|
| 1 | +// A stalled PENDING_EXECUTING snapshot must be requeued into a redis-primary run's durable residency. The |
| 2 | +// heartbeat payload carries no route, so on an undefined-dial engine the requeue's QUEUED snapshot would |
| 3 | +// otherwise take the Postgres shortcut and leave the MemoryDB head at PENDING_EXECUTING. Real infra. |
| 4 | +import { assertNonNullable, containerTest } from "@internal/testcontainers"; |
| 5 | +import { |
| 6 | + PostgresRunStore, |
| 7 | + RedisSnapshotStore, |
| 8 | + SnapshotResidencyResolver, |
| 9 | + TaskRunExecutionSnapshotStore, |
| 10 | + type SnapshotStoreDial, |
| 11 | +} from "@internal/run-store"; |
| 12 | +import { trace } from "@internal/tracing"; |
| 13 | +import { setTimeout } from "node:timers/promises"; |
| 14 | +import { expect } from "vitest"; |
| 15 | +import { RunEngine } from "../index.js"; |
| 16 | +import { createCompletedWaitpointResolver } from "../systems/completedWaitpointResolver.js"; |
| 17 | +import { setupAuthenticatedEnvironment, setupBackgroundWorker } from "./setup.js"; |
| 18 | + |
| 19 | +vi.setConfig({ testTimeout: 90_000 }); |
| 20 | + |
| 21 | +const ROUTE = "logical:1"; |
| 22 | +const PENDING_EXECUTING_TIMEOUT_MS = 500; |
| 23 | +const machines = { |
| 24 | + defaultMachine: "small-1x", |
| 25 | + machines: { |
| 26 | + "small-1x": { name: "small-1x" as const, cpu: 0.5, memory: 0.5, centsPerMs: 0.0001 }, |
| 27 | + }, |
| 28 | + baseCostInCents: 0.0001, |
| 29 | +}; |
| 30 | + |
| 31 | +function makeEngine( |
| 32 | + prisma: any, |
| 33 | + redisOptions: any, |
| 34 | + snapshotStore: RedisSnapshotStore, |
| 35 | + dial: () => SnapshotStoreDial | undefined, |
| 36 | + workerEnabled: boolean |
| 37 | +) { |
| 38 | + const delegate = new PostgresRunStore({ prisma, readOnlyPrisma: prisma }); |
| 39 | + const store = new TaskRunExecutionSnapshotStore(delegate, { |
| 40 | + store: snapshotStore, |
| 41 | + mode: "redis-only", |
| 42 | + resolveDial: dial, |
| 43 | + residencyResolver: new SnapshotResidencyResolver({ |
| 44 | + store: snapshotStore, |
| 45 | + taskRunExists: async (id: string) => (await prisma.taskRun.count({ where: { id } })) > 0, |
| 46 | + }), |
| 47 | + resolveCompletedWaitpoints: createCompletedWaitpointResolver(delegate), |
| 48 | + logicalRunStoreRoute: ROUTE, |
| 49 | + }); |
| 50 | + return new RunEngine({ |
| 51 | + prisma, |
| 52 | + store, |
| 53 | + worker: workerEnabled |
| 54 | + ? { redis: redisOptions, workers: 1, tasksPerWorker: 10, pollIntervalMs: 100 } |
| 55 | + : { redis: redisOptions, disabled: true }, |
| 56 | + queue: { |
| 57 | + redis: redisOptions, |
| 58 | + retryOptions: { maxTimeoutInMs: 50 }, |
| 59 | + masterQueueConsumersDisabled: true, |
| 60 | + processWorkerQueueDebounceMs: 50, |
| 61 | + }, |
| 62 | + runLock: { redis: redisOptions }, |
| 63 | + machines, |
| 64 | + heartbeatTimeoutsMs: { PENDING_EXECUTING: PENDING_EXECUTING_TIMEOUT_MS }, |
| 65 | + tracer: trace.getTracer("test", "0.0.0"), |
| 66 | + }); |
| 67 | +} |
| 68 | + |
| 69 | +describe("handleStalledSnapshot honors durable residency for a stalled PENDING_EXECUTING run", () => { |
| 70 | + containerTest( |
| 71 | + "a dial=undefined engine requeues a redis-primary run into its MemoryDB head", |
| 72 | + async ({ prisma, redisOptions }) => { |
