William R Tobin
William R Tobin
There is a failure in 2 mpm tests. I anticipate it being relatively straightforward to track down tomorrow. Once that is done I will promote this from draft and place...
I think I had some work on a branch related to this, I'll take a look first thing tomorrow.
If the workaround doesn't resolve your issue, please also try this branch: https://github.com/GEOS-DEV/GEOS/pull/3177 I altered our file and dataspace accesses to use MPIO collectives in all cases instead of baton...
Let me know if/when you're able to try that fix, since I haven't been able to recreate the failure internally yet there isn't much additional testing I can do.
There was also a missing close operation when the dataset was created, which should be handled when the file is subsequently closed, but possibly there are cases where that isn't...
I've updated the branch to fix the new bug, apologies. If you're able to try again please let me know if it resolves the issue.
@paveltomin @rrsettgast We were able to reproduce locally. The commit I just pushed fixed it on our end. Essentially it was possible for different ranks to be accessing the file...
This may have been improved after recent hdf5 output changes / improvements