Skip to content

Conversation

wkliao
Copy link
Collaborator

@wkliao wkliao commented Aug 21, 2025

OpenMPI 5.0.5 and priors contain a bug that can corrupt the
Darshan log files. The bug is related to file locking protocols
implemented when data sieving is enabled. Setting hint cb_nodes
to 1 can avoid the bug, but may make writing log files slow.
For more information, see PR #1070,

@wkliao wkliao added the OpenMPI label Aug 21, 2025
@wkliao wkliao force-pushed the check_opmi_506 branch 3 times, most recently from ca53aa9 to 891f45d Compare August 21, 2025 19:25
@wkliao
Copy link
Collaborator Author

wkliao commented Sep 15, 2025

In ad80b29 706b60b, the default I/O hint has been changed to just cb_nodes=1
when OpenMPI is detected.

This is because OpenMPI's implementation may serialize the I/O.
See open-mpi/ompi#13376
carns
carns previously approved these changes Sep 15, 2025
Copy link
Contributor

@carns carns left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

looks good to me. I see that some CI actions failed, but I don't know if it is related to these changes or not.

@wkliao
Copy link
Collaborator Author

wkliao commented Sep 15, 2025

I am encountered some errors related to configure option --log-path-by-env.
What is a proper form of this environment variable?

For example, if I configure with --log-path-by-env=DARSHAN_LOG_PATH, then
when I run my programs, should I set DARSHAN_LOG_PATH to a path
including date appended? For example,

export DARSHAN_LOG_PATH=/homes/wkliao/Darshan/LOG
or
export DARSHAN_LOG_PATH=/homes/wkliao/Darshan/LOG/2025/9/15

@carns
Copy link
Contributor

carns commented Sep 15, 2025

The --log-path-by-env= option specifies a flat directory (it will not traverse a year/month/day hierarchy from there)

Make sure it is defined in darshan-runtime/darshan-runtime-config.h
In this case, fall back to the value set in --log-path
@wkliao
Copy link
Collaborator Author

wkliao commented Sep 15, 2025

I think this PR is ready to be merged.

@carns carns merged commit e424328 into darshan-hpc:main Sep 16, 2025
7 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
Projects
None yet
Development

Successfully merging this pull request may close these issues.

2 participants