fix(qqbot): clean up resources after reconnect failure

[k176060444-lgtm]

Summary

Clean up stale QQBot WebSocket and aiohttp session resources when a reconnect attempt fails.

Context

The closed-WebSocket CPU busy-loop guard has already been merged upstream. This PR keeps the remaining focused resource-cleanup improvement:

• capture stale WebSocket and session references;
• clear adapter fields before awaiting cleanup;
• close the WebSocket and session independently;
• keep reconnect failure handling best-effort;
• ensure cleanup errors do not interrupt later reconnect attempts.

Tests

• python3 -m pytest tests/gateway/test_qqbot.py -q — 166 passed
• added regression coverage for:
  • stale WebSocket and session cleanup;
  • missing session handling;
  • session creation followed by WebSocket connect failure;
  • cleanup exceptions;
  • WebSocket close failure without skipping session cleanup.

Related

This extracts the remaining resource-cleanup improvement from #39430 after the primary busy-loop guard was merged upstream.

[liuhao1024]

Code Review — Clean ✅

Reviewed the diff (2 files, ~160 lines). Findings:

1. Resource cleanup pattern is correct — capture refs, null fields early, then close independently in try/except blocks. This prevents: (a) double-close from concurrent reconnect attempts, (b) one close failure skipping the other cleanup.

2. Test coverage is thorough — 5 test cases covering: normal cleanup, missing session, session created by _open_ws before failure, cleanup exception handling, and ws.close() failure not blocking session.close(). The test_cleanup_ws_close_failure_does_not_skip_session_close test is particularly valuable — it validates the independent cleanup guarantee.

3. Logging at debug level — cleanup failures logged at debug not warning, which is appropriate since these are best-effort operations during an already-failed reconnect.

No issues found. LGTM.

[teknium1]

Thanks for the focused reconnect cleanup. The premise remains valid on current main: gateway/platforms/qqbot/adapter.py:468 assigns _session before ws_connect(), while the failure handler at gateway/platforms/qqbot/adapter.py:684-686 only logs and returns. The proposed clear-before-close, independent best-effort cleanup addresses that path.

Problems

• tests/gateway/test_qqbot.py:2253 calls _reconnect(0) with the production delay still active. gateway/platforms/qqbot/constants.py:41 defines the first retry delay as two seconds; the five added tests therefore add about ten seconds of real waiting.

Suggested changes

• Stub asyncio.sleep or patch the backoff value for this test class.

Automated hermes-sweeper review.