Good catch. This was a result of the "Don't unnecessarily loop through all ports/interfaces" patch.[1]
Passing a list of names to "ovs-vsctl list Interface" with an element that doesn't match an existing port causes an error.[2] This happens when a port is deleted right after the list of interface names is gathered but before the "list Interface" command is run. I think the resulting runtime errors were interfering with the correct operation of the agent. There is a fix up[3] that adds the '--if-exists' flag that should restore the previous behavior of not requiring the requested names to exist. Hopefully that should clear up the issue. 1. https://github.com/openstack/neutron/commit/3f0bf6cfac2e151d5a4a7f076062b3365bdbf457 2. http://logs.openstack.org/12/142412/3/gate/gate-tempest-dsvm-neutron-full/9990c84/logs/screen-q-agt.txt.gz#_2015-01-02_15_42_40_999 3. https://review.openstack.org/#/c/144872/ On Sat, Jan 3, 2015 at 4:14 PM, Matt Riedemann <mrie...@linux.vnet.ibm.com> wrote: > https://bugs.launchpad.net/neutron/+bug/1407190 > > Looks like this started on 12/29. I checked the neutron commits around > that time and there are some related to ports/interfaces and openvswitch > but I'm not sure that they contributing to this issue. > > I've put a bunch of notes and log links in the bug report, basically there > is a race in a tempest test where it creates a server and adds two ports to > it (so a total of three), it waits for each port to be ACTIVE before > continuing, and then does a show port call by port ID and server ID and > expects the port to be the same as what it originally created. When this > fails, the expected port_status is 'ACTIVE' but the actual port_status is > 'BUILD'. > > I've found in the logs where nova/neutron are seeing the port going to > 'BUILD' state but I don't know enough about the neutron code to tell what > actually triggers this change, the only place I see 'BUILD' used is in the > nova event notifier code in neutron which is related to the vif plug/unplug > callback on nova server boot. > > This isn't the top gate bug, but it's got a significant number of hits and > seems like something that we should be able to reasonably get a quick > handle on given I think we know when it started (12/29 according to > logstash, I caught it before the 10 day index in logstash passed). > > -- > > Thanks, > > Matt Riedemann > > > _______________________________________________ > OpenStack-dev mailing list > OpenStack-dev@lists.openstack.org > http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev > -- Kevin Benton
_______________________________________________ OpenStack-dev mailing list OpenStack-dev@lists.openstack.org http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev