[Python-Dev] February bug day outcome

[Neal Norwitz]

On Mon, Feb 25, 2008 at 12:53 PM, Barry Warsaw <barry at python.org> wrote:
>
>  That's a really good idea.  At least for the alpha's I would like to
>  have a policy that the monthly release goes out unless
>
>  1) There are critical bugs open for 2.6 and/or 3.0
>  2) The important buildbots are red (maybe)
>
>  In either case, it should probably be a priority for bug day to repair
>  those failures.
>
>  On #2, clearly we can't wait for greens across the board, so which
>  platforms are important enough to hold up a monthly release?  I'd say
>  something representative of the source release and each of the binary
>  releases we make, so maybe:
>
>  2.6: source = x86 gentoo trunk, windows = x86 xp-4 trunk, mac = g4 osx.
>  4 trunk
>  3.0: source = x86 gentoo 3.0, windows = x86 xp-4 3.0, mac g4 osx.4 3.0
>
>  Unfortunately, basing release status on buildbot health doesn't look
>  very encouraging. :(

It's been pretty bad the last month or so.  Although it's getting
better now.  I would recommend these are the golden bots based on what
have traditionally been fairly stable help expose errors:

  sparc solaris10
  amd64 gentoo (this is really an ubuntu 6.10 box)
  x86 gentoo (*)
  g4 os x (this one has svn problems from time to time which is odd
given that it is the only box colocated with the svn server)
  some win xp box (#4 wfm, i think that has been the most stable recently)
  ia64 ubuntu
  ppc debian (this may be ubuntu also)
  ppc64 debian (ditto)

The biggest challenge will be having svn work on all the machines.
The tests are getting more stable.  I worked on many of them.  There
are still issues from time to time, but at this point I think more are
caused by bad checkins.  Sometimes these machines go away, if they are
unavailable at time of release, so be it.

If we can get more people watching the buildbots and ping those
responsible for a failure, we can keep the red to a minimum.  If we
can fix the ~5 flaky tests, we will be in good shape.  If we can fix
the svn problems, we'll be in great shape.  Nearly all of the flaky
tests are due to networking problems.  Sometimes transient conditions
like the host is unavailable, others due to races.

(*) x86 gentoo should not be used for 3.0.  There is a problem with
signal 32 that causes it to rarely work.  I don't know the cause or
how to fix this.

n