: input.providerSessionId, |
| 1164 | reason: input.reason, |
| 1165 | cause, |
| 1166 | }), |
| 1167 | ), |
| 1168 | ), |
| 1169 | ); |
| 1170 | |
| 1171 | // Annotated to break the releaseIfStillIdle <-> scheduleIdleReleaseInternal |
| 1172 | // inference cycle introduced by the pin re-arm below. |
| 1173 | const releaseIfStillIdle = (input: { |
| 1174 | readonly providerSessionId: ProviderSessionId; |
| 1175 | readonly generation: number; |
| 1176 | }): Effect.Effect<void> => |
| 1177 | Effect.gen(function* () { |
| 1178 | const current = yield* Ref.get(sessions); |
| 1179 | const key = sessionKey(input.providerSessionId); |
| 1180 | const entry = current.get(key); |
| 1181 | if ( |
| 1182 | entry === undefined || |
| 1183 | entry.busyTurns.size > 0 || |
| 1184 | entry.idleGeneration !== input.generation |
| 1185 | ) { |
| 1186 | return; |
| 1187 | } |
| 1188 | // Capture runtime identity before yielding: a replacement session |
| 1189 | // can reuse the same providerSessionId while this fiber is parked. |
| 1190 | const probedRuntime = entry.runtime; |
| 1191 | const hasPendingWork = |
| 1192 | probedRuntime.hasPendingBackgroundWork === undefined |
| 1193 | ? false |
| 1194 | : yield* probedRuntime.hasPendingBackgroundWork.pipe( |
| 1195 | Effect.catchCause(() => Effect.succeed(false)), |
| 1196 | ); |
| 1197 | if (hasPendingWork) { |
| 1198 | const now = yield* Clock.currentTimeMillis; |
| 1199 | const pinnedSinceMs = entry.pinnedSinceMs ?? now; |
| 1200 | if (now - pinnedSinceMs < maxIdlePinMs) { |
| 1201 | const shouldContinuePin = yield* Ref.modify(sessions, (latest) => { |
| 1202 | const latestEntry = latest.get(key); |
| 1203 | if ( |
| 1204 | latestEntry === undefined || |
| 1205 | latestEntry.busyTurns.size > 0 || |
| 1206 | latestEntry.idleGeneration !== input.generation || |
| 1207 | latestEntry.runtime !== probedRuntime |
| 1208 | ) { |
| 1209 | return [false, latest] as const; |
| 1210 | } |
| 1211 | const updated = new Map(latest); |
| 1212 | updated.set(key, { ...latestEntry, pinnedSinceMs }); |
| 1213 | return [true, updated] as const; |
| 1214 | }); |
| 1215 | if (!shouldContinuePin) { |
| 1216 | // Generation or runtime advanced while we probed pending work; |
| 1217 | // the current owner of the entry owns idle release. |
| 1218 | return; |
| 1219 | } |
| 1220 | yield* Effect.logInfo("orchestration-v2.driver-session.idle-release-deferred", { |
| 1221 | providerSessionId: input.providerSessionId, |
| 1222 | pinnedForMs: now - pinnedSinceMs, |
| 1223 | }); |
| 1224 | // Re-check on this fiber after another idle window. Do not call |
| 1225 | // scheduleIdleReleaseInternal: that cancels entry.idleFiber, which |
| 1226 | // is this fiber, and can self-deadlock on Fiber.interrupt. |
| 1227 | yield* Effect.sleep(Duration.millis(idleTimeoutMs)); |
| 1228 | return yield* releaseIfStillIdle(input); |
| 1229 | } |
| 1230 | yield* Effect.logWarning("orchestration-v2.driver-session.idle-release-pin-expired", { |
| 1231 | providerSessionId: input.providerSessionId, |
| 1232 | pinnedForMs: now - pinnedSinceMs, |
| 1233 | }); |
| 1234 | } |
| 1235 | // hasPendingBackgroundWork yields to the adapter, so the idle |
| 1236 | // decision above can go stale; the generation guard revalidates |
| 1237 | // busyTurns and idleGeneration inside releaseEntry's atomic |
| 1238 | // entry removal. |
| 1239 | yield* releaseEntry({ |
| 1240 | providerSessionId: input.providerSessionId, |
| 1241 | reason: "idle_timeout", |
| 1242 | cancelIdleFiber: false, |
| 1243 | onlyIfIdleGeneration: input.generation, |
| 1244 | }).pipe( |
| 1245 | Effect.catchCause((cause) => |
| 1246 | Effect.logWarning("orchestration-v2.driver-session.idle-release-failed", { |
| 1247 | providerSessionId: input.providerSessionId, |
| 1248 | cause, |
| 1249 | }), |
| 1250 | ), |
| 1251 | ); |
| 1252 | }); |
| 1253 | |
| 1254 | const withActivityError = <A, E, R>( |
| 1255 | providerSessionId: ProviderSessionId, |
| 1256 | effect: Effect.Effect<A, E, R>, |
| 1257 | ): Effect.Effect<A, ProviderSessionActivityError, R> => |
| 1258 | effect.pipe( |
| 1259 | Effect.catchCause((cause) => |
| 1260 | Effect.fail( |
| 1261 | new ProviderSessionActivityError({ |
| 1262 | providerSessionId, |
| 1263 | cause, |
| 1264 | }), |
| 1265 | ), |
| 1266 | ), |
| 1267 | ); |
| 1268 | |
| 1269 | // Provider activity restarts the pin budget, so background work that keeps |
| 1270 | // reporting progress (a long Claude workflow) is never cut off; only work |
| 1271 | // that went silent for maxIdlePinMs is released. |
| 1272 | const scheduleIdleReleaseInternal = ( |
| 1273 | providerSessionId: ProviderSessionId, |
| 1274 | options?: { readonly providerActivity?: boolean }, |
| 1275 | ) => |
| 1276 | Effect.gen(function* () { |
| 1277 | const key = sessionKey(providerSessionId); |
| 1278 | const current = yield* Ref.get(sessions); |
| 1279 | const entry = current.get(key); |
| 1280 | if (entry === undefined || entry.busyTurns.size > 0) { |
| 1281 | return; |
| 1282 | } |
| 1283 | |
| 1284 | yield* cancelIdleFiber(entry.idleFiber); |
| 1285 | const generation = entry.idleGeneration + 1; |
| 1286 | const idleFiber = yield* Effect.sleep(Duration.millis(idleTimeoutMs)).pipe( |
| 1287 | Effect.andThen(releaseIfStillIdle({ providerSessionId, generation })), |
| 1288 | Effect.forkIn(layerScope), |
| 1289 | ); |
| 1290 | const lastActivityAtMs = yield* Clock.currentTimeMillis; |
| 1291 | yield* Ref.update(sessions, (latest) => { |
| 1292 | const latestEntry = latest.get(key); |
| 1293 | if (latestEntry === undefined || latestEntry.busyTurns.size > 0) { |
| 1294 | return latest; |
| 1295 | } |
| 1296 | const updated = new Map(latest); |
| 1297 | updated.set(key, { |
| 1298 | ...latestEntry, |
| 1299 | idleGeneration: generation, |
| 1300 | idleFiber, |
| 1301 | lastActivityAtMs, |
| 1302 | ...(options?.providerActivity === true ? { pinnedSinceMs: null } : {}), |
| 1303 | }); |
| 1304 | return updated; |
| 1305 | }); |
| 1306 | }); |
| 1307 | |
| 1308 | const scheduleIdleRelease = (providerSessionId: ProviderSessionId) => |
| 1309 | withActivityError(providerSessionId, scheduleIdleReleaseInternal(providerSessionId)); |
| 1310 | |
| 1311 | const touchActivity = ( |
| 1312 | providerSessionId: ProviderSessionId, |
| 1313 | options?: { readonly providerActivity?: boolean }, |
| 1314 | ) => |
| 1315 | withActivityError( |
| 1316 | providerSessionId, |
| 1317 | Effect.gen(function* () { |
| 1318 | const lastActivityAtMs = yield* Clock.currentTimeMillis; |
| 1319 | yield* Ref.update(sessions, (current) => { |
| 1320 | const entry = current.get(sessionKey(providerSessionId)); |
| 1321 | if (entry === undefined) { |
| 1322 | return current; |
| 1323 | } |
| 1324 | const updated = new Map(current); |
| 1325 | updated.set(sessionKey(providerSessionId), { |
| 1326 | ...entry, |
| 1327 | lastActivityAtMs, |
| 1328 | }); |
| 1329 | return updated; |
| 1330 | }); |
| 1331 | yield* scheduleIdleReleaseInternal(providerSessionId, options); |
| 1332 | }), |
| 1333 | ); |
| 1334 | |
| 1335 | /** Returns the runtime the thread was attached to, or undefined if it already was. */ |
| 1336 | const attachThread = (input: { |
| 1337 | readonly providerSessionId: ProviderSessionId; |
| 1338 | readonly threadId: ThreadId; |
| 1339 | }) => |
| 1340 | withActivityError( |
| 1341 | input.providerSessionId, |
