[None][test] Don't force disable_overlap_scheduler for perf-sanity ctx_only tests

[chenfeiz0326]

Description

The ctx_only mode of the perf-sanity suite (aggr[_upload]-ctx_only-<disagg config>)
builds an aggregated server config out of the disaggregated yaml's worker_config.ctx
section. While doing so it unconditionally overwrote the ctx worker's
disable_overlap_scheduler with True:

# Disable overlap scheduler for ctx_only
ctx_config["disable_overlap_scheduler"] = True

The overlap scheduler works fine for a context-only run, so this override is removed and
the value comes from the yaml like every other worker_config.ctx knob.

90 of the 96 CI disagg configs set ctx disable_overlap_scheduler: false (since #17390),
so their e2e / gen_only tests already run the ctx worker with the overlap scheduler
enabled while the ctx_only test generated from the same config file ran it disabled. The
6 gb300_nemotron-ultra-v3-fp4_* configs that set ctx disable_overlap_scheduler: true
keep running with the overlap scheduler off.

When the key is absent from a yaml, ServerConfig already defaults it to False, which
matches both the LLM API default and the aggregated tests. Spec-decode modes that cannot
support the overlap scheduler are still handled by the runtime: py_executor_creator.py
sets disable_overlap_scheduler = True itself and logs a warning when
spec_config.spec_dec_mode.support_overlap_scheduler() is false.

Impact on baselines

This changes the measured configuration of the existing ctx_only cases (90 of 96 flip from
overlap-off to overlap-on), so their perf baselines are expected to shift once after merge.

Test Coverage

Harness-only change. Verified by driving _parse_disagg_config_file in ctx_only mode
over all 96 configs in tests/scripts/perf-sanity/disaggregated/ and inspecting the
generated extra-llm-api-config content:

ctx value in yaml	emitted disable_overlap_scheduler	configs
false	false	90
true	true	6
key absent	key omitted (engine default)	synthetic case

cache_transceiver_config is still dropped in every case.

PR Checklist

• PR description clearly explains what and why.

• PR follows the TRT-LLM coding guidelines.

• Harness-only change; verified against all 96 CI disagg configs (see Test Coverage). No API change.

• No new dependencies.

• No documentation change needed.

• Please check this after reviewing the above items as appropriate for this PR.

Note on pre-commit: ruff format --check, ruff check and the whitespace/EOL checks were
run against the modified file and pass. The full pre-commit run --all-files could not be
executed in this environment (hook virtualenv creation fails on a missing platformdirs),
so please flag anything the CI formatting stage catches.

GitHub Bot Help

To see a list of available CI bot commands, please comment /bot help.

Dev Engineer Review

• ctx_only now honors worker_config.ctx.disable_overlap_scheduler.
• The change removes the unconditional True override.
• Missing YAML values use the ServerConfig default of False.
• Explicit true values remain unchanged.
• The scope is limited to the intended performance-sanity test behavior.
• No API, error-handling, or test-list changes were identified.
• Review verdict: sufficient.

QA Engineer Review

• Modified test function: ctx_only.
• The test function is covered by the existing CI disaggregated configurations.
• Validation covered all 96 configurations.
• Ninety configurations now use the default or YAML-defined value.
• Six configurations explicitly retain disable_overlap_scheduler: true.
• Spec-decode runtime handling remains unchanged.
• Review verdict: sufficient.

[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: 361b6105-152a-4055-b38a-de2cfbcc798c

📥 Commits

Reviewing files that changed from the base of the PR and between c96a463 and 0df65b2.

📒 Files selected for processing (1)

• tests/integration/defs/perf/test_perf_sanity.py

💤 Files with no reviewable changes (1)

• tests/integration/defs/perf/test_perf_sanity.py

Included review availability: Your plan provides up to 12 included reviews per hour; 10 remain after this review.

---

Walkthrough

The performance test no longer forces disable_overlap_scheduler to True when ctx_only is enabled. ServerConfig now uses the YAML value, or False when the setting is absent.

Changes

ctx_only configuration handling

Layer / File(s)	Summary
Preserve YAML scheduler setting tests/integration/defs/perf/test_perf_sanity.py	The ctx_only path passes the YAML-defined disable_overlap_scheduler value to ServerConfig. YAML omission retains the False default.

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

Merge Risk: ⚪ Minimal · up to 0df65

This localized harness change lets ctx-only performance tests honor the scheduler setting from their configuration; no actionable merge-blocking risk remains after normal checks and review.

Suggested reviewers: bowenfu

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Docstring Coverage	✅ Passed	Docstring coverage is 100.00% which is sufficient. The required threshold is 80.00%. Docstring coverage is scoped to functions touched by this diff. Analyzed 1 functions across 1 files.
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 clearly identifies the test change: perf-sanity ctx_only tests no longer force disable_overlap_scheduler. It uses the expected ticket and type format.
Description check	✅ Passed	The description includes the required Description, Test Coverage, and PR Checklist sections. It explains the issue, the solution, validation across all 96 configurations, baseline impact, and checklis…

Full details: Description check

Explanation

The description includes the required Description, Test Coverage, and PR Checklist sections. It explains the issue, the solution, validation across all 96 configurations, baseline impact, and checklist status.

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

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

[fredricz-20070104]

Approved.

[chenfeiz0326]

/bot skip --comment "Only update config for ctx only perf tests, no need to run the whole CI pipeline"

[tensorrt-cicd]

PR_Github #69895 [ skip ] triggered by Bot. Commit: 0df65b2 Link to invocation

[tensorrt-cicd]

PR_Github #69895 [ skip ] completed with state SUCCESS. Commit: 0df65b2
Skipping testing for commit 0df65b2

Link to invocation