Merge pull request #23950 from ericgribkoff/directly_query_health

xds tests: check health service in wait_for_healthy_backends

tools/run_tests/run_xds_tests.py

@@ -966,6 +966,12 @@ def test_header_matching(gcp, original_backend_service, instance_group,
         set_validate_for_proxyless(gcp, True)
 
 
+def get_serving_status(instance, service_port):
+    with grpc.insecure_channel('%s:%d' % (instance, service_port)) as channel:
+        health_stub = health_pb2_grpc.HealthStub(channel)
+        return health_stub.Check(health_pb2.HealthCheckRequest())
+
+
 def set_serving_status(instances, service_port, serving):
     logger.info('setting %s serving status to %s', instances, serving)
     for instance in instances:
@@ -973,16 +979,13 @@ def set_serving_status(instances, service_port, serving):
                                    (instance, service_port)) as channel:
             logger.info('setting %s serving status to %s', instance, serving)
             stub = test_pb2_grpc.XdsUpdateHealthServiceStub(channel)
-            health_stub = health_pb2_grpc.HealthStub(channel)
-
             retry_count = 5
             for i in range(5):
                 if serving:
                     stub.SetServing(empty_pb2.Empty())
                 else:
                     stub.SetNotServing(empty_pb2.Empty())
-                serving_status = health_stub.Check(
-                    health_pb2.HealthCheckRequest())
+                serving_status = get_serving_status(instance, service_port)
                 logger.info('got instance service status %s', serving_status)
                 want_status = health_pb2.HealthCheckResponse.SERVING if serving else health_pb2.HealthCheckResponse.NOT_SERVING
                 if serving_status.status == want_status:
@@ -1550,14 +1553,23 @@ def wait_for_healthy_backends(gcp,
                               timeout_sec=_WAIT_FOR_BACKEND_SEC):
     start_time = time.time()
     config = {'group': instance_group.url}
-    expected_size = len(get_instance_names(gcp, instance_group))
+    instance_names = get_instance_names(gcp, instance_group)
+    expected_size = len(instance_names)
     while time.time() - start_time <= timeout_sec:
+        for instance_name in instance_names:
+            try:
+                status = get_serving_status(instance_name, gcp.service_port)
+                logger.info('serving status response from %s: %s',
+                            instance_name, status)
+            except grpc.RpcError as rpc_error:
+                logger.info('checking serving status of %s failed: %s',
+                            instance_name, rpc_error)
         result = gcp.compute.backendServices().getHealth(
             project=gcp.project,
             backendService=backend_service.name,
             body=config).execute(num_retries=_GCP_API_RETRIES)
         if 'healthStatus' in result:
-            logger.info('received healthStatus: %s', result['healthStatus'])
+            logger.info('received GCP healthStatus: %s', result['healthStatus'])
             healthy = True
             for instance in result['healthStatus']:
                 if instance['healthState'] != 'HEALTHY':
@@ -1565,7 +1577,9 @@ def wait_for_healthy_backends(gcp,
                     break
             if healthy and expected_size == len(result['healthStatus']):
                 return
-        time.sleep(2)
+        else:
+            logger.info('no healthStatus received from GCP')
+        time.sleep(5)
     raise Exception('Not all backends became healthy within %d seconds: %s' %
                     (timeout_sec, result))