[jira] [Comment Edited] (AIRFLOW-1614) Improve performance of DAG parsing when there are many subdags

["Gregory Benison (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/AIRFLOW-1614?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16167003#comment-16167003 ] 

Gregory Benison edited comment on AIRFLOW-1614 at 9/14/17 9:44 PM:
-------------------------------------------------------------------

Some profiling suggests this is largely due to the cost of constructing DAG objects; in particular the call to "inspect.stack" in the constructor is rather expensive.  It's also a bit unnecessary - we only need the caller's frame, but inspect.stack gives you the whole stack.  I have a [pull request|https://github.com/apache/incubator-airflow/pull/2610] with one possible fix that cuts the parsing time of the above file by more than half on my platform.


was (Author: gbenison):
Some profiling suggests this is largely due to the cost of constructing DAG objects; in particular the call to "inspect.stack" in the constructor is rather expensive.  It's also a bit unnecessary - we only need the caller's frame, but inspect.stack gives you the whole stack.  I have a pull request with one possible fix (coming soon) that cuts the parsing time of the above file by more than half on my platform.

> Improve performance of DAG parsing when there are many subdags
> --------------------------------------------------------------
>
>                 Key: AIRFLOW-1614
>                 URL: https://issues.apache.org/jira/browse/AIRFLOW-1614
>             Project: Apache Airflow
>          Issue Type: Improvement
>            Reporter: Gregory Benison
>
> DAGs can be very slow to parse when they contain many (100s or 1000s) of subdags.  This can be illustrated using the following DAG definition file:
> {code}from datetime import datetime, timedelta
> from airflow.models import DAG
> from airflow.operators.dummy_operator import DummyOperator
> from airflow.operators.subdag_operator import SubDagOperator
> dag = DAG(
>     'subdaggy-2',
>     schedule_interval=None,
>     start_date=datetime(2017,1,1)
> )
> def make_sub_dag(parent_dag, N):
>     dag = DAG(
>         '%s.task_%d' % (parent_dag.dag_id, N),
>         schedule_interval=parent_dag.schedule_interval,
>         start_date=parent_dag.start_date
>         )
>     DummyOperator(task_id='task1', dag=dag) >> DummyOperator(task_id='task2', dag=dag)
>     return dag
> downstream_task = DummyOperator(task_id='downstream', dag=dag)
> for N in range(20):
>     SubDagOperator(
>         dag=dag,
>         task_id='task_%d' % N,
>         subdag=make_sub_dag(dag, N)
>         ) >> downstream_task
> {code}
> When there are more than 50 or so subdags this file becomes slow enough to parse that it fails to load in the web UI on a modest platform such as a laptop.
> It would be nice to support such DAGs, since there are useful workflows involving 100s or 1000s of subdags.



--
This message was sent by Atlassian JIRA
(v6.4.14#64029)