Red Hat Bugzilla – Bug 177165
modclusterd: Allow kids to gracefully exit
Last modified: 2009-04-16 18:35:16 EDT
Description of problem:
(this isn't an rgmanager problem, actually - but there's no BZ entry for clumond
clumond forks children and waits (currently), 3000ms (3 seconds) for those
applications to complete. Under many circumstances, it may take >3sec for them
to complete, and it's perfectly reasonable to kill them if we're on a time
schedule. It is probably better, however, to kill them with SIGINT or SIGTERM
prior to sending them SIGKILL so they have a chance to clean up.
One particular instance where this has become a problem is when using clustat:
clustat opens a connection descriptor to ccsd, which is persistent until closed.
If clustat is taking to long to complete (which is often), a ccsd connection
descriptor is leaked. If ccsd runs out of connection descriptors, nothing will
be able to configure/reconfigure, and this message will appear in the logs:
"Error while processing connect: Resource temporarily unavailable"
Version-Release number of selected component (if applicable): CVS/Head
Steps to Reproduce:
Server shouldn't misbehave no matter what client does, so this is a ccsd's bug,
Clumond should be a good resident and issue SIGINT or SIGTERM before SIGKILL;
Since server side has been fixed, this issue is a non-issue any more.