|Ok, we'll assume you have to use CLVM then.|
Have you checked if multipathd is running when you kill the storage device?
What are the symptoms of the cluster freeze. (I will need some kind of logging output.)
On Apr 25, 2008, at 12:06 PM, gnia gnia wrote:
sorry my first message was a bit vague.
We have 2 HP EVA arrays : one being the backup device of the other in case a crash occurs.
So we make each node of the cluster (active/passive , multipath + clvmd + cmirror) connected to both EVAs.
We create one disk on each EVAs with same characteristics, plus one disk on one of the 2 EVAs as a log mirror disk.
Then we 'lvcreate -m1 disk1 disk2 disk3' and put our file system resource on this lv.
That is where our test (shutdown of an EVA) totally fails.
Unfortunately ha-lvm won't be usable because we intend to use GFS soon.
What we need is a solution that ensures clusterized services to carry on working even if on EVA burns...
Thanks a lot for helping.
On Fri, Apr 25, 2008 at 4:26 PM, Jonathan Brassow <jbrassow redhat com
I'm not sure what you mean by multi-SAN. Do you mean that you have two independent storage networks connecting the same machines to the same disks, or same machines to different disks, or something else? It sounds like you have 1 SAN with two storage devices? It also sounds like your top level services are active/passive, but that your storage is setup active/active. You may consider HA-LVM instead of CLVM in this case (http://sourceware.org/cluster/wiki/LVMFailover).
I've seen this scenario where things get frozen... Turns out that the user did not have multipathd running, so multipath was queueing all I/O until a path would come back - effectively freezing everything.
The problem could be a number of things... we may need more information if the above doesn't help.
On Apr 25, 2008, at 7:41 AM, gnia gnia wrote:
Situation : 2 nodes cluster (active/active) under RedHat 4.
Each clusterized service has a multipathed/mirrored LV (one PV on each storage bay -> HP EVA 8000) declared as an FS resource.
(We decided to implement a multi-SAN environment to prevent from one SAN failure)
But all clusterized mirrored LVs get frozen and become unusable when we simulate one SAN outage.
According to you experts, what is the best way to implement a storage redundant redhat cluster that wouldn't fail in case of one SAN outage ?
Is lvm mirroring a godd choice?
Linux-cluster mailing list
Linux-cluster redhat com