| 1342 | Ref.modify(sessions, (current) => { |
| 1343 | const entry = current.get(sessionKey(input.providerSessionId)); |
| 1344 | if (entry === undefined || entry.attachedThreadIds.has(input.threadId)) { |
| 1345 | return [undefined, current] as const; |
| 1346 | } |
| 1347 | const updated = new Map(current); |
| 1348 | updated.set(sessionKey(input.providerSessionId), { |
| 1349 | ...entry, |
| 1350 | attachedThreadIds: new Set([...entry.attachedThreadIds, input.threadId]), |
| 1351 | }); |
| 1352 | return [entry.runtime, updated] as const; |
| 1353 | }), |
| 1354 | ); |
| 1355 | |
| 1356 | /** |
| 1357 | * Undoes an attach to `runtime`. A replacement session that reopened under |
| 1358 | * the same id since is left alone. |
| 1359 | */ |
| 1360 | const removeThreadAttachment = (input: { |
| 1361 | readonly providerSessionId: ProviderSessionId; |
| 1362 | readonly threadId: ThreadId; |
| 1363 | readonly runtime: ProviderAdapter.ProviderAdapterV2SessionRuntime; |
| 1364 | }) => |
| 1365 | Ref.update(sessions, (current) => { |
| 1366 | const key = sessionKey(input.providerSessionId); |
| 1367 | const entry = current.get(key); |
| 1368 | if ( |
| 1369 | entry === undefined || |
| 1370 | entry.runtime !== input.runtime || |
| 1371 | !entry.attachedThreadIds.has(input.threadId) |
| 1372 | ) { |
| 1373 | return current; |
| 1374 | } |
| 1375 | const attachedThreadIds = new Set(entry.attachedThreadIds); |
| 1376 | attachedThreadIds.delete(input.threadId); |
| 1377 | const loadedProviderThreadKeyByThread = new Map(entry.loadedProviderThreadKeyByThread); |
| 1378 | loadedProviderThreadKeyByThread.delete(input.threadId); |
| 1379 | const updated = new Map(current); |
| 1380 | updated.set(key, { |
| 1381 | ...entry, |
| 1382 | attachedThreadIds, |
| 1383 | loadedProviderThreadKeyByThread, |
| 1384 | }); |
| 1385 | return updated; |
| 1386 | }); |
| 1387 | |
| 1388 | const isProviderThreadLoaded = (input: { |
| 1389 | readonly providerSessionId: ProviderSessionId; |
| 1390 | readonly threadId: ThreadId; |
| 1391 | readonly providerThreadKey: string; |
| 1392 | }) => |
| 1393 | Ref.get(sessions).pipe( |
| 1394 | Effect.map( |
| 1395 | (current) => |
| 1396 | current |
| 1397 | .get(sessionKey(input.providerSessionId)) |
| 1398 | ?.loadedProviderThreadKeyByThread.get(input.threadId) === input.providerThreadKey, |
| 1399 | ), |
| 1400 | ); |
| 1401 | |
| 1402 | const markProviderThreadLoaded = (input: { |
| 1403 | readonly providerSessionId: ProviderSessionId; |
| 1404 | readonly threadId: ThreadId; |
| 1405 | readonly providerThreadKey: string; |
| 1406 | }) => |
| 1407 | Ref.update(sessions, (current) => { |
| 1408 | const key = sessionKey(input.providerSessionId); |
| 1409 | const entry = current.get(key); |
| 1410 | if (entry === undefined) { |
| 1411 | return current; |
| 1412 | } |
| 1413 | const loadedProviderThreadKeyByThread = new Map(entry.loadedProviderThreadKeyByThread); |
| 1414 | loadedProviderThreadKeyByThread.set(input.threadId, input.providerThreadKey); |
| 1415 | const updated = new Map(current); |
| 1416 | updated.set(key, { ...entry, loadedProviderThreadKeyByThread }); |
| 1417 | return updated; |
| 1418 | }); |
| 1419 | |
| 1420 | const ensureThreadAttached = (input: { |
| 1421 | readonly providerSessionId: ProviderSessionId; |
| 1422 | readonly threadId: ThreadId; |
| 1423 | readonly providerInstanceId: ProviderInstanceId; |
| 1424 | }) => |
| 1425 | Effect.suspend(() => { |
| 1426 | let attachedTo: ProviderAdapter.ProviderAdapterV2SessionRuntime | undefined; |
| 1427 | let preparedForCleanup: PreparedMcpCredential | undefined; |
| 1428 | let reservationDropped = false; |
| 1429 | const dropReservation = () => { |
| 1430 | if (!reservationDropped && preparedForCleanup?.mcpCredentialId !== undefined) { |
| 1431 | reservationDropped = true; |
| 1432 | dropMcpCredentialReservation(input.threadId, preparedForCleanup.mcpCredentialId); |
| 1433 | } |
| 1434 | }; |
| 1435 | // The whole attach, including undoing a failed one, holds the |
| 1436 | // thread's lock: a concurrent attach of the same thread waits, so it |
| 1437 | // never sees an attachment that this call is about to roll back. |
| 1438 | const attach = Effect.gen(function* () { |
| 1439 | const attached = yield* attachThread(input).pipe( |
| 1440 | // Recorded with no gap for an interrupt: cleanup undoes only an |
| 1441 | // attach this call made, never one an earlier open made. |
| 1442 | Effect.tap((runtime) => Effect.sync(() => (attachedTo = runtime))), |
| 1443 | Effect.uninterruptible, |
| 1444 | ); |
| 1445 | if (attached !== undefined) { |
| 1446 | const prepared = yield* prepareMcpSession(input.threadId, input.providerInstanceId); |
| 1447 | preparedForCleanup = prepared; |
| 1448 | if (prepared.mcpCredentialId !== undefined) { |
| 1449 | const mcpCredentialId = prepared.mcpCredentialId; |
| 1450 | yield* Ref.update(sessions, (current) => { |
| 1451 | const key = sessionKey(input.providerSessionId); |
| 1452 | const entry = current.get(key); |
| 1453 | if (entry === undefined) return current; |
| 1454 | const mcpCredentialIdByThread = new Map(entry.mcpCredentialIdByThread); |
| 1455 | mcpCredentialIdByThread.set(input.threadId, mcpCredentialId); |
| 1456 | const updated = new Map(current); |
| 1457 | updated.set(key, { ...entry, mcpCredentialIdByThread }); |
| 1458 | return updated; |
| 1459 | }); |
| 1460 | } |
| 1461 | const entry = (yield* Ref.get(sessions)).get(sessionKey(input.providerSessionId)); |
| 1462 | if (entry !== undefined) { |
| 1463 | yield* withActivityError( |
| 1464 | input.providerSessionId, |
| 1465 | writeProviderSessionEvents({ |
| 1466 | runtime: entry.runtime, |
| 1467 | threadIds: [input.threadId], |
| 1468 | type: "provider-session.attached", |
| 1469 | payload: entry.runtime.providerSession, |
| 1470 | }), |
| 1471 | ); |
| 1472 | } |
| 1473 | } |
| 1474 | }).pipe( |
| 1475 | // An interrupted attach is undone too, so the next attach writes |
| 1476 | // the attachment instead of finding the thread already attached. |
| 1477 | Effect.onError(() => |
| 1478 | attachedTo === undefined |
| 1479 | ? Effect.void |
| 1480 | : removeThreadAttachment({ ...input, runtime: attachedTo }).pipe( |
| 1481 | Effect.andThen( |
| 1482 | Effect.suspend(() => { |
| 1483 | dropReservation(); |
| 1484 | // Revoke only a credential this attach freshly minted: a |
| 1485 | // REUSED credential is by definition held by another |
| 1486 | // live provider process, and revoking it thread-wide |
| 1487 | // would break that process's MCP client mid-conversation. |
| 1488 | if (preparedForCleanup?.issued !== true) return Effect.void; |
| 1489 | const mcpCredentialId = preparedForCleanup.mcpCredentialId; |
| 1490 | const attachedRuntime = attachedTo; |
| 1491 | // As in release: a replacement session (or an open |
| 1492 | // configuring one) may have taken the credential up. |
| 1493 | return Ref.get(sessions).pipe( |
| 1494 | Effect.flatMap((current) => |
| 1495 | (mcpCredentialId !== undefined && |
| 1496 | isMcpCredentialReserved(input.threadId, mcpCredentialId)) || |
| 1497 | Array.from(current.values()).some( |
| 1498 | (other) => |
| 1499 | other.runtime !== attachedRuntime && |
| 1500 | (other.attachedThreadIds.has(input.threadId) || |
| 1501 | (mcpCredentialId !== undefined && |
| 1502 | other.mcpCredentialIdByThread.get(input.threadId) === |
