On 5/16/2013 1:31 AM, Oliver-Rainer Wittmann wrote:
Hi,
On 16.05.2013 10:24, Oliver-Rainer Wittmann wrote:
Hi,
On 15.05.2013 21:51, Andrew Rist wrote:
On 5/15/2013 12:03 AM, Oliver-Rainer Wittmann wrote:
Hi,
On 14.05.2013 10:16, Oliver-Rainer Wittmann wrote:
Hi,
our buildbot building trunk nightly for windows has problems in modul
apr since a couple of days.
Unfortunately, the log does provide nothing for a reason not known to
me.
Thus, I will try to investigate the problem.
Hopefully, I can change the buildbot script to get the build output
directly on stdout instead of as html. The html output is
currently not
containing the corresponding information about the build of module
apr.
Done and Thx to Herbert triggering a clean build.
Unfortunately, the build was successful. Thus, I assume the reason
that we had no nightly windows builds from trunk since 2013-04-28 was
that no clean build had been performed.
This is not the case. The clean build is not the panacea you see it
as. As mentioned in several other communications, I went onto the box
and cleaned up some processes that were hung (win7, win7snap, and
win7ia2). All built successfully - even though the other two were
incremental. The hung processes tend to occur /more/ during clean
builds - not all the time, just more often. Thus, clean builds are
more
likely to create this type of build failure, they are not a fix as
you're suggesting.
Thanks for the information.
In order to have something more tangible for fixing this defect of
hanging build processes I propose to start an corresponding
investigation.
At least we should have a look after each build, esp. after each clean
build, if there are processes which hang.
It seems that this defect just occured with build #105 of aoo-w7ia2 -
see [1]. The build had been "killed". I assume that the one or the
other process of this build is still working.
Can somebody with corresponding karma check, if there are again
hanging processes?
[1] http://ci.apache.org/builders/aoo-w7ia2/builds/105
The build was not "killed" - the process that was running didn't report
back in 12000 sec = 200 min or 3+hours
At that point the buildbot tries to clean up, but this is the reaction,
not the root cause.
command timed out: 12000 seconds without output, killing pid 2472
SIGKILL failed to kill process
using fake rc=-1
program finished with exit code -1
When I checked it later, the process was still hung (thus it's unlikely
that our problem is just with the length of the timeout). This is what
one of these hung processes looks like, and any subsequent builds will
fail if it's not cleaned up, as the processes lock files and block
subsequent compiles of the same package.
Best regards, Oliver.
Andrew, can only you perform such an investigation, because (as far as I
know) you are the only who have direct access on the machine?
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscr...@openoffice.apache.org
For additional commands, e-mail: dev-h...@openoffice.apache.org
