06-14-2012 04:05 PM
we have a 5 node VCS cluster .We need to apply OS patches without any downtime.On all the nodes we have 2 or more service groups running.After patching the node will go for a reboot.The OS is linux
What would be the best recomended ways to peform this task without no downtime ?What all things i need to keep in mind when i am planning this.
Recomedations would be appreciated
Thanks and regards,
Siva
Solved! Go to Solution.
06-18-2012 01:22 PM
06-15-2012 01:44 AM
Suppose your nodes are A, B , C, D & E and A, B and C run 2 services and D and E run 3 services so A runs A1 and A2 services for example and D runs D1, D2 and D3. You may have somehting a bit different but the same principle in the method below can be used:
Start with: A B C D E A1 A2 B1 B2 C1 C2 D1 D2 D3 E1 E2 E3 Switch A1 and A2 to B and C and patch A: A B C D E Patch B1 B2 A1 C1 C2 A2 D1 D2 D3 E1 E2 E3 Switch B's services to A and patch B: A B C D E B1 B2 A1 Patch C1 C2 A2 D1 D2 D3 E1 E2 E3 Switch C's services to B and patch C: A B C D E B1 B2 A1 C1 C2 A2 Patch D1 D2 D3 E1 E2 E3 Switch D's services to C and patch D: A B C D E B1 B2 A1 C1 C2 A2 D1 D2 D3 Patch E1 E2 E3 Switch E's services to D and patch E: A B C D E B1 B2 A1 C1 C2 A2 D1 D2 D3 E1 E2 E3 Patch Switch A1 and A2 to E A B C D E B1 B2 C1 C2 D1 D2 D3 E1 E2 E3 A1 A2
This will involve some downtime of the services when you are switching services, but you cannot avoid this, but if the services switch quickly, this maybe transarent to the end user depending on the nature of your application.
Note the services on the server you start with will have the most downtime, so in example above, A1 and A2 are switched 3 times, where as all other services only switched once, so if some services are not as important then you should start with this server.
The method assumes that you don't want services to run on a particular node - if you want services to remain where they started you would need to swich services to another node, patch, then swich services back to same node and then repeat this for the other 4 nodes so each service has 2 outages (this is bad practice as you have less Availabilty, so really it shouldn't matter where services run).
Mike
06-15-2012 03:24 AM
Hi Mike ,
Thanks for your detailed explaination.
<<<<<<The method assumes that you don't want services to run on a particular node - if you want services to remain where they started you would need to swich services to another node, patch, then swich services back to same node and then repeat this for the other 4 nodes so each service has 2 outages (this is bad practice as you have less Availabilty, so really it shouldn't matter where services run).>>>>>>
Unfortunately in our 5 node cluster we have diffrent configuration(RAM cpu) on diffrent nodes.so in our setup service groups need to be remain where they are.So we would be having two outages
06-15-2012 05:56 AM
As aside, if you are not already, I would recommend using load based failover rather than default of priority - see section "Service group workload management" in the VCS user guide. With this you can assign a "capacity" for each server which can be based on CPU or RAM (or some algorithm of the 2 combined). Then you set a load for each service group which represents how much capacity it takes from the server and then if you set FailOverPolicy service group attribute to "load", then VCS will choose the best systems (those with the most available capacity) to fail too.
Mike
06-18-2012 10:51 AM
Hi Mike ,
To the extension to my Patching questions.say if we get a downtime for patching .The approach then would be shuting down the cluster on the nodes where the downtime is provided and implemet patching ?
Regards,
siva
06-18-2012 01:22 PM
Yes.
Mike