This post is more than 5 years old
6 Posts
0
1522
Help regarding PPME with Host copy
Hello Everyone,
We are facing a problem in regards of using PPME with hostcopy. Our majority of clusters are 4 node SQL servers with 3 active instances per cluster.
The PPME white paper says When migrating devices in cluster configuration only one node should remain active and move all failover goups to one active node and shut down the remanining cluster nodes.
But we are unable to move multiple SQL instances on one single node due to memory and CPU constraints.
Please advice!!
Thank you
Brion2
154 Posts
0
January 20th, 2011 13:00
OK. I just want to be clear. You have a 4-node cluster with 3 active nodes. So, the 3 active nodes can ONLY fail to the passive node.
Example:
Act_node_1 Only fails to Passive node_4
Act_node_2 Only fails to Passive node_4
Act_node_3 Only fails to Passive node_4
In this case, as long as the passive node is down, failover is impossible, correct?
Brion2
154 Posts
0
January 20th, 2011 08:00
Hello. Are each of three active instances processing reads and writes to the same LUNs? Are they operating as individual applications?
Malhar1
6 Posts
0
January 20th, 2011 10:00
Hi Brion,
No they are not.. they are accessing different LUNs.
Yes they are operating as individual applications.
Thanks.
Malhar1
6 Posts
0
January 23rd, 2011 14:00
Yes.. 3 active nodes only fails to only Passivenode_4.
My question is, for PPME we have to move all the failover groups to one acitve node and shut down all other nodes. In this scenario will one active node can take the load?
What would you advice?
Malhar1
6 Posts
0
January 24th, 2011 06:00
Brion,
If Act_node_1 Only fails to Passive node_4
Act_node_2 Only fails to Passive node_4
Act_node_3 Only fails to Passive node_4
In this case can we shut down the Passive node_4 and start the migration with three active nodes up?
Will this work?
Sam Claret
86 Posts
0
January 24th, 2011 07:00
Hello
The EMC perspective is quite clear - the hostcopy mode should be run on a single node with all other nodes in the cluster shutdown.
There are a couple of reasons for this.
1) the inability for the single node to possibly failover to another node - resolved by limiting ownership of the devices in MSCS
2) identification of the target devices on the other nodes after the migration.
It is this last part that has the potential to 'kill' nodes in your cluster after the migration if nodes are up at the time of commiting (and later cleanup) of the luns then the other active nodes may not get the update properly.
I would recommend that you do as you have said - shutdown the passive node (4) - limit the ownership of the specific instances - but migrate only one instance of SQL at a time and then reboot the entire cluster in an orderly fashion. Make sure any issues are resolved before moving onto the next instance.
This will have the negative impact of potentially taking much longer - however only one SQL instance will be exposed at any one time.
Zoning/masking/storage group management should be used at each step to limit or make available the LUNS as appropriate to minimize MSCS getting 'confused'
Sam Claret EMC TSE3
Sam Claret
86 Posts
0
January 26th, 2011 06:00
Hello
Given that the manual states that the other nodes have to be down for the migration I expected something like this - instead of rebooting you may be able to do something like restart the cluster service - though I doubt it.
Note that this is not supported what you are doing - just that it works if you have the reboot there.
Sam Claret EMC TSE3
Malhar1
6 Posts
0
January 26th, 2011 06:00
Sam,
We tested on the test cluster for PPME with 2active/passive node. In this test we shutdown the passive node, took cluster group offline and limiting the ownership of each node so that it wont failover. We were successfull in moving the data to target devices but when we were done with one node and scanned on the other active node, we were unable to see the drives. So we had to reboot the other active node so that we can see the changes.
In PPME nodes doesn't need a reboot. Can you suggest something for this?
Thanks,
Malhar
Sam Claret
86 Posts
0
January 26th, 2011 07:00
Hello
I am aware of this - however please bear in mind that you are venturing into unknown territory and if there are issues and problems with something that is basically not supported and contrary to the required method as stated in the manual then this is to be expected.
Thankyou
Sam Claret EMC TSE3
Malhar1
6 Posts
0
January 26th, 2011 07:00
Hi,
Actually in prod environment we have SQL cluster with 3 active/passive node.. So its very difficult to bring all the resources to one single node as stated in the manual. So we were testing with this.
Thanks
-Malhar