thinkharderdev opened a new issue, #663:
URL: https://github.com/apache/arrow-ballista/issues/663

   **Is your feature request related to a problem or challenge? Please describe 
what you are trying to do.**
   A clear and concise description of what the problem is. Ex. I'm always 
frustrated when [...] 
   (This section helps Arrow developers understand the context and *why* for 
this feature, in addition to  the *what*)
   
   Currently when an executor is lost we try and rollback the job and retry 
tasks (and possibly stages) to continue. In some cases this is not the desired 
behavior and it would be better to just let the job fail. For instance, in an 
interactive query system where a late result is just as bad as a failure, it 
would be better to fail fast than consume more cluster resources to re-compute 
results. 
   
   **Describe the solution you'd like**
   A clear and concise description of what you want to happen.
   
   Scheduler should have a configurable policy:
   
   ```
   enum TaskResetPolicy {
      Fail,
      Retry
   }
   ```
   
   If set to `Fail` then when an executor is lost we fail any jobs with tasks 
running on that executor (and try and cancel any running tasks for those jobs 
on other executors) instead of trying to reset stages. 
   
   We can combine this with the ideas outlined in 
https://github.com/apache/arrow-ballista/issues/655 to try and allow executors 
to finish tasks after receiving `SIGTERM` before cancelling everything to 
minimize disruption. For interactive query systems where individual tasks are 
generally short-lived, this might be a more appropriate behavior.
   
   Initially this should be a scheduler global config, but in principle it 
could be set on a per-query basis in the future. 
   
   
   **Describe alternatives you've considered**
   A clear and concise description of any alternative solutions or features 
you've considered.
   
   Do nothing
   
   **Additional context**
   Add any other context or screenshots about the feature request here.
   


-- 
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