On Wed, Dec 07, 2011 at 01:25:10PM +0100, Michal Novotny wrote:
On 12/07/2011 01:12 PM, Daniel P. Berrange wrote:
On Sun, Dec 04, 2011 at 04:10:34PM +0800, Daniel Veillard wrote:
On Fri, Dec 02, 2011 at 02:30:35PM +0100, Peter Krempa wrote:
On 12/02/2011 01:42 AM, Daniel Veillard wrote:
On Thu, Dec 01, 2011 at 02:11:24PM -0700, Eric Blake wrote:
But that means we really are committing to an rc2. Definitely. For example there is apparently a problem with commit fa9595003d043df9f2efe95521c00898cef27106 that we ough to fix quickly too to allow further testing :-)
Daniel
The problem was caused by two threads that both were thinking they are having the buck and entering poll() which caused some clients to hang. It was possible due to a race condition and therefore was not 100% reproducible. It should be fixed now with: http://www.redhat.com/archives/libvir-list/2011-December/msg00116.html Indeed, the current git head works fine again for me, thanks a lot for chasing this :-)
So I made an second release candidate available at:
ftp://libvirt.org/libvirt/libvirt-0.9.8-rc2.tar.gz
along with rpms, I also tagged git with it. Hopefully the builds on BSD and Windows should be fixed, it would be good if it could be tested on OsX and since there was a build done last month on Android, I wonder if this could be done again [1]. Thre is something very broken in the RPC code when an event loop is activated, which is resulting in frequent crashes. Fixing this is a release blocker IMHO.
Attaching a demo program which crashes 50% of the time or more with GIT head.
Daniel
Daniel, I've been trying it myself using the syntax like:
gcc -Wall -o libvirt-test `pkg-config --libs --cflags libvirt` -lpthread libvirt-test.c
to run it both as root and non-root and also with a guest running and not-running and it never failed for me on my Intel(R) Core(TM)2 Duo CPU T9600 @ 2.80GHz i686 system with 4 GB of RAM (just 2.9 GB really available because of 32-bit system).
I was thinking of adding this into the tests directory and putting it directly to the tests run by `make check`. What do you think about this?
This demo program isn't really suitable for running as a unit test, since it depends on a suitably running libvirt. It would be possible to write a unit test though, that directly uses the virNetClient and virNetServer APIs. Daniel -- |: http://berrange.com -o- http://www.flickr.com/photos/dberrange/ :| |: http://libvirt.org -o- http://virt-manager.org :| |: http://autobuild.org -o- http://search.cpan.org/~danberr/ :| |: http://entangle-photo.org -o- http://live.gnome.org/gtk-vnc :|