Hacktoberfest 2026：维护者为十月标记出来的 issue，仍然开放、适合新手。 浏览 Hacktoberfest issue

Service: a live webhook batch that fails in pipeline_sync is dropped after Stripe got a 200

未关闭
  • 哪里不对？
#353 0 条评论 0 个 reaction 已指派 0 人 在 GitHub 查看

还没有人认领这个 Issue。

评估

难度
4/5
预计耗时
1-2 天
新手友好度
48/100
Issue 类型
缺陷
描述清晰度
基本清楚
活跃度
活跃
技术栈
postgresql, typescript
领域
backend, distributed-systems

调研方向

Start with the live batching loop in apps/service/src/temporal/workflows/pipeline-lifecycle.ts and the pipelineSync activity in apps/service/src/temporal/activities/pipeline-sync.ts; the issue also points to apps/service/src/tests/workflow.test.ts for the Temporal workflow test setup. Reproduce the failed eof and errored-stream cases with the described stubbed activity, then verify the batch is retried while successful batches still run once. Decide how retries are bounded or delayed, since the issue identifies repeated permanent failures as a risk.

由索引模型根据 Issue 内容生成。

描述

What happens

POST /webhooks/:pipeline_id verifies the signature, signals the pipeline workflow and answers 200 (app.ts#L973-L1004). The workflow's live loop takes up to 10 events off inputQueue with splice and awaits pipelineSync, but does not look at the eof it returns (pipeline-lifecycle.ts#L83-L101).

A processing failure comes back as a returned eof, not as a thrown error: the engine turns a throw into eof { status: 'failed' } (engine.ts#L695-L712), the Postgres destination reports a flush error as stream_status: error (index.ts#L445-L466), and the activity throws only when there is no eof at all (pipeline-sync.ts#L26). So the activity's retry policy never applies, the batch is not put back, and Stripe does not resend it, because it only retries deliveries that did not get a 2xx (webhooks docs).

An invoice.paid or customer.subscription.deleted that arrives while Postgres restarts or hits max_connections never reaches the table. The weekly reconcile does not repair it: it lists only objects created since the previous backfill (pipeline-lifecycle.ts#L121-L128, src-list-api.ts#L262-L266), so an existing invoice stays open in Postgres.

Reproduction

The workflow code from dev as of 2026-10-20, types stripped and nothing else changed, under @temporalio/testing's local server as in apps/service/src/__tests__/workflow.test.ts. Only the activities are stubbed: pipelineSync returns a failed eof for the first live batch and succeeds afterwards. Then one source_input signal with evt_1, which is what the webhook endpoint sends before answering 200:
https://github.com/terminalstate/callback-audit/tree/main/repro/stripe-sync-engine

dev with the change below
successful batch (control) evt_1 processed once once
batch eof status: 'failed' evt_1 passed to pipelineSync once, then gone processed again
eof status: 'started' with a stream errored (flush error) once, then gone processed again

Possible fix

Keep a failed batch instead of dropping it. The change the test runs with puts the events back at the head of the queue and waits before the next attempt:

const { eof } = await pipelineSync(pipelineId, { input: events, run_id: workflowInfo().runId })
operationCount++

const batchFailed =
  eof.status === 'failed' ||
  Object.values(eof.run_progress.streams).some((stream) => stream.status === 'errored')
if (batchFailed) {
  inputQueue.unshift(...events)
  await sleep(LIVE_BATCH_RETRY_DELAY)
}

It needs a growing delay or a limit, or an event that always fails is retried forever. The other option is to throw a retryable ApplicationFailure from the activity for a failed live batch, so the existing retry policy handles it.

主要语言
TypeScript
星标
1.1k
派生
118
30 天内没有已合并 PR

环境准备

从这里开始

  1. 先读完整个 Issue，再读项目的贡献指南。
  2. 在 Issue 下留言说明你要接手 —— 这能避免两个人做同样的事。
  3. Fork 仓库，在一个分支上完成修改。
  4. 提交 Pull Request，并在描述里引用这个 Issue 编号。

stripe/sync-engine 的其他 Issue

查看 stripe/sync-engine 的全部 Issue

相似的 Issue

更多 TypeScript Issue

把新 issue 发到你的邮箱

精选适合新手参与的 GitHub issue 摘要。