Main CI failed: E2E Tests — cli/acp-cron.test.ts > … > cron job fires and streams results via sessionUpdate after prompt returns

[qwen-code-dev-bot]

A main-branch E2E Tests run failed on main.

Failing tests

• cli/acp-cron.test.ts > acp cron integration > cron job fires and streams results via sessionUpdate after prompt returns

This issue is labeled for autofix so the existing agent can create a repair PR.
It is deduped by failing test, so every later commit that hits the same
failure is appended below instead of opening another issue.

Recurrences

• 1d1bc494736c · 2026-08-01T02:03:40Z · run 30677083363
• 63780e0d2fbb · 2026-07-31T14:19:52Z · run 30635549911

[qwen-code-ci-bot]

✅ Qwen Triage finished — view run. See the stage comments in this thread for the result.

✅ Qwen Triage 已完成 —— 查看运行。结果见本线程中的各阶段评论。

[qwen-code-ci-bot]

Triage

• Type: bug — recurring flaky E2E timing failure on main CI (recurrence of Main CI failed: E2E Tests — cli/acp-cron.test.ts > … > cron job fires and streams results via sessionUpdate after prompt returns #8076)
• Labels: type/bug, scope/testing, category/integration, status/need-retesting, status/ready-for-agent, autofix/approved
• Next: verify the already-merged fix test(integration): migrate acp-cron to fake-openai-server (#8076) #8082 resolves this on the next main CI run before opening any new repair PR

Key findings

• Same test as Main CI failed: E2E Tests — cli/acp-cron.test.ts > … > cron job fires and streams results via sessionUpdate after prompt returns #8076 (closed COMPLETED today at 14:14Z, ~13 min before this issue was filed): cli/acp-cron.test.ts > cron job fires and streams results via sessionUpdate after prompt returns.
• This failure is on a stale, pre-fix commit. The failing run 30635549911 tested commit 63780e0d (fix(cli): add Option+V for clipboard image paste on macOS #8120, clipboard-paste — unrelated to cron). That commit's copy of the test still uses the real model path with the tight 15_000 ms agent_message_chunk timeout and has no fake-openai-server.
• The fix for Main CI failed: E2E Tests — cli/acp-cron.test.ts > … > cron job fires and streams results via sessionUpdate after prompt returns #8076 — PR test(integration): migrate acp-cron to fake-openai-server (#8076) #8082, commit e8d72673a "test(integration): migrate acp-cron to fake-openai-server" — merged to main at 14:13:42Z, before this issue was filed, but the stale commit's CI run only completed at 14:19:52Z and tripped this auto-issue.
• Failure is isolated to E2E Test - macOS - shard 1/2; all six Linux shards and macOS shard 2/2 passed — consistent with a timing flake, not a deterministic regression.

---

Stage 2 — Source Analysis

• Reproduced: not executed (unattended CI run — static analysis only, per the no-PR-code-execution rule). Evidence is from the failing run's own CI checks via the API.
• CI evidence: run 30635549911 on 63780e0d — E2E Test - macOS - shard 1/2 = failure; E2E Test (Linux) sandbox:none/docker shards 1–3 and E2E Test - macOS - shard 2/2 = success. (The job log is dominated by an unrelated telemetry test's multi-MB dump, so the exact assertion line wasn't cleanly extractable; the prior Main CI failed: E2E Tests — cli/acp-cron.test.ts > … > cron job fires and streams results via sessionUpdate after prompt returns #8076 run shows the same signature: Timed out waiting for sessionUpdate: agent_message_chunk after cron fire (after 15000ms, …).)
• Affected area: integration-tests/cli/acp-cron.test.ts (the 15 s agent_message_chunk wait) and the cron response-streaming path in packages/cli/src/acp-integration/session/Session.ts.
• Root cause (carried over from Main CI failed: E2E Tests — cli/acp-cron.test.ts > … > cron job fires and streams results via sessionUpdate after prompt returns #8076, confirmed against the failing commit): the cron job fires and the user_message_chunk with _meta.source='cron' is received, but the model's agent_message_chunk doesn't arrive within 15 s. On the pre-fix test the model call hits the real API, where CI latency/cold-starts easily exceed 15 s. Confidence: confirmed for the mechanism on the stale commit.
• Fix status: already merged via test(integration): migrate acp-cron to fake-openai-server (#8076) #8082 (e8d72673a), which migrates the test to fake-openai-server so model responses are deterministic and local — removing the latency that blew the 15 s window. The current main test still shows 15_000, but with the fake server there is no real network/model latency, so the window is now ample.

