Performance of L2 population

[shihanzhang]

when a compute node restarts, all ports on this host will trigger l2 pop again, even if these ports were not change, if there are many compute nodes restarting at the same time, the l2 pop will be much consumable, I think if a l2 agent restarts, other l2 agents should not flush the tunnel info to this restarted l2 agent.

[Li Ma (nick-ma-z)]

In linuxbridge-agent, l2-pop rules are not persistent. They should be learned again via RPC. I think here you are discussing about ovs-agent, right?

[shihanzhang (shihanzhang)]

ovs also not persistent the l2-po rules, in large scale, for example there are 1000 compute nodes, if 100 compute nodes restart, they will trigger other 900 compute nodes do l2-pop, so I think linuxbridge-agent also has this problem

[Li Ma (nick-ma-z)]

You are right. I discovered several problems in l2pop in our production environments.

1. binding:host_id changes like VM migration
2. L3 HA or other mechanism which has distributed ports
3. rules recreation via service restart

I'm not sure the progress in the upstream, as I just start to look into it.

[Li Ma (nick-ma-z)]

4. this situation as you stated.

[Armando Migliaccio (armando-migliaccio)]

This bug is > 180 days without activity. We are unsetting assignee and milestone and setting status to Incomplete in order to allow its expiry in 60 days.

If the bug is still valid, then update the bug status.

[Launchpad Janitor (janitor)]

[Expired for neutron because there has been no activity for 60 days.]