[Linux-cluster] is clvm a must all the time in a cluster ?

Fri Feb 29 01:39:48 UTC 2008

Hi,

You asked below: can I run the cluster with GFS without clvmd? The answer is yes. I believe in having the least number of components running, and found that clvmd had start up problems (and then refuses to stop) after doing an update of RHEL4 during December.  

It is clearly possible to use GFS directly on a SAN based LUN. The trick of course is how to deal with the /dev/sdb reference which will probably not be the same on all hosts. To fix that use udev rules that provide a symlink (using the serial number of the LUN, for example) by which the GFS file system can be referred to in /etc/fstab.

The conversion is simple and requires downtime. Create a new LUN sized as the old one, attach it to the cluster, create a GFS file system on the device /dev/sd-SN3030303 (for example), mount it, use whatever unix tool you would like to transfer the data from the existing GFS file system to the new one, modify the /etc/fstab files and you are done.

We have converted 2 clusters in the past few months because we have had real problems with clvmd misbehaving itself during startup. At this point it is a pleasure to let the cluster boot by itself and not to have to worry about GFS file systems not being mounted (ccsd, cman, fenced, iscsi, gfs).

Steffen

> -----Original Message-----
> From: linux-cluster-bounces at redhat.com 
> [mailto:linux-cluster-bounces at redhat.com] On Behalf Of Roger Peña
> Sent: Thursday, February 28, 2008 6:50 PM
> To: linux clustering
> Subject: Re: [Linux-cluster] is clvm a must all the time in a 
> cluster ?
> 
> 
> --- Jonathan Brassow <jbrassow at redhat.com> wrote:
> 
> >  From the description, I am skeptical that clvm is
> > your problem...   
> > (Also, are you using a proper fencing mechanism?)
> yes, fencing is ok
> everytime the cluster lose contact with a node, it is fenced 
> very quickly :-) but, in the cases I am describing, the nodes 
> remains members of the cluster so no need to fence anyone .
> just happen that cman get stuck because some cman service in 
> some node get stuck, and because cman is stuck we can't do 
> anything that interact with cman except with the node who 
> make the problem (we have to guess what node is :-) ) what I 
> had found is that usually, clvmd is the one that it is not 
> working as it should, but in other cases we just don't know 
> who is causing the problem.
> usually the reboot of the offending node provoque some 
> transicions in the cluster and cman get out of the stuck and 
> cluster come back to live :-)
> 
> the cluster is in production so I am very limited to do tests.
> right now the admin team agree into let clvmd work (started 
> at boot time) but if in the future we get more problems and 
> if we find that that problem somehow is related with clvmd 
> then we will try to revert the cluster lvm setting from our 
> SAN lv any downside with this approach except not be able to 
> run lvm commands ?
> 
> thanks
> roger
> 
> > 
> > We would probably need more information from the logs to determine 
> > what is going on.
> > 
> >   brassow
> > 
> > On Feb 25, 2008, at 3:06 PM, Roger Peña wrote:
> > 
> > > Hi
> > >
> > > I am running RHEL4 with csgfs from centos, all packages 
> updates in 
> > > both sets (RHEL and CS-GFS) but since the last updates my cluster 
> > > become more unstable, sometimes after a failed shutdown 
> of a service 
> > > (it failed because was unable to umount
> > the
> > > GFS) and the other two times just dont know
> > exactly
> > > why (I wasn't ther when it happen, just went to fixed).
> > > all of this in lest than two weeks
> > > and every time, the cman infrastructure just hand
> > or
> > > collapse.
> > > the first to times, I realice that one of the
> > nodes
> > > (and just one) was hanging the cman so the entire cluster 
> where in 
> > > stand-by, until I fence (reboot)
> > the
> > > node.
> > > from cman_tool services in that node, I guess clvm
> > was
> > > the guilty just because it were running (run
> > state):-)
> > > it shouldn't be in that state ;-) . it is just a guess, 
> ok? I am not 
> > > an expert in CS.
> > > also, a lot, and I mean a lot, of time my services can't start is 
> > > because the clustered LV are not Activated even when clvm 
> is running 
> > > and cman_tool services say it is running.
> > > just a clvmd restar fix that.
> > >
> > > so, my question is, can I run the cluster with GFS without clvmd 
> > > working?
> > > I don't have plans to work with lvm (extending for
> > > example) the sistem is working and should be in
> > the
> > > same state (from the filesystem point of view) for
> > a
> > > long time.
> > > so, can I "declusterice" the LV where GFS live?
> > >
> > > and if I can, what are the steps to do it?
> > >
> > > 1- vgchange -cn "VG Name"
> > > 2- cmand stop
> > > 3- lvmconf --disable-cluster
> > >
> > > I guess I don't need to unmount the LV to do that,
> > do
> > > I?
> > > nore the data in those FS should be lost because
> > of
> > > that operation (tha change), right?
> > >
> > >
> > > thanks in advance for the help
> > >
> > > cu
> > > roger
> > >
> > > __________________________________________
> > > RedHat Certified ( RHCE )
> > > Cisco Certified ( CCNA & CCDA )
> > >
> > >
> > >      Be smarter than spam. See how smart SpamGuard
> > is at giving junk
> > > email the boot with the All-new Yahoo! Mail. 
> > Click on Options in
> > > Mail and switch to New Mail today or register for
> > free at http://mail.yahoo.ca
> > >
> > > --
> > > Linux-cluster mailing list
> > > Linux-cluster at redhat.com
> > >
> >
> https://www.redhat.com/mailman/listinfo/linux-cluster
> > 
> > 
> > --
> > Linux-cluster mailing list
> > Linux-cluster at redhat.com
> >
> https://www.redhat.com/mailman/listinfo/linux-cluster
> > 
> 
> 
> __________________________________________
> RedHat Certified ( RHCE )
> Cisco Certified ( CCNA & CCDA )
> 
> 
>       Connect with friends from any web browser - no download 
> required. Try the new Yahoo! Canada Messenger for the Web 
> BETA at http://ca.messenger.yahoo.com/webmessengerpromo.php
> 
> --
> Linux-cluster mailing list
> Linux-cluster at redhat.com
> https://www.redhat.com/mailman/listinfo/linux-cluster
> 