dpdispatcher icon indicating copy to clipboard operation
dpdispatcher copied to clipboard

Improving Task Resubmission in DPDispatcher

Open Cloudac7 opened this issue 2 years ago • 2 comments

Occasionally, tasks in dpdispatcher might crash unexpectedly, causing the need for resubmission. However, this resubmission process can be inconvenient, for example if the original task was almost completed, and the resubmitted task would simply append to the outputs from zero. In order to avoid this in this issue an alternative solution is proposed - backing up the previous output files to another location and then starting a new run from scratch. This approach would improve efficiency and prevent the need for unnecessary resubmissions. Would it be OK to considering this as a potential user story for dpdispatcher?

Cloudac7 avatar Apr 10 '23 13:04 Cloudac7

This is what you need to handle in the command argument. Some program (such as LAMMPS) supports restarting the task from the existing checkpoint, so moving old files away will break the restarting process.

njzjz avatar Apr 10 '23 20:04 njzjz

This is what you need to handle in the command argument. Some program (such as LAMMPS) supports restarting the task from the existing checkpoint, so moving old files away will break the restarting process.

I knew, for what I am now dealing with is just about restarting in CP2K. However, it might be in risk of just using original CP2K overwriting, for it might cause repeated part in output files for tasks not end in the normal way. Using command might be rather complicated for situation that a lot of output files to be handled. I thought just add a choice instead of only one way might be an alternative solution.

Cloudac7 avatar Apr 11 '23 05:04 Cloudac7