Release v0.9.0

[liu-cong]

---

name: New Release
about: Propose a new release
title: Release v0.9.0
labels: kind/release
assignees: ''

---

---

• Introduction
• Prerequisites
• Release Process
• Announce the Release
• Final Steps

Introduction

This document defines the process for releasing llm-d-router.

Prerequisites

1. Permissions to push to the llm-d-router repository.

2. Membership in the @llm-d/router-release-managers team. Tag protection on
   refs/tags/v* restricts who can push release tags, which is what triggers
   the release build.

3. Choose whether you are releasing a release candidate or an official release, and set the environment variables accordingly:

   • For a Release Candidate (e.g. v0.9.0-rc.1):

     export VERSION=v0.9.0-rc.1
     export BRANCH_VERSION=0.9
     export REMOTE=origin

   • For an Official Release (e.g. v0.9.0):

     export VERSION=v0.9.0
     export BRANCH_VERSION=0.9
     export REMOTE=origin

4. (Optional) If the latency predictor release version does not align with the router version, also set the expected tag (refer to the latency predictor releases to find the latest valid release tag):

   export LATENCY_PREDICTOR_TAG=v0.8.0-rc.1

5. If needed, clone the llm-d-router repo.

   git clone -o ${REMOTE} git@github.com:llm-d/llm-d-router.git

Release Process

Create or Checkout branch

1. If you already have the repo cloned, ensure it's up-to-date and your local branch is clean.

2. Release Branch Handling:

   • For a Release Candidate:
     Create a new release branch from the main branch. The branch should be named release-${BRANCH_VERSION}, for example, release-0.9:

     git checkout -b release-${BRANCH_VERSION}

   • For a Major, Minor or Patch Release:
     A release branch should already exist. In this case, check out the existing branch:

     git checkout release-${BRANCH_VERSION} ${REMOTE}/release-${BRANCH_VERSION}

3. By default, LATENCY_PREDICTOR_TAG in the Makefile automatically resolves to the router release version (${BUILD_REF}). If the latency predictor tag does not align with the router version, update the default value of LATENCY_PREDICTOR_TAG in the Makefile to match your exported ${LATENCY_PREDICTOR_TAG}.
   Commit the change (if modified):

   # Update LATENCY_PREDICTOR_TAG ?= vX.Y.Z in Makefile
   git commit -a -s -m "release: set LATENCY_PREDICTOR_TAG to ${LATENCY_PREDICTOR_TAG}"

4. Push your release branch to the llm-d-router remote.

   git push ${REMOTE} release-${BRANCH_VERSION}

Tag commit and trigger image build

1. Tag the head of your release branch with the version:

   git tag -s -a ${VERSION} -m "llm-d-router ${VERSION} Release"

2. Push the tag to the llm-d-router repo:

   git push ${REMOTE} ${VERSION}

3. Pushing the tag triggers CI action to build and publish the EPP image (ghcr.io/llm-d/llm-d-router-endpoint-picker) and sidecar image (ghcr.io/llm-d/llm-d-router-disagg-sidecar) to the ghcr registry.

4. Verify the CI release workflow completed successfully before proceeding.

5. Test the steps in the tagged quickstart guide after the PR merges.

Create the release!

1. Create a new release:

   1. Choose the tag that you created for the release.

   2. Use the tag as the release title, e.g. v0.1.0.

   3. Click "Generate release notes" to auto-populate the list of PRs and contributors.

   4. Summarize the release notes using an LLM of your choice (e.g., Gemini, Copilot, ChatGPT). Provide the newly compiled release notes block from RELEASE-NOTES.md (or the unreleased fragments in release-notes.d/unreleased/) with the following prompt:

      Please summarize these release notes into three clear sections:
      1. Highlights (key features, performance wins, bug fixes)
      2. Upgrade Steps & Deprecations (configuration changes, deprecated flags/metrics)
      3. Known Issues (if any, otherwise omit)
      

      Review the generated content, edit it if necessary to ensure accuracy, and then copy and prepend this summary at the very top of the release description box on GitHub.

