[Linux-cluster] Questions about rgmanager
brem.belguebli at gmail.com
Wed Aug 19 20:50:26 UTC 2009
What is the behaviour of rgmanager supposed to be in case one adds a new LVM
and FS resources to an already running service?
Let's say I have a running database service on one node, and I need to add
some storage to this service.
After creating all the LVM things, the filesystems,, mount them on the right
I edit the cluster.conf, setup the lines concerning what I've added, and
upgrade file version.
Running then ccs_tool update /etc/cluster/cluster.conf will it restart my
modified service ?
2009/8/19 brem belguebli <brem.belguebli at gmail.com>
> Changing the cluster version reloads the rgmanager configuration, but
> it restarts my custom resources despite the fact that they were "started"
> (in clustat).
> Just changing the cluster.conf version, not doing anything to my custom
> resource makes this happen.
> In the log I get the message :
> <info> Stopping changed resources
> <info> Restarting changed resources
> <info> Starting changed resources
> which may be generic messages and then I see my resources being stopped and
> Sometimes, I also can notice that my service after being stopped, I have a
> monitor action that is triggered that fails (normal service is stopped) and
> that disables (status failed) completely the service.
> I'm running Rhel 5.3 (I saw the post about VM's on RHel 5.2 which seems to
> be similar).
> I've tried to run rgmanager in debug (clurgmgrd -d) mode but it seems to
> get stuck
> Any idea on what I may be missing ?
> 2009/8/19, brem belguebli <brem.belguebli at gmail.com>:
>> thx Lon
>> 2009/8/18 Lon Hohberger <lhh at redhat.com>
>>> On Tue, 2009-08-18 at 11:01 +0200, brem belguebli wrote:
>>> > Hello,
>>> > I have a few questions about rgmanager:
>>> > 1) How to make it take into account resource metadata changes without
>>> > restarting it. service rgmanager reload (or kill -HUP) doesn't.
>>> > In case one needs to change some timers in a resource metadata,
>>> > the only way to make it is to stop and start rgmanager which brings
>>> > down the resources.
>>> Need to change the configuration version in cluster.conf; rgmanager
>>> doesn't record for example file mtimes and run stat() to see if metadata
>>> changed. Doing this wouldn't work anyway, as some resources have
>>> external metadata.
>>> > 2) Is there a way to log what "rules" rgmanager succeded to load at
>>> > start time (or the ones it failed to), as having a "24 rules loaded"
>>> > message isn't very explicit.
>>> You can run:
>>> rg_test rules | grep ^Agent
>>> There's no way currently to have rgmanager log specifically which agent
>>> it loaded at run-time.
>>> > The last point concerns the potential replacement of rgmanager with
>>> > pacemaker (as stated at
>>> > http://sources.redhat.com/cluster/wiki/RGManager). Will this happen in
>>> > a near future ?
>>> Not immediately; we still have a bit to go:
>>> * need configuration file format conversion utility
>>> * need basic "work-alike" tools. Ex: clustat.
>>> * need to do a bit of work on rgmanager agents so they work
>>> on pacemaker - the standard by which rgmanager+pacemaker
>>> operate was inadequate and the two programs have diverged
>>> in slightly incompatible ways
>>> -- Lon
>>> Linux-cluster mailing list
>>> Linux-cluster at redhat.com
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the Linux-cluster