| 1503 | mcpCredentialId)), |
| 1504 | ) |
| 1505 | ? Effect.void |
| 1506 | : clearMcpSession(input.threadId, mcpCredentialId), |
| 1507 | ), |
| 1508 | ); |
| 1509 | }), |
| 1510 | ), |
| 1511 | ), |
| 1512 | ), |
| 1513 | ); |
| 1514 | return threadAttachment.withLock(threadAttachmentKey(input), attach).pipe( |
| 1515 | // The entry's own record (written above while the thread is |
| 1516 | // attached) guards the credential from here on; the reservation |
| 1517 | // is only needed until then. Ensuring covers defects/interrupts. |
| 1518 | Effect.ensuring(Effect.sync(dropReservation)), |
| 1519 | ); |
| 1520 | }); |
| 1521 | |
| 1522 | const markBusy = (providerSessionId: ProviderSessionId, turnKey: string) => |
| 1523 | withActivityError( |
| 1524 | providerSessionId, |
| 1525 | Effect.gen(function* () { |
| 1526 | const key = sessionKey(providerSessionId); |
| 1527 | const now = yield* Clock.currentTimeMillis; |
| 1528 | const idleFiber = yield* Ref.modify(sessions, (current) => { |
| 1529 | const entry = current.get(key); |
| 1530 | if (entry === undefined) { |
| 1531 | return [null, current] as const; |
| 1532 | } |
| 1533 | const updated = new Map(current); |
| 1534 | updated.set(key, { |
| 1535 | ...entry, |
| 1536 | busyTurns: new Set(entry.busyTurns).add(turnKey), |
| 1537 | idleFiber: null, |
| 1538 | lastActivityAtMs: now, |
| 1539 | pinnedSinceMs: null, |
| 1540 | }); |
| 1541 | return [entry.idleFiber, updated] as const; |
| 1542 | }); |
| 1543 | yield* cancelIdleFiber(idleFiber); |
| 1544 | }), |
| 1545 | ); |
| 1546 | |
| 1547 | // Clearing a turn that is not marked busy (one whose failed start already |
| 1548 | // cleared it, or a subagent turn the manager never started) only |
| 1549 | // records activity. |
| 1550 | const markIdle = ( |
| 1551 | providerSessionId: ProviderSessionId, |
| 1552 | providerThreadId: ProviderThreadId, |
| 1553 | runOrdinal: number, |
| 1554 | ) => |
| 1555 | withActivityError( |
| 1556 | providerSessionId, |
| 1557 | Effect.gen(function* () { |
| 1558 | const key = sessionKey(providerSessionId); |
| 1559 | const now = yield* Clock.currentTimeMillis; |
| 1560 | yield* Ref.update(sessions, (current) => { |
| 1561 | const entry = current.get(key); |
| 1562 | if (entry === undefined) { |
| 1563 | return current; |
| 1564 | } |
| 1565 | const busyTurns = new Set(entry.busyTurns); |
| 1566 | busyTurns.delete(busyTurnKey(providerThreadId, runOrdinal)); |
| 1567 | const updated = new Map(current); |
| 1568 | updated.set(key, { |
| 1569 | ...entry, |
| 1570 | busyTurns, |
| 1571 | lastActivityAtMs: now, |
| 1572 | }); |
| 1573 | return updated; |
| 1574 | }); |
| 1575 | yield* scheduleIdleReleaseInternal(providerSessionId); |
| 1576 | yield* scheduleThreadUnload(providerSessionId, providerThreadId); |
| 1577 | }), |
| 1578 | ); |
| 1579 | |
| 1580 | const hasBusyTurn = (entry: LiveSessionEntry, providerThreadId: ProviderThreadId) => { |
| 1581 | const prefix = busyTurnPrefix(providerThreadId); |
| 1582 | for (const turnKey of entry.busyTurns) { |
| 1583 | if (turnKey.startsWith(prefix)) return true; |
| 1584 | } |
| 1585 | return false; |
| 1586 | }; |
| 1587 | |
| 1588 | const updateIdleThreadUnload = ( |
| 1589 | providerSessionId: ProviderSessionId, |
| 1590 | threadId: ThreadId, |
| 1591 | update: (current: IdleThreadUnload | undefined) => IdleThreadUnload | undefined, |
| 1592 | ) => |
| 1593 | Ref.modify(sessions, (current) => { |
| 1594 | const key = sessionKey(providerSessionId); |
| 1595 | const entry = current.get(key); |
| 1596 | if (entry === undefined) return [undefined, current] as const; |
| 1597 | const previous = entry.idleThreadUnloads.get(threadId); |
| 1598 | const next = update(previous); |
| 1599 | const idleThreadUnloads = new Map(entry.idleThreadUnloads); |
| 1600 | if (next === undefined) idleThreadUnloads.delete(threadId); |
| 1601 | else idleThreadUnloads.set(threadId, next); |
| 1602 | const updated = new Map(current); |
| 1603 | updated.set(key, { ...entry, idleThreadUnloads }); |
| 1604 | return [previous, updated] as const; |
| 1605 | }); |
| 1606 | |
| 1607 | /** |
| 1608 | * Starts tracking the provider thread an app thread runs its turns on, and |
| 1609 | * stops any unload pending for it. Called before the thread is resumed or |
| 1610 | * given a turn, so an unload cannot land between a resume that found the |
| 1611 | * thread loaded and the turn that relies on it. |
| 1612 | */ |
| 1613 | const holdThreadLoaded = (input: { |
| 1614 | readonly providerSessionId: ProviderSessionId; |
| 1615 | readonly threadId: ThreadId; |
| 1616 | readonly providerThread: OrchestrationV2ProviderThread; |
| 1617 | }) => |
| 1618 | Effect.gen(function* () { |
| 1619 | const entry = (yield* Ref.get(sessions)).get(sessionKey(input.providerSessionId)); |
| 1620 | if ( |
| 1621 | entry === undefined || |
| 1622 | !entry.supportsMultipleProviderThreads || |
| 1623 | entry.exposedRuntime.unloadThread === undefined || |
| 1624 | input.providerThread.nativeThreadRef === null |
| 1625 | ) { |
| 1626 | return; |
| 1627 | } |
| 1628 | const previous = yield* updateIdleThreadUnload( |
| 1629 | input.providerSessionId, |
| 1630 | input.threadId, |
| 1631 | (current) => ({ |
| 1632 | providerThread: input.providerThread, |
| 1633 | generation: (current?.generation ?? 0) + 1, |
| 1634 | fiber: null, |
| 1635 | }), |
| 1636 | ); |
| 1637 | yield* cancelIdleFiber(previous?.fiber ?? null); |
| 1638 | }); |
| 1639 | |
| 1640 | /** |
| 1641 | * A shared runtime never goes idle while any of its threads is in use, so |
| 1642 | * its threads get the idle timeout one by one: a thread with no turn for |
| 1643 | * `idleTimeoutMs` is unloaded from the runtime, along with the native MCP |
| 1644 | * servers it started. Its next turn's resume loads it again. |
| 1645 | */ |
| 1646 | const scheduleThreadUnload = ( |
| 1647 | providerSessionId: ProviderSessionId, |
| 1648 | providerThreadId: ProviderThreadId, |
| 1649 | ) => |
| 1650 | Effect.gen(function* () { |
| 1651 | const entry = (yield* Ref.get(sessions)).get(sessionKey(providerSessionId)); |
| 1652 | if (entry === undefined || hasBusyTurn(entry, providerThreadId)) return; |
| 1653 | const tracked = Array.from(entry.idleThreadUnloads).find( |
| 1654 | ([, pending]) => pending.providerThread.id === providerThreadId, |
| 1655 | ); |
| 1656 | if (tracked === undefined) return; |
| 1657 | const [threadId, pending] = tracked; |
| 1658 | const generation = pending.generation + 1; |
| 1659 | const fiber = yield* Effect.sleep(Duration.millis(idleTimeoutMs)).pipe( |
| 1660 | Effect.andThen(unloadIdleThread({ providerSessionId, threadId, generation })), |
| 1661 | Effect.forkIn(layerScope), |
| 1662 | ); |
| 1663 | // A turn that started meanwhile already moved the generation on. |
| 1664 | const previous = yield* updateIdleThreadUnload(providerSessionId, threadId, (current) => |
| 1665 | current?.generation === pending.generation |
| 1666 | ? { ...current, generation, fiber } |
| 1667 | : current, |
| 1668 | ); |
| 1669 | yield* cancelIdleFiber( |
| 1670 | previous?.generation === pending.generation ? previous.fiber : fiber, |
| 1671 | ); |
| 1672 | }); |
| 1673 | |
| 1674 | const unloadIdleThread = (input: { |
| 1675 | readonly providerSessionId: ProviderSessionId; |
| 1676 | readonly threadId: ThreadId; |
| 1677 | readonly generation: number; |
