[Buildd-tools-devel] Bug#477788: why not use the device-mapper?

Ferenc Wagner wferi at niif.hu
Tue Feb 24 19:33:28 UTC 2009


martin f krafft <madduck at debian.org> writes:

> also sprach Ferenc Wagner <wferi at niif.hu> [2009.02.24.1922 +0100]:
>
>> The idea is to augment the lvm-snapshot schema by creating a fresh
>> LV, appending it to the snapshot and resizing the filesystem on
>> the resulting block device.  Currently I abuse the chroot
>> description to carry the size of the extension, lacking anything
>> better.  So:
>
> I totally don't understand the point of what you are doing.

Then surely I didn't understand your problem with the lvm-snapshot
type :).  Anyway, others have the problem I'm attacking here, and I
wanted to share my thoughts, if that explains my over-eagerness.

The problem: tgz based chroots are slow.  There are two solutions to
this, AFAIK:

 * cowbuilder style LD_PRELOAD fun: this is neat, but have its quirks,
   even you called it "hackery", which would be best to avoid.

 * lvm-snapshot: clean, but the chroot source LVs have to be large
   enough to accomodate the build process, even though this space is
   always unused by the minimal chroot environments themselves.  Thus
   it potentially wastes a lot of space.

Avoiding this waste is the point of my patches.  They make possible to
leave the source chroot LVs small (like 300 MB) but automatically (and
ephemerally/virtually) extended to an arbitrary size when
"instantiated" for doing a build.  And it's still fast.
-- 
Regards,
Feri.





More information about the Buildd-tools-devel mailing list