Split Brain prevention...

[Andrea_Colombo]

Hi all: I work on VCS4.0 MP1 running on AiX (2 nodes).
In the last month my clusters lost all the HB and public network for a couple of times causing a split brain with DB corruption!! (our network architecture is not so good...)
Is it possible to prevent such a split brain causing one of the 2 nodes to crash immediately after the 2 HB lost?
Otherwise do you know some other suggestion to prevent data corruption? Is it possible to prevent the VCS to launch a scsi reset making it fail when trying to mount the volumes?

thanks ANDREA

[Gene_Henriksen]

Check out the preonline_ipc trigger in the /opt/VRTSvcs/bin/sample_triggers directory. Copy this to the triggers directory, rename it preonline. Then turn on the attribute for preonline in the service groups. You can read thru the trigger, essentially it looks to see if there is an ip address in the SG, if so it pings for it, if it is up somewhere, SG does not come up. If it does not get a response, it checks to see if the system is isolated. If it is isolated (NICs down) then it does not bring up the SG.

The only option for panic that I know of is when you reconnect after an event like you experience, but that is too late.

[Carlos_Santos_T]

Are you using the VxVM? you can configure the I/O Fencing.

T+