| 1678 | }): Effect.Effect<void> => |
| 1679 | Effect.gen(function* () { |
| 1680 | const outcome = yield* threadAttachment.withLock( |
| 1681 | threadAttachmentKey(input), |
| 1682 | Effect.gen(function* () { |
| 1683 | const key = sessionKey(input.providerSessionId); |
| 1684 | const entry = (yield* Ref.get(sessions)).get(key); |
| 1685 | const pending = entry?.idleThreadUnloads.get(input.threadId); |
| 1686 | const unloadThread = entry?.exposedRuntime.unloadThread; |
| 1687 | if ( |
| 1688 | entry === undefined || |
| 1689 | pending === undefined || |
| 1690 | pending.generation !== input.generation || |
| 1691 | unloadThread === undefined || |
| 1692 | !entry.attachedThreadIds.has(input.threadId) || |
| 1693 | hasBusyTurn(entry, pending.providerThread.id) |
| 1694 | ) { |
| 1695 | return "skipped" as const; |
| 1696 | } |
| 1697 | // Unloading stops the native thread's background terminals, so a |
| 1698 | // thread still running background work stays loaded. |
| 1699 | const hasPendingWork = |
| 1700 | entry.runtime.hasPendingBackgroundWorkForThread === undefined |
| 1701 | ? false |
| 1702 | : yield* entry.runtime |
| 1703 | .hasPendingBackgroundWorkForThread(pending.providerThread) |
| 1704 | .pipe(Effect.catchCause(() => Effect.succeed(false))); |
| 1705 | if (hasPendingWork) return "deferred" as const; |
| 1706 | const unloading = yield* Ref.modify(sessions, (current) => { |
| 1707 | const latest = current.get(key); |
| 1708 | if ( |
| 1709 | latest?.runtime !== entry.runtime || |
| 1710 | latest.idleThreadUnloads.get(input.threadId)?.generation !== input.generation |
| 1711 | ) { |
| 1712 | return [false, current] as const; |
| 1713 | } |
| 1714 | const loadedProviderThreadKeyByThread = new Map( |
| 1715 | latest.loadedProviderThreadKeyByThread, |
| 1716 | ); |
| 1717 | loadedProviderThreadKeyByThread.delete(input.threadId); |
| 1718 | const idleThreadUnloads = new Map(latest.idleThreadUnloads); |
| 1719 | idleThreadUnloads.delete(input.threadId); |
| 1720 | const updated = new Map(current); |
| 1721 | updated.set(key, { |
| 1722 | ...latest, |
| 1723 | loadedProviderThreadKeyByThread, |
| 1724 | idleThreadUnloads, |
| 1725 | }); |
| 1726 | return [true, updated] as const; |
| 1727 | }); |
| 1728 | if (!unloading) return "skipped" as const; |
| 1729 | yield* unloadThread({ providerThread: pending.providerThread }).pipe( |
| 1730 | Effect.timeout(UNLOAD_THREAD_TIMEOUT_MS), |
| 1731 | Effect.catchCause((cause) => |
| 1732 | Effect.logWarning("orchestration-v2.driver-session.idle-unload-failed", { |
| 1733 | providerSessionId: input.providerSessionId, |
| 1734 | threadId: input.threadId, |
| 1735 | providerThreadId: pending.providerThread.id, |
| 1736 | cause, |
| 1737 | }), |
| 1738 | ), |
| 1739 | ); |
| 1740 | return "unloaded" as const; |
| 1741 | }), |
| 1742 | ); |
| 1743 | // Re-check on this fiber after another idle window, outside the |
| 1744 | // lock so the thread's next attach is not held up meanwhile. |
| 1745 | if (outcome === "deferred") { |
| 1746 | yield* Effect.sleep(Duration.millis(idleTimeoutMs)); |
| 1747 | return yield* unloadIdleThread(input); |
| 1748 | } |
| 1749 | }); |
| 1750 | |
| 1751 | const observeActivity = ( |
| 1752 | providerSessionId: ProviderSessionId, |
| 1753 | activity: Effect.Effect<void, ProviderSessionActivityError>, |
| 1754 | ) => |
| 1755 | activity.pipe( |
| 1756 | Effect.catchCause((cause) => |
| 1757 | Effect.logWarning("orchestration-v2.driver-session.activity-failed", { |
| 1758 | providerSessionId, |
| 1759 | cause, |
| 1760 | }), |
| 1761 | ), |
| 1762 | ); |
| 1763 | |
| 1764 | const makeEventSubscription = ( |
| 1765 | subscribers: Ref.Ref< |
| 1766 | ReadonlyMap<number, Queue.Queue<ProviderSessionEventSignal, Cause.Done>> |
| 1767 | >, |
| 1768 | ): Effect.Effect<ProviderAdapter.ProviderAdapterV2EventSubscription> => |
| 1769 | Effect.gen(function* () { |
| 1770 | const queue = yield* Queue.unbounded<ProviderSessionEventSignal, Cause.Done>(); |
| 1771 | const subscriberId = yield* Ref.getAndUpdate(nextSubscriberId, (value) => value + 1); |
| 1772 | yield* Ref.update(subscribers, (current) => { |
| 1773 | const updated = new Map(current); |
| 1774 | updated.set(subscriberId, queue); |
| 1775 | return updated; |
| 1776 | }); |
| 1777 | const close = Ref.modify(subscribers, (current) => { |
| 1778 | if (!current.has(subscriberId)) { |
| 1779 | return [false, current] as const; |
| 1780 | } |
| 1781 | const updated = new Map(current); |
| 1782 | updated.delete(subscriberId); |
| 1783 | return [true, updated] as const; |
| 1784 | }).pipe( |
| 1785 | Effect.flatMap((removed) => |
| 1786 | removed |
| 1787 | ? Queue.clear(queue).pipe(Effect.andThen(Queue.end(queue)), Effect.asVoid) |
| 1788 | : Effect.void, |
| 1789 | ), |
| 1790 | ); |
| 1791 | const events = Stream.fromQueue(queue).pipe( |
| 1792 | Stream.mapEffect((signal) => |
| 1793 | signal.type === "event" |
| 1794 | ? Effect.succeed(signal.event) |
| 1795 | : Effect.failCause(signal.cause), |
| 1796 | ), |
| 1797 | Stream.ensuring(close), |
| 1798 | ); |
| 1799 | return { events, close } satisfies ProviderAdapter.ProviderAdapterV2EventSubscription; |
| 1800 | }); |
| 1801 | |
| 1802 | const decorateRuntime = ( |
| 1803 | runtime: ProviderAdapter.ProviderAdapterV2SessionRuntime, |
| 1804 | eventSubscribers: Ref.Ref< |
| 1805 | ReadonlyMap<number, Queue.Queue<ProviderSessionEventSignal, Cause.Done>> |
| 1806 | >, |
| 1807 | ): ProviderAdapter.ProviderAdapterV2SessionRuntime => { |
| 1808 | const providerSessionId = runtime.providerSessionId; |
| 1809 | const subscribeEvents = makeEventSubscription(eventSubscribers); |
| 1810 | // Every provider's turn operations pass through here, so this is where they are |
| 1811 | // counted. Only turn starts are timed: until the provider accepts the turn. |
| 1812 | const turnMetrics = (operation: string, model?: string) => |
| 1813 | withMetrics({ |
| 1814 | counter: providerTurnsTotal, |
| 1815 | ...(operation === "send" ? { timer: providerTurnDuration } : {}), |
| 1816 | attributes: { |
| 1817 | provider: runtime.driver, |
| 1818 | operation, |
| 1819 | modelFamily: normalizeModelMetricLabel(model), |
| 1820 | }, |
| 1821 | }); |
| 1822 | return { |
| 1823 | ...runtime, |
| 1824 | subscribeEvents, |
| 1825 | events: Stream.unwrap( |
| 1826 | subscribeEvents.pipe(Effect.map((subscription) => subscription.events)), |
| 1827 | ), |
| 1828 | ensureThread: (input) => |
| 1829 | observeActivity( |
| 1830 | providerSessionId, |
| 1831 | ensureThreadAttached({ |
| 1832 | providerSessionId, |
| 1833 | threadId: input.threadId, |
| 1834 | providerInstanceId: runtime.instanceId, |
| 1835 | }), |
| 1836 | ).pipe( |
| 1837 | Effect.andThen(runtime.ensureThread(input)), |
| 1838 | Effect.tap((providerThread) => |
| 1839 | markProviderThreadLoaded({ |
| 1840 | providerSessionId, |
| 1841 | threadId: input.threadId, |
| 1842 | providerThreadKey: providerThreadLoadKey({ |
| 1843 | providerThread, |
| 1844 | modelSelection: input.modelSelection, |
| 1845 | runtimePolicy: input.runtimePolicy, |
| 1846 | }), |
| 1847 | }), |
| 1848 | ), |
| 1849 | ), |
| 1850 | resumeThread: (input) => { |
| 1851 | const threadId = input.threadId ?? input.providerThread.appThreadId; |
| 1852 | if (threadId === null || threadId === undefined) { |
| 1853 | return runtime.resumeThread(input); |
| 1854 | } |
