Re: Journalling idea

[Oliver Fromme <check+in66b200rs58nllq@xxxxxxxxxx>]

Emiel Kollof wrote:
 > Gabriel Ambuehl wrote:
 > > Oliver Fromme wrote:
 > > > The result would be that the backup filesystem represented
 > > > the state of the original filesystem as of one hour earlier.
 > > > It would be a "moving snapshot", so to speak, which could
 > > > be useful for file recovery (accidental rm) and debugging.
 > > > 
 > > > Would that work, or am I dreaming?
 > > 
 > > As far as I understand, that would work easily. However, it might not
 > > even be needed once UNDO is working (that's right, there's plan for TWO
 > > way journalling ;)

I'm aware that Matt has implemented undo records in the
journal stream.  But how would a user restore a single
file (or an entire directory tree) that existed one hour
before?

In the "delayed backup" scenario that I described, the
user would simply type ``cp -p /backup/home/foo/file .''
or ``cpdup /backup/home/foo/directory .'', respectively.

 > I believe Solaris has (or had) something like this. I used to work at SARA, 
 > where everyone had a .snapshot dir in their $HOME [...]

Yes, NetApp filers also support that natively.  But that's
not what I meant.

 > I believe something similar can be built using FreeBSD 5 and up with the UFS2 
 > snapshotting.

Well, in theory, yes.  You could write an hourly cronjob
which generates and dismisses snapshots, similar to what
you descriebd above under Solaris or what NetApp does.
(However, FreeBSD's snapshot feature has disadvantages,
particularly it is much to slow on large file systems.)

But that's not what I meant.  I wasn't thinking about real
snapshots.  Snapshots don't change, once they're created.
My idea was to apply the journal stream with a delay, so
the file system would be "live" and follow the original
filesystem with a given delay.  Calling it a "moving snap-
shot" was probably inappropriate; sorry for the confusion.

Maybe call it a "filesystem time machine".  ;-)

Best regards
   Oliver

--