openoffice-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Damjan Jovanovic <dam...@apache.org>
Subject Re: Let's fix the Windows build bots
Date Thu, 04 Feb 2016 18:11:34 GMT
On Thu, Feb 4, 2016 at 7:42 PM, Kay Schenk <kay.schenk@gmail.com> wrote:

>
> On 02/04/2016 06:05 AM, Damjan Jovanovic wrote:
> > On Wed, Feb 3, 2016 at 1:48 PM, Jochen Nitschke <j.nitschke@ok.de>
> wrote:
> >
> >> On Wed, 3 Feb 2016 13:15:23 +0200 Damjan Jovanovic wrote:
> >>> So builds@ replied saying the buildbot was broken by an earlier bad
> >> commit
> >>> (not mine) and should be working now, and it was, but I ended up
> needing
> >> to
> >>> commit another patch (cydrive -> cygdrive typo in the path to "svn
> >> info").
> >>> That 1 byte patch cannot be wrong, yet the buildbot isn't using it now,
> >>> despite committing it hours ago.
> >>>
> >>> The website changes also haven't taken.
> >> but hey, its progress :-)
> >>
> >> could you follow the flow?
> >> commit
> >> wait 5 mins
> >> pull changes
> >> buildbot checkconfig  (on your local config files?)
> >> buildbot reconfig (wouldn't know how)
> >>
> >>
> > After mailing builds@ and infrastructure@, I ran into other IRC users on
> > #asftest, one of which (pono) helped, restarting the buildmaster. That
> got
> > it using the latest commits, which finally fixed "svn info" on aoo-win7
> :-).
> >
> > I then made further commits with the same "svn info" improvement and
> > bootstrap improvement for aoo-w7snap, and another on aoo-win7 to use "svn
> > export" instead of rsync to generate the build directory to test the
> theory
> > that the use of rsync is what breaks apr later (
> > https://issues.apache.org/jira/browse/INFRA-10481), but the commits
> weren't
> > going through again. Pono investigated, eventually finding the buildbot
> > configuration was up to date, but the buildbot wasn't using it; by
> > reloading the config (not sure how) (which caused the ooo-bot to
> > temporarily disconnect from IRC) it started using them again.
> >
> > My aoo-w7snap fixes went through and "svn info" is now working there too,
> > but my aoo-win7 changes were less successful: my new svn export command
> was
> > preceded by "rm -rf build" which fails in the infamous apr module:
> >
> > rm: cannot remove `build/ext_libraries/apr/
> > wntmsci12.pro/misc/build/apr-1.4.5/Makefile.win': Device or resource
> busy
> > (
> >
> https://ci.apache.org/builders/aoo-win7/builds/173/steps/svn%20export/logs/stdio
> > )
> >
> > "Device or resource busy" is the same error that later breaks building
> apr,
> > and they seem related. I asked pono to look into what has that file
> opened
> > or locked and am waiting for that and other maintenance on aoo-win7.
> >
>
> OK, great. Yes, this locking/resource busy business continues to be
> a problem with this buildbot.
>
>
And the aoo-w7snap buildbot.


> Thank you SO much for trudging on with this. I see you also put a
> custom remove in for the linux-32 nightly buildbot with its own
> custom timeout so hopefully we can get past this snag as well.
>
>
Pleasure. Yes, the SVN() buildbot command doesn't apply its timeout to the
subcommands it runs internally, so its "rm -rf" only gets 120 seconds which
isn't long enough. The openoffice-linux64-nightly has a similar timeout
issue with "cp" every now and then, which I am not sure how to fix. Maybe
we should just do svn checkouts the aoo-win7 way: running "svn co"
ourselves?


> Did pono have any words of wisdom regarding how to make our commits
> actually *happen*, or do we need to request a restart of specific
> buildmaster each time?
>

No, but after I committed my openoffice-linux32-nightly change and started
a rebuild of that buildbot (5 minutes and 18 seconds later), that change
was being used, so I guess that's working now.

Mime
  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message