Cluster Fail to start automaticly

[banini]

We just install a new Mc ServiceGuard under HP-UX.
during the test, we power off both nodes and starting with only one.
We Remark that the Startup is not automatically.
and the service is not up.
but we can do it manually.
Any help.

[Michael Steele_2]

Automatic cluster startup will take place if the flag AUTOSTART_CMCLD is
set to 1 in the /etc/rc.config.d/cmcluster file.

[likid0]

You need to change the file:

/etc/rc.config.d/cmcluster

you can execute on both nodes:

#ch_rc -a -p AUTOSTART_CMCLD=1

[banini]

Many Thanks, but we have already done this procedure.
Any more suggestion pls.

[Turgay Cavdar]

If you start only one node then the cluster waits for the other nodes with a second which configured in cluster parameter "AUTO_START_TIMEOUT". The default is 10 Minutes.

[Matti_Kurkela]

Whether Serviceguard itself starts at system boot or not, is determined in the file /etc/rc.config.d/cmcluster.

If the AUTOSTART_CMCLD variable in that file is set to 0, the node won't join/form a cluster automatically. If the variable is set to 1, it will.

When the cluster is starting up, the package startup is determined by the auto_run setting in the package configuration (the name of the setting is AUTO_RUN in older Serviceguard versions).

Changing only the package configuration file is not enough: you must run "cmapplyconf -P " to make the new configuration take effect.

Using "cmmodpkg -e " or "cmmodpkg -d " is not quite the same thing: the package configuration file determines the _default_ state of the AUTO_RUN flag, while cmmodpkg modifies the _current_ state of the flag. When the cluster is started, all the package AUTO_RUN flags are reset to the default state.

MK

[Michael Steele_2]

Hi

Well, you're going to have to start debugging and the first place to look is always, always /var/adm/syslog/syslog.log. There will be something. Make sure you verify the TIMESTAMP.

There are other commands:

ps -ef | grep -i cmcld (* is this process started? *)

ps -ef | grep -i cm (* what other mc/sg processes do you see? There are 14 MC/SG daemons *)

cmgetconf -c cluster1 temp.ascii

cmcheckconf -C temp.ascii

cmviewcl (* will report if cmcld is up *)

cmscancl (* another debugging command *)

Question: Is this a new build?
Question: What did you do to cause this?
Question: Has this cluster ever been up and running or is this a new project?

[Horia Chirculescu]

>we power off both nodes and starting with only one.

When starting only one node from a 2 nodes cluster, the cluster itself won't start.

The packages will be moved from one node to another in the cluster only when both nodes were functional and something was wrong with one node.

In order to use only one node, you have to run force (-f) cluster forming with only one node, calling cmruncl -f -n node1


Horia.

[Stephen Doud]

/sbin/init.d/cmcluster runs the cmrunnode command at boot time, not cmruncl. cmrunnode will cause the booting system to either join a currently running cluster or form a cluster if the partner node is also performing cmrunnode within 10 minutes.

The standard cmruncl command will not cause a cluster to form if any member nodes are not able to respond to the cluster formation request.

You also stated that you will only be starting one server, so since none of the above conditions will occur, editing the /etc/rc.config.d/cmcluster AUTOSTART_CMCLD parameter will serve no purpose.

If you want a one-node cluster to form after booting, watch the syslog.log for a message indicating cluster formation attempts failed (10 minutes after the first cmrunnode is issued by /sbin/init.d/cmcluster), then run 'cmruncl -n -f'