What happened
I was exploring the metrics SDK and found that ExplicitBucketHistogramAggregation accepts unsorted boundary lists without any validation or warning. Internally, aggregate() calls bisect_left(self._boundaries, value) which requires sorted input when boundaries are unsorted, measurements land in the wrong buckets silently.
Reproducer
from opentelemetry.sdk.metrics import MeterProvider
from opentelemetry.sdk.metrics.export import InMemoryMetricReader
from opentelemetry.sdk.metrics.view import View
from opentelemetry.sdk.metrics._internal.aggregation import ExplicitBucketHistogramAggregation
reader = InMemoryMetricReader()
provider = MeterProvider(
metric_readers=[reader],
views=[View(
instrument_name="latency",
aggregation=ExplicitBucketHistogramAggregation(boundaries=[100, 10, 50]), # unsorted
)]
)
meter = provider.get_meter("test")
hist = meter.create_histogram("latency")
hist.record(25)
hist.record(75)
hist.record(150)
data = reader.get_metrics_data()
pt = data.resource_metrics[0].scope_metrics[0].metrics[0].data.data_points[0]
print("bounds:", pt.explicit_bounds) # (100, 10, 50)
print("counts:", pt.bucket_counts) # (0, 0, 1, 2) <-- wrong, 2/3 values misplaced
# with sorted [10, 50, 100] the correct output is (0, 1, 1, 1)Root cause
aggregation.py#L478-L511 stores boundaries as-is and passes them directly to bisect_left, which assumes sorted input. No validation happens anywhere in the chain.
How other SDKs handle it
Proposed fix
A 2-3 line check in ExplicitBucketHistogramAggregation.__init__ that validates boundaries are strictly increasing. Happy to open a PR if this looks good.
What happened
I was exploring the metrics SDK and found that
ExplicitBucketHistogramAggregationaccepts unsorted boundary lists without any validation or warning. Internally,aggregate()callsbisect_left(self._boundaries, value)which requires sorted input when boundaries are unsorted, measurements land in the wrong buckets silently.Reproducer
Root cause
aggregation.py#L478-L511stores boundaries as-is and passes them directly tobisect_left, which assumes sorted input. No validation happens anywhere in the chain.How other SDKs handle it
IllegalArgumentException("Bucket boundaries must be in increasing order")inExplicitBucketHistogramUtils.java#L55-L72"non-monotonic boundaries"inaggregation.goProposed fix
A 2-3 line check in
ExplicitBucketHistogramAggregation.__init__that validates boundaries are strictly increasing. Happy to open a PR if this looks good.