I have also seen it on my notebook. I can currently never get through a make check on my notebook. I have not yet seen it on RHEL 5 Realtime and on the n way machines I was testing on. I you want me to do something, shout as I can reproduce it 100% of the time
on my notebook.

Carl.

Andrew Stitcher wrote:
Is your usual build machine the 4 way box? I ask 'cos you seem to be the
only one seeing this and this may be a difference between us. I don't as
a regular thing during a build. I have occasionally seen tests hang, but
I've always thought I knew why for that particular run.

Another difference might be that I'm still not testing with your cluster
code, so is it possible that this makes a difference?

If you can reproduce this more orless at will, then running broker and
client -t will let you see if there have been any packets not sent.

Andrew

PS With some help from Gordon I'm about to replace the network code in
the reading thread in the C++ client. this may help, or then again it
might make things worse. :(

On Tue, 2007-08-07 at 17:36 -0400, Alan Conway wrote:
For a week or two now I am seeing the system tests hang regularly. It
usually hangs in one of the python tests, today I got hangs in:
test_example (tests_0-10.example.ExampleTest) ...
test_ack (tests_0-10.message.MessageTests) ...

I have seen other python tests hang, and more rarely client_test. I
suspect it coincides with the new epoll IO layer but I'm not sure of
that.

At the point of the hang the broker appears perfectly healthy - other
clients can talk to it no problem. It seems to be the client that's
hanging, but the fact that c++ and python clients are hanging makes me
suspect the broker. Possibly there's a race condition where the broker
drops an outgoing frame and clients are hanging waiting for it.

So far I haven't tried to pin it down, I'm not sure exactly how to.
It's frequent enough that make check hangs more often than not for me,
but usually in a python test so I can't attach a debugger to figure out
where it hung.

Anyone have any thoughts?

Cheers,
Alan.



Reply via email to