HACMP setup - nodes powering off

[unixfreak]

Hi, maybe I don't understand HA well enough (well, I don't)

Whenever I try to start-up the cluster services both nodes power off instantly. Wehn I run verification I get two warnings:

- Only one node on net_ether_01

- node01 and net_ether_01 are on the same subnet

Is it related? Why would the nodes power off?

Thanks

 

[RMGBELGIUM]

conflicts in the config of your HACMP.You should run a verification before starting up the cluster.This will tell you if you have a valid setup , or where stuff goes wrong

rgds,

R.

 

[khalidaaa]

did you try to view the /tmp/hacmp.out for error messages?

I think there is a halt -q command in one of the events that cause the nodes to go off but i'm not sure which one is it!

Any way, I hope you don't experience a half-brain cluster!

Regards,
Khalid

 

[unixfreak]

Nothing in the logs. It's sorted out now anyway, don't know how

 

[khalidaaa]

Just a question, do you use disk heartbeating?

 

[unixfreak]

Nope, HA 5.1

 

[khalidaaa]

So you don't have any means of a second heartbeating other than network? I think this what caused the nodes to go down because it was experiencing a split-brain cluster! When something goes wong in the network (even sometimes glitches) the cluster will think that the network is failing so each node will try to grap the disks (I think you are using external disks right? like san!?) so this will lead the cluster to be in a split-brain situation! and if you, by any change, having mirroring on the disks then each node might take a copy of the mirror and this might lead to a data corruption!?!

You got to have another means of heartbeating! think about using the current shared disks for that to avoid going thru all of that in the future!

Regards,
Khalid

 

[unixfreak]

The cluster has two ethernet networks and a serial network.

I think the problem was that two labels had been accidentally been put on the same network.