buildbot on Mac OS X - SVN stops working after "some time" due to name resolution issue

[parrothead]

Hi,

both the buildbot slave (source.SVN buildstep) and the master (SVNPoller) stop working on Mac OS 10.5 after usually a couple of hours with the following error message:

2008-03-31 21:32:28-0500 [-] _finished : [Failure instance: Traceback (failure with no frames): exceptions.IOError: got stderr: "svn: PROPFIND request failed on '/repos/saga/trunk'\nsvn: PROPFIND of '/repos/saga/trunk': Could not resolve hostname `svn.cct.lsu.edu': Temporary failure in name resolution ( https://svn.cct.lsu.edu)\n"

If I try to trigger SVN as the buildbot users in one of the build directories everything works fine. I also tried to add the SVN server to /etc/hosts to override a possible DNS problem. No success. Any ideas?

[warner]

omg, I've been having exactly the same problem with darcs checkouts. I have no idea what OS-X is doing. The source checkout command is being spawned as a separate process, so I can't imagine any sorts of libc -based or per-process DNS caching being the problem. I suppose one angle to pursue would be to strace the buildslave (or whatever the equivalent of strace is in the OS-X world.. maybe one of those magical dtrace things I've read about?) and see which name-resolution syscalls the child is invoking. Doing this for a checkout soon after buildslave restart (so that it works) and again a few hours later (so that it fails) and comparing the syscalls made might give us some clues.

I do note that this wasn't happening to me before I upgraded my buildslave to 10.5 .. I think 10.4 didn't suffer from this problem.

[dustin]

Name resolution depends on the Mach context you're running in. See UsingLaunchd and  http://www.google.com/search?hl=en&client=safari&rls=en&q=StartupItemContext&btnG=Search.