On Thu, Apr 02, 2015 at 06:44:08PM -0300, Flavio Leitner wrote: > On Thu, 2 Apr 2015 13:47:56 -0700 > Ben Pfaff <b...@nicira.com> wrote: > > > On Thu, Apr 02, 2015 at 05:45:34PM -0300, Flavio Leitner wrote: > > > On Thu, 2 Apr 2015 13:37:06 -0700 > > > Ben Pfaff <b...@nicira.com> wrote: > > > > > > > On Thu, Apr 02, 2015 at 05:33:25PM -0300, Flavio Leitner wrote: > > > > > On Thu, 2 Apr 2015 12:58:37 -0700 > > > > > Ben Pfaff <b...@nicira.com> wrote: > > > > > > > > > > > On Thu, Apr 02, 2015 at 03:57:20PM -0300, Flavio Leitner > > > > > > wrote: > > > > > > > The ofproto-dpif creates dummies backed by sockets so > > > > > > > depending on the order of execution when bridge is > > > > > > > reconfiguring, an active socket may run first and not find > > > > > > > the file. That is usually not a problem because it will > > > > > > > try to reconnect one second later. However, it breaks the > > > > > > > testsuite. > > > > > > > > > > > > > > This patch fixes the issue splitting active and passive > > > > > > > sockets in different vsctl-ctl commands that guarantees the > > > > > > > proper ordering between them. > > > > > > > > > > > > WAIT_FOR_DUMMY_PORTS is supposed to avoid this problem, by > > > > > > waiting until the ports have connected. Is it busted? > > > > > > > > > > No, but it takes at least one extra second for the port to > > > > > reconnect. If we consider the four tests fixed by the patch, it > > > > > can be 4 extra seconds to complete the same tests, so I'd just > > > > > go with the proposed patch. > > > > > > > > OK, I buy that, but in that case the commit message goes overboard > > > > when it says that the current form "breaks the testsuite". Can > > > > you rephrase this as an optimization rather than a bug fix, then? > > > > > > Not really because it is a bug fix. Those tests break like 7 out > > > of 10 times on a s390x. > > > > OK, so WAIT_FOR_DUMMY_PORTS is buggy then, can we figure out why and > > fix it? Optimizations are fine too but I'd like to get to the root of > > the problem. > > Ok, I missed that WAIT_FOR_DUMMY_PORTS is not available on branch-2.3 > where it breaks most of the times. So, I think the correct would be > to backport these patches to branch-2.3 first: > > 93fa0de tests: Fix race in 'balance-tcp bonding' test. > 60187ac test: Remove explicit sleeps from ofproto-dpif bond tests > d611e23 test: add WAIT_FOR_DUMMY_PORTS helper macro for writing tests > > and then hopefully my patch becomes just an optimization. Anyway, > I can do that next week if you or someone else didn't do it before me.
Thanks for the list of commits. I took care of the backport. I'd appreciate it if you'd verify it. I'm happy enough to take optimizations (master only, please) now that we've figured out the root of the problem. It's funny that s390x passes in the Debian buildds (see https://buildd.debian.org/status/package.php?p=openvswitch) but maybe that's just luck. _______________________________________________ dev mailing list dev@openvswitch.org http://openvswitch.org/mailman/listinfo/dev
