How to prepare for UAT of mcsg?

[zhaogui]

As end user we are praparing for a UAT(User Acceptance Test) of MCSG and we are told that a large amount of testing is necessary, prior to commisioning, to correctly determine when a failover should occur and configure MCSG. This involves things like turning off the power while a node is running, turning off the network, disabling cards, disabling disks, turning off cpu's etc. This is the only way to make it reliable.

But I was wondering a few things such as:
1) how to disable the disk while the package is still running?
2) turning off the power while a node/package is running may corrupt file system and cause data corruption, is that too risky?
3) How to disablie FC cards? what kind of effect will we have on the system? Will it cause MCSG hang or trigger TOC?
4) how to turn off cpu's ?
5) how to test disk mirroring/RAID 5? Unplug one of disks?

Any comments/suggestions before UAT?

Thanks,

[James R. Ferguson]

Hi:

Hava a look at this guideline from the Technical Knowledge Base (#UXSGLVKBAN00000470). I've attached a copy for your convenience.

Regards!

...JRF...

[Steven Sim Kok Leong]

Hi,

Below is just my personal opinion.

You must ALWAYS prepare for the unexpected:
1) ALWAYS have your make_recovery tapes ready.
2) ALWAYS have your latest tape backup of your user DATA ready.

>> 1) how to disable the disk while the package is still running?

Pull out a hot-swappable disk.

2) turning off the power while a node/package is running may corrupt file system and cause data corruption, is that too risky?

You have to similate a power failure, risky or not risky. That's the whole purpose of getting a HA cluster in the first place ie. to reduce risk.

3) How to disablie FC cards? what kind of effect will we have on the system? Will it cause MCSG hang or trigger TOC?

Use lanadmin.

4) how to turn off cpu's ?

Power off the entire system.

5) how to test disk mirroring/RAID 5? Unplug one of disks?

Yes. Note that RAID 5 takes a longer time to rebuild than RAID 1. Thus performance may be degraded for a period of time for RAID 5. If performance during the rebuild is a concern, then you would want to note this.

Hope this helps. Regards.

Steven Sim Kok Leong

[zhaogui]

Here FC cards refers to Fibre Channel Mass Storage Adapter which is similar to Fast-wide SCSI card. I have read some comments by other people saying that if you disconnect all SCSI cables, MCSG will just hang there. So again how to disable SCSI cards, have you ever done that?

So lanadmin definetely cannot be used here, it is meant for LAN cards. As for LAN cards, I believe ifconfig lanx ip_address down also cannot be used to disable tha LAN card. I have tried to unplug all the LAN cables and it seems work.

[Steven Sim Kok Leong · Accepted solution]

Hi,

I had mistaken you to be referring to FDDI cards in my previous reply. Guess I must start wearing spectacles 8-)

Well, for the FC adapters, you can also disconnect at the brocade switch ports (I believe you are using brocade)?

For LAN cards, if you wish to deconfigure an interface (remove its binding), use unplumb instead.

# ifconfig lan0 unplumb

However you cannot simulate NO_HW in ioscan with ifconfig. The hardware state will still show as CLAIMED. The only way to simulate a hardware failure is to pluck out the network interface card.

However, there is no need to go for extremes in your UAT tests. As mentioned, note that MC/ServiceGuard can only provide HA to a certain extent. It cannot provide HA for every single component in your system.

I think there are tons of varying failure states with all the varying interfacing system components (hardware, software and network) that you are unlikely to be able to simulate every failure state, and expect MC/ServiceGuard to provide HA for every single one of them.

Hope this helps. Regards.

Steven Sim Kok Leong

[zhaogui]

Thanks.