Recommended next step

1. Treat this as a stale recurrence of Main CI failed: E2E Tests — cli/acp-cron.test.ts > … > cron job fires and streams results via sessionUpdate after prompt returns #8076, not a new regression. Do not open a duplicate repair PR against the current tree.
2. Confirm the next main CI run on a commit containing e8d72673a passes cli/acp-cron.test.ts. If it does, close this as completed (duplicate of Main CI failed: E2E Tests — cli/acp-cron.test.ts > … > cron job fires and streams results via sessionUpdate after prompt returns #8076).
3. Only if the test still fails on a post-e8d72673a commit, revisit raising the 15 s agent_message_chunk timeout toward the 75 s used for the user_message_chunk wait.

中文说明

• 类型: bug — main CI 上反复出现的 flaky E2E 超时失败（Main CI failed: E2E Tests — cli/acp-cron.test.ts > … > cron job fires and streams results via sessionUpdate after prompt returns #8076 的复发）
• 标签: type/bug、scope/testing、category/integration、status/need-retesting、status/ready-for-agent、autofix/approved
• 下一步: 先验证已合并的修复 test(integration): migrate acp-cron to fake-openai-server (#8076) #8082 能否在下一次 main CI 中解决此问题，再决定是否开新的修复 PR

关键发现

• 与 Main CI failed: E2E Tests — cli/acp-cron.test.ts > … > cron job fires and streams results via sessionUpdate after prompt returns #8076 是同一个测试（Main CI failed: E2E Tests — cli/acp-cron.test.ts > … > cron job fires and streams results via sessionUpdate after prompt returns #8076 今天 14:14Z 以 COMPLETED 关闭，约 13 分钟后本 issue 被创建）。
• 本次失败发生在修复前的旧 commit 上。 失败运行 30635549911 测的是 commit 63780e0d（fix(cli): add Option+V for clipboard image paste on macOS #8120，剪贴板粘贴，与 cron 无关）。该 commit 的测试仍走真实模型路径、agent_message_chunk 超时仍是紧凑的 15_000 ms，且没有 fake-openai-server。
• Main CI failed: E2E Tests — cli/acp-cron.test.ts > … > cron job fires and streams results via sessionUpdate after prompt returns #8076 的修复 PR test(integration): migrate acp-cron to fake-openai-server (#8076) #8082（commit e8d72673a，迁移到 fake-openai-server）已于 14:13:42Z 合并进 main，早于本 issue 创建；但旧 commit 的 CI 运行直到 14:19:52Z 才完成并触发了本自动 issue。
• 失败仅出现在 E2E Test - macOS - shard 1/2；6 个 Linux 分片与 macOS shard 2/2 全部通过 —— 符合时序 flake，而非确定性回归。

根因与修复状态

• 复现: 未执行（无人值守 CI，仅静态分析）。证据来自该失败运行自身的 CI 检查（经 API 获取）。
• 根因沿用 Main CI failed: E2E Tests — cli/acp-cron.test.ts > … > cron job fires and streams results via sessionUpdate after prompt returns #8076 并已对失败 commit 核实：cron 成功触发、收到带 _meta.source='cron' 的 user_message_chunk，但模型的 agent_message_chunk 未在 15 秒内到达。旧测试走真实 API，CI 延迟/冷启动很容易超过 15 秒。置信度：针对旧 commit 上的机制为已确认。
• 修复已合并（test(integration): migrate acp-cron to fake-openai-server (#8076) #8082 / e8d72673a）：将测试迁移到 fake-openai-server，模型响应确定且本地化，消除了撑爆 15 秒窗口的延迟。

建议下一步

1. 视为 Main CI failed: E2E Tests — cli/acp-cron.test.ts > … > cron job fires and streams results via sessionUpdate after prompt returns #8076 的过期复发，而非新回归；不要对当前代码开重复的修复 PR。
2. 确认下一次包含 e8d72673a 的 main CI 能通过 cli/acp-cron.test.ts；若通过则按 Main CI failed: E2E Tests — cli/acp-cron.test.ts > … > cron job fires and streams results via sessionUpdate after prompt returns #8076 的重复项关闭。
3. 仅当在包含 e8d72673a 的 commit 上仍失败时，再考虑把 15 秒 agent_message_chunk 超时提高到接近 user_message_chunk 所用的 75 秒。

--- Qwen Code

[qwen-code-dev-bot]

🤖 The scheduled autofix agent is picking this issue up. It will attempt to establish the current behavior, implement the requested change, run E2E verification, and open a pull request linked to this issue. If the attempt fails, this claim will be withdrawn so a human can take over.

Maintainers: comment or assign someone to stop future automated attempts, or add the autofix/skip label.