Question

You maintain a Cloud Composer (Airflow) DAG that loads data from a partner's REST API each hour. The API occasionally returns HTTP 503 errors during peak traffic, causing sporadic task failures that page your on-call engineer. When the task is manually re-run a few minutes later it succeeds. You want to reduce noise and let the pipeline self-heal for these transient errors without masking genuine, persistent failures. What should you configure on the affected task?

Accepted Answer

Set retries with retry_exponential_backoff enabled and a bounded max_retry_delay, so retries space out and the task ultimately fails if the API stays unavailable.

Answer

Set a high number of retries with a short fixed retry_delay so the task keeps retrying immediately until it succeeds.

Answer

Wrap the task in a TriggerRule of all_done so downstream tasks continue regardless of whether the API call succeeds.

Answer

Disable SLA and alerting on the DAG so the on-call engineer no longer receives pages for this task.