topic Re: Backing up a snapshot File System in Operating System - HP-UX

Backing up a snapshot File System

Kristofferson Fagan — Wed, 04 Feb 2004 11:59:32 GMT

I am fairly new to the Unix environment. Can someone please comfirm whether or not Fbackup can be used to perform backup within an OnlineJFS environment?

Thanks

Re: Backing up a snapshot File System

Sridhar Bhaskarla — Wed, 04 Feb 2004 12:02:14 GMT

Hi,

I guess you meant backing up a snapshot filesystem (available with OnlineJFS) with fbackup.

If so, yes. You can use fbackup. fbackup can span multiple media and supports incremental backup.

-Sri

Re: Backing up a snapshot File System

Bernhard Mueller — Wed, 04 Feb 2004 12:04:42 GMT

Hi,

if you refer to OnlineJFS "snapshot" mounts which you want to backup, then the answer is no, you cannot use fbackup because it needs write access and your OnlineJFS snapshot is a read-only filesystem.

So you could use tar, cpio, dd ... (which all have their restrictions, e.g. 2GB wise)

Regards,
Bernhard

Re: Backing up a snapshot File System

Steven E. Protter — Wed, 04 Feb 2004 12:06:46 GMT

fbackup can not easily be used to back up any open "database". These snapshots have processes open on them and will not back up well with fbackup.

If there is a way to tell OnlineJFS to make new snapshots you might be able to get the previous version. I'm new with OnlineJFS and have yet to figure that out.

SEP

Re: Backing up a snapshot File System

A. Clay Stephenson — Wed, 04 Feb 2004 12:43:32 GMT

Fbackup can certainly be used to backup snapshot filesystems "read-only" or not.

It is possible to make multiple snapshots at different points in time on a vxfs filesystem. Each snapshot will be current as of its snapshot time. There is, of course, additional overhead associated with each snapshot image so for busy filesystems I would not exceed one without knowing the impact.

Snapshots, with regard to databases, are not intended to used as hotbackups. The idea is to shutdown; snapshot; restart the database and then backup the snapshot. No files are then active and fbackup works like a charm. You get almost all the uptime of a hotbackup with all the safety of a cold backup for about two minutes of downtime.

One thing that is not obvious is that you should mirror (or have alternate paths if a RAID) the snapshot buffer lvol. If you suffer a disk failure in the snapshot buffer device, the original filesystem can hang --- not often considered a desirable thing. A typical generous size for a snapshot buffer is about 15% of the original.

Re: Backing up a snapshot File System

John Waller — Thu, 05 Feb 2004 04:42:56 GMT

A few years ago Bernhards answer would have been correct, it was the case that you could not use fbackup to backup a snapshot volume. Following the previous convincing answer I've re-tried and yes it now does work. I have to presume that a patch has been released which fixed the problem so depending on your patch level you may have a problem. Best advice is to manually run a small test using a single file after making a snapshot of a temporary created lvol. Unlike windows it is unlikely you will crash your system, at worst you will just get an error informing you that the backup has failed.

Re: Backing up a snapshot File System

Tim D Fulford — Sat, 07 Feb 2004 16:46:53 GMT

tar will do it, fbackup wont as it modifies the i-nodes (unless your snap LV is the same size as the data on the original imaged LV).

The way a snap works is that it freezes ALL activity on a filesystem. The filesystem is allowed to continue, but every time a file is modified the original i-node is copied to a reserved piece of space. Usually 10% is chosen. Thus you can back-up the original "frozen" filesystem. Unfortunately fbackup "touches" each i-node as it backs it up. Thus the reserved LV will eventually contain as much data as the original.

Tim

Re: Backing up a snapshot File System

A. Clay Stephenson — Sat, 07 Feb 2004 16:58:19 GMT

Nonsense again. Fbackup will certainly backup snapshots and because snapshots operate at the block level only changed blocks are transferred to the snapshot buffer. It is extremely unlikely that all the blocks in all the files will be modified (and only the FIRST block update counts), a 15% or so snapshot buffer is usually more than adequate.