Skip to content

Investigate a failed run

Start with the terminal result, then retain it only when useful.

kazeflow run pipeline.py --yes --verbose --store .kazeflow/runs.sqlite3

A failed run exits 1 but still prints a structured terminal result. Text identifies failed tasks, portable exception type/message, skipped tasks, and dependency blockers. --verbose adds structurally limited attempt-level detail.

Inspect the exact saved record later:

kazeflow runs list
kazeflow runs show RUN_ID

Compare a later rerun without assuming dedicated raw-output or partition-key fields were stored:

kazeflow runs compare FAILED_RUN_ID RETRY_RUN_ID

For programmatic in-process diagnosis, inspect each TaskResult and AttemptResult on the returned RunResult; the Python value retains details intentionally omitted from portable storage.

Portable failure messages and tracebacks remain application-controlled and can repeat a partition key or other sensitive value. A stored record is not automatically redacted; review the portable-record trust boundary.