[ 
https://issues.apache.org/jira/browse/AIRFLOW-6361?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17205875#comment-17205875
 ] 

ASF GitHub Bot commented on AIRFLOW-6361:
-----------------------------------------

mik-laj edited a comment on pull request #6905:
URL: https://github.com/apache/airflow/pull/6905#issuecomment-702445420


   @ryw  It is not necessary to be included in Airflow 2.0. We can do it in any 
version. After that, I realized it would be a problem for my users. My users 
use Git/GCS sync to sync plugins, and this can be problematic after making this 
change. For this reason, if we would like to add this change, it should be 
activated by the feature gate.
   
   If you are interested you can take over this change. If you are interested 
you can take over this change as it is not currently in my task queue.


----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

For queries about this service, please contact Infrastructure at:
[email protected]


> Run LocalTaskJob directly in Celery task
> ----------------------------------------
>
>                 Key: AIRFLOW-6361
>                 URL: https://issues.apache.org/jira/browse/AIRFLOW-6361
>             Project: Apache Airflow
>          Issue Type: Improvement
>          Components: executors
>    Affects Versions: 1.10.6
>            Reporter: Kamil Bregula
>            Priority: Major
>              Labels: performance
>
> Hello,
> Celery runs the CLI first command, which contains LocalTaskJob. LocalTaskJob 
> is responsible for starting the next user-code process. This level of 
> isolation is redundant because LocalTaskJob doesn't execute unsafe code. The 
> first command is run by a new process creation, not by a fork, so this is an 
> expensive operation.
> According to preliminary measurements, this change results in an increase in 
> performance close to 30%.
> I will provide more information in PR.
> Best regards
> Kamil Bregula
> After:
> ```
> real  0m38.394s
> user  0m4.340s
> sys   0m1.600s
> real  0m38.355s
> user  0m4.700s
> sys   0m1.340s
> real  0m38.675s
> user  0m4.760s
> sys   0m1.530s
> real  0m38.488s
> user  0m4.770s
> sys   0m1.280s
> real  0m38.434s
> user  0m4.600s
> sys   0m1.390s
> real  0m38.378s
> user  0m4.500s
> sys   0m1.270s
> real  0m38.106s
> user  0m4.200s
> sys   0m1.100s
> real  0m38.082s
> user  0m4.170s
> sys   0m1.030s
> real  0m38.173s
> user  0m4.290s
> sys   0m1.340s
> real  0m38.161s
> user  0m4.460s
> sys   0m1.370s        
> ```
> Before:
> ```
> real  0m53.488s
> user  0m5.140s
> sys   0m1.700s
> real  1m8.288s
> user  0m6.430s
> sys   0m2.200s
> real  0m53.371s
> user  0m5.330s
> sys   0m1.630s
> real  0m58.939s
> user  0m6.470s
> sys   0m1.730s
> real  0m53.255s
> user  0m4.950s
> sys   0m1.640s
> real  0m58.802s
> user  0m5.970s
> sys   0m1.790s
> real  0m58.449s
> user  0m5.380s
> sys   0m1.580s
> real  0m53.308s
> user  0m5.120s
> sys   0m1.430s
> real  0m53.485s
> user  0m5.220s
> sys   0m1.290s
> real  0m53.387s
> user  0m5.020s
> sys   0m1.590s
> ```



--
This message was sent by Atlassian Jira
(v8.3.4#803005)

Reply via email to