Order of startup for 2 node cluster

[Morris Makuch]

I have a 2 node cluster that share one monitor console. After a power down of both nodes, is there a difference which node should be powered up first to ensure that the package starts up automatically and both nodes join the cluster? The autostart is set to 1 on both nodes and I want to be sure one node comes up all the way clean so I can check for any errors on the console before switching over the console to the other node and bringing it up.

[melvyn burnard]

If you do this the way I suspect you mean, your clustyer will not come up!
There is a timeout period of 10 minutes, when the first node that gets to th epoint of starting to form a cluster will be looking for the other node to be availablke.
If it is not ready to try to join the cluister in that time period, the first node will cease trying to form a cluster, and then once the 2nd node getst to the right stage, it will go through the same 10 minute period.
to start a cluster up from the beginning, 100% of the nodes knwon about in the configuration MUST be present.
If you do consider continuing with this, when the first node has booted, and has gone through the 10 minute period, you could issue the following:
cmrunnode -v -n node1_name

This will start th eclkuster on the first node, and allow you to then start th eboot up of the second node and have it auto-join the running cluster.
CAVEAT!!! All the packages will start up on the first node.

[James R. Ferguson]

Hi Morris:

I boot (power-up) both nodes simultaneously. I *expect* that *no* cluster formation will occur for 10-minutes (the value of 'AUTO_START_TIMEOUT') and in fact will fail.

I wait for the expiration of this period and then manually issue a 'cmruncl'.

For more information, see Knowledge Base document #UXSGLVKBAN00000009.

Regards!

...JRF...

[linuxfan]

Hi Morris,

Both Melvyn and James have already told you that when you reboot the first node the cluster will try to start but will fail since all the nodes in the cluster are not present.

If you don't want to wait for the 10 minutes (or whatever your AUTO_START_TIMEOUT variable is set to) what you could do is while booting up the first node, modify the /etc/rc.config.d/cmcluster file and set your AUTOSTART_CMCLD=1 to 0
Once your machine is up all the way, run
cmruncl -v -n node1 or
cmrunnode -v node1
(THis will start the cluster on node1)
MOdify the /etc/rc.config.d/cmcluster file to change the variable AUTOSTART_CMCLD to 1

Once your application is up and running on node1, you can switch the console to node2 and bring it up, you don't need to do anything on node2 since the cluster is already up and running on node1.

-HTH
Ramesh

[A. Clay Stephenson]

Hi,

I agree start up both nodes at once unless you want to set AUTOSTART_CMCLD to zero and start the cluster manually. You can always simply check you startup logs rather than watching the console. I really prefer to have a dedicated console for each box so that I don't have your situation and can see exactly what is happening on all the boxes in the cluster.

Regards, Clay

[Morris Makuch]

Thanks for all the replies. My whole idea is to have the cluster and package come up automatically with as little intervention as possible. The only intervention needed is just to walk to the servers and physically power the servers up and then walk away and hope that the nodes, cluster, package and switching are enabled.

[Sanjay_6]

Hi Morris,

You can start any node first. However normally what you should do is to start that system first on which you want the package to start first. Since your system is configured to start the cluster service automatically, it will form a cluster on its own. and will start the package configured for this node if the package parameters are set to start automatically.

However there should not be a problem if you start both the servers at the same time. Most of the time, you hardly face any problem. In case there is any problem, it will be logged into /var/adm/syslog/syslog.log file that you can look at afterwords.

Hope this helps.

thanks

[James R. Ferguson]

Hi Morris:

I'm afraid that you are not going to be able to "...physically power the servers up and then walk away and hope...".

In my opinion, the important fact is that you *know* that you are likely to have to manually interact with your cluster and in what timeframe. Powering off (and on) all nodes of a cluster, simultaneously, should be relatively rare. As an administrator, either I want to be available to control it and/or I want to have adequate operational documentation for someone else to do it.

Regards!

...JRF...