ianrinehart
commented
3 years ago This may not be a problem per se; the workaround is to basically take the backup files, create a new collection, and run with that collection, but this requires writing a new executable script instead of just passing failed structures to a separate backup for triage.
I think that with a larger collection of molecules where the timeframe for jobs is longer, it would be useful to have it complete what it can, and then separately submit the triaged structures to complete a library. This way, you won't expect a job to take 4 days, and realize that it essentially stopped after part of that time because as few as one sub-job failed.
When a concurrent instance for a crest conformer search is done, if a job has failed (one possibility being timeout), then the returned object has no attribute 'name,' but is instead an instance of the error. This means that one failed job from a bundle of them stops someone from directly running a screen on the structures that did work.
I would like to change this so that there is a condition for no name attribute in the output, and it basically just ignores that structure and raises an error or writes the failed files to a directory for inspection/troubleshooting.