fullstack OVS agent in native openflow mode sometimes fails to bind socket

[Assaf Muller]

Plenty of hits in the last few days, currently the top issue affecting fullstack stability.

Example paste:
http://paste.openstack.org/show/495797/

Example logs:
http://logs.openstack.org/18/276018/21/check/gate-neutron-dsvm-fullstack/c0761dc/logs/TestOvsConnectivitySameNetwork.test_connectivity_VLANs,openflow-native_ovsdb-native_/

[Miguel Angel Ajo (mangelajo)]

Are we sending the individual fullstack log outputs to logstash/elasticsearch? I don't get to make a decent query on logstash.. (I tried http://logstash.openstack.org/#dashboard/file/logstash.json?query=message%3A%20%5C%22error%3A%20%5BErrno%2098%5D%20Address%20already%20in%20use%5C%22 ).

I guess this is to concurrent runs of openvswitch agents trying to use the same port number?

If that's the case, could we retry tests runs on these cases, with random increments of port numbers? Or, it'd still be better if we could pass down to workers an incremental port-start-number to use when that is needed.

[IWAMOTO Toshihiro (iwamoto)]

Should be fixed by commit 9319b1a8366e00ec2ea9a763b5c3558649758f51 .
Can we close this?

[Assaf Muller (amuller)]

Looking at bug https://bugs.launchpad.net/neutron/+bug/1551288 you said you saw more failures after the patch you mentioned was merged, and that we need patch https://review.openstack.org/#/c/319807/ still. Is this still the case?

[Ihar Hrachyshka (ihar-hrachyshka)]

It's long fixed.