apache/airflow version: 2.1.3

other 2.X versions might also work, just hasn't tested

Offical Doc: Run Airflow in Docker

There are 3 kinds of executors provided for local testing. Aims to quickly test DAGs for the parsing error.

• SequentialExecutor: scheduler can only run tasks one by one, since sqlite can not accept multiple connections. All the airflow components are in one container.
  • data in sqlite is not preserved, shutdown airflow cleans all the data.
    • create empty ./airflow.db for the first time and use volume to preserve data if needed.
• LocalExecutor: workers executes tasks concurrently. Scheduler and workers are in same container.
  • data in posgresql is preserved with volume
• CeleryExecutor: scheduler sends task to redis queue, celery workers pull tasks from queue and execute, which is same as offcial docker-compose file
  • data in posgresql is preserved with volume

On Linux, the mounted volumes in container use the native Linux filesystem user/group permissions, so make sure the container and host computer have matching file permissions.

echo -e "AIRFLOW_UID=$(id -u)\nAIRFLOW_GID=0" > .env

Run for the first time, which helps to

• Check resources and print airflow version
• Create logs and plugins folders
• Run database migrations and create the first user account
  • Since table schema has been evolved, scheduler can not run successfully without migrations first

End log message as below

airflow-init_1       | Upgrades done
airflow-init_1       | [2021-09-06 02:35:47,963] {manager.py:788} WARNING - No user yet created, use flask fab command to do it.
airflow-init_1       | Admin user airflow created
airflow-init_1       | 2.1.3
airflow-examples_airflow-init_1 exited with code 0

SequentialExecutor does not need this step, since the commands (without checking resources and airflow version) in airflow-init is merged to one container which runs when up.

• LocalExecutor

docker-compose -f docker-compose-local.yaml up airflow-init

• CeleryExecutor

docker-compose -f docker-compose-celery.yaml up airflow-init

It takes some time for airflow to start, wait patiently for UI in localhost:8080

• SequentialExecutor

docker-compose -f docker-compose-sequentail.yaml up -d
docker-compose -f docker-compose-sequentail.yaml down

• LocalExecutor

docker-compose -f docker-compose-local.yaml up -d
docker-compose -f docker-compose-local.yaml down

• CeleryExecutor

docker-compose -f docker-compose-celery.yaml up -d
docker-compose -f docker-compose-celery.yaml down

default user and pwd are both airflow, which can be set by _AIRFLOW_WWW_USER_USERNAME and _AIRFLOW_WWW_USER_PASSWORD

All the dags in dags folder contains sh label which can be used to filter DAGs with UI. Check README.md or markdown in DAG for how to trigger and observe the result.

Since AIRFLOW__CORE__LOAD_EXAMPLES is true in docker-compose-*.yaml, airflow example dags will also show in UI which is useful for learning how to contruct DAGs.

dag folder path in container is /opt/airflow/dags, use docker cp command to copy dag files in local to the path in scheduler container after docker-compose up. Click refresh DAG button from UI and it is expected to change DAG without restarting all docker containers

Where is scheduler?

• docker-compose-sequential.yaml: airflow-service
  • container name: airflow-examples_airflow-service_1
• docker-compose-local.yaml: airflow-scheduler
  • container name: airflow-examples_airflow-scheduler_1
• docker-compose-celery.yaml: airflow-scheduler
  • container name: airflow-examples_airflow-scheduler_1

Note: If using airflow < 1.10.7 without dag serialization, dag files should be copied to both webserver and scheduler. Check dag-serialization for more details.