feat: graceful shutdown to prevent lost LLM API calls #173

Closed

pook wants to merge 2 commits from feat/graceful-shutdown into main

pook commented

2026-04-08 23:42:40 -04:00

Owner

Summary

SIGTERM/SIGINT triggers graceful drain instead of immediate kill
Active request counter tracks in-flight requests via middleware
30-second timeout prevents hanging forever, exits 1 if exceeded
/health and /health/ready return 503 during shutdown
New requests rejected with 503 once shutdown begins

Files changed

packages/api/src/middleware/graceful-shutdown.ts - new module
packages/api/src/index.ts - switch to Bun.serve(), wire signal handlers
packages/api/src/routes/health.ts - 503 during shutdown
packages/api/tests/unit/graceful-shutdown.test.ts - 8 unit tests

Closes #55

## Summary - SIGTERM/SIGINT triggers graceful drain instead of immediate kill - Active request counter tracks in-flight requests via middleware - 30-second timeout prevents hanging forever, exits 1 if exceeded - /health and /health/ready return 503 during shutdown - New requests rejected with 503 once shutdown begins ## Files changed - packages/api/src/middleware/graceful-shutdown.ts - new module - packages/api/src/index.ts - switch to Bun.serve(), wire signal handlers - packages/api/src/routes/health.ts - 503 during shutdown - packages/api/tests/unit/graceful-shutdown.test.ts - 8 unit tests Closes #55

pook added 1 commit

2026-04-08 23:42:40 -04:00

feat: add graceful shutdown to prevent lost API calls during deploy 1814f0be85

SIGTERM/SIGINT now triggers a drain sequence: stop accepting new connections,
wait up to 30s for in-flight requests (e.g. LLM generation) to complete,
then exit. Health endpoint returns 503 during shutdown for load balancer awareness.

Closes #55

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

agent-bot added 1 commit

2026-04-08 23:42:50 -04:00

feat: issue #171 issue-55-exists-but-is-stale-with-no-pr (agent task agent-ta) 97205e5430

pook commented

2026-04-10 15:08:10 -04:00

Author

Owner

Closed 2026-04-10 during pipeline triage.

Merge conflicts with current main were blocking the CEO agent's backlog view. The compliancebot repo had ~60 open PRs and 141 open agent-task issues. CEO couldn't see progress and kept duplicating work due to a git-push race in agent-worker (now fixed — runId threaded through dispatch pipeline for unique branch names).

Reopen / resubmit against current main if the work is still relevant. Shim /shim/ceo route now injects open issues + PRs into the CEO prompt and refuses dispatch when backlog exceeds 20.

Closed 2026-04-10 during pipeline triage. Merge conflicts with current main were blocking the CEO agent's backlog view. The compliancebot repo had ~60 open PRs and 141 open agent-task issues. CEO couldn't see progress and kept duplicating work due to a git-push race in agent-worker (now fixed — runId threaded through dispatch pipeline for unique branch names). Reopen / resubmit against current main if the work is still relevant. Shim `/shim/ceo` route now injects open issues + PRs into the CEO prompt and refuses dispatch when backlog exceeds 20.