Ocfs2 users - Mar 2010 - node B reboots when node A is isolated from the network

A and B are identical machines. Network has lots of redundancy. They 
both access same OCFS2 volumes over Fiber Channel on a SAN.

Red Hat Enterprise Linux Server release 5.3 (Tikanga)
2.6.18-128.el5 x86_64
OCFS2 1.2.9

I log in to the switch and isolate node A from the network (shutdown all 
ports to node A on the switch), but the node remains connected to SAN 
over fiber.

In that case, node B will reset itself and then boot up. While booting 
up, node B will not remount the OCFS2 volumes on the SAN. If I try to 
mount them manually on B, mount fails.

All this time, the OCFS2 volumes are still mounted on node A - I can 
access them if I go on A through the console and do some filesystem 
tests (df, mount, touch, rm ... ).

I can only remount the OCFS2 volumes on node B after I bring node A back 
online on the Ethernet switch.

I assume this is actually normal behavior. Am I correct?

-- 
Florin Andrei
http://florin.myip.org/

If node A is a lower node number than node B, then the behavior
is correct. In a 2 node cluster, if the two nodes cannot talk to each
other, the higher node number will fence itself. Also, when a node
mounts a volume, it initiates connections to other live nodes. If any
connection fails, the mount will fail.

Florin Andrei wrote:
> A and B are identical machines. Network has lots of redundancy. They 
> both access same OCFS2 volumes over Fiber Channel on a SAN.
>
> Red Hat Enterprise Linux Server release 5.3 (Tikanga)
> 2.6.18-128.el5 x86_64
> OCFS2 1.2.9
>
> I log in to the switch and isolate node A from the network (shutdown all 
> ports to node A on the switch), but the node remains connected to SAN 
> over fiber.
>
> In that case, node B will reset itself and then boot up. While booting 
> up, node B will not remount the OCFS2 volumes on the SAN. If I try to 
> mount them manually on B, mount fails.
>
> All this time, the OCFS2 volumes are still mounted on node A - I can 
> access them if I go on A through the console and do some filesystem 
> tests (df, mount, touch, rm ... ).
>
> I can only remount the OCFS2 volumes on node B after I bring node A back 
> online on the Ethernet switch.
>
> I assume this is actually normal behavior. Am I correct?
>
>