Comments on: Heartbeat2 Xen cluster with drbd8 and OCFS2

By: Kommy

Kommy — Thu, 22 Oct 2009 19:18:02 +0000

I have a problem a with this command: sudo cibadmin -C crm_config -x /root/cluster/bootstrap.xml, if i run this the help page i see. I use openSUSE 11.1 x64 operation system. Can anybody help me, what is do wrong? I think is do wrong the heartbeat 2 installation / configuration. Can anybody write the installation /configuration step by step?

By: Angel Genchev

Angel Genchev — Thu, 25 Jun 2009 14:37:24 +0000

The SSH stonith won`t work if the other node`s network fails. I`ve tested it - puled the plug. The heartbeat will wait and the resources won`t take over. It`s better to disable it.

By: Fantastic Sid

Fantastic Sid — Mon, 22 Jun 2009 10:17:20 +0000

This is totally the wrong way to roll. With this setup, any partition across the DRBD interface will cause (at best) fencing (assuming that your fencing via ssh hack is using a different interface) leading to (at least some) downtime, and at worst two completely different copies of your OCFS, one of which will need to be thrown out (and those VM’s changes lost entirely.)

If you used one drbd per VM disk, and drbd’s block-drbd script for Xen, you would completely ameliorate this problem AND do away with any need for disk-related pseudofencing, because (except for a handful of milliseconds during live migration) the volume is only primary in one spot at once, so any partition will be resolved automatically by drbd on reconnection.

Also, I can only imagine that OCFS2 wouldn’t exactly speed things up…

By: organetic

organetic — Thu, 18 Jun 2009 20:23:56 +0000

Anyone knows how to apply this solution to KVM instead of XEN? I guess it’s just a matter of changing the vm01.xml script bu I don’t have the idea how to do it.

By: Angel Genchev

Angel Genchev — Tue, 28 Apr 2009 22:45:36 +0000

I like it the way it is. OCFS2 is necessary, because we want DomU images to reside in files (easy backup, migration to new hardware,live migration requires rw accesss on both nodes). If we don`t want files, we end up with multiple drbd primary/primary parttions for each DomU.
What I miss here is a more complicated example - w 2 DomUs and resource stickiness, auto failback/fail forward to spread the load.
Also drbd startup, ocfs2 startup & mount setup as Clone resources. It`s interesting also what happens if you type # halt on a node. (graceful shutdown requires to migrate domUs, stop heartbeat, umount ocfs2, stop drbd (detach disks) & continue w shutdown…

By: Christoph

Christoph — Wed, 25 Feb 2009 11:45:03 +0000

good howto.

but I’d rather transfer the SSH key using the ssh-copy-id command. The ‘scp’ way presented here overwrites other RSA IDs that are permitted to log in on the other host. The ssh-copy-id command _adds_ the RSA ID on the other host, it doesn’t overwrite the IDs already stored in the authorized_keys file.

By: Daniel

Daniel — Sat, 25 Oct 2008 09:50:18 +0000

Another way is to skip OCFS2 as I believe that is only complicating the setup and reduces performance. Using DRBD in your DomU config files, i.e. using drbd: instead of phy:, will also simplify the management of DRBD as the block device driver will take care of the DRBD states. It also does not require OCFS2 or GFS as only one node will have unconditional access. So using for example EXT3 works fine. Have a look at: http://www.asplund.nu/xencluster/xen-cluster-howto.html for an alternative solution. Cheers, Daniel

By: Jukka

Jukka — Thu, 21 Aug 2008 13:52:40 +0000

Thank you for the howto!
Though I did run to some problems while installing this system on hardy. For example with the dpkg-reconfigure o2cb I had to use ocfs2-tools instead of o2cb also the configuration file for the ocfs cluster did not work as such but needed some tuning. The parameters in the config need a tab in front:
node:
ip_port = 7777
ip_address = 192.168.0.128
number = 0
name = node1
cluster = ocfs2

By: Ali

Ali — Thu, 03 Jul 2008 13:24:05 +0000

pretty good howto although one thing i’m missing: you never mention how the xen guests are set up in this environment. What kind of disk devices do they have? the only thing that comes to mind would be the file-backend, but then i’ve heard some bad things about the performance of it. I’d appreciate it if you could clear this up.

By: travis

travis — Tue, 12 Feb 2008 14:43:40 +0000

great post! Im gonna give this a try with some 5750s I got laying around. Thanks.