Hi Han Thanks for your feedback.
Inactivity probe kicks in as the test is heavy (4 hv, each claiming hundreds of ports) + the fact that github might run multiple of those tests in parallel (monitor-all true / false, parallel northd). I saw very long poll interval in the controller (> 5 seconds), due to sb_port_binding handlers. When inactivity probe kicked in, commits to sb from controller failed as connection has dropped, causing recompute in the next run.Isn't this the expected behavior? Thanks Xavier On Mon, Aug 29, 2022 at 9:30 PM Han Zhou <[email protected]> wrote: > > > On Mon, Aug 29, 2022 at 9:13 AM Xavier Simonart <[email protected]> > wrote: > > > > Recompute test might fail on some slow systems due to sb inactivity probe > > causing recomputes. > > Disable inactivity probe to avoid this issue. > > > Hi Xavier, > > Thanks for making the test case stable, but before that shall we check why > inactivity probe triggers recompute? Inactivity probe doesn't change any > input, so it is not expected to cause recompute. If that's not the case, > I'd fix the I-P engine instead of the test case. > > Thanks, > Han > > > Fixes: a7c7d451 ("controller: avoid recomputes triggered by SBDB > Port_Binding updates") > > > > Signed-off-by: Xavier Simonart <[email protected]> > > --- > > tests/ovn.at | 2 ++ > > 1 file changed, 2 insertions(+) > > > > diff --git a/tests/ovn.at b/tests/ovn.at > > index 5d73c3379..53519b92b 100644 > > --- a/tests/ovn.at > > +++ b/tests/ovn.at > > @@ -32611,6 +32611,8 @@ for i in $(seq 1 $n_hv); do > > ovn_attach n1 br-phys 192.168.0.$i 24 geneve > > done > > > > +check ovn-sbctl set connection . inactivity_probe=0 > > + > > add_switch_ports() { > > start_port=$1 > > end_port=$2 > > -- > > 2.31.1 > > > > _______________________________________________ > > dev mailing list > > [email protected] > > https://mail.openvswitch.org/mailman/listinfo/ovs-dev > _______________________________________________ dev mailing list [email protected] https://mail.openvswitch.org/mailman/listinfo/ovs-dev
