Post latency metrics whether or not tracker is configured (#774)

Author: deliahu

pkg/workloads/cortex/lib/type/api.py

@@ -64,20 +64,21 @@ def upload_class(self, class_name):
     def metric_dimensions(self):
         return [{"Name": "APIName", "Value": self.name}, {"Name": "APIID", "Value": self.id}]
 
-    def post_request_metrics(self, status_code, start_time, prediction_value=None):
-        metrics_list = []
-        metrics_list.append(self.status_code_metric(status_code))
+    def post_latency_metrics(self, status_code, start_time):
+        metrics = [self.status_code_metric(status_code), self.latency_metric(start_time)]
+        self.post_metrics(metrics)
 
+    def post_tracker_metrics(self, prediction_value=None):
         if prediction_value is not None:
-            metrics_list.append(self.prediction_metrics(prediction_value))
-
-        metrics_list.append(self.latency_metric(start_time))
+            metrics = [self.prediction_metrics(prediction_value)]
+            self.post_metrics(metrics)
 
+    def post_metrics(self, metrics):
         try:
             if self.statsd is None:
                 raise CortexException("statsd client not initialized")  # unexpected
 
-            for metric in metrics_list:
+            for metric in metrics:
                 tags = ["{}:{}".format(dim["Name"], dim["Value"]) for dim in metric["Dimensions"]]
                 if metric.get("Unit") == "Count":
                     self.statsd.increment(metric["MetricName"], value=metric["Value"], tags=tags)

pkg/workloads/cortex/serve/serve.py

@@ -120,9 +120,11 @@ def after_request(response):
         prediction = g.prediction
 
     try:
+        api.post_latency_metrics(response.status_code, g.start_time)
+
         if api.tracker is not None:
             predicted_value = api.tracker.extract_predicted_value(prediction)
-            api.post_request_metrics(response.status_code, g.start_time, predicted_value)
+            api.post_tracker_metrics(predicted_value)
             if predicted_value is not None and predicted_value not in local_cache["class_set"]:
                 api.upload_class(predicted_value)
                 local_cache["class_set"].add(predicted_value)