toil icon indicating copy to clipboard operation
toil copied to clipboard

Do not submit empty jobs

Open jsmedmar opened this issue 3 years ago • 3 comments
trafficstars

Thanks again for such a great tool.

Here I'm proposing a change, which I don't think its production ready, for a leader.py TODO item. When Toil finishes jobs, it walks up the DAG submitting empty jobs (jobs that actually do no work) requesting the original resources. Provisioning cloud machines for these jobs is not great.

Here I'm proposing to delete these jobs directly in leader.py. Full disclaimer, my knowledge of Toil's inner workings is limited, so this should be considered as a starting point.

All the best


I had created an issue about this before: https://github.com/DataBiosphere/toil/issues/3131

jsmedmar avatar Aug 03 '22 20:08 jsmedmar

Thank you for your PR, @jsmedmar ! I've copied it into our repo for testing at https://github.com/DataBiosphere/toil/tree/issues/3131-do-not-submit-empty-jobs , so check https://ucsc-ci.com/databiosphere/toil/-/pipelines/2001 for results from the CI

mr-c avatar Aug 05 '22 06:08 mr-c

It looks like this is a patch submitted from a long-running fork that branched off before the JobNode/JobGraph -> Job/JobDescription refactor, and so the patch doesn't work anymore. We ought to be able to port it over though.

adamnovak avatar Aug 10 '22 18:08 adamnovak

Adam you are right, I made these changes in my v3.18.0 branch. I think the overall idea holds as I saw the TODO is still there in master

jsmedmar avatar Aug 10 '22 18:08 jsmedmar

I think #4259 integrated this but Github didn't pick it up for some reason.

adamnovak avatar Dec 07 '22 15:12 adamnovak