Uh oh!
There was an error while loading. Please reload this page.
Fully clean up buffers/memory resources in tests - #2243
Conversation
Add explicit sync points and notes around `mr.close()` making sure that there is no MR shutdown with pending async buffer freeing. These can currently cause occasional dead-locks when we have many MRs in parallel. Signed-off-by: Sebastian Berg <sebastianb@nvidia.com>
mdboom
commented
Jun 23, 2026
Somewhat adjacent issue: #1719 |
mdboom
left a comment
There was a problem hiding this comment.
I see no reason not to merge this.
As an aside, though... How did you find all of these and is there a way to enforce that this kind of thing doesn't creep back in in the future?
seberg
commented
Jun 23, 2026
That'll be the full changes at #2194 to run with pytest-run-parallel, I was just splitting things out since it seemed a bit easier to move forward. To be fair, cleanup eventually happens so with single threaded (no pytest-run-parallel) I am not sure this can fail (at least in practice). /ok to test b478199 |
seberg
commented
Jun 23, 2026
/ok to test b478199 |
This comment has been minimized.
This comment has been minimized.
Uh oh!
There was an error while loading. Please reload this page.
This comment has been minimized.
This comment has been minimized.
1 similar comment
|
Maybe the final (small) split-out for the free-threading test PR.
Add explicit sync points and notes around
mr.close()making sure that there is no MR shutdown with pending async buffer freeing. These can currently cause occasional dead-locks when we have many MRs in parallel. I believe this is a small upstream issue (625181).