[core][metric] improve histogram metrics midpoint calculation (ray-project#57948)

There’s a potential risk in the current midpoint calculation. It can be
wrong when values are negative.

Line 167: lower_bound + buckets[0] / 2.0
Line 171: (buckets[i] + buckets[i - 1]) / 2.0

I improved the formula and added a test to make sure it works.

Signed-off-by: justwph <2732352+wph95@users.noreply.github.com>
Signed-off-by: peterxcli <peterxcli@gmail.com>

Author: wph95

python/ray/_private/telemetry/open_telemetry_metric_recorder.py

@@ -164,7 +164,7 @@ def register_histogram_metric(
                 if i == 0:
                     lower_bound = 0.0 if buckets[0] > 0 else buckets[0] * 2.0
                     self._histogram_bucket_midpoints[name].append(
-                        lower_bound + buckets[0] / 2.0
+                        (lower_bound + buckets[0]) / 2.0
                     )
                 else:
                     self._histogram_bucket_midpoints[name].append(

python/ray/tests/test_open_telemetry_metric_recorder.py

@@ -119,6 +119,16 @@ def test_register_histogram_metric(
     )
     mock_logger_warning.assert_not_called()
 
+    mock_meter.create_histogram.return_value = NoOpHistogram(name="neg_histogram")
+    recorder.register_histogram_metric(
+        name="neg_histogram",
+        description="Histogram with negative first boundary",
+        buckets=[-5.0, 0.0, 10.0],
+    )
+
+    mids = recorder.get_histogram_bucket_midpoints("neg_histogram")
+    assert mids == pytest.approx([-7.5, -2.5, 5.0, 20.0])
+
 
 @patch("opentelemetry.metrics.set_meter_provider")
 @patch("opentelemetry.metrics.get_meter")