Red Hat Bugzilla – Bug 147828
cman initscript in_cluster function needs rework after latest cman changes
Last modified: 2009-04-24 10:32:19 EDT
Description of problem:
The latest changes in cman regarding node naming have made it
necessary to change how the in_cluster() function works. Currently it
looks for the node's shortname in /proc/cluster/nodes. I was testing
private network stuff so had link-10-pvt in the cluster.conf. They
joined just fine, but I was sitting around waiting 5 minutes for it to
fail because link-10-pvt != link-10. I can't think of a clean way to
implement this check. Hope someone can.
Version-Release number of selected component (if applicable):
Steps to Reproduce:
Makes stopping a beeeyatch too.
Well, I definately can't use uname or shortname now since Dave's
recent changes break cman's behavior, and I'm not about to implelement
any ip resolving hacks to determine hostname. Thankfully, Patrick
added the wait feauture to the `cman_tool join` command, so we're not
totally hosed. Another option would be to force cman_tool join to use
a predetermined string (such as "hostname" or "uname -n") instead of
allowing cman_tool to try to guess it's value. That way the startup
script will know what to look for in /proc/cluster/nodes.
I saw something in one of Patrick's commits about `cman_tool leave`
now having a wait option too. I'll try to pull that into the init
script refresh as well.
patrial fix has been checked in. leaving this still assigned because
I think I need to use the wait option, but am not getting it to work
like I expect at the moment.
Closing because the current initscript has been in the packages for ages and no-one has complained ;-)