What problem does it solve? When a DAG run or task instance fails in Managed Service for Apache Airflow (formerly Cloud Composer), finding the root cause requires digging through Cloud Logging, task metadata, and remote DAG source code. This Skill provides a deterministic, evidence-based troubleshooting workflow that pinpoints the exact failure cause instead of offering generic advice. ## Core Features & Use Cases - Log-driven diagnosis: Fetches task and scheduler logs with gcloud logging read, matching error patterns like Python tracebacks, API error codes, and Airflow signals such as AirflowTaskTimeout. - Task and DAG metadata inspection: Retrieves task states, try numbers, and execution details via gcloud composer environments run commands to identify the critical failing task. - Remote code correlation: Downloads DAG source code from the environment's GCS bucket and compares configuration values (table IDs, URIs, disk sizes) against log errors. - Known platform issue detection: Matches failures against six documented failure modes including DAG_RUN_TIMEOUT, DETECTED_ZOMBIE, and WORKER_OUT_OF_POD_STORAGE. - Orchestration Pipeline support: Resolves generated DAGs back to their pipeline YAML definitions and logical action names. - Use Case: A BigQuery task in your nightly DAG fails with a 404 error. The Skill fetches the worker logs, downloads the DAG source, and shows that the dataset ID in the code does not match the actual BigQuery dataset, then suggests the corrected line. ## Quick Start Ask the assistant to troubleshoot why a specific DAG run failed in your Managed Airflow environment, providing the environment name, location, DAG ID, and run ID.
