Skip to content

fix: switch api-server probes from exec to httpGet#679

Draft
Connor Braa (cwlbraa) wants to merge 2 commits into
mainfrom
cwlbraa/httpget-probes
Draft

fix: switch api-server probes from exec to httpGet#679
Connor Braa (cwlbraa) wants to merge 2 commits into
mainfrom
cwlbraa/httpget-probes

Conversation

@cwlbraa
Copy link
Copy Markdown

Summary

  • Replace exec probes (python /api/healthcheck.py) with native httpGet probes on the api-server container in the langgraph-cloud standalone chart
  • Exec probes spawn a full Python interpreter every probe interval, which can exceed the 1s timeout on resource-constrained containers and cause crashloops
  • httpGet probes are performed by kubelet directly with near-zero overhead — matching what the lgp-operator deployment template already does
  • Startup probe uses check_db=1 (verify DB connectivity once, with generous failureThreshold=30), readiness/liveness use check_db=0

Test plan

  • Deploy to a staging cluster and verify probes pass
  • Confirm startup probe waits for DB before marking ready
  • Verify no crashloops on resource-constrained pods

Release Notes: None

🤖 Generated with Claude Code

Connor Braa (cwlbraa) and others added 2 commits April 15, 2026 14:45
The exec probes spawn a full Python process (`python /api/healthcheck.py`)
every probe interval, which can easily exceed the 1s timeout on
resource-constrained containers — causing crashloops. Switch to native
httpGet probes that kubelet performs directly, matching the lgp-operator
deployment template. Also differentiate probe behavior:
- startupProbe: check_db=1 with generous failureThreshold (30)
- readiness/liveness: check_db=0 (just confirm server is responsive)

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant