Troubleshooting Dagster

DON'T PANIC!

Pipeline failures

Check #dagster-alerts

Check #dagster-alerts on Slack for failed job run notifications. Most failures are transient framework errors and can be retried. Failures that recur or follow a code change require investigation and a fix via pull request.

Diagnose the failure

Click the link to the failed run.
Determine whether it is a framework error or a programming error.

Framework error (most common) — infrastructure or transient issue unrelated to the code:

Re-execute the run From Failure using the button in the Dagster Cloud UI.

Programming error — a bug in the asset code or dbt model:

Python assets: find the STEP_FAILURE event in the run logs and read the traceback.
dbt assets: go to Raw compute logs and search for the dbt error message.

Failed or missing partitions

Check the Asset Health page for assets with failed materializations.

Unpartitioned asset: click the asset, then click Materialize.
Partitioned asset: go to the Partitions tab, filter for Failed partitions, and materialize only those.

"Unsynced" views

View assets may show as Unsynced in the Dagster UI after an upstream table materializes. This is expected — see Automation Conditions for the explanation.

See also: Architecture · Automation Conditions · Dagster Guide