Merge pull request #3673 from rooftopcellist/metrics_uuids

add uuids to ping and metrics

Reviewed-by: https://github.com/softwarefactory-project-zuul[bot]

awx/api/views/root.py

@@ -158,11 +158,12 @@ class ApiV1PingView(APIView):
             'ha': is_ha_environment(),
             'version': get_awx_version(),
             'active_node': settings.CLUSTER_HOST_ID,
+            'install_uuid': settings.INSTALL_UUID,
         }
 
         response['instances'] = []
         for instance in Instance.objects.all():
-            response['instances'].append(dict(node=instance.hostname, heartbeat=instance.modified,
+            response['instances'].append(dict(node=instance.hostname, uuid=instance.uuid, heartbeat=instance.modified,
                                               capacity=instance.capacity, version=instance.version))
             sorted(response['instances'], key=operator.itemgetter('node'))
         response['instance_groups'] = []

awx/main/analytics/metrics.py

@@ -37,10 +37,10 @@ USER_SESSIONS = Gauge('awx_sessions_total', 'Number of sessions', ['type',])
 CUSTOM_VENVS = Gauge('awx_custom_virtualenvs_total', 'Number of virtualenvs')
 RUNNING_JOBS = Gauge('awx_running_jobs_total', 'Number of running jobs on the Tower system')
 
-INSTANCE_CAPACITY = Gauge('awx_instance_capacity', 'Capacity of each node in a Tower system', ['type',])
-INSTANCE_CPU = Gauge('awx_instance_cpu', 'CPU cores on each node in a Tower system', ['type',])
-INSTANCE_MEMORY = Gauge('awx_instance_memory', 'RAM (Kb) on each node in a Tower system', ['type',])
-INSTANCE_INFO = Info('awx_instance', 'Info about each node in a Tower system', ['type',])
+INSTANCE_CAPACITY = Gauge('awx_instance_capacity', 'Capacity of each node in a Tower system', ['instance_uuid',])
+INSTANCE_CPU = Gauge('awx_instance_cpu', 'CPU cores on each node in a Tower system', ['instance_uuid',])
+INSTANCE_MEMORY = Gauge('awx_instance_memory', 'RAM (Kb) on each node in a Tower system', ['instance_uuid',])
+INSTANCE_INFO = Info('awx_instance', 'Info about each node in a Tower system', ['instance_uuid',])
 INSTANCE_LAUNCH_TYPE = Gauge('awx_instance_launch_type_total', 'Type of Job launched', ['node', 'launch_type',])
 INSTANCE_STATUS = Gauge('awx_instance_status_total', 'Status of Job launched', ['node', 'status',])
 
@@ -48,7 +48,7 @@ INSTANCE_STATUS = Gauge('awx_instance_status_total', 'Status of Job launched', [
 def metrics():
     license_info = get_license(show_key=False)
     SYSTEM_INFO.info({
-        'system_uuid': settings.SYSTEM_UUID,
+        'install_uuid': settings.INSTALL_UUID,
         'insights_analytics': str(settings.INSIGHTS_DATA_ENABLED),
         'tower_url_base': settings.TOWER_URL_BASE,
         'tower_version': get_awx_version(),
@@ -87,10 +87,10 @@ def metrics():
 
     instance_data = instance_info(None)
     for uuid in instance_data:
-        INSTANCE_CAPACITY.labels(type=uuid).set(instance_data[uuid]['capacity'])
-        INSTANCE_CPU.labels(type=uuid).set(instance_data[uuid]['cpu'])
-        INSTANCE_MEMORY.labels(type=uuid).set(instance_data[uuid]['memory'])
-        INSTANCE_INFO.labels(type=uuid).info({
+        INSTANCE_CAPACITY.labels(instance_uuid=uuid).set(instance_data[uuid]['capacity'])
+        INSTANCE_CPU.labels(instance_uuid=uuid).set(instance_data[uuid]['cpu'])
+        INSTANCE_MEMORY.labels(instance_uuid=uuid).set(instance_data[uuid]['memory'])
+        INSTANCE_INFO.labels(instance_uuid=uuid).info({
             'enabled': str(instance_data[uuid]['enabled']),
             'last_isolated_check': getattr(instance_data[uuid], 'last_isolated_check', 'None'),
             'managed_by_policy': str(instance_data[uuid]['managed_by_policy']),
@@ -100,7 +100,6 @@ def metrics():
     instance_data = job_instance_counts(None)
     for node in instance_data:
         # skipping internal execution node (for system jobs) 
-        # TODO: determine if we should exclude execution_node from instance count
         if node == '':
             continue
         types = instance_data[node].get('launch_type', {})

awx/main/tests/functional/analytics/test_metrics.py

@@ -60,7 +60,6 @@ def test_metrics_counts(organization_factory, job_template_factory, workflow_job
 
 @pytest.mark.django_db 
 def test_metrics_permissions(get, admin, org_admin, alice, bob, organization):
-
     assert get(reverse('api:metrics_view'), user=admin).status_code == 200
     assert get(reverse('api:metrics_view'), user=org_admin).status_code == 403
     assert get(reverse('api:metrics_view'), user=alice).status_code == 403
@@ -75,7 +74,6 @@ def test_metrics_permissions(get, admin, org_admin, alice, bob, organization):
 
 @pytest.mark.django_db 
 def test_metrics_http_methods(get, post, patch, put, options, admin):
-
     assert get(reverse('api:metrics_view'), user=admin).status_code == 200
     assert put(reverse('api:metrics_view'), user=admin).status_code == 405
     assert patch(reverse('api:metrics_view'), user=admin).status_code == 405

awx/settings/development.py

@@ -116,6 +116,7 @@ INSTALLED_APPS += ('rest_framework_swagger',)
 
 # Configure a default UUID for development only.
 SYSTEM_UUID = '00000000-0000-0000-0000-000000000000'
+INSTALL_UUID = '00000000-0000-0000-0000-000000000000'
 
 # Store a snapshot of default settings at this point before loading any
 # customizable config files.