Add stackdriver underflow bucket (#406)

Author: c24t

opencensus/stats/aggregation_data.py

@@ -133,16 +133,14 @@ def __init__(self,
             bounds = []
         else:
             assert bounds == list(sorted(set(bounds)))
+            assert all(bb > 0 for bb in bounds)
 
         if counts_per_bucket is None:
             counts_per_bucket = [0 for ii in range(len(bounds) + 1)]
         else:
             assert all(cc >= 0 for cc in counts_per_bucket)
             assert len(counts_per_bucket) == len(bounds) + 1
 
-        assert bounds == sorted(bounds)
-        assert all(bb > 0 for bb in bounds)
-
         self._counts_per_bucket = counts_per_bucket
         self._bounds = bucket_boundaries.BucketBoundaries(
             boundaries=bounds).boundaries

opencensus/stats/exporters/stackdriver_exporter.py

@@ -219,9 +219,14 @@ def create_time_series_list(self, v_data, option_resource_type,
                 # point.value.distribution_value.range.min = agg_data.min
                 # point.value.distribution_value.range.max = agg_data.max
                 bounds = dist_value.bucket_options.explicit_buckets.bounds
-                bounds.extend(list(map(float, agg_data.bounds)))
-
                 buckets = dist_value.bucket_counts
+
+                # Stackdriver expects a first bucket for samples in (-inf, 0),
+                # but we record positive samples only, and our first bucket is
+                # [0, first_bound).
+                bounds.extend([0])
+                buckets.extend([0])
+                bounds.extend(list(map(float, agg_data.bounds)))
                 buckets.extend(list(map(int, agg_data.counts_per_bucket)))
             else:
                 convFloat, isFloat = as_float(tag_value[0])

tests/unit/stats/exporter/test_stackdriver_stats.py

@@ -19,6 +19,7 @@
 
 from opencensus.__version__ import __version__
 from opencensus.stats import aggregation as aggregation_module
+from opencensus.stats import aggregation_data as aggregation_data_module
 from opencensus.stats import measure as measure_module
 from opencensus.stats import stats as stats_module
 from opencensus.stats import view as view_module
@@ -621,6 +622,43 @@ def test_create_timeseries_float_tagvalue(self, monitor_resource_mock):
                           "custom.googleapis.com/opencensus/view-name2")
         self.assertIsNotNone(time_series)
 
+    def test_create_timeseries_from_distribution(self):
+        """Check for explicit 0-bound bucket for SD export."""
+
+        v_data = mock.Mock(spec=view_data_module.ViewData)
+        v_data.view.name = "example.org/test_view"
+        v_data.start_time = '2018-11-21T00:12:34.56Z'
+        v_data.end_time = '2018-11-21T00:23:45.67Z'
+
+        # Aggregation over (10 * range(10)) for buckets [2, 4, 6, 8]
+        dad = aggregation_data_module.DistributionAggregationData(
+            mean_data=4.5,
+            count_data=100,
+            min_=0,
+            max_=9,
+            sum_of_sqd_deviations=825,
+            counts_per_bucket=[20, 20, 20, 20, 20],
+            bounds=[2, 4, 6, 8],
+            exemplars={mock.Mock() for ii in range(5)}
+        )
+        v_data.tag_value_aggregation_data_map = ({'tag_key': dad})
+
+        exporter = stackdriver.StackdriverStatsExporter(
+            options=mock.Mock(),
+            client=mock.Mock(),
+        )
+        time_series = exporter.create_time_series_list(v_data, "", "")
+
+        self.assertEqual(len(time_series.points), 1)
+        [point] = time_series.points
+        dv = point.value.distribution_value
+        self.assertEqual(100, dv.count)
+        self.assertEqual(4.5, dv.mean)
+        self.assertEqual(825.0, dv.sum_of_squared_deviation)
+        self.assertEqual([0, 20, 20, 20, 20, 20], dv.bucket_counts)
+        self.assertEqual([0, 2, 4, 6, 8],
+                         dv.bucket_options.explicit_buckets.bounds)
+
     def test_create_metric_descriptor_count(self):
         client = mock.Mock()
         option = stackdriver.Options(