[luv-main] pacemaker, iscsi, multipath, and filesystem
I'm testing multipath with iscsi now, and having the multipath infrastructure present means that the actual iscsi device is unusable, you have to go through the multipath device node. This makes perfect sense exact that pacemaker wants to mount the filesystem before multipath has created the device node. I notice there is a Delay agent, but that seems like a bit of a hack - how much of a delay is enough, without waiting unnecessarily long? Any suggestions? This is a followup to my nfs or ocfs2 question... it turns out that nfs failover doesn't work that smoothly due to hanging tcp connections and holding the block device open so I've gone with ocfs2. Thanks James
On Wed, 14 Sep 2011, James Harper wrote:
This is a followup to my nfs or ocfs2 question... it turns out that nfs failover doesn't work that smoothly due to hanging tcp connections and holding the block device open so I've gone with ocfs2.
Really? I, just a few moments ago, failed over the app servers that some of the cluster nodes in www.bom.gov.au use to mount /web. What might be tricky is things like fsid and the like. We using nfs ver 3, tcp, hard mounts, retrans=2 (ie, defaults as far as I know). The backend filesystem is vxfs but probably doesn't matter. I don't know how veritas works behind the scenes, but I suspect it just removes the IP address, unexports the filesystems then unmounts before reassigning the IP address to the destination host. I wonder how many tens of thousands of dollars were paying for the nice gui? -- Tim Connors
On 14/09/11 14:17, James Harper wrote:
I'm testing multipath with iscsi now, and having the multipath infrastructure present means that the actual iscsi device is unusable, you have to go through the multipath device node. This makes perfect sense exact that pacemaker wants to mount the filesystem before multipath has created the device node.
I notice there is a Delay agent, but that seems like a bit of a hack - how much of a delay is enough, without waiting unnecessarily long?
Any suggestions?
This sounds like something I should know the answer to, but don't. I'd never recommend the Delay agent, because it's just a fixed delay, which is... ugly. It may perhaps be better to (somehow) not start Pacemaker at all until after multipath had settled down. You might try #linux-ha or #linux-cluster on freenode in a few hours time (after the Europeans wake up). Otherwise you'll just be chatting with me :) Regards, Tim -- tim@wirejunkie.com http://www.wirejunkie.com/
On 14/09/11 14:17, James Harper wrote:
I'm testing multipath with iscsi now, and having the multipath infrastructure present means that the actual iscsi device is
you have to go through the multipath device node. This makes perfect sense exact that pacemaker wants to mount the filesystem before multipath has created the device node.
I notice there is a Delay agent, but that seems like a bit of a hack
unusable, -
how much of a delay is enough, without waiting unnecessarily long?
Any suggestions?
This sounds like something I should know the answer to, but don't.
I'd never recommend the Delay agent, because it's just a fixed delay, which is... ugly. It may perhaps be better to (somehow) not start Pacemaker at all until after multipath had settled down.
Pacemaker does the iscsi for me in this case, so that's not an option in my current configuration. It could be an option though as the volumes I'm working with right now are ocfs2 and cloned to all nodes. Once I start setting up VM's I'll want the iSCSI volumes added and removed on demand though so it is still a problem that needs solving.
You might try #linux-ha or #linux-cluster on freenode in a few hours time (after the Europeans wake up). Otherwise you'll just be chatting with me :)
I never got the hang of IRC. It always seemed too busy. James
Pacemaker does the iscsi for me in this case, so that's not an option
in my
current configuration. It could be an option though as the volumes I'm working with right now are ocfs2 and cloned to all nodes. Once I start setting up VM's I'll want the iSCSI volumes added and removed on demand though so it is still a problem that needs solving.
I think I've fixed it. A few 'meta is-managed="false"' entries hiding away were obscuring the problem. My initial diagnosis was incorrect - it's not a delay problem just a misconfiguration :) Thanks James
On 14/09/11 15:56, James Harper wrote:
Pacemaker does the iscsi for me in this case, so that's not an option
in my
current configuration. It could be an option though as the volumes I'm working with right now are ocfs2 and cloned to all nodes. Once I start setting up VM's I'll want the iSCSI volumes added and removed on demand though so it is still a problem that needs solving.
I think I've fixed it. A few 'meta is-managed="false"' entries hiding away were obscuring the problem. My initial diagnosis was incorrect - it's not a delay problem just a misconfiguration :)
Oh, sweet. I love those problems :) Regards, Tim -- tim@wirejunkie.com http://www.wirejunkie.com/
participants (3)
-
James Harper -
Tim Connors -
Tim Serong