| 1855 | const providerThreadKey = providerThreadLoadKey({ |
| 1856 | providerThread: input.providerThread, |
| 1857 | ...(input.modelSelection === undefined |
| 1858 | ? {} |
| 1859 | : { modelSelection: input.modelSelection }), |
| 1860 | ...(input.runtimePolicy === undefined ? {} : { runtimePolicy: input.runtimePolicy }), |
| 1861 | }); |
| 1862 | return observeActivity( |
| 1863 | providerSessionId, |
| 1864 | ensureThreadAttached({ |
| 1865 | providerSessionId, |
| 1866 | threadId, |
| 1867 | providerInstanceId: runtime.instanceId, |
| 1868 | }), |
| 1869 | ).pipe( |
| 1870 | Effect.andThen( |
| 1871 | holdThreadLoaded({ |
| 1872 | providerSessionId, |
| 1873 | threadId, |
| 1874 | providerThread: input.providerThread, |
| 1875 | }), |
| 1876 | ), |
| 1877 | Effect.andThen( |
| 1878 | isProviderThreadLoaded({ providerSessionId, threadId, providerThreadKey }), |
| 1879 | ), |
| 1880 | Effect.flatMap((loaded) => |
| 1881 | loaded ? Effect.succeed(input.providerThread) : runtime.resumeThread(input), |
| 1882 | ), |
| 1883 | Effect.tap((providerThread) => |
| 1884 | markProviderThreadLoaded({ |
| 1885 | providerSessionId, |
| 1886 | threadId, |
| 1887 | providerThreadKey: providerThreadLoadKey({ |
| 1888 | providerThread, |
| 1889 | ...(input.modelSelection === undefined |
| 1890 | ? {} |
| 1891 | : { modelSelection: input.modelSelection }), |
| 1892 | ...(input.runtimePolicy === undefined |
| 1893 | ? {} |
| 1894 | : { runtimePolicy: input.runtimePolicy }), |
| 1895 | }), |
| 1896 | }), |
| 1897 | ), |
| 1898 | ); |
| 1899 | }, |
| 1900 | forkThread: (input) => |
| 1901 | observeActivity( |
| 1902 | providerSessionId, |
| 1903 | ensureThreadAttached({ |
| 1904 | providerSessionId, |
| 1905 | threadId: input.targetThreadId, |
| 1906 | providerInstanceId: runtime.instanceId, |
| 1907 | }), |
| 1908 | ).pipe( |
| 1909 | Effect.andThen(runtime.forkThread(input)), |
| 1910 | Effect.tap((providerThread) => |
| 1911 | markProviderThreadLoaded({ |
| 1912 | providerSessionId, |
| 1913 | threadId: input.targetThreadId, |
| 1914 | providerThreadKey: providerThreadLoadKey({ |
| 1915 | providerThread, |
| 1916 | ...(input.modelSelection === undefined |
| 1917 | ? {} |
| 1918 | : { modelSelection: input.modelSelection }), |
| 1919 | ...(input.runtimePolicy === undefined |
| 1920 | ? {} |
| 1921 | : { runtimePolicy: input.runtimePolicy }), |
| 1922 | }), |
| 1923 | }), |
| 1924 | ), |
| 1925 | ), |
| 1926 | startTurn: (input) => |
| 1927 | observeActivity( |
| 1928 | providerSessionId, |
| 1929 | ensureThreadAttached({ |
| 1930 | providerSessionId, |
| 1931 | threadId: input.threadId, |
| 1932 | providerInstanceId: runtime.instanceId, |
| 1933 | }), |
| 1934 | ).pipe( |
| 1935 | // A start that fails or is stopped may never emit turn.terminal, |
| 1936 | // so it clears its own turn or the session never goes idle. If |
| 1937 | // the adapter emits the terminal anyway, clearing the same turn |
| 1938 | // again changes nothing, so another thread's turn on a shared |
| 1939 | // session stays busy either way. |
| 1940 | Effect.andThen( |
| 1941 | holdThreadLoaded({ |
| 1942 | providerSessionId, |
| 1943 | threadId: input.threadId, |
| 1944 | providerThread: input.providerThread, |
| 1945 | }), |
| 1946 | ), |
| 1947 | Effect.andThen( |
| 1948 | Effect.acquireUseRelease( |
| 1949 | observeActivity( |
| 1950 | providerSessionId, |
| 1951 | markBusy( |
| 1952 | providerSessionId, |
| 1953 | busyTurnKey(input.providerThread.id, input.runOrdinal), |
| 1954 | ), |
| 1955 | ), |
| 1956 | () => |
| 1957 | runtime.startTurn(input).pipe(turnMetrics("send", input.modelSelection.model)), |
| 1958 | (_, exit) => |
| 1959 | Exit.isFailure(exit) |
| 1960 | ? observeActivity( |
| 1961 | providerSessionId, |
| 1962 | markIdle(providerSessionId, input.providerThread.id, input.runOrdinal), |
| 1963 | ) |
| 1964 | : Effect.void, |
| 1965 | ), |
| 1966 | ), |
| 1967 | ), |
| 1968 | steerTurn: (input) => |
| 1969 | observeActivity(providerSessionId, touchActivity(providerSessionId)).pipe( |
| 1970 | Effect.andThen(runtime.steerTurn(input).pipe(turnMetrics("steer"))), |
| 1971 | ), |
| 1972 | interruptTurn: (input) => |
| 1973 | observeActivity(providerSessionId, touchActivity(providerSessionId)).pipe( |
| 1974 | Effect.andThen(runtime.interruptTurn(input).pipe(turnMetrics("interrupt"))), |
| 1975 | ), |
| 1976 | respondToRuntimeRequest: (input) => |
| 1977 | observeActivity(providerSessionId, touchActivity(providerSessionId)).pipe( |
| 1978 | Effect.andThen( |
| 1979 | runtime |
| 1980 | .respondToRuntimeRequest(input) |
| 1981 | .pipe(turnMetrics("runtime-request-response")), |
| 1982 | ), |
| 1983 | ), |
| 1984 | }; |
| 1985 | }; |
| 1986 | |
| 1987 | const persistProviderSessionUpdate = ( |
| 1988 | entry: LiveSessionEntry, |
| 1989 | event: Extract< |
| 1990 | ProviderAdapter.ProviderAdapterV2Event, |
| 1991 | { readonly type: "provider_session.updated" } |
| 1992 | >, |
| 1993 | ) => |
| 1994 | Effect.gen(function* () { |
| 1995 | const current = (yield* Ref.get(sessions)).get( |
| 1996 | sessionKey(entry.runtime.providerSessionId), |
| 1997 | ); |
| 1998 | if (current?.runtime !== entry.runtime) { |
| 1999 | return; |
| 2000 | } |
| 2001 | yield* writeProviderSessionEvents({ |
| 2002 | runtime: entry.runtime, |
| 2003 | threadIds: current.attachedThreadIds, |
| 2004 | type: "provider-session.updated", |
| 2005 | payload: event.providerSession, |
| 2006 | }); |
| 2007 | }).pipe( |
| 2008 | Effect.catchCause((cause) => |
| 2009 | Effect.logWarning("orchestration-v2.driver-session.status-persist-failed", { |
| 2010 | providerSessionId: entry.runtime.providerSessionId, |
| 2011 | cause, |
| 2012 | }), |
| 2013 | ), |
| 2014 | ); |
| 2015 | |
| 2016 | const startEventPump = (entry: LiveSessionEntry) => { |
| 2017 | let stoppedByProvider = false; |
| 2018 | return entry.runtime.events.pipe( |
| 2019 | Stream.runForEach((event) => { |
| 2020 | if (shutdownSignal.received) return Effect.void; |
| 2021 | if ( |
| 2022 | event.type === "provider_session.updated" && |
| 2023 | event.providerSession.status === "stopped" |
| 2024 | ) { |
| 2025 | stoppedByProvider = true; |
| 2026 | } |
| 2027 | return observeActivity( |
| 2028 | entry.runtime.providerSessionId, |
| 2029 | event.type === "turn.terminal" |
| 2030 | ? markIdle( |
| 2031 | entry.runtime.providerSessionId, |
| 2032 | event.providerThreadId, |
| 2033 | event.runOrdinal, |
| 2034 | ) |
| 2035 | : touchActivity(entry.runtime.providerSessionId, { providerActivity: true }), |
| 2036 | ).pipe( |
| 2037 | Effect.andThen( |
| 2038 | event.type === "provider_session.updated" |
| 2039 | ? persistProviderSessionUpdate(entry, event) |
| 2040 | : Effect.void, |
| 2041 | ), |
| 2042 | Effect.andThen( |
| 2043 | Effect.gen(function* () { |
| 2044 | // Some providers can block before a run subscriber exists |
| 2045 | // (project trust, login, or session-switch hooks). Persist |
| 2046 | // their runless request artifacts directly so the normal T3 |
| 2047 | // request UI can answer them and unblock session setup. |
| 2048 | const threadId = sessionScopedRuntimeRequestThreadId(event); |
| 2049 | if (threadId !== undefined) { |
| 2050 | yield* Effect.gen(function* () { |
| 2051 | const current = (yield* Ref.get(sessions)).get( |
| 2052 | sessionKey(entry.runtime.providerSessionId), |
| 2053 | ); |
| 2054 | if (current?.runtime !== entry.runtime) return; |
