topic Re: SG two node cluster - startup(boot) in Operating System - HP-UX

SG two node cluster - startup(boot)

edi_4 — Fri, 30 Jan 2009 11:06:34 GMT

Hi! We are running 2 node serviceguard linux cluster (SLES). Suppose this scenario:
One node fail for a long time. Packages are fail over to second node. This works OK.
Now I am going to reboot second node. After reboot the cluster is not running. I can start it using cmruncl -n node2.
Is there any solutions to form cluster automatic - without manual intervention. (cmcluster.rc AUTOSTART_CMCLD=1).
Perhaps I must delete failing node from configuration? Is there more elegant solution. Thank's!

Re: SG two node cluster - startup(boot)

Duncan Edmonstone — Fri, 30 Jan 2009 12:07:31 GMT

if a failed node is down and is likely to stay down for a long time, then yes it can make sense to delete it from the cluster configuration.

this is because initial cluster formation can *only* occur when all cluster nodes are present. if you think about it, this makes sense as otherwise the cluster isn't aware of whether a node has failed or whether it has just lost communication. you need to avoid split brain situations, so this rule is always the case.

don't get tempted to code the "cmruncl -n node2" into a startup script - that would be a *very* bad idea!

HTH

Duncan

Re: SG two node cluster - startup(boot)

Stephen Doud — Fri, 30 Jan 2009 14:05:46 GMT

If the failed node will be out of commission for an extended period and you want the running (functional) node to be able to auto-start Serviceguard at boot time, then you must remove the down node from the cluster. You can do that by removing all references to the down node in the package configuration file(s) and the cluster configuration file (comment the references out for now). Then run cmapplyconf on ALL the configuration files at the same time. eg:

# cmapplyconf -f -v -C -P -P ...

After the cluster is reconfigured down to a 1-node cluster, the remaining node should auto-start the cluster at boot time because there are no other nodes it will be waiting on to join with.

Re: SG two node cluster - startup(boot)

Emil Velez — Sat, 31 Jan 2009 00:42:42 GMT

check for a /etc/rc.config.d/cmcluster
or something like that with the AUTOSTART_CMCLD variable.

I forget what the path is on linux.

script that starts serviceguard at boot time where the node will join a existing clsuter.

Re: SG two node cluster - startup(boot)

edi_4 — Mon, 02 Feb 2009 06:53:06 GMT

Thank's for answers. I came from Tru64 - there is a litle diference... If I can understand I have to delete failed member from configuration in order to form cluster on boot, there are not other solutions. Yhank you!