| 1 | import { CommandId, MessageId, type OrchestrationV2Command } from "@t3tools/contracts"; |
| 2 | import * as DateTime from "effect/DateTime"; |
| 3 | import * as Effect from "effect/Effect"; |
| 4 | import * as Layer from "effect/Layer"; |
| 5 | import * as Scheduler from "../scheduling/Scheduler.ts"; |
| 6 | import * as ServerSettings from "../serverSettings.ts"; |
| 7 | import * as ProjectionStore from "./ProjectionStore.ts"; |
| 8 | import * as ThreadManagement from "./ThreadManagementService.ts"; |
| 9 | |
| 10 | /** The persisted run and reset form the identity of one recovery opportunity. */ |
| 11 | export function limitRecoveryCommand( |
| 12 | thread: ProjectionStore.ProjectionLimitRecoveryCandidate, |
| 13 | autoResume: boolean, |
| 14 | nowMs: number, |
| 15 | snooze = false, |
| 16 | ): OrchestrationV2Command | null { |
| 17 | if ( |
| 18 | thread.status !== "failed" || |
| 19 | thread.lastErrorClass !== "usage_limit" || |
| 20 | !thread.latestRunId || |
| 21 | !thread.usageLimitResetAt || |
| 22 | thread.archivedAt !== null || |
| 23 | thread.settledOverride === "settled" || |
| 24 | thread.pendingRuntimeRequest !== null |
| 25 | ) |
| 26 | return null; |
| 27 | const resetMs = Date.parse(thread.usageLimitResetAt); |
| 28 | // An already-expired window reported with a fresh failure cannot start a retry loop. |
| 29 | if ( |
| 30 | !Number.isFinite(resetMs) || |
| 31 | resetMs <= DateTime.toEpochMillis(thread.latestRunCompletedAt ?? thread.updatedAt) |
| 32 | ) |
| 33 | return null; |
| 34 | const identity = `${thread.id}:${thread.latestRunId}:${resetMs}`; |
| 35 | const recovery = thread.limitRecovery; |
| 36 | if (recovery?.runId !== thread.latestRunId || recovery.resetAt !== thread.usageLimitResetAt) { |
| 37 | if (!autoResume && (!snooze || resetMs <= nowMs)) return null; |
| 38 | return { |
| 39 | type: "thread.metadata.update", |
| 40 | commandId: CommandId.make(`limit-arm:${identity}`), |
| 41 | threadId: thread.id, |
| 42 | limitRecovery: { |
| 43 | runId: thread.latestRunId, |
| 44 | resetAt: thread.usageLimitResetAt, |
| 45 | autoResume, |
| 46 | snooze: snooze && resetMs > nowMs, |
| 47 | }, |
| 48 | }; |
| 49 | } |
| 50 | if ( |
| 51 | !recovery.autoResume || |
| 52 | resetMs > nowMs || |
| 53 | (thread.snoozedUntil != null && DateTime.toEpochMillis(thread.snoozedUntil) > nowMs) |
| 54 | ) |
| 55 | return null; |
| 56 | const deliveryIdentity = `${identity}:${recovery.requestId ?? "legacy"}`; |
| 57 | return { |
| 58 | type: "message.dispatch", |
| 59 | commandId: CommandId.make(`limit-resume:${deliveryIdentity}`), |
| 60 | messageId: MessageId.make(`limit-resume:${deliveryIdentity}`), |
| 61 | threadId: thread.id, |
| 62 | usageLimitContinuationOfRunId: thread.latestRunId, |
| 63 | ...(recovery.requestId === undefined |
| 64 | ? {} |
| 65 | : { usageLimitRecoveryRequestId: recovery.requestId }), |
| 66 | text: "Continue where you left off.", |
| 67 | attachments: [], |
| 68 | dispatchMode: { type: "start_immediately" }, |
| 69 | createdBy: "user", |
| 70 | creationSource: "server", |
| 71 | }; |
| 72 | } |
| 73 | |
| 74 | const makeSweep = Effect.gen(function* () { |
| 75 | const projections = yield* ProjectionStore.ProjectionStoreV2; |
| 76 | const threads = yield* ThreadManagement.ThreadManagementService; |
| 77 | const settings = yield* ServerSettings.ServerSettingsService; |
| 78 | return Effect.fn("UsageLimitRecoveryWorker.sweep")(function* () { |
| 79 | const preferences = yield* settings.getSettings; |
| 80 | const now = yield* DateTime.now; |
| 81 | const candidates = yield* projections.getLimitRecoveryCandidates({ |
| 82 | now, |
| 83 | autoResume: preferences.autoResumeLimitedThreads, |
| 84 | snooze: preferences.snoozeLimitedThreads, |
| 85 | }); |
| 86 | const nowMs = DateTime.toEpochMillis(now); |
| 87 | for (const thread of candidates) { |
| 88 | const command = limitRecoveryCommand( |
| 89 | thread, |
| 90 | preferences.autoResumeLimitedThreads, |
| 91 | nowMs, |
| 92 | preferences.snoozeLimitedThreads, |
| 93 | ); |
| 94 | if (command === null) continue; |
| 95 | yield* threads.dispatch(command).pipe( |
| 96 | Effect.catchCause((cause) => |
| 97 | Effect.logWarning("orchestration-v2.limit-recovery.dispatch-failed", { |
| 98 | threadId: thread.id, |
| 99 | cause, |
| 100 | }), |
| 101 | ), |
| 102 | ); |
| 103 | } |
| 104 | }); |
| 105 | }); |
| 106 | |
| 107 | // The shared scheduler derives due work from persisted failures and recovery |
| 108 | // choices, so restarts need no timer restoration or connected client. |
| 109 | export const layer = Layer.effectDiscard( |
| 110 | Effect.gen(function* () { |
| 111 | const sweep = yield* makeSweep; |
| 112 | const scheduler = yield* Scheduler.Scheduler; |
| 113 | yield* scheduler.register("usage-limit-recovery", sweep()); |
| 114 | }), |
| 115 | ); |