How control DAG concurrency in airflow

Question

I use airflow v1.7.1.3

I have two DAG, dag_a and dag_b. I set up 10 dag_a tasks at one time, which theoretically should be execution one by one. In reality, the 10 dag_a tasks are executed in parallel. The concurrency parameter doesn't work. Can anyone tell me why?

Here's the pseudocode：

in dag_a.py

dag = DAG('dag_a',
            start_date=datetime.now(),
            default_args=default_args,
            schedule_interval=None,
            concurrency=1,
            max_active_runs=1)

in dag_b.py

from fabric.api import local

dag = DAG('dag_b',
            start_date=datetime.now(),
            default_args=default_args,
            schedule_interval='0 22 */1 * *',
            concurrency=1,
            max_active_runs=1)


def trigger_dag_a(**context):

    dag_list = []
    for rec in rang(1,10):
        time.sleep(2)
        cmd = "airflow trigger_dag dag_a"

        log.info("cmd:%s"%cmd)
        msg = local(cmd)    #"local" is function in fabric
        log.info(msg)


trigger_dag_a_proc = PythonOperator(python_callable=trigger_dag_a,
                          provide_context=True,
                          task_id='trigger_dag_a_proc',
                          dag=dag)

Answer 1

You can limit your task instances by specifying a pool.

1. Create a pool in the UI:

2.Then setup your dags to use this pool:

        default_args = {
                    'email_on_failure': False,
                    'email_on_retry': False,
                    'start_date': datetime(2017, 12, 16),
                    'pool': 'my_pool'
                }

        dag = DAG(
            dag_id='foo',
            schedule_interval='@daily',
            default_args=default_args,
        )