alkismavridis opened a new issue, #57487:
URL: https://github.com/apache/airflow/issues/57487

   ### Description
   
   Implement a new executor type that executes tasks locally, but unlike 
LocalExecutor, it starts multiple threads instead of spawning multiple 
processes 
   
   ### Use case/motivation
   
   Our usecase is pretty straight forward. We use airflow to orchestrate our 
tasks that will run in different servers. To achieve that we use exclusively 
SSHOpeators. The actual heavy lifting of the task happens on the target 
machines (the tech stack of which may vary and is completely independent of 
airflow)
   
   At the moment we use LocalExecutor with a parallelism of 32.
   But unfotunately, the spawned processes of LocalExecutor eat way too much 
RAM, which causes significant memory issues on our infrastructure.
   As mentioned 
[here](https://github.com/apache/airflow/issues/56641#issuecomment-3460906655), 
each one of the 32 workers will eventually occupy about 200MB. 
   
   I think that in the case of SSHOperatos (and other super-lightweight tasks 
that) the spawning of multiple processes can be an overkill. It would maybe 
make sense for such usecases to have a very simple executor that will run as a 
single process and will allocate a thread-pool of N threads instead of spawning 
N processes.
   
   If you think this is a good idea, will happily do an attempt to implement 
this myself and submit a PR. But I might need some support since I am not a 
python expert :)
   If you think that this is not a good idea, please suggest an alternative 
because for us having 5+ Gigabytes being allocated just to open couple of SSH 
Connections is a real pain.
   
   ### Related issues
   
   https://github.com/apache/airflow/issues/56641#issuecomment-3460906655
   
   ### Are you willing to submit a PR?
   
   - [x] Yes I am willing to submit a PR!
   
   ### Code of Conduct
   
   - [x] I agree to follow this project's [Code of 
Conduct](https://github.com/apache/airflow/blob/main/CODE_OF_CONDUCT.md)
   


-- 
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.

To unsubscribe, e-mail: [email protected]

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

Reply via email to