4 nodes clusters

[Samuel Bouhnik]

Hi All,

I have a four nodes cluster using FC and with SG 11.13 with the latest patch.

If I'm switching off all four nodes and I'm restarting only 3 the cluster doesn't go up.

I read that it's normal due that cmrunnode is waiting that all nodes are up and running. To fix this problem, cmruncl -f -n should be used. But this command can be only ran on one node.

So my question is, how to fix that problem ?

Samuel

[Sanjay_6]

Hi Samuel,

Try,

cmruncl -f -n node1 -n node2 -n node3

don't include the name of the fourth node that is down.

Hope this helps.

Regds

[Samuel Bouhnik]

Hi Sanjay,

Ok ... it should work like that but how to modify the /sbin/init.d/cmcluster scripts to do it ... I want that thing will be automatic.

Samuel

[Christopher McCray_1]

Hello,

The cmcluster script is designed to do exactly what it does... Allows the rejoining of a node automatically into and EXISTING cluster. The reason for this is to give a warning to the SA that a problem exists by not allowing a cluster reformation due lack of 100% node attendance. I don't believe that this type of modification is supported by HP (although I could be wrong). I don't think this should be done. If this bouncing of your environment is planned and necessary ( what about mnual failover/ maintenance one node at a time?), I would turn off AUTO_CMCLD and halt the cluster manually, reboot and manually start the cluster in this instance.

Just my $0.02 and apologies if I didn't tell you what you want to hear.

Chris

[Steve Lewis]

The reason it doesn't come up is this:
The 3 remaining servers will not be able to communicate with the 4th, so will not know if

(a) the 4th server is down OR
(b) the 4th server is running packages OK and that the other 3 have a network problem.

So by default it refuses to run packages if it cannot detect everything in the cluster when it boots up.
That is why you have to manually intervene in this case. It is set up this way to avoid any chance of data corruption.
If you change this it may not be supported.

[Sanjay_6]

Hi,

you cannot configure the /etc/rc.config.d/cmcluster to do start the three nodes in a four node cluster when one of the nodes is down. The best option would be diable the automatic custer startup on boot, AUTOSTART_CMCLD=0.

One thing you should remeber is that once the cluster is up, even if it is up on one node, the AUTOSTART_CMCLD=1 on another node that has booted will start the cluster services on that node.

Hope this helps.

Regds

[Samuel Bouhnik]

Hi,

Thanks all for your answers ... but I can't agree with you. Let me explain one case : if there is a power cut during the night and the power is coming back few seconds after and unfortunately one server is not able to restart, my cluster should be able to run to keep the service available.

So I think, it's a bug in SG. Does some HP expert online could answer to me ?

Regards,

Samuel

[Christopher McCray_1]

Hello again,

No that is not a bug, It is drawing your attention to the fact that you had a power outage in which all four (??!) of your servers had lost power, one could not make it back up, and you definately have a problem that you need to come in and investigate. MCSG is set up to not take any chances at all and will not start if you have less than 100% node attendance at cluster start attempt.

One thing I might want to point out, although I'm just guessing; I hope you are not in a situation that all your power is from one source, but we do with what we can. Remember that HA is more tahn MCSG.

Good luck
Chris

[melvyn burnard]

You are wrong in your assumption. SG is designed to work in this way, as already pointed out in the thread you need 100% of nodes ready to start the cluster.
There is no way of changing this, it is the way it works.

[Samuel Bouhnik]

Thanks to all

Bye

Samuel