non-deterministic sparse sum failure #3115
Assignees
Labels
bug
Something isn't working
Comments
stress-tess
changed the title
sparse_sum_helper race condition
stress-tess
added a commit
to stress-tess/arkouda
that referenced
this issue
Apr 24, 2024
This PR (closes Bears-R-Us#3115) fixes the non-determinstic `sparse_sum_helper` failure. I was seeing runs where both the `if` and the `if else` would execute. Since the fetch conditions are mutually exclusive, we can use an `if` for both. I didn't realize that `on` statements were non-blocking, so there's pontential for them toexecute concurrently on different locales. Since my logic requires the previous locale to finish before the next begins, I've added an atomic helper to enforce this (thanks jeremiah for helping with that!)
stress-tess
added a commit
to stress-tess/arkouda
that referenced
this issue
Apr 24, 2024
This PR (closes Bears-R-Us#3115) fixes the non-determinstic `sparse_sum_helper` failure. I was seeing runs where both the `if` and the `if else` would execute. Since the fetch conditions are mutually exclusive, we can use an `if` for both. I didn't realize that `on` statements were non-blocking, so there's pontential for them toexecute concurrently on different locales. Since my logic requires the previous locale to finish before the next begins, I've added an atomic helper to enforce this (thanks jeremiah for helping with that!)
stress-tess
added a commit
to stress-tess/arkouda
that referenced
this issue
Apr 24, 2024
This PR (closes Bears-R-Us#3115) fixes the non-determinstic `sparse_sum_helper` failure. I was seeing runs where both the `if` and the `if else` would execute. Since the fetch conditions are mutually exclusive, we can use an `if` for both. I didn't realize that `on` statements were non-blocking, so there's pontential for them toexecute concurrently on different locales. Since my logic requires the previous locale to finish before the next begins, I've added an atomic helper to enforce this (thanks jeremiah for helping with that!)
github-merge-queue bot
pushed a commit
that referenced
this issue
Apr 24, 2024
This PR (closes #3115) fixes the non-determinstic `sparse_sum_helper` failure. I was seeing runs where both the `if` and the `if else` would execute. Since the fetch conditions are mutually exclusive, we can use an `if` for both. I didn't realize that `on` statements were non-blocking, so there's pontential for them toexecute concurrently on different locales. Since my logic requires the previous locale to finish before the next begins, I've added an atomic helper to enforce this (thanks jeremiah for helping with that!) Co-authored-by: Tess Hayes <[email protected]>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
The sparse sum helper failed the nightly with the following:
Even with runtime checks enabled I wasn't able to reproduce until I made an address sanitizer (
asan) build. And even then it seems to be non-deterministic and sometimes passes. This leads me to believe there is a race conditionThe text was updated successfully, but these errors were encountered: