image.png

Retries tell Flyte how many times to rerun a step that fails before giving up on the whole pipeline — only the failed step reruns, not the ones that already finished

In the console, a failed run shows [1/1] currentAttempt done — that's 1 attempt of 1 allowed, meaning no retries are set; adding @task(retries=3) to a flaky step (one that talks to Blob or MLflow, say) gives it three more tries before Flyte marks the run FAILED