gcp-composer-troubleshooting

Start by gathering the DAG ID, run ID, and task ID, then use gcloud logging read with severity>=ERROR filters scoped to the cloud_composer_environment resource. Correlate the log stack traces with the actual DAG code downloaded from the environment's GCS bucket to pinpoint the failing line.

Use gcloud logging read with a filter like resource.type="cloud_composer_environment" combined with labels.dag_id and severity>=ERROR. Add --limit and a time range to narrow results, and format output with timestamp, severity, task_id, and textPayload fields.

The Composer environment runs code from its GCS bucket, which may not match your local files if syncs were missed. Download the remote DAG with gcloud storage cp to a temporary folder and perform the RCA on the actual running code before applying fixes.

No, it follows a read-only-first approach: it analyzes logs and remote code to prove the root cause and proposes a fix, but does not modify code unless you explicitly ask. This prevents untested changes from being applied to production pipelines.

Provide the DAG ID and ideally the run ID, task ID, or an approximate failure time. If you only have a vague error, the agent will ask for a DAG ID or time range to search logs effectively.