| 2055 | yield* providerEventIngestor |
| 2056 | .ingestNormalized({ |
| 2057 | providerSessionId: entry.runtime.providerSessionId, |
| 2058 | providerInstanceId: entry.runtime.instanceId, |
| 2059 | threadId, |
| 2060 | event, |
| 2061 | }) |
| 2062 | .pipe( |
| 2063 | Effect.mapError( |
| 2064 | (cause) => |
| 2065 | new ProviderAdapter.ProviderAdapterEventStreamError({ |
| 2066 | driver: entry.runtime.driver, |
| 2067 | providerSessionId: entry.runtime.providerSessionId, |
| 2068 | cause, |
| 2069 | }), |
| 2070 | ), |
| 2071 | ); |
| 2072 | }).pipe(entry.requestEventPermit.withPermits(1)); |
| 2073 | return; |
| 2074 | } |
| 2075 | yield* publishToSubscribers(entry.eventSubscribers, { type: "event", event }); |
| 2076 | }), |
| 2077 | ), |
| 2078 | ); |
| 2079 | }), |
| 2080 | Effect.exit, |
| 2081 | Effect.flatMap((exit) => |
| 2082 | Effect.gen(function* () { |
| 2083 | // A provider that exits on the shutdown signal is released by shutdown. |
| 2084 | if (shutdownSignal.received) return; |
| 2085 | const current = (yield* Ref.get(sessions)).get( |
| 2086 | sessionKey(entry.runtime.providerSessionId), |
| 2087 | ); |
| 2088 | if (current?.runtime !== entry.runtime) { |
| 2089 | return; |
| 2090 | } |
| 2091 | if (stoppedByProvider && Exit.isSuccess(exit)) { |
| 2092 | yield* releaseEntry({ |
| 2093 | providerSessionId: entry.runtime.providerSessionId, |
| 2094 | reason: "manual_shutdown", |
| 2095 | gracefulSubscribers: true, |
| 2096 | }).pipe(logReleaseFailure(entry.runtime.providerSessionId)); |
| 2097 | return; |
| 2098 | } |
| 2099 | const cause = Exit.isFailure(exit) |
| 2100 | ? exit.cause |
| 2101 | : Cause.fail( |
| 2102 | new ProviderAdapter.ProviderAdapterEventStreamError({ |
| 2103 | driver: entry.runtime.driver, |
| 2104 | providerSessionId: entry.runtime.providerSessionId, |
| 2105 | cause: "Provider event stream ended unexpectedly.", |
| 2106 | }), |
| 2107 | ); |
| 2108 | yield* publishToSubscribers(entry.eventSubscribers, { |
| 2109 | type: "failure", |
| 2110 | cause, |
| 2111 | }); |
| 2112 | yield* Ref.set(entry.eventSubscribers, new Map()); |
| 2113 | yield* releaseEntry({ |
| 2114 | providerSessionId: entry.runtime.providerSessionId, |
| 2115 | reason: "runtime_error", |
| 2116 | detail: Cause.pretty(cause), |
| 2117 | }).pipe(logReleaseFailure(entry.runtime.providerSessionId)); |
| 2118 | }), |
| 2119 | ), |
| 2120 | Effect.forkIn(layerScope), |
| 2121 | ); |
| 2122 | }; |
| 2123 | |
| 2124 | // Parent of every session scope. On layer close, shutdown releases the |
| 2125 | // live sessions first, then closes any session whose open is still in |
| 2126 | // flight, time-boxed so a stuck adapter cannot hold up server shutdown. |
| 2127 | // Both steps run in parallel. The service launcher kills the server |
| 2128 | // 5 s after it asks it to stop, and a busy server can hold 15 sessions. |
| 2129 | const sessionScopes = yield* Scope.make("parallel"); |
| 2130 | const shutdown = Effect.gen(function* () { |
| 2131 | const activeSessions = [...(yield* Ref.get(sessions)).values()]; |
| 2132 | yield* Effect.forEach( |
| 2133 | activeSessions, |
| 2134 | (entry) => |
| 2135 | releaseEntry({ |
| 2136 | providerSessionId: entry.runtime.providerSessionId, |
| 2137 | reason: "server_shutdown", |
| 2138 | }).pipe( |
| 2139 | Effect.catchCause((cause) => |
| 2140 | Effect.logWarning("orchestration-v2.driver-session.shutdown-release-failed", { |
| 2141 | providerSessionId: entry.runtime.providerSessionId, |
| 2142 | cause, |
| 2143 | }), |
| 2144 | ), |
| 2145 | ), |
| 2146 | { concurrency: "unbounded", discard: true }, |
| 2147 | ); |
| 2148 | }); |
| 2149 | yield* Effect.addFinalizer(() => |
| 2150 | shutdown.pipe( |
| 2151 | Effect.ensuring(closeScopeWithin(sessionScopes, { reason: "server_shutdown" })), |
| 2152 | ), |
| 2153 | ); |
| 2154 | |
| 2155 | return ProviderSessionManagerV2.of({ |
| 2156 | shutdown, |
| 2157 | open: (input) => |
| 2158 | sessionOpen.withLock( |
| 2159 | input.providerSessionId, |
| 2160 | Effect.gen(function* () { |
| 2161 | const cwd = input.runtimePolicy.cwd; |
| 2162 | if (cwd !== null) { |
| 2163 | const workspaceIsDirectory = yield* fileSystem.stat(cwd).pipe( |
| 2164 | Effect.map((stat) => stat.type === "Directory"), |
| 2165 | Effect.catch((error) => Effect.succeed(error.reason._tag !== "NotFound")), |
| 2166 | ); |
| 2167 | if (!workspaceIsDirectory) { |
| 2168 | return yield* new ProviderWorkspaceMissingError({ |
| 2169 | threadId: input.threadId, |
| 2170 | cwd, |
| 2171 | }); |
| 2172 | } |
| 2173 | } |
| 2174 | const key = sessionKey(input.providerSessionId); |
| 2175 | const existing = (yield* Ref.get(sessions)).get(key); |
| 2176 | if (existing !== undefined) { |
| 2177 | const attached = existing.attachedThreadIds.has(input.threadId); |
| 2178 | if (!attached && !existing.supportsMultipleProviderThreads) { |
| 2179 | return yield* new ProviderSessionOpenError({ |
| 2180 | instanceId: input.modelSelection.instanceId, |
| 2181 | providerSessionId: input.providerSessionId, |
| 2182 | cause: `Provider ${existing.runtime.driver} does not support attaching multiple app threads to one session.`, |
| 2183 | }); |
| 2184 | } |
| 2185 | // Joining a shared session is a new session for this thread. |
| 2186 | if (!attached) yield* agentScope.clear(input.threadId); |
| 2187 | yield* ensureThreadAttached({ |
| 2188 | providerSessionId: input.providerSessionId, |
| 2189 | threadId: input.threadId, |
| 2190 | providerInstanceId: existing.runtime.instanceId, |
| 2191 | }); |
| 2192 | yield* touchActivity(input.providerSessionId); |
| 2193 | return existing.exposedRuntime; |
| 2194 | } |
| 2195 | |
| 2196 | const adapter = yield* registry.get(input.modelSelection.instanceId).pipe( |
| 2197 | Effect.mapError( |
| 2198 | (cause) => |
| 2199 | new ProviderSessionOpenError({ |
| 2200 | instanceId: input.modelSelection.instanceId, |
| 2201 | providerSessionId: input.providerSessionId, |
| 2202 | cause, |
| 2203 | }), |
| 2204 | ), |
| 2205 | ); |
| 2206 | const prepared = yield* prepareMcpSession( |
| 2207 | input.threadId, |
| 2208 | input.modelSelection.instanceId, |
| 2209 | ); |
| 2210 | const mcpCredentialId = prepared.mcpCredentialId; |
| 2211 | // The reservation from prepare protects the credential (which |
| 2212 | // eager adapters bake into the provider process during |
| 2213 | // openSession) from racing releases until this session's entry |
| 2214 | // is recorded below. Dropped exactly once on every path. |
| 2215 | let reservationDropped = mcpCredentialId === undefined; |
| 2216 | const dropReservation = Effect.sync(() => { |
| 2217 | if (!reservationDropped && mcpCredentialId !== undefined) { |
| 2218 | reservationDropped = true; |
| 2219 | dropMcpCredentialReservation(input.threadId, mcpCredentialId); |
| 2220 | } |
| 2221 | }); |
| 2222 | const sessionScope = yield* Scope.fork(sessionScopes); |
| 2223 | // The previous session's agent scope must not explain this |
| 2224 | // session's failures, even when this provider runs without one. |
| 2225 | yield* agentScope.clear(input.threadId); |
| 2226 | const runtime = yield* adapter |
| 2227 | .openSession({ |
| 2228 | threadId: input.threadId, |
| 2229 | providerSessionId: input.providerSessionId, |
| 2230 | modelSelection: input.modelSelection, |
| 2231 | runtimePolicy: input.runtimePolicy, |
