MirrorDisk UX testing and failover

[Richard Cooper_1]

I'm working with a customer to help test MirrorDisk UX failover. I've been unable to find a list of error conditions that will cause a mirror to break and IO to be directed to one side. I can force the mirror to fail by taking one drive off-line. Can anyone help with details of SCSI error conditions (check conditions) that will cause failover? In particular, what should happen when a hard read error is reported by the disk?

[Ajitkumar Rane]

hi,

the best way to test is plug the disk out that is I guess the only error condtn u can simulate.
Basically the mirror will never break, it is just that one of the disks will become unavailable. If one disk completely blows out or like u plug it out the system will just flas mesg on ur console and the syslog tha one of the PVs is unavailable.
Next if you get any bad blocks on a disk the system will still continue to use bothe the disks until the stale extents reach a point that the PV will be considered unavaliable and it will bw show as unavaliable in the vgdisplay.
I have seen this a couple of days back that the system continues with the mount point like one disk has some extents stale the mirror disk had some other extents stale, and the filesystem could be used.

hope this helps

goodluck.

[KapilRaj]

Hi ,

Error conditions

01. Unplug the disk (if it is hotswappable or do it when the machine has brought down)

02. cat /stand/vmunix > /dev/rdsk/cXtXdX this will distroy LVM information on one of the disk . so mirroring will not happen

Hope this helps you

kaps

[Marcin Wicinski]

Hi,

Simple way to destroy LVM structures on a PV is to use just dd command.

Good luck.