Open jprotze opened 1 year ago
I had a quick look at the problem using a few examples. It turns out that the done and error file check happens more often than expected and needed, and even more often than described here. However, due to the complexities involved, we need a little more time to analyse and fix the problem.
jube continue seems to check done/error files in all work package directories while processing one work package (according to strace). This results in at least 2n^2 file accesses when only 2n should be necessary. For a larger experiment space and slow meta data server, this can result in significant execution time.