arkouda
arkouda copied to clipboard
Bears-R-Us
Segmentation violation in parquet unit test
We've seen signal 11 failures happen on UnitTestParquetCpp for those with arm64 architecture (verified by those on the team with apple silicon). But we're now seeing the same failure with a user running a linux desktop with x86_64 architecture. So I figured it would be a good idea to track this in an issue
@Ethan-DeBandi99 @bmcdonald3, for their awareness (and since both know more about hardware and parquet than me :) )
I managed to track down where exactly this is coming from. The final function call, errors += testReadStrings(strFilename, strDsetname);.
I need to see if I can determine why this is causing a seg fault. I validated that removing this test and running all the others successfully completes.
I took the test code and moved it into a arkouda module and it ran fine without seg faulting. Maybe it has to do with the Chapel testing architecture? Going to step away from this for a while and come back.
Just curious, did you run all the unit tests in that module or just the failing one? It is possible that the stack being manipulated via the other functions is kind of creating a perfect storm where the tests only fail with the stack looking a certain way that requires the particular order of the tests being run, but this is something I haven't looked into at all and so that is just a wild speculation.
I did both. I ran it with everything, each individual and things grouped together. The only thing that fails (including the singular test) is when testReadStrings is run. I could also try adjusting the CHPL_TASKS and see if that does anything to alter the results.
I think I am satisfied with the investigation you have done, so don't feel the need to do it on my behalf, just a thought that I had that sounds like you have already considered.