Hello,
Just a few quick, but not very simple questions...:
Do any Linux filesystems (besides XFS) support freezing? (ie. in conjuction with LVM snapshots this can allow a mounted filesystem to be frozen [freezing all processes writing to this filesystem] in a valid state (with possibly dangling unlinked files), the device can be snapshotted via LVM, and the original filesystem unfrozen - the snapshot now contains a valid snapshot in time of the original filesystem - with no corruption.
Can this (freezing) be used for the root filesystem or other important filesystems? ie. might it cause something vitally important to freeze?
Can a snapshotted LVM (the snapshot, not the original filesystem) be modified (I assume snapshots are purely readonly...) - so as to remove the dangling deleted/unlinked files and possibly modify the uuid (I realize there is a nouuid mount option for XFS).
And the last question: is there some sort of filesystem which supports generating 'in-filesystem' snapshots (ie. some sort of copy-on-write block semantics applicable both to files and directories combined with both a root rw directory to access current files and a ro /snapshot/YYMMDD-HHMMSS hierarchy to access snapshots from past dates with the possibility of deleting them at will (ie. generate a snapshot every hour, after 24 hours leave only every 6th one, after 7 days leave only dailies, after a month leave only weeklies etc... - much too complicated, but it's easier to get the basic idea across on a larger example - probably organized via cronjob).
I realise such a solution would not be foolproof (if the disk goes bad or the filesystem goes haywire it's pretty much useless) so a hard-copy would still need to be maintained (probably via freeze/snapshot/unfreeze/dump it somewhere else) - but this would be useful for the users (the ability to restore past deleted files...)
In a way this could possibly be realized via some twidling with rsync (although for space efficiency this would require copy-on-write hardlinks or something like that) but I see no way of doing this without wasting a lot of space (at least 1 copy of the snapshoted filesystem, the rest could probably be hardlinked). And to maintain full R/O status, this would need to be a seperate partition, etc...
Any comments/ideas?
Cheers, MaZe.
Maciej Żenczykowski wrote:
And the last question: is there some sort of filesystem which supports generating 'in-filesystem' snapshots (ie. some sort of copy-on-write block semantics applicable both to files and directories combined with both a root rw directory to access current files and a ro /snapshot/YYMMDD-HHMMSS hierarchy to access snapshots from past dates with the possibility of deleting them at will (ie. generate a snapshot every hour, after 24 hours leave only every 6th one, after 7 days leave only dailies, after a month leave only weeklies etc... - much too complicated, but it's easier to get the basic idea across on a larger example - probably organized via cronjob).
WAFL does this, unfortunatly it only runs on NetApps.
-jim
On Mon, 2005-11-28 at 11:23 -0500, Jim Bartus wrote:
WAFL does this, unfortunatly it only runs on NetApps.
Give LVM2/DeviceMapper enough time on Linux (like 2 more years), and you'll see more and more features and capabilities.
But yes, you are correct. A generic OS will _never_ have the features like a filer-assuming, NVRAM, controller, etc... hardware-software design symbios with Data OnTap its WAFL filesystem.
The PC would need to have some radical changes to even compete, because the software very much requires certain hardware. Intel is now starting to put IOP33x (XScale) logic in the southbridge. If they ever pair that with battery-backed main memory in a semi-commodity design, then it might be possible for generic OSes to offer far more.
Until then, if you want absolutely consistent filesystems, snapshots and services, it's hard to beat NetApp because of the hardware-software design symbios. Generic systems and OSes can't match it.
On 11/24/05, Maciej Żenczykowski maze@cela.pl wrote:
Do any Linux filesystems (besides XFS) support freezing? (ie. in conjuction with LVM snapshots this can allow a mounted filesystem to be frozen [freezing all processes writing to this filesystem] in a valid state (with possibly dangling unlinked files), the device can be snapshotted via LVM, and the original filesystem unfrozen - the snapshot now contains a valid snapshot in time of the original filesystem - with no corruption.
I'm reasonably certain that LVM+ext3 automatically makes consistent snapshots (no freezing necessary). See http://www.tldp.org/HOWTO/LVM-HOWTO/snapshotintro.html for (slightly) more details.
Can a snapshotted LVM (the snapshot, not the original filesystem) be modified (I assume snapshots are purely readonly...) - so as to remove the dangling deleted/unlinked files and possibly modify the uuid (I realize there is a nouuid mount option for XFS).
Under LVM2, snapshots are read-write; however, LVM2 isn't currently reliable on CentOS 4 or the upstream release. (See the thread starting at http://marc.theaimsgroup.com/?l=centos&m=113155162316509 for details.)
And the last question: is there some sort of filesystem which supports generating 'in-filesystem' snapshots
It looks like there's a research project underway at http://www.ext3cow.com/, and there's the Wayback filesystem for FUSE at http://wayback.sourceforge.net/, but I've no idea how usable either is.
On Fri, Nov 25, 2005 at 01:27:14AM +0100, Maciej ?enczykowski wrote:
Hello,
Just a few quick, but not very simple questions...:
Do any Linux filesystems (besides XFS) support freezing? (ie. in conjuction with LVM snapshots this can allow a mounted filesystem to be frozen [freezing all processes writing to this filesystem] in a valid state (with possibly dangling
Not a linux filesystem, but sun's new ZFS filesystem has great snapshotting support, as well as a million other incredible features.
It's more than what we think of as a traditional filesystem - it incorporates software raid & volume management functionality as well.
danno
-
Bryan J. Smith -
Dan Pritts -
Jim Bartus -
Josh Kelley -
Maciej Żenczykowski