I am presently using ocfs2 managed by a pacemaker cluster on Debian Wheezy, and while it works great while everything is up, rebooting a node has mixed results, things really start to go south as on reboot another node will get upset and crash which has a bit of a cascade effect sometimes requiring manual intervention to resolve. Can anyone comment on their experience with gfs and/or ocfs2 and reliability? I'm using ocfs2 to store xen configs, and iso images for xen vm's, so performance really isn't a consideration while reliability is. Thanks James
Hi, On 15/10/2012, at 7:26 PM, James Harper <james.harper@bendigoit.com.au> wrote:
I am presently using ocfs2 managed by a pacemaker cluster on Debian Wheezy, and while it works great while everything is up, rebooting a node has mixed results,
Any reason why you're not using the O2CB cluster stack for OCFS2? I have no personal experience using non-Oracle cluster stacks with OCFS2.
Can anyone comment on their experience with gfs and/or ocfs2 and reliability?
I'm using ocfs2 to store xen configs, and iso images for xen vm's, so performance really isn't a consideration while reliability is.
OCFS2/O2CB form the basis of the cluster filesystem and stack for Oracle VM, which provides both reliability and performance. Check how recent your ocfs2-tools are, because mkfs.ocfs2 now has a -T vmstore option which is optimised for storing Xen-like virtual disk images alongside small configuration files. It also enables reflink support (for copy-on-write clones) and inode packing. Cheers, Avi
Hi,
On 15/10/2012, at 7:26 PM, James Harper <james.harper@bendigoit.com.au> wrote:
I am presently using ocfs2 managed by a pacemaker cluster on Debian Wheezy, and while it works great while everything is up, rebooting a node has mixed results,
Any reason why you're not using the O2CB cluster stack for OCFS2? I have no personal experience using non-Oracle cluster stacks with OCFS2.
Pacemaker manages o2cb
Can anyone comment on their experience with gfs and/or ocfs2 and reliability?
I'm using ocfs2 to store xen configs, and iso images for xen vm's, so performance really isn't a consideration while reliability is.
OCFS2/O2CB form the basis of the cluster filesystem and stack for Oracle VM, which provides both reliability and performance. Check how recent your ocfs2-tools are, because mkfs.ocfs2 now has a -T vmstore option which is optimised for storing Xen-like virtual disk images alongside small configuration files. It also enables reflink support (for copy-on-write clones) and inode packing.
I've read about that but don't have a platform to test it on yet. James
On 15/10/2012, at 7:54 PM, James Harper <james.harper@bendigoit.com.au> wrote:
Any reason why you're not using the O2CB cluster stack for OCFS2? I have no personal experience using non-Oracle cluster stacks with OCFS2.
Pacemaker manages o2cb
Right, ok. Can you provide more information about what happens during a reboot that causes problems? I'm most interested in seeing the /var/log/messages on both the node being rebooted and the remaining nodes, to make sure you see the rebooting node properly leave the cluster. Also, what happens when the node comes back, i.e. the events that cause the non-rebooted nodes to fence themselves. If Pacemaker manages o2cb, it may be as simple as adjusting the o2cb cluster timeout values to something more appropriate for your particular storage and network. Though, I'm not sure what Pacemaker *does* to o2cb to manage it: it should just manage itself. I may have to read more about that. :) Cheers, Avi
participants (2)
-
Avi Miller -
James Harper