[mccalpink]

• I have added tests that cover my changes.
• If adding a new instrumentation or changing an existing one, I've added screenshots from some observability platform showing the change. (N/A: SDK metric view, no instrumentation change; see the test output below.)
• PR name follows conventional commits format: fix(traceloop-sdk): ....
• (If applicable) I have updated the documentation accordingly. (N/A)

Summary

#3129 (following #3104) dropped the LLM_OPERATION_DURATION View from metric_views() so that the histogram uses the OpenTelemetry SDK default boundaries. That default, [0, 5, 10, 25, ..., 10000], is sized for milliseconds, while gen_ai.client.operation.duration is recorded in seconds (every one of the 14 instrumentations creates it with unit="s"). Almost every LLM call therefore lands in the first bucket, and e.g. histogram_quantile(0.95, ...) is a constant ~4.75 s. This PR adds the View back with the GenAI semconv boundaries [0.01, 0.02, ..., 81.92] seconds. The le labels of this already exported histogram will change, which is an expected series change.

Root cause

Before #3129 the boundaries were swapped between the two histograms: LLM_TOKEN_USAGE had the sub-second [0.01 ... 81.92] list and LLM_OPERATION_DURATION had the token-sized [1, 4, 16, ...] list (#3104). #3129 put the token boundaries from opentelemetry-python-contrib on LLM_TOKEN_USAGE, and deliberately removed the LLM_OPERATION_DURATION View to fall back to the SDK default (see the #3129 description and the discussion in #3104). That was a reasonable way to fix the swap, but the SDK default is a millisecond-oriented list, which does not fit a histogram whose unit is s; the GenAI semconv for gen_ai.client.operation.duration recommends [0.01 ... 81.92]. This PR restores only the duration View; the token boundaries are untouched.

Fix

Add a View(instrument_name=Meters.LLM_OPERATION_DURATION, aggregation=ExplicitBucketHistogramAggregation([0.01, ..., 81.92])) to metric_views() in packages/traceloop-sdk/traceloop/sdk/metrics/metrics.py. The token-usage boundaries are unchanged. New tests in packages/traceloop-sdk/tests/test_metric_views.py build a MeterProvider with InMemoryMetricReader + metric_views() and assert explicit_bounds for both histograms.

How I reproduced it

With only the new test added (no fix), on main:

cd packages/traceloop-sdk
uv run pytest tests/test_metric_views.py -v

test_operation_duration_uses_semconv_second_buckets fails (bounds are the SDK defaults instead of 0.01 ... 81.92); test_token_usage_uses_semconv_token_buckets passes (it guards the token boundaries against a future swap).

How I verified it

In packages/traceloop-sdk:

• uv run pytest tests/test_metric_views.py -v: 2 passed (after the fix).
• uv run pytest tests/: 446 passed.
• uv run ruff check .: All checks passed.
• uv run mypy traceloop/sdk: no issues found in 50 source files.

Closes #4702

This PR was prepared with AI assistance (Claude Code). I reproduced the bug, reviewed the change, and ran the tests myself.

Summary by CodeRabbit

• New Features
  • Operation-duration metrics now use explicit histogram boundaries, ranging from 0.01 to 81.92, for more consistent duration reporting.

[CLAassistant]

All committers have signed the CLA.

[coderabbitai]

No actionable comments were generated in the recent review.

Walkthrough

metric_views() now adds explicit histogram boundaries for LLM_OPERATION_DURATION. New tests collect and verify the duration and token-usage histogram boundaries.

Changes: Duration histogram boundaries

Layer / File(s)	Summary
Configure and verify histogram boundaries packages/traceloop-sdk/traceloop/sdk/metrics/metrics.py, packages/traceloop-sdk/tests/test_metric_views.py	metric_views() adds explicit duration histogram boundaries from 0.01 to 81.92. Tests collect the instrument bounds and compare duration and token-usage bounds with their expected tuples.

Estimated code review effort: 2 (Simple) | ~8 minutes

Change: Bug fix · Severity of issue fixed: Medium

Merge Risk: Minimal · up to ecb1b

The change applies GenAI-appropriate histogram boundaries to the operation-duration metric, so duration observations fall into meaningful buckets. The exported bucket labels change, which is intended. No actionable merge-blocking risk remains.

Pre-merge checks: 4

Check name	Status	Explanation
Description Check		Check skipped - CodeRabbit’s high-level summary is enabled.
Title check		The title names the main change: restoring the gen_ai.client.operation.duration histogram view in the traceloop-sdk. The raw summary confirms metric_views() now adds an explicit-bucket view for Meters…
Linked Issues check		Issue #4702 requires the GenAI semconv boundaries [0.01, 0.02, 0.04, ..., 40.96, 81.92] for gen_ai.client.operation.duration. At the reviewed head, packages/traceloop-sdk/traceloop/sdk/metrics/metrics…
Out of Scope Changes check		The PR changes two files. metrics.py adds only the LLM_OPERATION_DURATION View, which is the subject of #4702. The other Views (token usage, Pinecone duration, Pinecone scores) are present at the head…

Full details: Title check

Explanation

The title names the main change: restoring the gen_ai.client.operation.duration histogram view in the traceloop-sdk. The raw summary confirms metric_views() now adds an explicit-bucket view for Meters.LLM_OPERATION_DURATION with boundaries from 0.01 to 81.92, matching the objectives. The title is specific and uses a conventional fix() prefix. It omits the bucket-boundary detail, which is acceptable.

Full details: Linked Issues check

Explanation

Issue #4702 requires the GenAI semconv boundaries [0.01, 0.02, 0.04, ..., 40.96, 81.92] for gen_ai.client.operation.duration. At the reviewed head, packages/traceloop-sdk/traceloop/sdk/metrics/metrics.py metric_votes() includes a View for Meters.LLM_OPERATION_DURATION with ExplicitBucketHistogramAggregation using exactly these 14 boundaries, which matches the issue. The token-usage View is unchanged. The new test packages/traceloop-sdk/tests/test_metric_views.py asserts explicit_bounds for both histograms, covering the requirement with automated tests. The author reports the test suite, ruff, and mypy results; I did not run them. The metrics.py diff against the merge base could not be read, so the change was verified from the head file content only.

Full details: Out of Scope Changes check

Explanation

The PR changes two files. metrics.py adds only the LLM_OPERATION_DURATION View, which is the subject of #4702. The other Views (token usage, Pinecone duration, Pinecone scores) are present at the head and are not altered by this PR per the summary. test_metric_views.py is a new test file that directly verifies the new View and the unchanged token-usage View. No unrelated changes were identified.

Finishing Touches

Generate unit tests (beta)

• Create a new PR

• Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

Recent review info

Run configuration

• Source mode: Auto
• Repository: Unavailable
• Branch: Unavailable
• File: No configuration file from this repository applied
• Commit: Unavailable
• Layers: defaults
• Review profile: CHILL
• Plan: Advanced
• Run ID: 7abd367a-38f5-4c29-a2ea-3e78d89c8b32

Commits

Reviewing files that changed from the base of the PR and between e3d651a and ecb1b3c.

Files selected for processing: 2

• packages/traceloop-sdk/tests/test_metric_views.py
• packages/traceloop-sdk/traceloop/sdk/metrics/metrics.py

Included review availability: This review used your included allowance. Your plan provides up to 8 included reviews per hour; 7 remain after this review.