Friday Feb 16, 2007
Friday Feb 16, 2007
The new command "clresourcegroup evacuate" switches all resource groups off of a given node. This is a narrower form of the command "clnode evacuate", which switches both resource groups and disk device groups off of a given node. In the old command line interface, the "scswitch -S" command performed the same function as "clnode evacuate" does in the new command line interface. However, the old command line interface had no exact counterpart to "clresourcegroup evacuate".
Recently, a co-worker asked me this question about node evacuation:
I have an nfs resource group configured and online on node1. I evacuate node1 first, and the resource group switches over to node2. Immediately after the nfs resource group goes online on node2, I evacuate node2. But then the nfs resource group is offline on both nodes.
If I wait for some time after the first evacuation to do the 2nd one, then the nfs resource group will be online on node1.
Is this expected? Does it happen for all agents?
The answer is "yes" and "yes". This is expected behavior, and it applies to all agents. In the new command line interface, it is controlled by the -T option of clresourcegroup(1CL). Here is an excerpt from the man page:
-T seconds
--time=seconds
--time seconds
Specifies the number of seconds to keep resource groups from switching back onto a node or zone after you have evacuated resource groups from the node or zone.
You can use this option only with the 'evacuate' subcommand. You must specify an integer value between 0 and 65535 for seconds. If you do not specify a value, 60 seconds is used by default.
Resource groups cannot fail over or automatically switch over onto a node or zone while that node or zone is being evacuated. The -T option specifies that resource groups are not to be brought online by the RGM on the evacuated node or zone for a period of _seconds_ seconds after the evacuation has completed. You can override the -T timer by switching a resource group onto the evacuated node or zone, using the clrg 'switch' or 'online' subcommand with the -n option. When such a switch is done, the -T timer is immediately considered to have expired for that node or zone. However, switchover commands such as 'clrg online' or 'clrg remaster' without the -n flag will continue to respect the -T timer and will avoid switching any resource groups onto the evacuated node.
To modify the evacuate behavior, you can specify a different value of -T than the default of 60 seconds.
In the old command-line interface, the feature corresponding to -T is the the -K option, used with the scswitch -S command.
Martin Rattner
Sun Cluster Engineering
Posted by guna on March 22, 2007 at 07:57 AM PDT #
Posted by Suraj Verma on March 22, 2007 at 11:44 PM PDT #
Posted by Tim Read on March 23, 2007 at 02:57 AM PDT #
Posted by Martin Rattner on March 23, 2007 at 10:43 AM PDT #
Posted by Martin Rattner on March 23, 2007 at 10:44 AM PDT #
Hi Sun gurus,
I have four node sun cluster of which
I want to take 3 nodes offline (break cluster) all resources and device groups and perform reinstall of web applications and rejoin them the cluster(ONLINE) with 4th node.
Note that our four node cluster is configured as below.
runtime 1 and 2 - active-active
manager 1 and 2 - active-standby
I want runtime 1 and 2 and manager 1 nodes off the cluster and reinstall applications and join the cluster back.
Our applications are installed locally on disks and media-data stored on SAN disks.
Can anyone suggest a solution.
Thanks,
Shiva
Posted by Shiva Duddeda on August 07, 2009 at 01:44 PM PDT #