| 2232 | ...(input.resumeFromSession === undefined |
| 2233 | ? {} |
| 2234 | : { resumeFromSession: input.resumeFromSession }), |
| 2235 | ...(input.initialNativeThreadId === undefined |
| 2236 | ? {} |
| 2237 | : { initialNativeThreadId: input.initialNativeThreadId }), |
| 2238 | ...(input.initialProviderItemIdentityVersion === undefined |
| 2239 | ? {} |
| 2240 | : { |
| 2241 | initialProviderItemIdentityVersion: |
| 2242 | input.initialProviderItemIdentityVersion, |
| 2243 | }), |
| 2244 | }) |
| 2245 | .pipe( |
| 2246 | Effect.provideService(Scope.Scope, sessionScope), |
| 2247 | // Any failure, including a Stop that interrupts a slow |
| 2248 | // handshake, stops the provider process this open started. |
| 2249 | // The session cleanup runs first, and the close is |
| 2250 | // time-boxed: this runs under the session's open lock, so an |
| 2251 | // adapter finalizer that never finishes must not hold the |
| 2252 | // interrupter, the lock, or later opens. |
| 2253 | Effect.onError(() => |
| 2254 | dropReservation.pipe( |
| 2255 | // Clear only a session this open freshly set up: a reused |
| 2256 | // one is held by another live provider process and must |
| 2257 | // survive this open's failure. |
| 2258 | Effect.andThen( |
| 2259 | prepared.issued |
| 2260 | ? clearMcpSession(input.threadId, mcpCredentialId) |
| 2261 | : Effect.void, |
| 2262 | ), |
| 2263 | Effect.ensuring( |
| 2264 | closeScopeWithin(sessionScope, { |
| 2265 | providerSessionId: input.providerSessionId, |
| 2266 | reason: "open_failed", |
| 2267 | }), |
| 2268 | ), |
| 2269 | ), |
| 2270 | ), |
| 2271 | Effect.mapError( |
| 2272 | (cause) => |
| 2273 | new ProviderSessionOpenError({ |
| 2274 | instanceId: input.modelSelection.instanceId, |
| 2275 | providerSessionId: input.providerSessionId, |
| 2276 | cause, |
| 2277 | }), |
| 2278 | ), |
| 2279 | withMetrics({ |
| 2280 | counter: providerSessionsTotal, |
| 2281 | attributes: { provider: adapter.driver, operation: "open" }, |
| 2282 | }), |
| 2283 | ); |
| 2284 | const eventSubscribers = yield* Ref.make< |
| 2285 | ReadonlyMap<number, Queue.Queue<ProviderSessionEventSignal, Cause.Done>> |
| 2286 | >(new Map()); |
| 2287 | const exposedRuntime = decorateRuntime(runtime, eventSubscribers); |
| 2288 | const now = yield* Clock.currentTimeMillis; |
| 2289 | const entry: LiveSessionEntry = { |
| 2290 | attachedThreadIds: new Set([input.threadId]), |
| 2291 | loadedProviderThreadKeyByThread: new Map(), |
| 2292 | mcpCredentialIdByThread: |
| 2293 | mcpCredentialId === undefined |
| 2294 | ? new Map() |
| 2295 | : new Map([[input.threadId, mcpCredentialId]]), |
| 2296 | supportsMultipleProviderThreads: |
| 2297 | runtime.providerSession.capabilities.sessions |
| 2298 | .supportsMultipleProviderThreadsPerSession, |
| 2299 | runtime, |
| 2300 | exposedRuntime, |
| 2301 | eventSubscribers, |
| 2302 | requestEventPermit: yield* Semaphore.make(1), |
| 2303 | scope: sessionScope, |
| 2304 | idleGeneration: 0, |
| 2305 | busyTurns: new Set(), |
| 2306 | lastActivityAtMs: now, |
| 2307 | idleFiber: null, |
| 2308 | pinnedSinceMs: null, |
| 2309 | idleThreadUnloads: new Map(), |
| 2310 | }; |
| 2311 | yield* Ref.update(sessions, (current) => { |
| 2312 | const updated = new Map(current); |
| 2313 | updated.set(key, entry); |
| 2314 | return updated; |
| 2315 | }); |
| 2316 | // The entry now guards the credential via its recorded id, so |
| 2317 | // the pre-open reservation can be dropped. |
| 2318 | yield* dropReservation; |
| 2319 | yield* withActivityError( |
| 2320 | input.providerSessionId, |
| 2321 | writeProviderSessionEvents({ |
| 2322 | runtime, |
| 2323 | threadIds: [input.threadId], |
| 2324 | type: "provider-session.attached", |
| 2325 | payload: runtime.providerSession, |
| 2326 | }), |
| 2327 | ).pipe( |
| 2328 | // Released on interrupt too: this entry has no event pump or |
| 2329 | // idle timer yet, so nothing else would ever release it. |
| 2330 | Effect.onError((cause) => |
| 2331 | releaseEntry( |
| 2332 | Cause.hasInterruptsOnly(cause) |
| 2333 | ? { |
| 2334 | providerSessionId: input.providerSessionId, |
| 2335 | reason: "manual_shutdown", |
| 2336 | detail: "The provider session start was interrupted.", |
| 2337 | } |
| 2338 | : { |
| 2339 | providerSessionId: input.providerSessionId, |
| 2340 | reason: "runtime_error", |
| 2341 | detail: "Failed to persist the provider-session attachment.", |
| 2342 | }, |
| 2343 | ).pipe(logReleaseFailure(input.providerSessionId)), |
| 2344 | ), |
| 2345 | ); |
| 2346 | yield* startEventPump(entry); |
| 2347 | yield* scheduleIdleRelease(input.providerSessionId); |
| 2348 | return exposedRuntime; |
| 2349 | }), |
| 2350 | ), |
| 2351 | isLive: (providerSessionId) => |
| 2352 | Ref.get(sessions).pipe( |
| 2353 | Effect.map((current) => current.has(sessionKey(providerSessionId))), |
| 2354 | ), |
| 2355 | get: (providerSessionId) => |
| 2356 | Effect.gen(function* () { |
| 2357 | const entry = (yield* Ref.get(sessions)).get(sessionKey(providerSessionId)); |
| 2358 | if (entry === undefined) { |
| 2359 | return Option.none<ProviderAdapter.ProviderAdapterV2SessionRuntime>(); |
| 2360 | } |
| 2361 | yield* touchActivity(providerSessionId); |
| 2362 | return Option.some(entry.exposedRuntime); |
| 2363 | }).pipe( |
| 2364 | Effect.mapError( |
| 2365 | (cause) => |
| 2366 | new ProviderSessionLookupError({ |
| 2367 | providerSessionId, |
| 2368 | cause, |
| 2369 | }), |
| 2370 | ), |
| 2371 | ), |
| 2372 | close: (providerSessionId) => |
| 2373 | releaseEntry({ providerSessionId, reason: "manual_shutdown" }).pipe( |
| 2374 | Effect.mapError( |
| 2375 | (cause) => |
| 2376 | new ProviderSessionCloseError({ |
| 2377 | providerSessionId, |
| 2378 | cause, |
| 2379 | }), |
| 2380 | ), |
| 2381 | ), |
| 2382 | closeInstance: (instanceId) => |
| 2383 | Effect.gen(function* () { |
| 2384 | const active = [...(yield* Ref.get(sessions)).values()].filter( |
| 2385 | (entry) => entry.runtime.instanceId === instanceId, |
| 2386 | ); |
| 2387 | const outcomes = yield* Effect.forEach( |
| 2388 | active, |
| 2389 | (entry) => |
| 2390 | releaseEntry({ |
| 2391 | providerSessionId: entry.runtime.providerSessionId, |
| 2392 | reason: "manual_shutdown", |
| 2393 | detail: `Provider instance ${instanceId} logged out.`, |
| 2394 | }).pipe(Effect.exit), |
| 2395 | { concurrency: "unbounded" }, |
| 2396 | ); |
| 2397 | const failure = outcomes.find(Exit.isFailure); |
| 2398 | if (failure !== undefined && Exit.isFailure(failure)) { |
| 2399 | return yield* Effect.failCause(failure.cause); |
| 2400 | } |
| 2401 | }).pipe( |
| 2402 | Effect.mapError( |
| 2403 | (cause) => |
| 2404 | new ProviderSessionCloseError({ |
| 2405 | providerSessionId: ProviderSessionId.make( |
| 2406 | `provider-session:provider-instance:${instanceId}`, |
| 2407 | ), |
| 2408 | cause, |
| 2409 | }), |
| 2410 | ), |
| 2411 | ), |
| 2412 | release: releaseEntry, |
| 2413 | detach: (input) => |
| 2414 | Effect.gen(function* () { |
| 2415 | const key = sessionKey(input.providerSessionId); |
| 2416 | const currentEntry = (yield* Ref.get(sessions)).get(key); |
| 2417 | let detachedProviderThreads: ReadonlyArray<OrchestrationV2ProviderThread> = []; |
| 2418 | if (currentEntry?.supportsMultipleProviderThreads === true) { |
| 2419 | const projection = yield* Effect.option( |
| 2420 | projectionStore.getThreadRecords(input.threadId, [ |
