[Lustre-discuss] OST recovery

Mailer PH mail at powhosts.com
Wed Feb 20 05:38:32 PST 2008


Hi ,
Im trying to figure out what is the best way to recover a failed OST , basicly we have 10 OST's
each has DRBD + HA on top of raid 6 so its kind of redundent and suppose to be solid
just want to notice for the other post that asked of that configuration that its working ok and the
performance is fairly ok considering that redundency is more important then full speed of the cluster at list in this case .

Regarding the backup strategy , we make a client backups to tapes of all the important stuff
and also a seperate backup of the OST files only to a USB HD (daily on each OST)
that backup is made possible by mounting the OST with -t ldiskfs insted of lustre the by running rsync to the USB HD
so the main thing i dont understand is if an OST failed as in hardware problem then 
to avoid full file system recovery from tapes there is a need to restore the OST only data from the USB drive or tapes
to the new OST , then the lustre procedure e.g
e2fsck -n -v --mdsdb /tmp/ostdb /dev/{ostdev}
on all OST's then
lfsck -n -v ............. /mnt/mainfs
the only things i see possibly is to write zero holes on files that were changed for example seens the last backup 
of the OST file system itslef with rsync to the USB drive or tapes
so baicly what will happen to a mysql table file that has inconcitency on its tripes
how is possible to restore it the best way possible , i realize that it must suffer some kind of data lose
but its better then loading the entire lustre file system backup wich will take days is some cases .

Thanks for any help .


----------------------------------------------------------
Outgoing messages are virus free checked by NOD32 system 
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.lustre.org/pipermail/lustre-discuss-lustre.org/attachments/20080220/aadda6a0/attachment.htm>


More information about the lustre-discuss mailing list