steps to test out a new SG cluster

[Jim Smith]

Hi,

were comissioning a new 2 node K class cluster with EMC's running SRDF across buildings. What are the tests we should do to ensure its all working ?

[Stefan Farrelly]

These are some of the tests your should do regarding your EMC setup;

1. Break the SRDF link between your EMC frames. Do your 2 nodes keep running ?
2. Failover node A to B, break the SRDF link, does it keep running ?
3. Repeat failover but from node B to A, break SRDF Link, does it keep running ?
4. Switch off one of the EMC frames, do your nodes continue running ?
5. Try same when failed over.

Before more tests can be listed we need more info on yur SG setup;

a. are your 2 nodes running packages of their own or one is a standby for the other ?
b. What lan cards are they using and do you have backup lan cards ?

[Jim Smith]

a. Yes, each node is running its own package and plenty of resource to run both on the one node if needed.

b. 10Mb core i/o lan card for heartbeat between the 2 nodes, and fiber as main lan and a fiber as backup. all configured.

[James R. Ferguson · Accepted solution]

Hi Jim:

If you are also interested in some general MC/SVG testing guidelines too, these two Knowledge Base documents are good:

#UXSGLVKBAN00000470 and #UXSGKBAN00000471.

...JRF...

[Stefan Farrelly]

Right, now you need to test our all your lancards.

1. On each node test if the failover to the backup fiber card works by unplugging the primary one. Do the packages continue to run ? When both packages are on the same node repeat these tests.

2. If you unplug the lan cable from the heartbeat lan, what happens ? if both nodes are up and the heartbeat is unplugged do both nodes try to 'grab' the other nodes packages and import them and run them ? (dont want that to happen at the same time!) maybe you should run manual failover in this case.

3. If you unplug the primary fiber and run on the backup aok, then re-plug the main fiber does the package switch back ? then if you unplug the backup does it move back ? It should in order for the packages to continue running, although your users may temporarily lose their connections while you are mucking around with unplugging the lan connections.


What sore of SCSI/Fiber disk connections are you using ? you should do similar tests with these as long as all your disk devices are using alternate links across controllers. This will protect against cable or controller failure.