Continue check execution when only a few hosts are unhealthy (#6954)

rabbitmq/datadog_checks/rabbitmq/rabbitmq.py

@@ -674,7 +674,12 @@ def _check_aliveness(self, base_url, vhosts, custom_tags):
             # We need to urlencode the vhost because it can be '/'.
             path = u'aliveness-test/{}'.format(quote_plus(vhost))
             aliveness_url = urljoin(base_url, path)
-            aliveness_response = self._get_data(aliveness_url)
+            aliveness_response = {}
+            try:
+                aliveness_response = self._get_data(aliveness_url)
+            except Exception as e:
+                self.log.debug("Couldn't get aliveness status from vhost, %s: %s", vhost, e)
+
             message = u"Response from aliveness API: {}".format(aliveness_response)
 
             if aliveness_response.get('status') == 'ok':

rabbitmq/tests/test_unit.py

@@ -72,12 +72,12 @@ def test__check_aliveness(check, aggregator):
     check._get_data = mock.MagicMock()
 
     # only one vhost should be OK
-    check._get_data.side_effect = [{"status": "ok"}, {}]
-    check._check_aliveness('', vhosts=['foo', 'bar'], custom_tags=[])
+    check._get_data.side_effect = [{"status": "ok"}, {}, {"status": "not_ok"}, Exception("foo")]
+    check._check_aliveness('', vhosts=['foo', 'bar', 'baz', 'xyz'], custom_tags=[])
     sc = aggregator.service_checks('rabbitmq.aliveness')
-    assert len(sc) == 2
+    assert len(sc) == 4
     aggregator.assert_service_check('rabbitmq.aliveness', status=RabbitMQ.OK)
-    aggregator.assert_service_check('rabbitmq.aliveness', status=RabbitMQ.CRITICAL)
+    aggregator.assert_service_check('rabbitmq.aliveness', status=RabbitMQ.CRITICAL, count=3)
 
     # in case of connection errors, this check should stay silent
     check._get_data.side_effect = RabbitMQException