Hard drive imaging
in [UK Linux]
|
Prev: eeepc+wifi+Wetherspoon
Next: Breaking into Ubuntu
From: Ian Rawlings on 28 Jun 2008 14:44 On 2008-06-28, Nix <nix-razor-pit(a)esperi.org.uk> wrote: > I wouldn't go back to non-LVM systems unless forced at gunpoint. The > only non-boot partitions I've got are containers for md arrays which > then have LVM PVs inside them. Hmm, not sure about using LVM for raid systems, if you use RAID then use LVM to build it on, you can end up breaking journalled filesystems so you get a nice RAID filesystem that can have its metadata destroyed by an unfortunately timed power-down. IIRC there are times when LVM does not prevent journalling from happening, I believe when a LV only consists of one physical disc, but don't quote me on that. -- Blast off and strike the evil Bydo empire! http://youtube.com/user/tarcus69 http://www.flickr.com/photos/tarcus/sets/
From: Andy Burns on 28 Jun 2008 15:06 On 28/06/2008 19:44, Ian Rawlings wrote: > Hmm, not sure about using LVM for raid systems, if you use RAID then > use LVM to build it on, you can end up breaking journalled filesystems > so you get a nice RAID filesystem that can have its metadata destroyed > by an unfortunately timed power-down. IIRC there are times when LVM > does not prevent journalling from happening, I believe when a LV only > consists of one physical disc, but don't quote me on that. LVM doesn't support write-barriers, XFS on top of LVM notices this and disables its own write-barriers, I don't know what ext3 does for its journals, just write and hope I guess? That said, I've used LVM on top of various RAID0/1/5 md raid volumes and never met any problems.
From: Ian Rawlings on 28 Jun 2008 15:45 On 2008-06-28, Andy Burns <usenet.april2008(a)adslpipe.co.uk> wrote: > LVM doesn't support write-barriers, XFS on top of LVM notices this and > disables its own write-barriers, I don't know what ext3 does for its > journals, just write and hope I guess? LVM *does* support write barriers, however not when there are multiple "entities" involved, whether this involves multiple real things like discs, or virtual things like partitions, I don't remember for sure. From what I do remember, translating write barriers for one device into write barriers for multiple devices is the stumbling block, but when the LV is mapped to one device then it's a 1:1 mapping so it's supported. If someone is using LVM to stripe two discs together then that's not going to support write barriers. All the above from memory, and also the changes may have been very recent. > That said, I've used LVM on top of various RAID0/1/5 md raid volumes and > never met any problems. Sure but I've never used RAID 5 and have never met any problems, doesn't mean RAID 5's not a good idea ;-) -- Blast off and strike the evil Bydo empire! http://youtube.com/user/tarcus69 http://www.flickr.com/photos/tarcus/sets/
From: Nix on 29 Jun 2008 09:04
On 28 Jun 2008, Andy Burns verbalised: > On 28/06/2008 19:44, Ian Rawlings wrote: > >> Hmm, not sure about using LVM for raid systems, if you use RAID then >> use LVM to build it on, you can end up breaking journalled filesystems >> so you get a nice RAID filesystem that can have its metadata destroyed >> by an unfortunately timed power-down. IIRC there are times when LVM >> does not prevent journalling from happening, I believe when a LV only >> consists of one physical disc, but don't quote me on that. > > LVM doesn't support write-barriers, XFS on top of LVM notices this and md's RAID-5/RAID-6 also doesn't support write barriers, so we have a perfect storm of lack of safety. :) > disables its own write-barriers, I don't know what ext3 does for its > journals, just write and hope I guess? Write and sync, unless turned off (I've turned it off, it ruins performance and is minimally useful). > That said, I've used LVM on top of various RAID0/1/5 md raid volumes > and never met any problems. Same here. To be blunt if I suffer a power failure or big kernel panic, the difference between `oops you lost this stuff because the journal wasn't synced and we didn't support barriers' and `oops you lost this stuff because the kupdated hadn't flushed it to disk at all' is kind of academic, and damaging performance to fix *part* of that is silly. This is the sort of situation where I'd fsck anyway, and it's really rare, on my systerms at least... -- `If you are having a "ua luea luea le ua le" kind of day, I can only assume that you are doing no work due [to] incapacitating nausea caused by numerous lazy demons.' --- Frossie |