Set OVS inactivity_probe to vsctl_timeout when adding manager

If the vsctl_timeout > OVS's inactivity probe interval and a
transaction execution time exceeds the probe interval, OVS will
disconnect and the transaction will return TRY_AGAIN and most
likely repeat failing until the vsctl_timeout is reached. This
change ensures that the "failsafe" creation of the manager also
sets the inactivity probe to the vsctl_timeout value.

Currently the patch doesn't override the probe_interval on an
existing Manager since it is possible that connection is used by
outside applications and it theoretically should be handled at
deployment.

Related-Bug: #1627106
Change-Id: I76fa0a0cf04a166edf062086fceb2fd90960ad6b

neutron/agent/common/ovs_lib.py

@@ -55,6 +55,8 @@ OVS_DEFAULT_CAPS = {
     'iface_types': [],
 }
 
+_SENTINEL = object()
+
 
 def _ofport_result_pending(result):
     """Return True if ovs-vsctl indicates the result is still pending."""
@@ -107,8 +109,21 @@ class BaseOVS(object):
         self.vsctl_timeout = cfg.CONF.ovs_vsctl_timeout
         self.ovsdb = ovsdb.API.get(self)
 
-    def add_manager(self, connection_uri):
-        self.ovsdb.add_manager(connection_uri).execute()
+    def add_manager(self, connection_uri, timeout=_SENTINEL):
+        """Have ovsdb-server listen for manager connections
+
+        :param connection_uri: Manager target string
+        :param timeout: The Manager probe_interval timeout value
+                        (defaults to ovs_vsctl_timeout)
+        """
+        if timeout is _SENTINEL:
+            timeout = cfg.CONF.ovs_vsctl_timeout
+        with self.ovsdb.transaction() as txn:
+            txn.add(self.ovsdb.add_manager(connection_uri))
+            if timeout:
+                txn.add(
+                    self.ovsdb.db_set('Manager', connection_uri,
+                                      ('inactivity_probe', timeout * 1000)))
 
     def get_manager(self):
         return self.ovsdb.get_manager().execute()

neutron/agent/ovsdb/native/helpers.py

@@ -28,10 +28,15 @@ def _connection_to_manager_uri(conn_uri):
         return 'p%s:%s' % (proto, addr)
 
 
-def enable_connection_uri(conn_uri):
+def enable_connection_uri(conn_uri, set_timeout=False):
     class OvsdbVsctlContext(object):
         vsctl_timeout = cfg.CONF.ovs_vsctl_timeout
 
     manager_uri = _connection_to_manager_uri(conn_uri)
     api = ovsdb.API.get(OvsdbVsctlContext, 'vsctl')
-    api.add_manager(manager_uri).execute(check_error=False, log_errors=True)
+    with api.transaction() as txn:
+        txn.add(api.add_manager(manager_uri))
+        if set_timeout:
+            timeout = cfg.CONF.ovs_vsctl_timeout * 1000
+            txn.add(api.db_set('Manager', manager_uri,
+                               ('inactivity_probe', timeout)))

neutron/agent/ovsdb/native/idlutils.py

@@ -126,7 +126,7 @@ def get_schema_helper(connection, schema_name, retry=True):
             if not retry:
                 ctx.reraise = True
             # We may have failed due to set-manager not being called
-            helpers.enable_connection_uri(connection)
+            helpers.enable_connection_uri(connection, set_timeout=True)
 
             # There is a small window for a race, so retry up to a second
             @tenacity.retry(wait=tenacity.wait_exponential(multiplier=0.01),

neutron/tests/functional/agent/test_ovs_lib.py

@@ -510,6 +510,9 @@ class OVSLibTestCase(base.BaseOVSLinuxTestCase):
         self.addCleanup(self.ovs.remove_manager, conn_uri)
         self.ovs.add_manager(conn_uri)
         self.assertIn(conn_uri, self.ovs.get_manager())
+        self.assertEqual(self.ovs.db_get_val('Manager', conn_uri,
+                                             'inactivity_probe'),
+                         self.ovs.vsctl_timeout * 1000)
         self.ovs.remove_manager(conn_uri)
         self.assertNotIn(conn_uri, self.ovs.get_manager())