build(vllm): consume protocol SDKs from Buf

.cargo/config.toml

@@ -19,3 +19,6 @@ rustflags = ["-C", "target-cpu=neoverse-n1", "-C", "force-frame-pointers=yes", "
 [env]
 PCRE2_SYS_STATIC = "1"
 
+[registries.buf]
+index = "sparse+https://buf.build/gen/cargo/"
+credential-provider = "cargo:token"

.github/actions/compliance-extract/action.yml

@@ -61,6 +61,9 @@ inputs:
     description: 'SCCACHE region (= aws_default_region the build passed). See sccache_bucket.'
     required: false
     default: ''
+  buf_token:
+    description: 'Buf Schema Registry token for cache-missed Cargo rebuilds.'
+    required: true
   diff_base_sha:
     description: |
       Commit SHA of the baseline build to diff this build's OSRB CSV against
@@ -126,6 +129,7 @@ runs:
       env:
         SCCACHE_BUCKET: ${{ inputs.sccache_bucket }}
         SCCACHE_REGION: ${{ inputs.sccache_region }}
+        BUF_TOKEN: ${{ inputs.buf_token }}
         GIT_SHA: ${{ inputs.git_sha }}
         EPP_IMAGE: ${{ inputs.epp_image }}
       run: |
@@ -149,6 +153,12 @@ runs:
         # Only forward the credentials when S3 sccache is actually configured
         # (bucket set) — a bucket-less caller's build has no use for them.
         SECRET_ARGS=""
+        if [ -n "${BUF_TOKEN:-}" ]; then
+          SECRET_ARGS+=" --secret id=buf_token,env=BUF_TOKEN"
+        else
+          echo "::error::BUF_TOKEN is required for Cargo access to the Buf registry"
+          exit 1
+        fi
         if [ -n "${SCCACHE_BUCKET:-}" ]; then
           TOKEN_FILE="${AWS_WEB_IDENTITY_TOKEN_FILE:-}"
           if [ -n "$TOKEN_FILE" ] && [ -f "$TOKEN_FILE" ] && [ -n "${AWS_ROLE_ARN:-}" ]; then
@@ -344,6 +354,7 @@ runs:
       env:
         SCCACHE_BUCKET: ${{ inputs.sccache_bucket }}
         SCCACHE_REGION: ${{ inputs.sccache_region }}
+        BUF_TOKEN: ${{ inputs.buf_token }}
         GIT_SHA: ${{ inputs.git_sha }}
         EPP_IMAGE: ${{ inputs.epp_image }}
       run: |
@@ -353,6 +364,12 @@ runs:
         # as the compliance_artifact extract above). Same IRSA secrets too, so a
         # cache-missed stage rebuild can still authenticate sccache (see above).
         SECRET_ARGS=""
+        if [ -n "${BUF_TOKEN:-}" ]; then
+          SECRET_ARGS+=" --secret id=buf_token,env=BUF_TOKEN"
+        else
+          echo "::error::BUF_TOKEN is required for Cargo access to the Buf registry"
+          exit 1
+        fi
         if [ -n "${SCCACHE_BUCKET:-}" ]; then
           TOKEN_FILE="${AWS_WEB_IDENTITY_TOKEN_FILE:-}"
           if [ -n "$TOKEN_FILE" ] && [ -f "$TOKEN_FILE" ] && [ -n "${AWS_ROLE_ARN:-}" ]; then

.github/actions/docker-remote-build/action.yml

@@ -25,6 +25,9 @@ inputs:
   sccache_s3_bucket:
     description: 'SCCache S3 Bucket'
     required: false
+  buf_token:
+    description: 'Buf Schema Registry token'
+    required: true
   no_cache:
     description: 'Disable Docker build cache'
     required: false
@@ -66,6 +69,7 @@ runs:
       env:
         AWS_DEFAULT_REGION: ${{ inputs.aws_default_region }}
         SCCACHE_S3_BUCKET:  ${{ inputs.sccache_s3_bucket }}
+        BUF_TOKEN: ${{ inputs.buf_token }}
         PLATFORM: ${{ inputs.platform }}
         GITHUB_RUN_ID: ${{ github.run_id }}
         GITHUB_JOB: ${{ github.job }}
@@ -137,6 +141,12 @@ runs:
         # AWS_ROLE_ARN. We pass the token file and role ARN to BuildKit so sccache
         # can authenticate via STS AssumeRoleWithWebIdentity -- no static keys needed.
         SECRET_ARGS=""
+        if [ -n "${BUF_TOKEN:-}" ]; then
+          SECRET_ARGS+=" --secret id=buf_token,env=BUF_TOKEN"
+        else
+          echo "::error::BUF_TOKEN is required for Cargo access to the Buf registry"
+          exit 1
+        fi
         if [ "${{ inputs.use_sccache }}" == "true" ]; then
           TOKEN_FILE="${AWS_WEB_IDENTITY_TOKEN_FILE:-}"
           if [ -n "$TOKEN_FILE" ] && [ -f "$TOKEN_FILE" ] && [ -n "${AWS_ROLE_ARN:-}" ]; then

