yellow team mailing list archive

[Gary Poster <gary.poster@xxxxxxxxxxxxx>]

(copying Francis because of xvfb discussion)

Hi all.  Benji has branches of everything that start the buildbot slave 
and master and run the launchpad tests.  yay!  we'll land this tomorrow 
as an incremental step.

It's incremental because it seems a number of systemic things are making 
many of the tests not work properly.

The one we are working on now, and that I'm going to write about here, 
is that xvfb does not start.

We traced this down so far to the fact that the following will succeed 
in a real lxc instance (where xvfb starts and works) and fails in an 
ephemeral instance.

cd /tmp
touch 1
chmod 0444 1
ln 1 2

That mirrors something that xvfb is trying to do.

I'm suspicious of overlayfs.  Tomorrow our plan is this:

- try to dupe the problem by mounting an overlayfs (without lxc).  File 
a bug in kernel or lxc depending on the results and escalate.

- Look for temporary workarounds.  We have at least one crazy idea in 
the wings.

Francis, here's your cue. :-) I think that running the browser tests is 
an essential aspect of success, and we should pursue it now so that we 
have a better chance of whatever fixes we need going into Precise.  Do 
you agree, or should we dump xvfb and circle around once we have the 
rest of the suite running?

(Of course, if anyone else has an opinion, please weigh in.)

Thanks

Gary