Cloud Composer is Google's managed Apache Airflow service. Google runs the Airflow control plane (webserver, scheduler, metadata DB) on GKE under the hood; you write DAGs and deploy them.
Problem first. You still need orchestration: "wait for GCS file → run Dataflow → run dbt in BigQuery → notify Slack." Airflow is the standard DAG tool; Composer removes most of the cluster babysitting.
Composer (managed Airflow)
DAG: daily_sales
1. GCSSensor
2. DataflowStartFlexTemplateOperator
3. BigQueryInsertJobOperator
4. Email / Pub/Sub notifyWhy DE teams use it
- Native GCP operators (BigQuery, Dataflow, GCS, Dataproc)
- IAM + VPC integration with the rest of GCP
- Less DIY Airflow ops than self-hosting
Trade-offs
- Cost of a managed environment (even when idle-ish)
- Version upgrades / environment sizing still need attention
- Same Airflow concepts: DAGs, tasks, retries, sensors, XCom
Interview tip: "Composer = managed Airflow on GCP." Map one DAG step to BigQuery/Dataflow so it sounds concrete.