Do not set different cache directories per job#4460
Conversation
This should not be necessary as the disk caches are now safer in how they handle race conditions. Using a unified cache directory should improve performance.
…per-job-cache-dir
|
This appears to work fine. The latest failure is just the macro solve one. |
|
9fa112d mostly. Though I have touched
I think so. This is all very hard to reason about but I don't think there should be any issues provided that each process/rank can only add cache entries (i.e. I imagine one process running With our test suite there is a lot of contention for disk access between processes so I think we are stress testing this quite well. When I originally added this functionality we were reliably getting hangs so the fact that this hasn't been an issue for a couple of runs bodes well. This change also only affects CI so we can be a little more gung ho about things. |
Ok, this looks ok then. Is sounds like if this change does end up harming things then we'll see if very quickly at least. I'm happy for this to go in once the stokes fix is merged in and CI is passing. |
…per-job-cache-dir
* Do not set different cache directories per job This should not be necessary as the disk caches are now safer in how they handle race conditions. Using a unified cache directory should improve performance.
2 participants
This should not be necessary as the disk caches are now safer in how they handle race conditions. Using a unified cache directory should improve performance.