Make association_refresh configurable

The provider-tree refresh in the SchedulerReportClient() instance
of each compute node happens every five minutes as it is hard coded.
This patch adds this update interval as a new config option which
can be set/changed on each compute node.

Change-Id: I00f92aac44d7b0169f94940ef389796c782b0cc1
Closes-Bug: #1767309

nova/conf/compute.py

@@ -640,6 +640,21 @@ Possible values:
 Related options:
 
 * ``shutdown_timeout``
+"""),
+    cfg.IntOpt('resource_provider_association_refresh',
+        default=300,
+        min=1,
+        help="""
+Interval for updating nova-compute-side cache of the compute node resource
+provider's aggregates and traits info.
+
+This option specifies the number of seconds between attempts to update a
+provider's aggregates and traits information in the local cache of the compute
+node.
+
+Possible values:
+
+* Any positive integer in seconds.
 """)
 ]
 

nova/scheduler/client/report.py

@@ -45,9 +45,6 @@ _RE_INV_IN_USE = re.compile("Inventory for (.+) on resource provider "
                             "(.+) in use")
 WARN_EVERY = 10
 PLACEMENT_CLIENT_SEMAPHORE = 'placement_client'
-# Number of seconds between attempts to update a provider's aggregates and
-# traits
-ASSOCIATION_REFRESH = 300
 POST_RPS_RETURNS_PAYLOAD_API_VERSION = '1.20'
 NESTED_PROVIDER_API_VERSION = '1.14'
 POST_ALLOCATIONS_API_VERSION = '1.13'
@@ -778,8 +775,8 @@ class SchedulerReportClient(object):
         sharing providers for the specified resource provider uuid.
 
         Only refresh if there has been no refresh during the lifetime of
-        this process, ASSOCIATION_REFRESH seconds have passed, or the force arg
+        this process, CONF.compute.resource_provider_association_refresh
-        has been set to True.
+        seconds have passed, or the force arg has been set to True.
 
         Note that we do *not* refresh inventories.  The reason is largely
         historical: all code paths that get us here are doing inventory refresh
@@ -848,10 +845,12 @@ class SchedulerReportClient(object):
         "recently".
 
         Associations are stale if association_refresh_time for this uuid is not
-        set or is more than ASSOCIATION_REFRESH seconds ago.
+        set or is more than CONF.compute.resource_provider_association_refresh
+        seconds ago.
         """
         refresh_time = self._association_refresh_time.get(uuid, 0)
-        return (time.time() - refresh_time) > ASSOCIATION_REFRESH
+        return ((time.time() - refresh_time) >
+                CONF.compute.resource_provider_association_refresh)
 
     def _update_inventory_attempt(self, context, rp_uuid, inv_data):
         """Update the inventory for this resource provider if needed.

nova/tests/unit/scheduler/client/test_report.py

@@ -2273,14 +2273,16 @@ class TestAssociations(SchedulerReportClientTestCase):
 
         with mock.patch('time.time') as mock_future:
             # Not called a second time because not enough time has passed.
-            mock_future.return_value = now + report.ASSOCIATION_REFRESH / 2
+            mock_future.return_value = (now +
+                CONF.compute.resource_provider_association_refresh / 2)
             self.client._refresh_associations(self.context, uuid)
             mock_agg_get.assert_not_called()
             mock_trait_get.assert_not_called()
             mock_shr_get.assert_not_called()
 
             # Called because time has passed.
-            mock_future.return_value = now + report.ASSOCIATION_REFRESH + 1
+            mock_future.return_value = (now +
+                CONF.compute.resource_provider_association_refresh + 1)
             self.client._refresh_associations(self.context, uuid)
             mock_agg_get.assert_called_once_with(self.context, uuid)
             mock_trait_get.assert_called_once_with(self.context, uuid)

releasenotes/notes/add-association-refresh-config-opt-d1ca1af238d10c9a.yaml

@@ -0,0 +1,9 @@
+---
+fixes:
+  - |
+    The nova-compute service now allows specifying the interval for updating
+    nova-compute-side cache of the compute node resource provider's aggregates
+    and traits info via a new config option called
+    ``[compute]/resource_provider_association_refresh`` which defaults to 300.
+    This was previously hard-coded to run every 300 seconds which may be too
+    often in a large deployment.