| 2421 | "providerThreads", |
| 2422 | "providerTurns", |
| 2423 | ]), |
| 2424 | ); |
| 2425 | if (Option.isSome(projection)) { |
| 2426 | const providerThreads = new Map( |
| 2427 | projection.value.providerThreads |
| 2428 | .filter((thread) => thread.providerSessionId === input.providerSessionId) |
| 2429 | .map((thread) => [thread.id, thread] as const), |
| 2430 | ); |
| 2431 | detachedProviderThreads = [...providerThreads.values()]; |
| 2432 | const activeTurns = projection.value.providerTurns.filter( |
| 2433 | (turn) => turn.status === "running" && providerThreads.has(turn.providerThreadId), |
| 2434 | ); |
| 2435 | yield* Effect.forEach( |
| 2436 | activeTurns, |
| 2437 | (turn) => |
| 2438 | currentEntry.exposedRuntime |
| 2439 | .interruptTurn({ |
| 2440 | providerThread: providerThreads.get(turn.providerThreadId)!, |
| 2441 | providerTurnId: turn.id, |
| 2442 | }) |
| 2443 | .pipe( |
| 2444 | Effect.catchCause((cause) => |
| 2445 | Effect.logWarning( |
| 2446 | "orchestration-v2.driver-session.detach-interrupt-failed", |
| 2447 | { |
| 2448 | providerSessionId: input.providerSessionId, |
| 2449 | threadId: input.threadId, |
| 2450 | providerTurnId: turn.id, |
| 2451 | cause, |
| 2452 | }, |
| 2453 | ), |
| 2454 | ), |
| 2455 | ), |
| 2456 | { concurrency: 1, discard: true }, |
| 2457 | ); |
| 2458 | } |
| 2459 | } |
| 2460 | const detachResult = yield* Ref.modify(sessions, (current) => { |
| 2461 | const entry = current.get(key); |
| 2462 | if (entry === undefined || !entry.attachedThreadIds.has(input.threadId)) { |
| 2463 | return [ |
| 2464 | Option.none<{ |
| 2465 | readonly entry: LiveSessionEntry; |
| 2466 | readonly idleUnloadFiber: Fiber.Fiber<void, never> | null; |
| 2467 | }>(), |
| 2468 | current, |
| 2469 | ] as const; |
| 2470 | } |
| 2471 | const attachedThreadIds = new Set(entry.attachedThreadIds); |
| 2472 | attachedThreadIds.delete(input.threadId); |
| 2473 | const loadedProviderThreadKeyByThread = new Map( |
| 2474 | entry.loadedProviderThreadKeyByThread, |
| 2475 | ); |
| 2476 | loadedProviderThreadKeyByThread.delete(input.threadId); |
| 2477 | // For a plain (workspace-change) detach, the credential id stays |
| 2478 | // recorded: the thread may re-attach and reuse it, and |
| 2479 | // releaseEntry revokes it when the provider process finally goes |
| 2480 | // away. A terminal detach (archive/delete) prunes the record so |
| 2481 | // nothing vetoes the revocation below. |
| 2482 | const mcpCredentialIdByThread = |
| 2483 | input.revokeMcpCredential === true |
| 2484 | ? (() => { |
| 2485 | const pruned = new Map(entry.mcpCredentialIdByThread); |
| 2486 | pruned.delete(input.threadId); |
| 2487 | return pruned; |
| 2488 | })() |
| 2489 | : entry.mcpCredentialIdByThread; |
| 2490 | // The detach unloads the thread itself below. |
| 2491 | const idleThreadUnloads = new Map(entry.idleThreadUnloads); |
| 2492 | idleThreadUnloads.delete(input.threadId); |
| 2493 | const updatedEntry = { |
| 2494 | ...entry, |
| 2495 | attachedThreadIds, |
| 2496 | loadedProviderThreadKeyByThread, |
| 2497 | mcpCredentialIdByThread, |
| 2498 | idleThreadUnloads, |
| 2499 | }; |
| 2500 | const updated = new Map(current); |
| 2501 | updated.set(key, updatedEntry); |
| 2502 | return [ |
| 2503 | Option.some({ |
| 2504 | entry: updatedEntry, |
| 2505 | idleUnloadFiber: entry.idleThreadUnloads.get(input.threadId)?.fiber ?? null, |
| 2506 | }), |
| 2507 | updated, |
| 2508 | ] as const; |
| 2509 | }); |
| 2510 | if (Option.isSome(detachResult)) { |
| 2511 | yield* cancelIdleFiber(detachResult.value.idleUnloadFiber); |
| 2512 | } |
| 2513 | const detached = Option.map(detachResult, (result) => result.entry); |
| 2514 | // Plain detaches deliberately do not revoke: a detached thread's |
| 2515 | // provider process may still be alive (shared multi-thread codex |
| 2516 | // session across a workspace handoff) and holds its MCP client's |
| 2517 | // credential for the thread it will re-attach with. Credentials |
| 2518 | // are revoked when the session entry is released (process gone) |
| 2519 | // or rotated on the next attach if they stopped resolving. |
| 2520 | // Terminal detaches (thread archived or deleted) revoke the |
| 2521 | // thread's credentials immediately, even on a retry where the |
| 2522 | // entry is already gone: there is no legitimate future re-attach, |
| 2523 | // and the token must not outlive the thread. |
| 2524 | if (input.revokeMcpCredential === true) { |
| 2525 | yield* clearMcpSession(input.threadId); |
| 2526 | } |
| 2527 | if (Option.isNone(detached)) { |
| 2528 | return; |
| 2529 | } |
| 2530 | if ( |
| 2531 | detached.value.attachedThreadIds.size === 0 && |
| 2532 | !detached.value.supportsMultipleProviderThreads |
| 2533 | ) { |
| 2534 | yield* releaseEntry({ |
| 2535 | providerSessionId: input.providerSessionId, |
| 2536 | reason: "manual_shutdown", |
| 2537 | ...(input.detail === undefined ? {} : { detail: input.detail }), |
| 2538 | detachedThreadId: input.threadId, |
| 2539 | }); |
| 2540 | return; |
| 2541 | } |
| 2542 | // The shared runtime stays up for other threads, so unload this |
| 2543 | // thread's native state rather than leaving it (and its MCP |
| 2544 | // servers) resident until the whole runtime is released. |
| 2545 | const unloadThread = detached.value.exposedRuntime.unloadThread; |
| 2546 | if (detached.value.supportsMultipleProviderThreads && unloadThread !== undefined) { |
| 2547 | // Serialized with re-attachment: a thread whose next turn |
| 2548 | // attaches first stays loaded, and one that attaches during the |
| 2549 | // unload waits for it, so its resume reloads the native thread. |
| 2550 | yield* threadAttachment.withLock( |
| 2551 | threadAttachmentKey(input), |
| 2552 | Effect.gen(function* () { |
| 2553 | const entry = (yield* Ref.get(sessions)).get(key); |
| 2554 | if ( |
| 2555 | entry?.runtime !== detached.value.runtime || |
| 2556 | entry.attachedThreadIds.has(input.threadId) |
| 2557 | ) { |
| 2558 | return; |
| 2559 | } |
| 2560 | yield* Effect.forEach( |
| 2561 | detachedProviderThreads.filter((thread) => thread.nativeThreadRef !== null), |
| 2562 | (providerThread) => |
| 2563 | unloadThread({ providerThread }).pipe( |
| 2564 | // Bounded so a wedged provider cannot hold up the |
| 2565 | // thread's next attach. |
| 2566 | Effect.timeout(UNLOAD_THREAD_TIMEOUT_MS), |
| 2567 | Effect.catchCause((cause) => |
| 2568 | Effect.logWarning( |
| 2569 | "orchestration-v2.driver-session.detach-unload-failed", |
| 2570 | { |
| 2571 | providerSessionId: input.providerSessionId, |
| 2572 | threadId: input.threadId, |
| 2573 | providerThreadId: providerThread.id, |
| 2574 | cause, |
| 2575 | }, |
| 2576 | ), |
| 2577 | ), |
| 2578 | ), |
| 2579 | { concurrency: 1, discard: true }, |
| 2580 | ); |
| 2581 | }), |
| 2582 | ); |
| 2583 | } |
| 2584 | yield* scheduleIdleRelease(input.providerSessionId); |
| 2585 | }).pipe( |
| 2586 | Effect.catchCause((cause) => |
| 2587 | Effect.fail( |
| 2588 | new ProviderSessionReleaseError({ |
| 2589 | providerSessionId: input.providerSessionId, |
| 2590 | reason: "manual_shutdown", |
| 2591 | cause, |
| 2592 | }), |
| 2593 | ), |
| 2594 | ), |
| 2595 | ), |
| 2596 | } satisfies ProviderSessionManagerV2Shape); |
| 2597 | }), |
| 2598 | ); |
| 2599 | |
| 2600 | export const layer = layerWithOptions(); |