docs(profiler): clarify when planner-profile-data ConfigMap is emitted [DYN-2751]#8486
Conversation
Document the conditions under which `planner-profile-data-XXXX` is generated, and flag the agg + throughput-scaling gap with a TODO: - assemble_final_config: the profile-data ConfigMap only lands when the picked config is disagg AND NPZ sweep ran (thorough only); rapid deployments never emit it. - build_aic_interpolation_spec: the spec carries prefill_pick + decode_pick only, so agg picks on rapid have no AIC fallback and the planner relies on DYN_BENCHMARK_MODE runtime endpoint only. - profile_sla.py: TODO at the agg-skip site pointing at the wider fix (extend AICInterpolationSpec + run_interpolation with an agg_pick). No behavior change — docstrings and TODO only. Signed-off-by: hongkuanz <hongkuanz@nvidia.com>
|
No actionable comments were generated in the recent review. 🎉 ℹ️ Recent review info⚙️ Run configurationConfiguration used: Path: .coderabbit.yaml Review profile: CHILL Plan: Pro Run ID: 📒 Files selected for processing (2)
WalkthroughTwo profiler files updated to clarify interpolation handling: a conditional branch added to skip interpolation building for aggregated configurations, and docstrings expanded to document when interpolation-data ConfigMaps are emitted and the coverage limitations of interpolation specs. Changes
Estimated code review effort🎯 1 (Trivial) | ⏱️ ~5 minutes 🚥 Pre-merge checks | ✅ 5✅ Passed checks (5 passed)
✏️ Tip: You can configure your own custom pre-merge checks in the settings. Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
Summary
Documents the conditions under which
planner-profile-data-XXXXis emitted, and flags the agg + throughput-scaling gap that showed up in QA ticket DYN-2751.assemble_final_config— profile-data ConfigMap only lands when the picked config is disagg AND NPZ sweep ran (thorough only). Rapid deployments never emit it.build_aic_interpolation_spec— spec only carriesprefill_pick+decode_pick, so agg picks on rapid have no AIC bootstrap fallback. Planner currently relies solely onDYN_BENCHMARK_MODEruntime endpoint.profile_sla.py—TODOat the agg-skip site pointing at the wider fix (extendAICInterpolationSpec+run_interpolationwith anagg_pick).No behavior change — docstrings and
TODOonly. The bigger rework (plumb agg through the profiler + planner AIC handoff + mocker worker flags) is deliberately out of scope for a release QA bug fix.Related
release/1.1.0after this lands.Test plan
pre-commit runon the two touched files passes (isort / black / flake8 / ruff / codespell).Summary by CodeRabbit
Documentation
Chores