GitHub Actions container job killed with OOMKilled

By Kaveh Alemi·Latchkey

A process in a container job allocated more memory than the runner provided, so the kernel killed it. The job ends with exit code 137 (128 + SIGKILL). On Latchkey managed runners, an OOM from a transient memory spike is auto-retried, but a workload that consistently needs more memory should move to a larger runner.

What this error means

A memory-heavy step (bundler, compiler, test suite) dies abruptly with exit code 137 and an OOMKilled indication.

##[error]The operation was canceled.
Container exited with code 137 (OOMKilled)

Common causes

Workload exceeds available memory

A build or test run requests more RAM than the runner has, triggering the OOM killer.

Memory limit set too low on the container

A --memory option capped the container below what the workload needs.

How to fix it

Cap tool memory or use a larger runner

1. Bound peak memory in the tool itself (for example Node's --max-old-space-size or limiting test workers).
2. Raise or remove an overly low container --memory option.
3. If the workload truly needs more RAM, select a larger runner.

- name: Build with bounded heap
  run: node --max-old-space-size=4096 ./node_modules/.bin/build
  env:
    NODE_OPTIONS: --max-old-space-size=4096

How to prevent it

• Bound peak memory for memory-hungry build and test tools.
• Match the runner size to the workload's real memory footprint.

Frequently asked questions

Related guides

References

• GitHub Actions documentation
• Workflow syntax for GitHub Actions