[
https://issues.apache.org/jira/browse/AIRFLOW-1559?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16275603#comment-16275603
]
ASF subversion and git services commented on AIRFLOW-1559:
----------------------------------------------------------
Commit 6bf1a6edaf13d3e255c47488f2747a2b8ebeff6c in incubator-airflow's branch
refs/heads/master from [~StephanErb]
[ https://git-wip-us.apache.org/repos/asf?p=incubator-airflow.git;h=6bf1a6e ]
[AIRFLOW-1559] Dispose SQLAlchemy engines on exit
When a forked process or the entire interpreter terminates, we have
to close all pooled database connections. The database can run out
of connections otherwise. At a minimum, it will print errors in its
log file.
By using an atexit handler we ensure that connections are closed
for each interpreter and Gunicorn worker termination. Only usages
of multiprocessing.Process require special handling as those
terminate via os._exit() which does not run finalizers.
This commit is based on a contribution by @dhuang
https://github.com/apache/incubator-airflow/pull/2767
> MySQL warnings about aborted connections, missing engine disposal
> -----------------------------------------------------------------
>
> Key: AIRFLOW-1559
> URL: https://issues.apache.org/jira/browse/AIRFLOW-1559
> Project: Apache Airflow
> Issue Type: Bug
> Components: db
> Reporter: Daniel Huang
> Assignee: Daniel Huang
> Priority: Minor
> Fix For: 1.9.1
>
>
> We're seeing a flood of warnings about aborted connections in our MySQL logs.
> {code}
> Aborted connection 56720 to db: 'airflow' user: 'foo' host: 'x.x.x.x' (Got an
> error reading communication packets)
> {code}
> It appears this is because we're not performing [engine
> disposal|http://docs.sqlalchemy.org/en/latest/core/connections.html#engine-disposal].
> The most common source of this warning is from the scheduler, when it kicks
> off new processes to process the DAG files. Calling dispose in
> https://github.com/apache/incubator-airflow/blob/master/airflow/jobs.py#L403
> greatly reduced these messages. However, the worker is still causing some of
> these, I assume from when we spin up processes to run tasks. We do call
> dispose in
> https://github.com/apache/incubator-airflow/blob/master/airflow/models.py#L1394-L1396,
> but I think it's a bit early. Not sure if there's a place we can put this
> cleanup to ensure it's done everywhere.
> Quick script to reproduce this warning message:
> {code}
> from airflow import settings
> from airflow.models import Connection
> session = settings.Session()
> session.query(Connection).count()
> session.close()
> # not calling settings.engine.dispose()
> {code}
> Reproduced with Airflow 1.8.1, MySQL 5.7, and SQLAlchemy 1.1.13.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
