[
https://issues.apache.org/jira/browse/AIRFLOW-246?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15336365#comment-15336365
]
Kengo Seki commented on AIRFLOW-246:
------------------------------------
bq. Looks good to me. Query only takes ~4 seconds to run on our db (~750k rows).
I'm glad to hear that. Thanks for confirming ;)
bq. maybe it should also be looked at whether or not the whole thing should be
rearchitected... both of the queries (original, and the one you wrote) are
pretty ugly.
Well, I'm not sure whether it should be, since I'm a newbie to Airflow. I've
just started evaluation and contribution to Airflow a month ago. I hope Airflow
experts give us some advice, workarounds, etc.
I agree that these queries are not so readable, but I also think it's somewhat
inevitable as far as using ORM such as SQLAlchemy. The original python code
which generates the query in question is not so difficult to read to me.
bq. Let me know if you get a code fix for this and I'd be happy to test it.
Sure. I hope I can submit the first PR this weekend.
> 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
>
> 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)
