[
https://issues.apache.org/jira/browse/AIRFLOW-2325?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Fang-Pen Lin updated AIRFLOW-2325:
----------------------------------
Description:
In many cases, it's ideal to use remote logging while running Airflow in
production, as the worker could be easily scale down or scale up. Or the worker
is running in containers, where the local storage is not meant to be there
forever. In that case, the S3 task logging handler could be used
[https://github.com/apache/incubator-airflow/blob/master/airflow/utils/log/s3_task_handler.py]
However, it comes with drawback. S3 logging handler only uploads the log when
the task completed or failed. For long running tasks, it's hard to know what's
going on with the process until it finishes.
To make more real-time logging, I built a logging handler based on AWS
CloudWatch. It uses a third party python package `watchtower`
[https://github.com/kislyuk/watchtower/tree/master/watchtower]
I created a PR here [https://github.com/apache/incubator-airflow/pull/3229],
basically I just copy-pasted the code I wrote for my own project, it works fine
with 1.9 release, but never tested with master branch. Also, there is a bug in
watchtower causing task runner to hang forever when it completes. I created an
issue in their repo
[https://github.com/kislyuk/watchtower/issues/57]
And a PR for addressing that issue
[https://github.com/kislyuk/watchtower/pull/58]
The PR is still far from ready to be reviewed, but I just want to get some
feedback before I spend more time on it. I would like to see if youguys want
this cloudwatch handler goes into the main repo, or do youguys prefer it to be
a standalone third-party module. If it's that case, I can close this ticket and
create a standalone repo on my own.
was:
In many cases, it's ideal to use remote logging while running Airflow in
production, as the worker could be easily scale down or scale up. Or the worker
is running in containers, where the local storage is not meant to be there
forever. In that case, the S3 task logging handler could be used
[https://github.com/apache/incubator-airflow/blob/master/airflow/utils/log/s3_task_handler.py]
However, it comes with drawback. S3 logging handler only uploads the log when
the task completed or failed. For long running tasks, it's hard to know what's
going on with the process until it finishes.
To make more real-time logging, I built a logging handler based on AWS
CloudWatch. It uses a third party python package `watchtower`
[https://github.com/kislyuk/watchtower/tree/master/watchtower]
> Task logging with AWS Cloud watch
> ---------------------------------
>
> Key: AIRFLOW-2325
> URL: https://issues.apache.org/jira/browse/AIRFLOW-2325
> Project: Apache Airflow
> Issue Type: New Feature
> Components: logging
> Reporter: Fang-Pen Lin
> Priority: Minor
>
> In many cases, it's ideal to use remote logging while running Airflow in
> production, as the worker could be easily scale down or scale up. Or the
> worker is running in containers, where the local storage is not meant to be
> there forever. In that case, the S3 task logging handler could be used
> [https://github.com/apache/incubator-airflow/blob/master/airflow/utils/log/s3_task_handler.py]
> However, it comes with drawback. S3 logging handler only uploads the log when
> the task completed or failed. For long running tasks, it's hard to know
> what's going on with the process until it finishes.
> To make more real-time logging, I built a logging handler based on AWS
> CloudWatch. It uses a third party python package `watchtower`
>
> [https://github.com/kislyuk/watchtower/tree/master/watchtower]
>
> I created a PR here [https://github.com/apache/incubator-airflow/pull/3229],
> basically I just copy-pasted the code I wrote for my own project, it works
> fine with 1.9 release, but never tested with master branch. Also, there is a
> bug in watchtower causing task runner to hang forever when it completes. I
> created an issue in their repo
> [https://github.com/kislyuk/watchtower/issues/57]
> And a PR for addressing that issue
> [https://github.com/kislyuk/watchtower/pull/58]
>
> The PR is still far from ready to be reviewed, but I just want to get some
> feedback before I spend more time on it. I would like to see if youguys want
> this cloudwatch handler goes into the main repo, or do youguys prefer it to
> be a standalone third-party module. If it's that case, I can close this
> ticket and create a standalone repo on my own.
>
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
