Help with LVM snapshot scripts for Samba Shadow Copy support

[Geoff]

I need a hand here, my scripting skills are insufficent to pull this one off. However the end result is likely to benifit others in the future. But first, some background info.

I have recently been creating a new file server running Ubuntu 8.04 on a HP DL140G3 with an external MSA60 Disk Array. Obviously to service the Windows clients this is running Samba (with all the ACL, quota, etc bells and whistles).

The underlying file system for the user storage is RAID-5. This is then allocated with LVM2 into various partitions. These are formatted as JFS and mounted some place sensible (/home/samba/blah normally) and then shared via samba as required.

I wish to use LVM based snapshots to provide 'shadow copies' to Windows clients. To do this I need a suite of scripts to do the following:

1) create LVM snapshots on a regular basis of some LVM paritions. New snapshots must be mounted in the way sambas shadow copy VFS module expects.
2) remount LVM snapshots cleanly in the case of a server reboot/crash/etc.
2) delete LVM snapshots after a predetermined time and/or when disk space is low.

A couple of links with background info:

Chapter23.Stackable VFS modules
Samba Shadow Copy Howto - Waikato Linux Users Group
Samba and MS Shadow Copy

[Ric_]

IIRC AMANDA (see Amanda Open Source Backup ) pretty much does this for you... have you checked it out before re-inventing the wheel?

[Geoff]

I searched the site and only came up with a couple of articles mentioning snapshotting MySQL.

[Ric_]

I have been and had a look... since I last looked at it a couple of years ago, a commercial arm has broken off and the web-interface features that I thought could have been used to solve your problem are only in the paid-for version.

[dhicks]

Create LVM snapshots on a regular basis of some LVM paritions. New snapshots must be mounted in the way sambas shadow copy VFS module expects.

This is just the example script given in "Chapter 23: Stackable VFS Modules", run every day(?) from a cron job.

remount LVM snapshots cleanly in the case of a server reboot/crash/etc.

Just write a script to read all the subdirectories in your mount directory and mount them all, then run that script from your rc.local?

delete LVM snapshots after a predetermined time and/or when disk space is low.

Just figure out the modified time of each subdirectory, compare it with the current date, delete it if you need to. For disk space, read all subdirectories, order by date, delete the first, loop until you have enough space.

How do you mean you're having difficulty writing a script for the above? Does it have to be done as a bash script, can't you simply write it in Perl or Python or something?

--
David Hicks

[monkeyx]

I have a couple of test systems that use openfiler. This support iscsi, snapshots etc.

Not sure about support for Mysql though.

[ahuxham]

I have a couple of test systems that use openfiler. This support iscsi, snapshots etc.

Not sure about support for Mysql though.

Nice bit of software, thanks for the link!

[Geoff]

This is just the example script given in "Chapter 23: Stackable VFS Modules", run every day(?) from a cron job.

Well yes, that does work on a single LVM2 volume. However I do not have a single volume. I have 20. Of which 14 require shadow copy support.

Plus there's no error handling.

Just write a script to read all the subdirectories in your mount directory and mount them all, then run that script from your rc.local?

Er no, I have other volumes that are not on LVM2 (the raid1 arrays on the boot drive(s) for example) and I have LVM2 volumes that do not require shadow copy support. So I can't simply blindly assume every volume in /dev/disk/ or /dev/mapper/ is a LVM2 volume and/or snapshot.

So I need to be keeping track of which LVM2 volumes I wish to snapshot, and keep track of all their snapshots and the creation time of the snapshots.

Additionally, if the system has crashed, I need to fsck the volumes prior to mounting. Otherwise the mount fails.

Just figure out the modified time of each subdirectory, compare it with the current date, delete it if you need to. For disk space, read all subdirectories, order by date, delete the first, loop until you have enough space.

My scripting ability fails at this point.

How do you mean you're having difficulty writing a script for the above? Does it have to be done as a bash script, can't you simply write it in Perl or Python or something?

I think this problem is too complex for a bash script. I was thinking Python, but as I said, my skills are limited.

[dhicks]

Well yes, that does work on a single LVM2 volume. However I do not have a single volume. I have 20. Of which 14 require shadow copy support.

You could simply create a list of volumes you want to snapshot, or create a list of volumes you don't want to snapshot and have your script parse the output of "lvs" and skip the ones in that list.

Er no, I have other volumes that are not on LVM2 (the raid1 arrays on the boot drive(s) for example) and I have LVM2 volumes that do not require shadow copy support. So I can't simply blindly assume every volume in /dev/disk/ or /dev/mapper/ is a LVM2 volume and/or snapshot.

Sorry, I meant as in how that example script works - it makes a new directory to stick mount points in, so you can assume that all subdirectories in that directory can be mounted.

I think this problem is too complex for a bash script. I was thinking Python, but as I said, my skills are limited.

Darned if I can remember how to do loops in bash, anyhow. Wouldn't a Python script, very roughly, be something along the lines of:

#!/usr/bin/python
# Script to be run from a cron job, once a day, to create snapshots.
import os
import time
volumes = ["volOne", "volTwo", "volEtc"]
snapName = time.strftime("%Y.%m.%d-%H.%M.%S", time.gmtime())
for volume in volumes:
	os.system("xfs_freeze -f /data/shadow_share/")
	os.system("lvcreate -L10M -s -n "+volume+snapName+" /dev/shadowvol/sh_test")
	os.system("xfs_freeze -u /data/shadow_share/")
	os.system("mkdir /data/shadow_share/"+volume+snapName)
	os.system("mount /dev/shadowvol/"+volume+snapName+" /data/shadow_share/"+volume+snapName+" -onouuid,ro")

Might want to replace the above calls to os.system with a function that prints the commands out before actually running them, letting you see what the script is going to do before actually doing it.

--
David Hicks

[Geoff]

Well that's a start. I shall have a tinker and see how far I get. Thanks.

[apaton]

I'm might come across as a Solaris bigot . But if you want a filesystem that can do snapshots & CIFS/NFS/iSCSI support you must look at OpenSolaris ZFS.

The guys in the Hardware Forum rave on about Sun Unified Storage S7000 range, which is under the covers is basically OpenSolaris + BUI.

Snapshot are simple in ZFS and very efficient with space.
To access a snapshot just simply follow the hidden ".zfs" directory which access all the read only snapshots for the share.

EXAMPLE ONLY (Assuming zpool and Domain/Workgroup setup has been done)

# Create CIFS share

$ zfs create zpool/UserHomeDirs
$ zfs set sharesmb=on zpool/UserHomeDirs

Create snapshot

# Snapshot for Monday
$ zfs snapshot zpool/UserHomeDirs@UserHomeDirs-Monday

# Snapshot for Tuesday
$ zfs snapshot zpool/UserHomeDirs@UserHomeDirs-Tuesday

Create many CIFS shares and snapshot them in one command

$ zfs create zpool/cifs-shares
$ zfs set sharesmb=on zpool/cifs-shares
$ zfs create zpool/cifs-shares/share1
$ zfs create zpool/cifs-shares/share2
........
$ zfs create zpool/cifs-shares/share99

## Snapshot shares 1 to 99
$ zfs snapshot -r zpool/cifs-shares@cifs-shares-Monday

I can't do ZFS functionality justice but it well worth the time researching OpenSolaris & ZFS. A blog that may be helpfull. A Home File server using ZFS