Fix: Incorrect replication status on clusters

When running a replicated service in a cluster, the replication_status
field in the Cluster DB table doesn't get updated in some cases, making
the cluster look like it's non replicated:

- When starting a new clustered replicated service cluster.
- When enabling replication on an already clustered service.

The reason for that is that the service and cluster creation happens on
the Service's __init__ method, while the replication status isn't known
until later, when the start method in the service calls the manager's
init_host_with_rpc, where the manager updates the service DB entry.

This patch adds an update for the replication field to the cluster table
in that same place, making sure we have the right data.

Closes-Bug: #1758024
Change-Id: I0d2fded62372a2fc5f49538133c5b66694dccc77

cinder/tests/fake_driver.py

@@ -397,3 +397,11 @@ class FakeGateDriver(lvm.LVMVolumeDriver):
             snapshot_model_updates.append(snapshot_model_update)
 
         return model_update, snapshot_model_updates
+
+
+class FakeHAReplicatedLoggingVolumeDriver(FakeLoggingVolumeDriver):
+    SUPPORTS_ACTIVE_ACTIVE = True
+
+    @utils.trace_method
+    def failover_completed(self, context, active_backend_id=None):
+        pass

cinder/tests/unit/volume/test_replication_manager.py

@@ -663,3 +663,33 @@ class ReplicationTestCase(base.BaseVolumeTestCase):
                                           out_volumes, out_snapshots,
                                           [], [], [],
                                           self.manager.FAILBACK_SENTINEL)
+
+    @mock.patch('cinder.utils.log_unsupported_driver_warning', mock.Mock())
+    @mock.patch('cinder.utils.require_driver_initialized', mock.Mock())
+    def test_init_host_with_rpc_clustered_replication(self):
+        # These are not OVOs but ORM instances
+        cluster = utils.create_cluster(self.context)
+        service = utils.create_service(self.context,
+                                       {'cluster_name': cluster.name,
+                                        'binary': cluster.binary})
+        self.assertNotEqual(fields.ReplicationStatus.ENABLED,
+                            cluster.replication_status)
+        self.assertNotEqual(fields.ReplicationStatus.ENABLED,
+                            service.replication_status)
+
+        vol_manager = manager.VolumeManager(
+            'cinder.tests.fake_driver.FakeHAReplicatedLoggingVolumeDriver',
+            host=service.host, cluster=cluster.name)
+        vol_manager.driver = mock.Mock()
+        vol_manager.driver.get_volume_stats.return_value = {
+            'replication_enabled': True
+        }
+        vol_manager.init_host_with_rpc()
+
+        cluster_ovo = objects.Cluster.get_by_id(self.context, cluster.id)
+        service_ovo = objects.Service.get_by_id(self.context, service.id)
+
+        self.assertEqual(fields.ReplicationStatus.ENABLED,
+                         cluster_ovo.replication_status)
+        self.assertEqual(fields.ReplicationStatus.ENABLED,
+                         service_ovo.replication_status)

cinder/volume/manager.py

@@ -522,14 +522,23 @@ class VolumeManager(manager.CleanableManager,
             with excutils.save_and_reraise_exception():
                 LOG.error("Service not found for updating replication_status.")
 
-        if service.replication_status != (
-                fields.ReplicationStatus.FAILED_OVER):
+        if service.replication_status != fields.ReplicationStatus.FAILED_OVER:
             if stats and stats.get('replication_enabled', False):
-                service.replication_status = fields.ReplicationStatus.ENABLED
+                replication_status = fields.ReplicationStatus.ENABLED
             else:
-                service.replication_status = fields.ReplicationStatus.DISABLED
+                replication_status = fields.ReplicationStatus.DISABLED
+
+            if replication_status != service.replication_status:
+                service.replication_status = replication_status
+                service.save()
+
+        # Update the cluster replication status if necessary
+        cluster = service.cluster
+        if (cluster and
+                cluster.replication_status != service.replication_status):
+            cluster.replication_status = service.replication_status
+            cluster.save()
 
-        service.save()
         LOG.info("Driver post RPC initialization completed successfully.",
                  resource={'type': 'driver',
                            'id': self.driver.__class__.__name__})