[None][infra] Unwaive 5 disagg gen only cases

[chenfeiz0326]

Description

Removes five perf-sanity waivers from tests/integration/test_lists/waives.txt,
spanning two bugs.

https://nvbugs/6535767 — fix merged

The ValueError: The V2 Mamba state pool has only X slots but needs at least N ...
startup failure was fixed by #17163 ("Drain in-flight requests before clearing
the KV cache reuse state"), merged to main on 2026-08-06 as b7a9d6f5d7.
That PR did not clean up the waivers it obsoleted, which were added in
post-merge 2869 (#17041):

• perf/test_perf_sanity.py::test_e2e[aggr_upload-qwen3_5_397b_fp4_blackwell-qwen3_5_397b_fp4_dep8_8k1k]
• perf/test_perf_sanity.py::test_e2e[aggr_upload-qwen3_5_397b_fp4_blackwell-qwen3_5_397b_fp4_dep8_mtp3_8k1k]
• perf/test_perf_sanity.py::test_e2e[aggr_upload-qwen3_5_397b_fp4_blackwell-qwen3_5_397b_fp4_tp4_8k1k]

https://nvbugs/6490049 — verified case by case

This bug tracks an intermittent gen_only harness metric-extraction failure
(missing 'prev_device_step_time' / missing Average Per Iter Device Step Time)
across 13 cases, and is still open. Its waivers are being retired individually
as each case is verified — the same pattern as #16920 ("Unwaive GB300 Kimi
disagg e2e", merged 2026-08-06), which unwaived one case while the bug
remained open. This PR retires two more:

• perf/test_perf_sanity.py::test_e2e[disagg_upload-gen_only-gb200_deepseek-v32-fp4_8k1k_con4096_ctx1_dep4_gen1_dep32_eplb256_mtp0_ccb-NIXL]
• perf/test_perf_sanity.py::test_e2e[disagg_upload-gen_only-gb200_gpt-oss-120b-fp4_8k1k_con4_ctx1_tp1_gen1_tp4_eplb0_mtp0_ccb-NIXL]

The remaining 6490049 waiver (gb300_kimi-k25-thinking-fp4_8k1k_con1024_...)
is intentionally left in place.

Test Coverage

Waiver removal only — no source changes. All five cases remain listed in their
test-db lists, so removing the waivers re-enables them rather than leaving them
unrun:

Case	Test-db list
3× qwen3_5_397b_fp4	l0_b200_multi_gpus_perf_sanity.yml
deepseek-v32-fp4_8k1k_con4096	l0_gb200_multi_nodes_perf_sanity_ctx1_node1_gpu4_gen1_node8_gpu32.yml
gpt-oss-120b-fp4_8k1k_con4	l0_gb200_multi_nodes_perf_sanity_ctx1_node1_gpu1_gen1_node1_gpu4.yml

PR Checklist

• I have read the Contribution Guidelines
• Commit is DCO signed-off
• Each waiver removal references the fix or verification that justifies it

---

🤖 Generated with Claude Code

Dev Engineer Review

• Removed three qwen3_5_397b_fp4 perf-sanity waivers from tests/integration/test_lists/waives.txt.
• The removals match the fix for nvbug 6535767 from PR #17163.
• Five other waivers for nvbug 6535767 remain unchanged.
• The waiver file change is correctly scoped and contains no public API changes.

QA Engineer Review

• No test-db/ or qa/ files were modified.
• Removed three B200 multi-GPU perf-sanity waiver entries.
• The three cases remain in the test database for post-merge re-enablement.
• CBTS coverage data is unavailable.

Verdict: needs follow-up

[coderabbitai]

Walkthrough

The change removes three skipped performance sanity test entries for Qwen3.5 397B FP4 Blackwell configurations from the integration waiver list.

Changes

Performance waiver cleanup

Layer / File(s)	Summary
Remove Qwen3.5 performance waivers tests/integration/test_lists/waives.txt	Removes skip entries for dependency, MTP, and tensor-parallel configurations.

Estimated code review effort: 1 (Trivial) | ~2 minutes

Possibly related PRs

• NVIDIA/TensorRT-LLM#17350: Removes other obsolete performance-test waiver entries.
• NVIDIA/TensorRT-LLM#17386: Removes skipped performance-test waivers for different configurations.
• NVIDIA/TensorRT-LLM#17418: Removes obsolete skipped tests for other Qwen3.5 configurations.

Suggested labels: ci: full pre-merge approved

Suggested reviewers: qijune, xinhe-nv

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
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 describes two real waiver removals but omits the three Qwen3.5 waivers identified as the primary change.
Description check	✅ Passed	The description includes the required sections and clearly explains the five waiver removals, justification, test coverage, and checklist.

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

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

[chenfeiz0326]

/bot run --disable-fail-fast --stage-list "DGX_B200-8_GPUs-PyTorch-PerfSanity-Post-Merge-2,DGX_B200-8_GPUs-PyTorch-PerfSanity-Post-Merge-4,GB200-36_GPUs-9_Nodes-PyTorch-Disagg-PerfSanity-CTX1-NODE1-GPU4-GEN1-NODE8-GPU32-Post-Merge-5,GB200-8_GPUs-2_Nodes-PyTorch-Disagg-PerfSanity-CTX1-NODE1-GPU1-GEN1-NODE1-GPU4-Post-Merge-3"

[tensorrt-cicd]

PR_Github #64842 [ run ] triggered by Bot. Commit: 0e8d1b9 Link to invocation

[tensorrt-cicd]

PR_Github #64842 [ run ] completed with state FAILURE. Commit: 0e8d1b9
/LLM/main/L0_MergeRequest_PR pipeline #52683 (Partly Tested) completed with status: 'UNSTABLE'

CI Report

⚠️ Multi-GPU Label Required:
Multi-GPU tests require the ci: full pre-merge approved label on this PR. Ask a member of NVIDIA/trt-llm-ci-approvers to add the label, then re-trigger CI with the same bot command (no rebase needed).

⚠️ 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

Link to invocation

[chenfeiz0326]

/bot skip --comment "Only unwaive perf tests, no need to run the whole CI pipeline"

[tensorrt-cicd]

PR_Github #64927 [ skip ] triggered by Bot. Commit: 0e8d1b9 Link to invocation

[tensorrt-cicd]

PR_Github #64927 [ skip ] completed with state SUCCESS. Commit: 0e8d1b9
Skipping testing for commit 0e8d1b9

Link to invocation