Page MenuHomeClusterLabs Projects
Create Task
Maniphest T988

Running under callgrind doesn't work
Closed (Merged)Public
Actions

Description

If you enable callgrind in /etc/sysconfig/pacemaker, the cluster daemons fail to start in a way that's not especially helpful:

Jun 04 14:53:35.763 rhel9-scratch-1 pacemakerd          [1904] (pcmk_child_exit@pcmkd_subdaemons.c:296)         error: pacemaker-based[1905] exited with status 1 (Error occurred)

There's no indication of a core dump or any other failure, and the exec call that spawns the process doesn't return any error code. It's also very difficult to get stdout/stderr here to see what callgrind might be having a problem with. Note that running under just valgrind still works.

Revisions and Commits

rP Pacemaker
rP11a068ec22ce Low: sysconfig: Remove --leak-check=full from default VALGRIND_OPTS.

Event Timeline

clumens triaged this task as Low priority.Jun 4 2025, 2:55 PM
clumens created this task.
clumens created this object with edit policy "Pacemaker (Project)".
clumens added a comment.Fri, Aug 15, 5:54 PM

If you capture stdout and stderr in start_child just before calling execlp, you'll see the following:

valgrind: Unknown option: --leak-check=full
valgrind: Use --help for more information or consult the user manual.

In other words, the callgrind tool doesn't support this memcheck-specific option. These values come from VALGRIND_OPTS in /etc/sysconfig/pacemaker. So, our defaults do not work for callgrind. I think probably the best way to fix this is just to document it since callgrind is a less well known and less frequently used tool.

clumens closed this task as Merged by committing rP11a068ec22ce: Low: sysconfig: Remove --leak-check=full from default VALGRIND_OPTS..Mon, Aug 18, 7:32 PM
clumens claimed this task.
clumens added a commit: rP11a068ec22ce: Low: sysconfig: Remove --leak-check=full from default VALGRIND_OPTS..