The sync_routers interface takes too long

[liujinxin]

ENV: stable/victoria

In a large scale cloud deployment, when restart neutron l3-agent, neutron server side the _routers_to_sync interface is called twice. And this will cause the agent restart time to be too long to operate.

<_routers_to_sync> Elapsed:32.242
<_routers_to_sync> Elapsed:52.720s
<sync_routers> Elapsed:85.427s

https://github.com/openstack/neutron/blob/master/neutron/api/rpc/handlers/l3_rpc.py #128

Can it be changed to：

        if extensions.is_extension_supported(
                self.plugin, constants.PORT_BINDING_EXT_ALIAS):
            self._ensure_host_set_on_ports(context, host, routers)
            # refresh the data structure after ports are bound
            routers = self._routers_to_sync(context, router_ids, host)
        else:
            routers = self._routers_to_sync(context, router_ids, host)

[LIU Yulong (dragon889)]

The change looks reasonable, so please upload a patch to gerrit. Let's see if there are failures on CI pipelines.

[Brian Haley (brian-haley)]

I'm not sure the proposal is valid, as the routers used in the first call to _routers_to_sync() are used inside the 'if' clause:

        routers = self._routers_to_sync(context, router_ids, host)
        if extensions.is_extension_supported(
                self.plugin, constants.PORT_BINDING_EXT_ALIAS):
            self._ensure_host_set_on_ports(context, host, routers)
            # refresh the data structure after ports are bound
            routers = self._routers_to_sync(context, router_ids, host)

Perhaps, and this is just a thought, _ensure_host_set_on_port() can return a status indicating it did some work and we can skip the second sync call if it didn't?

[OpenStack Infra (hudson-openstack)]

Fix proposed to branch: master
Review: https://review.opendev.org/c/openstack/neutron/+/844172

[OpenStack Infra (hudson-openstack)]

Change abandoned by "liuyulong <email address hidden>" on branch: master
Review: https://review.opendev.org/c/openstack/neutron/+/844172