Romanos, our ODL CSIT has a similar environment with ovs 2.5 (not compiled by fuel, though) and we don't see this issue. But, the tests I'm looking at create more than 1 VM. six, actually, but they do delete *all* of them. We collect the vswitchd logs at the end of the jobs and no seg fault on either of the two compute nodes or one control node.
I would assume the seg fault is a bug in ovs, but getting the steps to more easily recreate would be useful. There is probably a port delete coming down to ovs when the VM is deleted. not sure what else. You could grab the full config of ovs before the VM delete and then recreate that in a standalone environment and see if you can find the right steps to recreate. As for the recreation of the interface, I'm not totally sure, but it's no openflowplugin doing that. probably ovsdb. JamO On 12/07/2016 02:51 AM, Skiadas Romanos wrote: > Hi, > > I'm seeing a twofold issue with Boron SR2 (latest), odl-netvirt-openstack, > OpenStack Mitaka and ovs 2.5 as compiled by > fuel-plugin-ovs. Whenever I delete all the VMs in OpenStack, ovs segfaults in > the compute node. This causes odl to > reconnect and push the config again. However, the config pushed is wrong and > breaks the deployment. > > To reproduce this: - Create only one VM - Delete the VM If there are eg 2 VMs > and only one is deleted, this bug does not > happen. All VMs have to be deleted. > > For the segfault part, this is what is shown in the ovs logs right when I > delete the VM: > > 2016-12-07T10:23:16.135Z|00046|bridge|WARN|could not open network device > tap47730c42-b1 (No such device) > 2016-12-07T10:23:16.387Z|00002|daemon_unix(monitor)|ERR|1 crashes: pid 1255 > died, killed (Segmentation fault), core > dumped, restarting > > Then the config is pushed again, ODL tries to recreate a port that patches > br-floating to an interface in a linux bridge > and breaks it: > > 30 Bridge br-floating 31 Port "p_ff798dba-0" > 32 Interface "p_ff798dba-0" 33 type: internal > 34 error: "could not open network device p_ff798dba-0 (File > exists)" This port is where the external > traffic comes from. There is a port between br-floating and br-int that puts > the traffic in br-int, the provider mapping > is physnet1:br-floating in the Fuel setup. > > Concerning the segfault part, is that an issue with netvirt, openflowplugin > or ovs? Where do I open the bug? Is there > anything further I have to provide? The recreating the interface issue, is > that a problem in netvirt or openflowplugin? > Which component is responsible for keeping track of the state and re-pushing > the configuration? I'd guess openflowplugin, > but I don't know. > > Please find attached the full ovs and odl logs with > org.opendaylight.openflowplugin set to DEBUG > > BR, Romanos Skiadas > > > > _______________________________________________ openflowplugin-dev mailing > list [email protected] > https://lists.opendaylight.org/mailman/listinfo/openflowplugin-dev > _______________________________________________ openflowplugin-dev mailing list [email protected] https://lists.opendaylight.org/mailman/listinfo/openflowplugin-dev
