On 4/5/22 17:22, Mark Michelson wrote: > On 3/30/22 06:40, Xavier Simonart wrote: >> Hi Dumitru >> >> On Mon, Mar 21, 2022 at 12:02 PM Dumitru Ceara <[email protected]> wrote: >> >>> On 3/21/22 10:56, Xavier Simonart wrote: >>>> - send gratuitous arp on localnet >>>> - send gratuitous arp for nat ips in localnet >>>> - dns lookup : 1 HV, 2 LS, 2 LSPs/LS -- ovn-northd -- dp-groups=yes >>>> - send gratuitous arp for NAT rules on distributed router >>>> - 2 HVs, 1 lport/HV, localport ports >>>> >>>> Signed-off-by: Xavier Simonart <[email protected]> >>>> --- >>> >>> Hi Xavier, >>> >>>> tests/ovn.at | 20 +++++++++++++------- >>>> 1 file changed, 13 insertions(+), 7 deletions(-) >>>> >>>> diff --git a/tests/ovn.at b/tests/ovn.at >>>> index 166b5f72e..cf027703f 100644 >>>> --- a/tests/ovn.at >>>> +++ b/tests/ovn.at >>> >>> [...] >>> >>>> @@ -10391,7 +10395,7 @@ test_dns6() { >>>> } >>>> >>>> AT_CAPTURE_FILE([ofctl_monitor0.log]) >>>> -as hv1 ovs-ofctl monitor br-int resume --detach --no-chdir \ >>>> +as hv1 ovs-ofctl -t 300 monitor br-int resume --detach --no-chdir \ >>>> --pidfile=ovs-ofctl0.pid 2> ofctl_monitor0.log >>>> >>> >>> Is this really needed? We set OVS_CTL_TIMEOUT=30 in atlocal.in. >>> >>> Thanks for looking into this. >> >> Yes :-) >> The issue here is that we start an ovs-ofctl monitor, then run a few >> times >> (send packet / check monitor.log). >> The test is quite long, and on slow systems (e.g. VM running s390 >> architecture on a x86_64 host), the 30 seconds are not enough for all >> tests, and hence monitoring stops. Then we check (for 30 seconds) whether >> monitor.log contains what we expect ... >> 300 seconds might be a little much... just depends on how slow the system >> is... Requires at least one minute on my system... > > Thanks for the clarification on this. > > Acked-by: Mark Michelson > > I think this should be committed as-is since it's fixing the flakiness > of the tests. >
Sounds good to me too, the CI has been quite unhappy lately. > With regards to the DNS test in particular, my assumption is that the > long timeout is needed because the DNS queries are falling back to > system default behavior when OVN is not able to answer the queries > successfully. This can take a very long time since the system will wait > a certain amount of time for a response and may retry the query a > multiple times as well. If there were some way to temporarily override > default system behavior it may end up speeding up the test and > eliminating the need to set the timeout so large. > > It's also possible I'm completely wrong in the above paragraph :) > >> >> Thanks >> Xavier >> >> Thanks, >>> Dumitru >>> >>> >> _______________________________________________ >> 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
