We have a pre-UAT environment comprising of 4 zones on a Sun V490 with a cheapo 2 Tb NAS. The zones comprise of Two App servers, 1 Oracle DB server and a web server.

The client doesn't want to spend any money on expensive software, but they want the ability to do live snapshots of the envrionment.

I'm proposing using Solaris 10's built in ffsnap and ufsdump tools. Anybody know of any problems with this when it comes to doing an offline restore of everything (my worry is mainly Oracle db consistency).

Cheers

Can it do the snapshot while the file system is in use.

Also the cheap 2TB NAS, is it RAID and if so don't forget that you need to flush the RAID cache to disk before you try and do any snapshots.

Since they are on Sun anyway why not try using ZFS as they then get an effective rollback, or do they specifically want snapshot and dump?

Also which discs are they trying to snapshot, the NAS, the servers local drives or the Oracle file system?

The system has been in and running since may last year. So using ZFS is a no go. The ffsnap creates a virtual device on the host OS for ufsdump to backup whilst the zones are running. The snapshot will be a consistent "grab" of all four zones in one go, pre new application deployments.

No need to flush RAID cache on NAS (V490 mounts to this via an NFS share). The NAS is being used as a holder for the 90gig snapshot/ufsdump files.

fssnap is a software level built in Solaris 10 tool to do backups of mounted filesystems.

The system has been in and running since may last year. So using ZFS is a no go. The ffsnap creates a virtual device on the host OS for ufsdump to backup whilst the zones are running. The snapshot will be a consistent "grab" of all four zones in one go, pre new application deployments.

No need to flush RAID cache on NAS (V490 mounts to this via an NFS share). The NAS is being used as a holder for the 90gig snapshot/ufsdump files.

So you are dumping from the NAS device but or to the NAS device.or from it to another chunk on it?

Mounting the item as an NFS share won't guarantee safety and I'd call a file system commit before taking your image to be 100% sure if this is being used for backup purposes.

(my worry is mainly Oracle db consistency).

Cheers

I'm no Oracle expert, but I think you are right to be concerned. If the DB is in use while the filesystem snapshot is taken, you are unlikely to be able to keep the DB consistent. Certainly you can't guarantee it.

Are there any oracle tools that can mirror the DB in realtime?

Or is there a tool that can pause the DB? How long does the snapshot take? I'd guess transferring 2TB of data, even over gigabit ethernet (at best, probably) will take some time.

I just think that if you want live snapshots you have to be prepared to pay for the software and/or hardware to do it.

The NAS is just being used as a replacement for a tape drive. The zones (this is the important bit --- read solaris virtual machines) run on the same box sharing the same local disks. The zones are mounted on /data so I'd just run

fssnap -o /data1 /data

then

ufsdump 0uf /dev/fssnap/0 /nfs share on NAS/backup.file

This provides a consitent grab of all 4 zones.

Just tested now and works fine. So answered my own questions

Cheers

WHOOOOOOOOOOOOOOOOOOSH

that was the sound of this thread going over my head :P

Just tested now and works fine.

Personally I would have some concern over Oracle's consistency doing a fssnap on a running DB - while you'll probably be lucky in most cases (especially if pre-UAT means it's not overly busy) and it won't encounter any conditions that it can't auto-recover from, there is a risk of inconsistency.

If you can accept this risk and are not too *rsed about having to restore from an Oracle dump if it can't recover (and the corresponding downtime and data loss), I wouldn't too much sleep over it.

If you need consistency, I'd be inclined to stop the database while doing the fssnap, although I guess that depends how long the snap takes (can't imagine it would be that long on a V490).

Rob.

I'm no Oracle expert, but I think you are right to be concerned. If the DB is in use while the filesystem snapshot is taken, you are unlikely to be able to keep the DB consistent. Certainly you can't guarantee it.

This is certainly the case with Ingres on Solaris, if you backup the databases using ufsdump and then restore them, they are flagged as inconsistant and need to also have a checkpoint rolled back to get them running again.

Your dump/snapshot may have worked, but you need to fully test the backup, chances are your oracle database will be inconsistant and need some kind of additional restore to get it to behave correctly.

My worries are also with the fact that it sounds like this is for disaster recovery.

If that is the case I'd get up some stats showing what happens to a company when they lose their data and put in writing that this is not 100% as mentioned the database might not be consistent, but then the FS might not be too.

Forgot to mention I put the db into hotbackup mode via RMAN during the script.. I'm on my 4th auto restore test now and all is well.... with various levels of load on the system whilst doing backups.

The snapshots are purely so we can quickly roll back deployments of new code if UAT fails.

Always a problem when doing agile development.... is the time for deployments and rollbacks.

Use ZFS next time