   5. If this is a release candidate, select the "This is a pre-release" checkbox.

2. If you find any bugs in this process, create an issue.

Announce the Release

Use the following steps to announce the release.

1. Generate the announcement email content by running the following block in your terminal (make sure ${VERSION} is set in your current shell):

   cat <<EOF
   Subject: [ANNOUNCE] llm-d-router ${VERSION} is released
   
   Hi all,
   
   We are pleased to announce the release of llm-d-router ${VERSION}!
   
   ### Container Images
   * Endpoint Picker: ghcr.io/llm-d/llm-d-router-endpoint-picker:${VERSION}
   * Disaggregated Sidecar: ghcr.io/llm-d/llm-d-router-disagg-sidecar:${VERSION}
   
   ### Helm Charts (OCI)
   * Standalone Chart: oci://ghcr.io/llm-d/charts/llm-d-router-standalone (version ${VERSION})
   * Gateway Chart: oci://ghcr.io/llm-d/charts/llm-d-router-gateway (version ${VERSION})
   
   ### Release Notes
   For more details, please see the GitHub release notes: https://github.com/llm-d/llm-d-router/releases/tag/${VERSION}
   EOF

2. Copy the generated subject and body, and send an email to llm-d-contributors@googlegroups.com.

3. Add a link to the final release in this issue.

4. Close this issue.

[liu-cong]

Currently blocked by missing latency predictor release: https://github.com/llm-d/llm-d-router/actions/runs/27630292613/job/81724468433

[liu-cong]

llm-d-router v0.9.0-rc.2 is functionally correct with no performance regressions on benchmarked guides. All 8 guides pass documentation, ecosystem, and dry-run verification. 7/8 pass E2E deployment. 4 benchmarked on matching hardware — all within or exceeding documented baselines.

Guide	E2E	Benchmark Performance vs Baseline
optimized-baseline	PASS (8r/16 H100)	Throughput ±5% at all rates, TTFT 1-40% better. NO REGRESSION
precise-prefix-cache	PASS (8r/16 H100)	Throughput +1.0%, TTFT -0.8% better. NO REGRESSION
predicted-latency	PASS (10r/20 H100)	Output tok/s exact match at C=30. NO REGRESSION
tiered-prefix-cache	PASS (8r/16 H100)	Matches at 5-10 QPS; config mismatch at 20+ QPS. NO REGRESSION
pd-disaggregation	PASS (B200)	Skipped — needs H200+InfiniBand
wide-ep-lws (default)	PASS (with workaround)	4,201 tok/s at C=500 after targetPorts workaround. Default guide uses hybrid-lb (single port) but router config has 8 ports (for dp-aware)
wide-ep-lws (dp-aware)	PASS (32 H200, standalone + gateway)	Multi-port DP-aware routing verified in both modes. 32 rank endpoints discovered. Image llm-d-cuda:v0.8.0 not yet published (used dev:latest)
agentic-serving	SKIPPED (no TPU v7)	Static verification (T1-T3.1) only
multimodal/opt-baseline	PASS (2r)	No benchmark steps in guide

[liu-cong]

Checklist before cutting the release

• Test well-lit path guides
• Cherry pick llm-d-kv-cache 0.9 release deps(go): bump llm-d-kv-cache to v0.9.0 #1691 (review)
• Cherry pick fix: resolve mismatched yq path for nested router subchart values #1683
• CP fix: bump golang.org/x/crypto to v0.52.0 to resolve Trivy CVEs #1720
• CP ci: publish CRD manifests to release assets #1713

[ahg-g]

looks good to me, thanks for the thorough work here @liu-cong

[liu-cong]

Need to cherry pick #1713 for a new patch release

[liu-cong]

Official release https://github.com/llm-d/llm-d-router/releases/tag/v0.9.0

[mtparet]

We used some containers v0.9.1 version but these have been removed..

[liu-cong]

We used some containers v0.9.1 version but these have been removed..

0.9.0 is the officially released and supported image. Please use the 0.9.0 container. Thanks.

@mtparet