airflow-commits mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Matthias Huschle (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (AIRFLOW-246) dag_stats endpoint has a terrible query
Date Thu, 30 Jun 2016 12:21:10 GMT

    [ https://issues.apache.org/jira/browse/AIRFLOW-246?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15356998#comment-15356998
] 

Matthias Huschle commented on AIRFLOW-246:
------------------------------------------

Do I understand the query correctly, that it delivers different results than before, as it
double counts the TI belonging to the last DAG run if it is still running? Not that I'd consider
this a bad trade-off, but it might be worth noting it.

> dag_stats endpoint has a terrible query
> ---------------------------------------
>
>                 Key: AIRFLOW-246
>                 URL: https://issues.apache.org/jira/browse/AIRFLOW-246
>             Project: Apache Airflow
>          Issue Type: Bug
>          Components: webserver
>    Affects Versions: Airflow 1.7.1
>         Environment: MySQL Backend through sqlalchemy
>            Reporter: Neil Hanlon
>            Assignee: Kengo Seki
>
> Hitting this endpoint creates a series of queries on the database which take over 20
seconds to run, causing the page to not load for that entire time. Luckily the main page (which
includes this under "Recent Statuses") loads this synchronously, but still... waiting almost
half a minute (at times more) to see the statuses for dags is really not fun.
> We have less than a million rows in the task_instance table--so it's not even a problem
with that.
> Here's a query profile for the query:
> https://gist.github.com/NeilHanlon/613f12724e802bc51c23fca7d46d28bf
> We've done some optimizations on the database, but to no avail.
> The query:
> {code:sql}
> SELECT task_instance.dag_id AS task_instance_dag_id, task_instance.state AS task_instance_state,
count(task_instance.task_id) AS count_1 FROM task_instance LEFT OUTER JOIN (SELECT dag_run.dag_id
AS dag_id, dag_run.execution_date AS execution_date FROM dag_run WHERE dag_run.state = 'running')
AS running_dag_run ON running_dag_run.dag_id = task_instance.dag_id AND running_dag_run.execution_date
= task_instance.execution_date LEFT OUTER JOIN (SELECT dag_run.dag_id AS dag_id, max(dag_run.execution_date)
AS execution_date FROM dag_run GROUP BY dag_run.dag_id) AS last_dag_run ON last_dag_run.dag_id
= task_instance.dag_id AND last_dag_run.execution_date = task_instance.execution_date WHERE
task_instance.task_id IN ... AND (running_dag_run.dag_id IS NOT NULL OR last_dag_run.dag_id
IS NOT NULL) GROUP BY task_instance.dag_id, task_instance.state;
> {code}



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

Mime
View raw message