return to PRS Technologies website
I need to restore an Informix mirror
(under HP/UX).
HP has a feature called "predictive index" that attempts to predict problems on the system.
This program alerted us to disk reporting non-terminal EEC errors, so HP wanted to replace the disk.
This disk happened to be used as an Informix mirror for a chunk.
Informix mirroring isn't used much these days. There are better solutions to mirroring, such as RAID systems, hardware controller mirroring or HP mirror software built into the OS. But this was an older system, configured when those solutions were not as popular, so Informix mirroring had been configured.
As a result of the above, plus the reliability of HP equipment in general, I know in theory how to restore a mirrored disk, but I hadn't actually ever really DONE it. So, this is how it went. It was a breeze!
Note: this disk hadn't offically failed (it was still marked as UP by Informix), but I expect to
wouldn't be much different if it had been marked down to begin with.
- First, we did a full Informix backup and ran HP's vgcfgbackup to make sure that the latest LVM
(logical volume manager) info was saved in the /etc/lvmconf/VGXX.conf file. In this case, the chunk to be restored resided in /dev/vg03 volume group.
- We changed the startup procedure so that Informix would not automatically come up at startup.
Then we then did an orderly shutdown of the Informix engine and Unix.
HP physically replaced the suspect disk with a new one and we restarted the machine.
- The manual recommends to boot Unix into single user mode to insure that the disk is not being used.
We didn't do this, but it's probably a good idea. We logged on as 'root' at the console.
- First, we needed to restore the HP LVM info to the disk from the LVM backup file.
- vgcfgrestore -n /dev/vg03 /dev/rdsk/c0txd0
- Then we need to reactivate the volume group since it wasn't done at boot time for this disk.
- Note: if you happened to be using HP Mirroring, you would need to re-sync the disk. This might 30 minutes to hours based on size. Once this is started, you could go ahead with any remaining steps while this is processing.
- That's it for the HP part. Then we started the Informix part.
- We started up the Informix engine.
Informix issued assert errors and marked the chunk down as expected.
We simply issued the 'onspaces' command to restore the mirrored chunk.
- onspaces -s dbs_name -p /dev/vg03/rmirror_chunk_name -o 0 -O -y
- This took about 15 minutes for a 1-gig chunk.
- Like I said, it was a breeze. Everything worked perfectly and we were back up in no time.