Ocfs2 users - Nov 2006 - about 2 nodes enviroment and metalink note 394827.1

Hi to all:

In 2 nodes environment I've 'suffered' the 'reboot 1st node
hangs 2nd
one', has described in metalink note 394827.1

Exactly this note says that this occurs when interconnect fails. Then i
understand that if interconnect fails the idea is that node 1 stay up
and running and node 2 'kills' itself to avoid split-brain. 

When 1st node reboots ( planned reboot )  ocfs2 thinks that interconnect
has failed? If this is true, the cluster is condemned to die, cause node
1 is rebooting and node 2 kills itself, isn't it?

Under a well know 2 nodes environment, does not exist some type of
message like '2nd node,I'm rebooting, don't panic and stay tuned
' ? :)

Any tip to avoid this behaviour ?

I think that one way ( not optimal in any way ) could be adding another
node, only for ocfs2,  to help second node to think that it is in the
max nodes group, when the 1st node reboots...


Regards and TIA

D.

I would imagine you are using RHEL4. If so, upgrade the ocfs2-tools
to 1.2.2. The previous version of the ocfs2 init script did not always
umount ocfs2 volumes on clean shutdowns leading to this problem.

dballester.david@gmail.com wrote:
> Hi to all:
>
> In 2 nodes environment I've 'suffered' the 'reboot 1st node
hangs 2nd
> one', has described in metalink note 394827.1
>
> Exactly this note says that this occurs when interconnect fails. Then i
> understand that if interconnect fails the idea is that node 1 stay up
> and running and node 2 'kills' itself to avoid split-brain. 
>
> When 1st node reboots ( planned reboot )  ocfs2 thinks that interconnect
> has failed? If this is true, the cluster is condemned to die, cause node
> 1 is rebooting and node 2 kills itself, isn't it?
>
> Under a well know 2 nodes environment, does not exist some type of
> message like '2nd node,I'm rebooting, don't panic and stay
tuned ' ? :)
>
> Any tip to avoid this behaviour ?
>
> I think that one way ( not optimal in any way ) could be adding another
> node, only for ocfs2,  to help second node to think that it is in the
> max nodes group, when the 1st node reboots...
>
>
> Regards and TIA
>
> D. 
>
>
>
>
> _______________________________________________
> Ocfs2-users mailing list
> Ocfs2-users@oss.oracle.com
> http://oss.oracle.com/mailman/listinfo/ocfs2-users
>

If you use correct startup scripts, then shutdown should stop  OCFS and then
O2CB, and so other node knows that first node rebooted.

If first node panic, I don't understand why second node decided to reboot -
it can easily see that there are not heartbeat on disk anymore, so network
connection loss is normal and it do not have a reasons to reboot. It should
reboot, if it don't have network connection but see heartbeat on the disk
(so it knows that other node is alive, but they are not working together).


----- Original Message ----- 
From: <ballester.david@gmail.com>
To: <ocfs2-users@oss.oracle.com>
Sent: Thursday, November 09, 2006 3:03 AM
Subject: [Ocfs2-users] about 2 nodes enviroment and metalink note 394827.1

> Hi to all:
>
> In 2 nodes environment I've 'suffered' the 'reboot 1st node
hangs 2nd
> one', has described in metalink note 394827.1
>
> Exactly this note says that this occurs when interconnect fails. Then i
> understand that if interconnect fails the idea is that node 1 stay up
> and running and node 2 'kills' itself to avoid split-brain.
>
> When 1st node reboots ( planned reboot )  ocfs2 thinks that interconnect
> has failed? If this is true, the cluster is condemned to die, cause node
> 1 is rebooting and node 2 kills itself, isn't it?
>
> Under a well know 2 nodes environment, does not exist some type of
> message like '2nd node,I'm rebooting, don't panic and stay
tuned ' ? :)
>
> Any tip to avoid this behaviour ?
>
> I think that one way ( not optimal in any way ) could be adding another
> node, only for ocfs2,  to help second node to think that it is in the
> max nodes group, when the 1st node reboots...
>
>
> Regards and TIA
>
> D.
>
>
>
>
> _______________________________________________
> Ocfs2-users mailing list
> Ocfs2-users@oss.oracle.com
> http://oss.oracle.com/mailman/listinfo/ocfs2-users
>