[buildd-tools-devel] Filesystem union v3

[Roger Leigh]

On Wed, Jul 01, 2009 at 05:57:14PM +0200, Jan-Marek Glogowski wrote:
> On Wed, 1 Jul 2009, Roger Leigh wrote:
> 
> > On Tue, Jun 30, 2009 at 08:05:38PM +0200, Jan-Marek Glogowski wrote:
> > >
> > > This version incorporates most changes from
> > >
> > > http://git.debian.org/?p=users/rleigh/schroot.git;a=shortlog;h=refs/heads/fs-union
> > >
> > > It's based on my v2 patch-set. The patches from 02 to 08 just got a little
> > > cleanup based on the comments from Roger Leigh at the beginning of this month.
> >
> > Thanks for the new patches, but I'm afraid I just finished my work of
> > merging most of the changes in your v2 patch set with my changes in the
> > fs-union branch.  As a result, I'm afraid they can't be applied as-is,
> > since the bulk of the changes are now present on the master branch.  I
> > literally *just* finished three days of merging work when your patches
> > arrived!
> 
> Seems we did the same work the last days :-( Especially splitting and
> merging the "rework union" and "sync master" patch took some time.

OK, I think we are almost done.

Over the weekend, I've finished the last remaining obvious bugfixing
and cleanup.  There were some subtle bugs in the chroot session code
which are now hopefully fixed, and also in some of the setup scripts.
This required further refactoring of the interface (chroot_plain
split into chroot_directory_base to allow correct dynamic_cast
failure for chroot_directory)

I've checked with directory and lvm_snapshot chroot types that it's
working to at least run with automatic session::OPERATION_AUTOMATIC
operation (i.e. no session persistence).

Could you check that the union support is functioning correctly, and
let me know if there are any further changes needed to get unions
working?  [For some reason my 2.6.30 kernel is oopsing on me when I
insert aufs.ko; not sure if it's directly related since I've since
had an oops without it installed.  This is hindering my own testing.]


The testing has exposed one current weakness in our testsuite: we
don't test the keyfile serialisation and deserialisation logic
before and after starting a session, so critical keys might be
missing that will mean the session can't be restored or ended.
We really need to add these tests in a similar manner to the
environment checks (observed and expected key-value maps).


Regards,
Roger

-- 
  .''`.  Roger Leigh
 : :' :  Debian GNU/Linux             http://people.debian.org/~rleigh/
 `. `'   Printing on GNU/Linux?       http://gutenprint.sourceforge.net/
   `-    GPG Public Key: 0x25BFB848   Please GPG sign your mail.