Re: [rdiff-backup-users] atomic increment files?

[Marcel (Felix) Giannelia]

Andrew Ferguson wrote:
> Whoops, you're right, I was thinking backwards.
You mean you were thinking forwards? :)  I've found it a bit hard to 
think about the way rdiff-backup works sometimes, even though I really 
like the idea of it -- backwards-facing incremental backups are indeed cool.
> But, yeah, rdiff-backup stores a snapshot file after every ~10 
> increments, so that it doesn't have to apply more than 10 increments 
> during the restore. Also, that is helpful for recovery if an increment 
> gets corrupted.
That's interesting -- and it means that rdiff-backup is a lot less 
dependent on the chain of increments being intact, than I thought.
> Plus, I do want to see a merge increments operation one day. I feel 
> like keeping the increments as distinct files makes that easier.
If this operation were used to make backups sparser the farther back in 
time you go (which seems to be what people on the wiki are after), and 
rdiff-backup already keeps snapshots every 10 increments or so, maybe 
just discard the ones that aren't snapshots? And then you might be able 
to shrink those snapshots by in turn diff'ing some of them with each 
other, as in:
Before consolidation: [current mirror] <10 diffs> [snapshot] <10 diffs> 
[snapshot] <10 diffs> <snapshot> <10 diffs>
becomes: [current mirror] <10 diffs> [snapshot] [snapshot] [snapshot]
and then: [current mirror] <10 diffs> [snapshot] <2 diffs>

So that'd leave you with the last 10 days' worth of backups, then one 
from 20, 30, and 40 days ago.

~Felix.