Skip to content

feat(stargate): add configurable Pulsar input TPS weights - #1002

Draft
barrygreengus wants to merge 2 commits into
mainfrom
codex/linf-240-windowed-tps-weight
Draft

feat(stargate): add configurable Pulsar input TPS weights#1002
barrygreengus wants to merge 2 commits into
mainfrom
codex/linf-240-windowed-tps-weight

Conversation

@barrygreengus

@barrygreengus barrygreengus commented Aug 19, 2026

Copy link
Copy Markdown
Contributor

TL;DR

Add an optional generation maximum input TPS measurement from Pylon through Stargate, and let Pulsar rendezvous routing select either the existing mean or that maximum as its capacity weight. Mean input TPS remains the default.

Additional Details

Pylon publishes the maximum already maintained by its generation-owned input TPS distribution. The value persists for that model generation and resets when the generation is replaced. There is no time window, bucket state, expiration path, or related runtime configuration.

Stargate sums the optional maximum only when every backend in the cluster supplies a valid positive finite value. Pulsar and Pulsar wait-and-widen accept last-mean-input-tps or max-input-tps for rendezvous weighting. Max mode does not silently fall back to mean when peak data is incomplete. Admission and work estimation continue using mean.

The ranking cache keys include the selected weight, so changing the unused statistic does not invalidate cached rankings. No dependencies or hash-version behavior changed.

For the Reviewer

Please focus on generation ownership and reset behavior, complete-only aggregation across mixed versions, and consistency between ranking cache invalidation and weight selection.

For QA

QA is not required beyond automated coverage. Verified with:

  • cargo fmt --all --check
  • cargo test -p pylon-lib --quiet
  • cargo test -p pylon --quiet
  • cargo test -p stargate --quiet
  • Bazel tests for pylon-lib, pylon, and stargate

Issues

Closes #1001

Checklist

  • I am familiar with the Contributing Guidelines.
  • I have signed off my commits for Developer Certificate of Origin compliance.
  • New or existing tests cover these changes.
  • The documentation is up to date with these changes.

Signed-off-by: Barry Greengus <bgreengus@nvidia.com>
@coderabbitai

coderabbitai Bot commented Aug 19, 2026

Copy link
Copy Markdown

Important

Review skipped

Draft detected.

Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Enterprise

Run ID: f1f5eb96-f3b1-4b44-943c-0557105078c2

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

  • 🔍 Trigger review

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

@github-actions

Copy link
Copy Markdown
Contributor

🛡️ CodeQL Analysis

🚨 Found 2 issue(s)

Severity Breakdown:

  • 🔴 Errors: 0
  • 🟡 Warnings: 0
  • 🔵 Notes: 0
📋 Top Issues

🔗 View full details in Security tab

🕐 Last updated: 2026-08-19 15:47:50 UTC | Commit: f00fe01

Signed-off-by: Barry Greengus <bgreengus@nvidia.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

Support generation peak throughput for Pulsar capacity weighting

1 participant