A desktop kernel wishlist - Big folder copy
A desktop kernel wishlist - Big folder copy
Posted Nov 25, 2014 22:37 UTC (Tue) by nix (subscriber, #2304)In reply to: A desktop kernel wishlist - Big folder copy by blujay
Parent article: A desktop kernel wishlist
real-auto-save is... more than slightly risky. Not *everything* I touch is in git: I don't want absolutely everything really saved unless I ask for it.
I used obnam for a while, but its insistence on rolling its own data structures means that corruption is still a problem years on, and it's *fearfully* slow -- by which I mean that it took more than a week to back up a filesystem that took bup a day, and takes twenty hours or so to do the sort of routine backup run that I run every night (it takes bup an hour to do the filesystem tree walk and on the order of three minutes to do the backup).
I back up *everything* in bup. OS, virtual machine disk images (once a week, providing --smaller=500M to other runs), $HOME including things like build trees and object files from build runs, everything. The only stuff I exclude is things like DVD rips which I can just re-rip on disk failure and which never change nor benefit from deduplication.
The dedup really does save your bacon in a lot of areas: e.g. I can flip between git branches and completely ignore that this will lead to the working directory changing over and over again: the branches have been flipped to before, so they're already in the backup, so bup will deduplicate them almost entirely. Most of my virtual machine images disappear: they're neither compressed nor encrypted, so the binaries on them have largely been seen in other VM images, and vanish. Renames of huge filesystem trees just get deduplicated out of existence; huge swathes of inode changes caused by things like cp -al get almost entirely deduplicated (the ctime change is recorded, but the content gets deduplicated). And of course an ordinary cp -a won't use any more space on the backup either.
As for not being able to expire backups, y'know, that's not really a problem. As I mentioned above, even backing all that stuff up, I estimate that I'll run out of space on my backup device in ten to twelve *years*. By that point the disk will have died of old age -- and I rotate them every so often anyway for offsite backups, so it's no trouble wiping one and doing a new full backup every decade or so!
btw, expiry is being worked on: several promising approaches have been posted to the bup list of late. Personally I'd not expect you'd have to wait too long for it. I just don't think it's all that useful... your backups fill up so much more slowly than they ever did with rdiff-backup and the like.