feat sigterm-safe deployment (#619)

veckatimest commented 5 months ago

✅ DoD

[x] Resolves #619
[ ] API docs updated na
[ ] Release notes updated na
[ ] Deployment notes updated na
[ ] Unit or integration tests added na
[ ] DB migrations tested na

(use na when API docs (Release notes, etc) do not need to be updated)

📝 Summary

This change main part is the deployment.yaml

WHY: when deployment has only args, and no "command", then the command is "bash -c". So inside the pod we have something like "bash -c yarn start:core"

When kubernetes wants to gracefully stop the pod, it sends SIGTERM signal to the main process. But signal does not come into the node.js, it just terminates the bash command. So if we call "node path/file.js" then the main process is the node.js, and it can handle the SIGTERM.

I have tested it. After this change, logs got new messages regarding "Received signal...". And I have also tested that now after sigterm I receive started request. (And that the pod can finish it's busyness).

🛑 Problems

Write any discovered & unresolved problems (link to created Jira issues)

💡 More ideas

Write any more ideas you have

antonyharfield commented 5 months ago

Super!

antonyharfield commented 5 months ago

You could have changed it one time in the Dockerfile and then removed the command/args from the Kubernetes files. https://github.com/rfcx/rfcx-api/blob/6f71d67a15a0056b4b42428f47a36aacbcff0d94/build/Dockerfile#L33

rfcx / rfcx-api