| 73 | + const env = await setupAuthenticatedEnvironment(prisma, "PRODUCTION"); |
| 74 | + const snapshotStore = new RedisSnapshotStore({ redisOptions, completedTtlMs: 60_000 }); |
| 75 | + const producer = makeEngine(prisma, redisOptions, snapshotStore, () => "redis-only", false); |
| 76 | + const consumer = makeEngine(prisma, redisOptions, snapshotStore, () => undefined, true); |
| 77 | + |
| 78 | + try { |
| 79 | + await setupBackgroundWorker(producer, env, "test-task"); |
| 80 | + |
| 81 | + async function dequeueOnConsumer() { |
| 82 | + for (let i = 0; i < 25; i++) { |
| 83 | + await producer.runQueue.processMasterQueueForEnvironment(env.id, 5); |
| 84 | + const dequeued = await consumer.dequeueFromWorkerQueue({ |
| 85 | + consumerId: "stall_consumer", |
| 86 | + workerQueue: "main", |
| 87 | + }); |
| 88 | + if (dequeued.length > 0) return dequeued[0]; |
| 89 | + await setTimeout(300); |
| 90 | + } |
| 91 | + throw new Error("run never reached the worker queue"); |
| 92 | + } |
| 93 | + |
| 94 | + const run = await producer.trigger( |
| 95 | + { |
| 96 | + number: 1, |
| 97 | + friendlyId: "run_stallroute", |
| 98 | + environment: env, |
| 99 | + taskIdentifier: "test-task", |
| 100 | + payload: "{}", |
| 101 | + payloadType: "application/json", |
| 102 | + context: {}, |
| 103 | + traceContext: {}, |
| 104 | + traceId: "t-stall", |
| 105 | + spanId: "s-stall", |
| 106 | + workerQueue: "main", |
| 107 | + queue: "task/test-task", |
| 108 | + isTest: false, |
| 109 | + tags: [], |
| 110 | + delayUntil: new Date(Date.now() + 60_000), |
| 111 | + }, |
| 112 | + prisma |
| 113 | + ); |
| 114 | + const runId = run.id; |
| 115 | + |
| 116 | + await prisma.taskRun.update({ |
| 117 | + where: { id: runId }, |
| 118 | + data: { delayUntil: null, queueTimestamp: new Date() }, |
| 119 | + }); |
| 120 | + const runRow = await prisma.taskRun.findFirstOrThrow({ where: { id: runId } }); |
| 121 | + await producer.enqueueSystem.enqueueRun({ run: runRow, env, enableFastPath: true }); |
| 122 | + |
| 123 | + const first = await dequeueOnConsumer(); |
| 124 | + assertNonNullable(first); |
| 125 | + expect(first.snapshotRoute?.residency).toBe("redis-primary"); |
| 126 | + expect(first.snapshot.executionStatus).toBe("PENDING_EXECUTING"); |
| 127 | + |
| 128 | + let latest = await consumer.getRunExecutionData({ runId }); |
| 129 | + for (let i = 0; i < 40 && latest?.snapshot.executionStatus !== "QUEUED"; i++) { |
| 130 | + await setTimeout(250); |
| 131 | + latest = await consumer.getRunExecutionData({ runId }); |
| 132 | + } |
| 133 | + assertNonNullable(latest); |
| 134 | + expect(latest.snapshot.executionStatus).toBe("QUEUED"); |
| 135 | + const head = await snapshotStore.getLatest(runId); |
| 136 | + assertNonNullable(head); |
| 137 | + expect(head.id).toBe(latest.snapshot.id); |
| 138 | + expect(await prisma.taskRunExecutionSnapshot.count({ where: { runId } })).toBe(0); |
| 139 | + |
| 140 | + const redelivered = await dequeueOnConsumer(); |
| 141 | + assertNonNullable(redelivered); |
| 142 | + expect(redelivered.run.id).toBe(runId); |
| 143 | + expect(redelivered.snapshot.executionStatus).toBe("PENDING_EXECUTING"); |
| 144 | + } finally { |
| 145 | + await producer.quit(); |
| 146 | + await consumer.quit(); |
| 147 | + await snapshotStore.quit(); |
| 148 | + } |
| 149 | + } |
| 150 | + ); |
| 151 | +}); |
0 commit comments