Count NaN histogram observations in the overflow bucket

Signed-off-by: Zhe Chen <zhechen.cs@connect.hku.hk>

Author: zhechen-cs

prometheus_client/metrics.py

@@ -680,6 +680,8 @@ def observe(self, amount: float, exemplar: Optional[Dict[str, str]] = None) -> N
         observations. See
         https://prometheus.io/docs/practices/histograms/#count-and-sum-of-observations
         for details.
+
+        NaN observations are counted in the +Inf bucket and make the sum NaN.
         """
         self._raise_if_not_observable()
         self._sum.inc(amount)
@@ -690,6 +692,8 @@ def observe(self, amount: float, exemplar: Optional[Dict[str, str]] = None) -> N
                     _validate_exemplar(exemplar)
                     self._buckets[i].set_exemplar(Exemplar(exemplar, amount, time.time()))
                 break
+        else:
+            self._buckets[-1].inc(1)
 
     def time(self) -> Timer:
         """Time a block of code or function, and observe the duration in seconds.

tests/test_core.py

@@ -1,4 +1,5 @@
 from concurrent.futures import ThreadPoolExecutor
+import math
 import os
 from threading import Barrier, Event
 import time
@@ -445,6 +446,18 @@ def test_histogram(self):
         self.assertEqual(3, self.registry.get_sample_value('h_count'))
         self.assertEqual(float("inf"), self.registry.get_sample_value('h_sum'))
 
+    def test_histogram_nan(self):
+        self.histogram.observe(2)
+        self.histogram.observe(float('nan'))
+        self.histogram.observe(3)
+
+        self.assertEqual(0, self.registry.get_sample_value('h_bucket', {'le': '1.0'}))
+        self.assertEqual(1, self.registry.get_sample_value('h_bucket', {'le': '2.5'}))
+        self.assertEqual(2, self.registry.get_sample_value('h_bucket', {'le': '5.0'}))
+        self.assertEqual(3, self.registry.get_sample_value('h_bucket', {'le': '+Inf'}))
+        self.assertEqual(3, self.registry.get_sample_value('h_count'))
+        self.assertTrue(math.isnan(self.registry.get_sample_value('h_sum')))
+
     def test_histogram_not_observable(self):
         """.observe() must fail if the Summary is not observable."""
         assert_not_observable(self.labels.observe, 1)

tests/test_multiprocess.py

@@ -1,4 +1,5 @@
 import glob
+import math
 import os
 import shutil
 import tempfile
@@ -109,6 +110,18 @@ def test_histogram_adds(self):
         self.assertEqual(3, self.registry.get_sample_value('h_sum'))
         self.assertEqual(2, self.registry.get_sample_value('h_bucket', {'le': '5.0'}))
 
+    def test_histogram_nan(self):
+        h1 = Histogram('h', 'help', registry=None)
+        values.ValueClass = MultiProcessValue(lambda: 456)
+        h2 = Histogram('h', 'help', registry=None)
+        h1.observe(float('nan'))
+        h2.observe(2)
+
+        self.assertEqual(1, self.registry.get_sample_value('h_bucket', {'le': '5.0'}))
+        self.assertEqual(2, self.registry.get_sample_value('h_bucket', {'le': '+Inf'}))
+        self.assertEqual(2, self.registry.get_sample_value('h_count'))
+        self.assertTrue(math.isnan(self.registry.get_sample_value('h_sum')))
+
     def test_gauge_all(self):
         g1 = Gauge('g', 'help', registry=None)
         values.ValueClass = MultiProcessValue(lambda: 456)