[https://nvbugs/6721558][fix] Only record first token time when collecting perf metrics

[trtllm-agent]

Summary

• Root cause: e11905f5c6 swapped the stateless get_steady_clock_now_in_seconds() for the instance-state self._adjusted_steady_clock.now() at three streaming first-token sites without applying the self._collect_perf_metrics gate it added at the equivalent _extract_metrics read.
• Fix: Route all three streaming sites through a new _record_server_first_token_time() helper carrying the same _collect_perf_metrics gate as line 1762.
• Original test: pytest tests/unittest/llmapi/test_llm.py::test_openai_completion_list_prompt_stream_reuses_stream_metadata -v
• Automated fix generated by repair-bot

Test plan

• Verify fix on the same GPU type as the original failure
• Check for regressions in related tests

Links

• Bug: https://nvbugs/6721558

Reproduction comparison

• Failed commit: repro_on_failed_commit at 53bb31f
  Signature: AttributeError: 'OpenAIServer' object has no attribute '_adjusted_steady_clock'
• ToT: repro_on_tot at 17803f6
  Signature: AttributeError: 'OpenAIServer' object has no attribute '_adjusted_steady_clock'
• Signature relation: needs_agent

Dev Engineer Review

• Added _record_server_first_token_time() to centralize first-token timestamp recording.
• The helper checks _collect_perf_metrics, request availability, and existing timestamp state before recording.
• Chat, completion, and Harmony streaming paths now use the gated helper.
• Non-streaming metric extraction uses the helper as a fallback.
• The change prevents ungated _adjusted_steady_clock access and fixes the reported completion-streaming AttributeError regression.
• No public API, configuration, or test-list changes were identified.
• The implementation is consistent across streaming paths and avoids duplicate timestamp writes.
• Verdict: sufficient.

QA Engineer Review

No test files are touched.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Enterprise

Run ID: 8439646d-48d6-4cce-b538-ddc1def831c9

📥 Commits

Reviewing files that changed from the base of the PR and between fc8969e and 60086d4.

📒 Files selected for processing (1)

• tensorrt_llm/serve/openai_server.py

Included review availability: Your plan provides up to 12 included reviews per hour; 9 remain after this review.

---

Walkthrough

The server centralizes guarded first-token timestamp recording. Chat, completion, Harmony, and non-streaming metric paths now use the shared helper.

Changes

First-token metrics

Layer / File(s)	Summary
Timestamp helper and request-path integrations tensorrt_llm/serve/openai_server.py	The new helper records the first-token timestamp only when metrics are enabled, a request exists, and no timestamp is stored. Streaming and non-streaming paths use the helper.

Estimated code review effort: 2 (Simple) | ~10 minutes

Merge Risk: ⚪ Minimal · up to 60086

This change prevents streaming requests from accessing the performance clock when metrics are disabled while preserving first-token timing when metrics are enabled. The affected request paths are consistently routed through the guarded helper, with no remaining merge-blocking risk identified.

Suggested reviewers: bowenfu

🚥 Pre-merge checks | ✅ 4 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 22.22% which is insufficient. The required threshold is 80.00%. Docstring coverage is scoped to functions touched by this diff. Analyzed 9 functions across 1 files.	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (4 passed)

Check name	Status	Explanation
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Title check	✅ Passed	The title follows the required NVBugs and type format and clearly describes the fix to gate first-token timing on performance metrics.
Description check	✅ Passed	The description explains the root cause, fix, affected behavior, bug link, reproduction details, and test plan. It uses Summary and Test plan headings instead of the template's Description and Test Co…

• Fix all pre-merge checks with AI

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

_{Comment @coderabbitai help to get the list of available commands.}

[tongyuantongyu]

/bot run

[tensorrt-cicd]

PR_Github #72083 [ run ] triggered by Bot. Commit: 60086d4 Link to invocation

[tensorrt-cicd]

PR_Github #72083 [ run ] completed with state FAILURE. Commit: 60086d4
/LLM/main/L0_MergeRequest_PR pipeline #59136 completed with status: 'FAILURE'

CI Report

⚠️ Action Required:

• Please check the failed tests and fix your PR
• If you cannot view the failures, ask the CI triggerer to share details
• Once fixed, request an NVIDIA team member to trigger CI again

CI Agent Failure Analysis

Link to invocation

[JunyiXu-nv]

Original issue has been fixed and test unwaived by #18686