DVR: fix router rescheduling on server side

This fixes router rescheduling for dvr routers to
correctly handle CSNAT portion rescheduling on server side.

Next patch will make L3 agent aware of possible SNAT role
rescheduling to/from it.

Partial-Bug: #1472205
Change-Id: I069500a2f1de9a734124544a3f2d793e9fa2d27c

diff --git a/neutron/db/l3_dvrscheduler_db.py b/neutron/db/l3_dvrscheduler_db.py
index 605a9356352665d0ff53006240a6ae171b25ac83..5e937611a5213d6d64317af33cb56101594a5baa 100644 (file)
--- a/neutron/db/l3_dvrscheduler_db.py
+++ b/neutron/db/l3_dvrscheduler_db.py
@@ -31,6 +31,7 @@ from neutron.db import agents_db
 from neutron.db import l3_agentschedulers_db as l3agent_sch_db
 from neutron.db import model_base
 from neutron.db import models_v2
+from neutron.extensions import l3agentscheduler
 from neutron.i18n import _LI, _LW
 from neutron import manager
 from neutron.plugins.common import constants as service_constants
@@ -342,7 +343,10 @@ class L3_DVRsch_db_mixin(l3agent_sch_db.L3AgentSchedulerDbMixin):
             return
         snat_candidates = self.get_snat_candidates(sync_router,
                                                    active_l3_agents)
-        if snat_candidates:
+        if not snat_candidates:
+            LOG.warn(_LW('No candidates found for SNAT'))
+            return
+        else:
             try:
                 chosen_agent = self.bind_snat_servicenode(
                     context, router_id, snat_candidates)
@@ -353,6 +357,43 @@ class L3_DVRsch_db_mixin(l3agent_sch_db.L3AgentSchedulerDbMixin):
                 context, router_id, chosen_agent)
             return chosen_agent
 
+    def reschedule_router(self, context, router_id, candidates=None):
+        """Reschedule router to new l3 agents
+
+        Remove the router from l3 agents currently hosting it and
+        schedule it again
+        """
+        router = self.get_router(context, router_id)
+        is_distributed = router.get('distributed', False)
+        if not is_distributed:
+            return super(L3_DVRsch_db_mixin, self).reschedule_router(
+                context, router_id, candidates)
+
+        old_agents = self.list_l3_agents_hosting_router(
+            context, router_id)['agents']
+        with context.session.begin(subtransactions=True):
+            for agent in old_agents:
+                self._unbind_router(context, router_id, agent['id'])
+                self.unbind_snat_servicenode(context, router_id)
+
+            self.schedule_router(context, router_id, candidates=candidates)
+            new_agents = self.list_l3_agents_hosting_router(
+                context, router_id)['agents']
+            if not new_agents:
+                raise l3agentscheduler.RouterReschedulingFailed(
+                    router_id=router_id)
+
+        l3_notifier = self.agent_notifiers.get(n_const.AGENT_TYPE_L3)
+        if l3_notifier:
+            old_hosts = [agent['host'] for agent in old_agents]
+            new_hosts = [agent['host'] for agent in new_agents]
+            for host in set(old_hosts) - set(new_hosts):
+                l3_notifier.router_removed_from_agent(
+                    context, router_id, host)
+            for host in new_hosts:
+                l3_notifier.router_added_to_agent(
+                    context, [router_id], host)
+
     def _get_active_l3_agent_routers_sync_data(self, context, host, agent,
                                                router_ids):
         if n_utils.is_extension_supported(self, n_const.L3_HA_MODE_EXT_ALIAS):

diff --git a/neutron/tests/unit/plugins/ml2/drivers/openvswitch/agent/test_agent_scheduler.py b/neutron/tests/unit/plugins/ml2/drivers/openvswitch/agent/test_agent_scheduler.py
index e512b102fb7e81357ac7e7093b3d14e0dcf84fb2..96bfbcdbebe75ca09a54f40ad206a3d82e8643c1 100644 (file)
--- a/neutron/tests/unit/plugins/ml2/drivers/openvswitch/agent/test_agent_scheduler.py
+++ b/neutron/tests/unit/plugins/ml2/drivers/openvswitch/agent/test_agent_scheduler.py
@@ -997,6 +997,34 @@ class OvsAgentSchedulerTestCase(OvsAgentSchedulerTestCaseBase):
                          set([a['configurations']['agent_mode'] for a in
                               l3agents['agents']]))
 
+    def test_dvr_router_csnat_rescheduling(self):
+        helpers.register_l3_agent(
+            host=L3_HOSTA, agent_mode=constants.L3_AGENT_MODE_DVR_SNAT)
+        helpers.register_l3_agent(
+            host=L3_HOSTB, agent_mode=constants.L3_AGENT_MODE_DVR_SNAT)
+        with self.subnet() as s:
+            net_id = s['subnet']['network_id']
+            self._set_net_external(net_id)
+
+            router = {'name': 'router1',
+                      'external_gateway_info': {'network_id': net_id},
+                      'admin_state_up': True,
+                      'distributed': True}
+            r = self.l3plugin.create_router(self.adminContext,
+                                            {'router': router})
+            self.l3plugin.schedule_router(
+                    self.adminContext, r['id'])
+            l3agents = self._list_l3_agents_hosting_router(r['id'])
+            self.assertEqual(2, len(l3agents['agents']))
+            csnat_agent_host = self.l3plugin.get_snat_bindings(
+                self.adminContext, [r['id']])[0]['l3_agent']['host']
+            self._take_down_agent_and_run_reschedule(csnat_agent_host)
+            l3agents = self._list_l3_agents_hosting_router(r['id'])
+            self.assertEqual(1, len(l3agents['agents']))
+            new_csnat_agent_host = self.l3plugin.get_snat_bindings(
+                self.adminContext, [r['id']])[0]['l3_agent']['host']
+            self.assertNotEqual(csnat_agent_host, new_csnat_agent_host)
+
     def test_router_sync_data(self):
         with self.subnet() as s1,\
                 self.subnet(cidr='10.0.2.0/24') as s2,\