Yeah I've seen that as well, but it depends on when livecd-creator fails. Generally if it fails in a totally unexpected way (like /dev/urandom being missing) it won't clean up. But if it fails in an expected way, one that has been coded for, then it will.

I'll talk with the livecd-tools upstream maintainers to find out if they know more about the issue and are working on it. If active research/development isn't happening in that space I'll contribute a patch to Koji that forcibly gets it to clean up. It's not an easy problem though: I'm not aware of a good way to identify which mounts are specific to a given task. Do you have any ideas?

- Jay

----- "Doug Reiland" <dreiland@gmail.com> wrote:
> OK, thanks.
> FYI, I noticed the loop devices weren't being cleaned up when it
> fails. I noticed because I eventually ran out.
> I checked after my successful run and everything was OK there.
>
>
> On Thu, Mar 11, 2010 at 10:38 AM, Jay Greguske <jgregusk@redhat.com> wrote:
> > They are bugs, and a patch was submitted upstream not long ago:
> >
> > https://fedorahosted.org/koji/changeset?old_path=builder&old=3deb8358e67ff901f147900c2b230d5e0f3a8e0b&new_path=builder&new=06a885fe47f10742c0616799bcee727f93628a88
> >
> > You'll also want to move the broot.expire() later in the task too. There is
> > a race condition where the output is getting cleaned up before it is moved
> > to a final resting place. The patch I sited above fixes that too.
> >
> > This code hasn't seen too much exposure, so don't hesitate to ask questions
> > here or to me directly. I'm hanging out on Freenode IRC as jgreguske in
> > #koji or #fedora-devel as well.
> >
> > Thanks for checking in!
> >
> > - Jay
> >
> >
> > ----- "Doug Reiland" <dreiland@gmail.com> wrote:
> >> I got  a minimal livecd to build (not tested yet) but a to make some
> >> kojid changes. Are these really bugs or should be handled elsewhere?
> >>
> >> I had to include '/dev/shm' in bind_opts
> >> Also, livecd-creator (imgcreator python package) was calling
> >> os.random() and that wanted '/dev/urandom', so I had to a mknod for
> >> that where loop? nodes are created.
> >>
> >> Also, I had to include device-mapper in the livecd-build group.
> >>
> >> Doug
> >> --
> >> buildsys mailing list
> >> buildsys@lists.fedoraproject.org
> >> https://admin.fedoraproject.org/mailman/listinfo/buildsys
> >>
> > --
> > buildsys mailing list
> > buildsys@lists.fedoraproject.org
> > https://admin.fedoraproject.org/mailman/listinfo/buildsys
> >
>