Heartbeat with dual SCSI config

Ragnar Kjørstad linux-ha at ragnark.vestdata.no
Wed Mar 6 15:46:47 MST 2002


On Wed, Mar 06, 2002 at 11:56:31AM -0700, Alan Robertson wrote:
> 	However perhaps it is legal to switch on an already connected
> 	device. It is perhaps not guaranteed this device doesn't corrupt
> 	an ongoing data transfer.

Yes - avoid active devices and everything should be fine. And even if it
says it is a "beta feature" it has been there for ages, so I think it
should have the worst bugs washed out by now.

However, it may not work the way you intend to! Specificly, I think the
device-names can be different after disabling and re-enabling a device.

Example 1:
* devices sda, sdb and sdc. Disable the device know as sdb and enable it
  again. I think it will show up as sdd.

Example 2:
* devices sda, sdb and sdc. Disable sdb and sdc. Then enable sdc and
  sdb, and I believe they will switch names. 

NB! I'm not saying this is the way it will happen - I'm just saying this
is something you would want to test first, before relaying on the
feature.


Second, having a shared scsi-bus between the two servers kind of defeats
the purpose of ha-clustering. The disk and the scsi-channels are both
single points of failure. The scsi bus may infact be dependent on _both_
servers working correctly (depending on how termination is done)! In
addition there is the added complexity of the failover software. 

IMHO if a problem is not worth HA hardware it's probably not worth HA
software either. Of course it's a great over-simplification and may not
apply to this particular problem, but I think it's important to point
out that ha-clustering isn't the answer to all problems.



-- 
Ragnar Kjørstad
Big Storage



More information about the Linux-HA mailing list