Hi,

I encountered a bug with the OCF script with DRBD. 

See my post from 2008-11-14, maybe you suffer from it too ?

Le Friday 05 December 2008 15:17:12 Michael Schwartzkopff, vous avez écrit :
> Am Freitag, 5. Dezember 2008 15:00 schrieb Tobias Appel:
> > Hi,
> >
> > I remember the survey on the mailinglist not long ago but I don't really
> > remember the results.
> > Is DRBD 8 currently working with Heartbeat 2.14.x ?
> > I got the book from Dr. Schwartzkopff in which he explains an example
> > with DRBD 0.7 and I tried to use this example with DRBD 8 but it fails
> > for (yet) unknown reasons.
>
> As far as I I interprete the results on my posting there are two problems
> with DRBD8 in heartbeat:
> - Latency problem in cluster filesystems (esp. OCFS2) that HB cannot deal
> with correctly. OCFS has a separate fencing mechanism that dos not go
> together well with HA RA.
> - HA RA allows only 1 (in words one) master of a DRBD resource. This is
> hard coded.
>
> From this I concluded that it makes no sense to use DRBD8 with a cluster
> filesystem in heartbeat cluster. But DRBD8 has the advantage of the dopd
> service. So my latest setup is:
> DRBD8, DRBD RA and normal ext3 filesystem. I have two DRBDs in my setup and
> use one on each node for load sharing reasons.
>
> > I got DRBD 8 to work with Heartbeat V1 just easily. The only difference
> > is that I had the DRBD service started on both nodes automatically at
> > startup and Heartbeat just mounted the filesystem and made the active
> > node primary. I used drbddisk and the Filesystem script on this
> > environment.
> >
> > Now for Heartbeat V2 I created a Master/Slave Resource for the drbd
> > service then a primitive for the Filesystem within my resource group and
> > an order and colocation constraint, just like in the book. But right now
> > it fails to start and I can't my resource group anymore.
>
> Fix it manually. drbdadmin and cat /proc/drbd is your friend. It should
> really work. DO NOT start DRBD from init, but only from heartbeat.
> Bringing DRBD with the RA to life is a little bit tricky but works
> afterwards like a charm.
>
> Also use the "cleanup button" or crm_resource -C -r<resource_drbd> -U
> <node> often! Beware with open files on a drbd device, that you opened
> manually. This prohibits unmounting and gives the cluster a headache.
>
> > So if there are known problems with the OCF Agent and DRBD 8 please let
> > me know. I don't know yet how to 'downgrade' to DRBD 0.7 because I used
> > yum to install it on RHEL5 but I guess I will have to manually compile
> > it then (unless there is a way to tell yum to install version 0.7
> > somehow).
>
> My advise: Use DRBD8 since this is packaged in the most distros now, but do
> not use a cluster FS now. Any other experiences?
>
> Michael.


_______________________________________________
Linux-HA mailing list
[email protected]
http://lists.linux-ha.org/mailman/listinfo/linux-ha
See also: http://linux-ha.org/ReportingProblems

Reply via email to