1

We have a Public API that includes a nightly metadata updater run via Cron in a pod deployed with Github Actions and Kubernetes. Overnight the updater pulls fresh metadata from our private API, and then commits those changes and pushes them to itself to refresh both the Github repo and trigger a lint/test/build/deployment.

We're using actions/checkout which fetches a single commit by default.

After a few weeks of updating our ~43,000 record ~700mb repo we started seeing failures at the git push step: error: pack-objects died of signal 9

Enumerating objects: 11754, done.
Counting objects: 100% (10881/10881), done.
error: pack-objects died of signal 9
error: pack-objects died of signal 9
error: remote unpack failed: eof before pack header was fully read

Increasing pod resources didn't help.

sighmon
  • 531
  • 4
  • 16
  • Tonight we're going to try doing a `git fetch --unshallow` to see if that allows the `git push` command to compare changes with less memory use. See PR: https://github.com/ACMILabs/acmi-api/pull/63/files And also this answer: https://stackoverflow.com/a/36189893 – sighmon Mar 25 '22 at 01:02

1 Answers1

0

After using git fetch --unshallow the pod was able to complete a git commit and git push to re-deploy itself.

sighmon
  • 531
  • 4
  • 16