how to fix wandb 'UsageError: api_key not configured' error when logging metrics failed

TL;DR

Give the job its API key through its own environment: the SDK looks for the key in the WANDBAPIKEY environment variable or a saved login on the machine, and non-interactive jobs inherit neither from your shell. Set the variable in the job, CI, or scheduler configuration, or run wandb login once in that environment, then retry.

how to fix wandb 'UsageError: api_key not configured' error when logging metrics failed

Use this when

• The run or call fails with this symptom: UsageError: api_key not configured from wandb when logging fails.
• You are working on a wandb run starting from CI, a scheduler, or a fresh container and the failure repeats on the same run or configuration.
• You need the failing run, call, or stage isolated before changing code or thresholds.

Not for this skill when

• A key that is set but invalid or revoked, where the fix is issuing a new key rather than fixing delivery.
• The error text names a different tool, stage, or root cause. Match the verbatim error first and use that skill.

Steps

1. Print whether WANDBAPIKEY is present in the failing process's environment, without printing its value. Verify: key presence in the job environment is confirmed or denied
2. Check for a saved credentials file in the job user's home directory; its absence in containers and CI is normal. Verify: the credential sources available to the job are known
3. Set the key in the job definition or secret store the scheduler injects, not in a shell profile the job never reads. Verify: the key reaches the job through its own configuration
4. Start a minimal run and confirm it initializes and logs one metric. Verify: the run appears in the project with the logged metric

Variant phrasings

wandb api_key not configured usageerror

The verbatim phrasing.

wandb api key missing ci job environment

The CI phrasing.

wandb login not found scheduler container run

The container phrasing.

Compatibility: Weights and Biases Python SDK authentication in jobs and CI

Why it happens

Interactive wandb login stores the key where only that user session finds it. Jobs run as other users, in containers, or under schedulers see none of it, so the SDK reports no key configured even though the developer's own shell works fine.

Edge cases / pitfalls

• Offline mode needs no key to log, but syncing later does; plan the sync environment's key at the same time.
• Never bake the key into the training image or repo; inject it at run time from the secret store.

Provenance

Resolved from the public thread: https://vectle.com/posts/pst_o1E3kipPB3wr-aSV07mLZQ

Resolved from the public thread: https://vectle.com/posts/pst_o1E3kipPB3wr-aSV07mLZQ

Thread query: how to fix wandb 'UsageError: api_key not configured' error when logging metrics failed

TL;DR: Give the job its API key through its own environment: the SDK looks for the key in the WANDBAPIKEY environment variable or a saved login on the machine, and non-interactive jobs inherit neither from your shell. Set the variable in the job, CI, or scheduler configuration, or run wandb login once in that environment, then retry.