.github/filters.yaml

@@ -310,9 +310,6 @@ rust:
   - '**/Cargo.toml'
   - '**/Cargo.lock'
   - 'deny.toml'
-  # Sidecar protobuf contracts are compiled by crate build scripts and need the
-  # same workspace checks as sidecar Rust sources.
-  - 'lib/sidecar/**/*.proto'
 
 benchmarks:
   - 'benchmarks/**'

.github/scripts/test-filters.js

@@ -94,16 +94,16 @@ const testCases = [
     desc: 'vllm component triggers only vllm'
   },
 
-  // Sidecar Rust and proto files should trigger Rust checks without unrelated E2E
+  // Sidecar Rust files should trigger Rust checks without unrelated E2E
   {
     file: 'lib/sidecar/common/src/lib.rs',
     expect: { sidecar: true, rust: true, core: false, frontend: false, vllm: false, sglang: false, trtllm: false },
     desc: 'common sidecar source avoids unrelated build and E2E filters'
   },
   {
-    file: 'lib/sidecar/vllm/proto/vllm_grpc.proto',
+    file: 'lib/sidecar/vllm/build.rs',
     expect: { sidecar: true, rust: true, core: false, frontend: false, vllm: false, sglang: false, trtllm: false },
-    desc: 'vllm sidecar proto triggers Rust checks without backend E2E'
+    desc: 'vllm sidecar build script triggers Rust checks without backend E2E'
   },
   {
     file: 'lib/sidecar/sglang/src/lib.rs',

.github/workflows/copyright-check.ps1

@@ -84,7 +84,7 @@ $global:copyright_results = @{
 
 $ignored_files = @('.clang-format', '.gitattributes', '.gitignore', '.gitkeep', '.patch', 'Cargo.lock', 'LICENSE', 'uv.lock', 'rust-toolchain.toml', 'codespell.txt', 'exclusions.txt')
 write-debug "<copyright-check> ignored_files = ['$($ignored_files -join "','")']."
-$ignored_paths = @('.github', '.mypy_cache', '.pytest_cache', 'lib/llm/tests/data/sample-models', 'lib/llm/tests/data/deepseek-v3.2', 'lib/llm/tests/data/deepseek-v4', 'container/compliance/spdx_licenses', 'lib/sidecar/vllm/proto/control.proto', 'lib/sidecar/vllm/proto/inference.proto')
+$ignored_paths = @('.github', '.mypy_cache', '.pytest_cache', 'lib/llm/tests/data/sample-models', 'lib/llm/tests/data/deepseek-v3.2', 'lib/llm/tests/data/deepseek-v4', 'container/compliance/spdx_licenses')
 write-debug "<copyright-check> ignored_paths = ['$($ignored_paths -join "','")']."
 $ignored_types = @('.bat', '.gif', '.ico', '.ipynb', '.jpg', '.jpeg', '.patch', '.png', '.pyc', '.pyi', '.rst', '.zip', '.md', '.json')
 write-debug "<copyright-check> ignored_types = ['$($ignored_types -join "', '")']."

.github/workflows/dynamo-pipeline.yml

@@ -73,6 +73,8 @@ on:
         required: false
       HF_TOKEN:
         required: false
+      BUF_TOKEN:
+        required: true
 
 
 jobs:
@@ -113,6 +115,7 @@ jobs:
         # "Permission denied (os error 13)" while downloading crates. Redirect
         # CARGO_HOME to the runner-writable workspace so cargo owns its cache.
         CARGO_HOME: /__w/dynamo/dynamo/.cargo
+        CARGO_REGISTRIES_BUF_TOKEN: "Bearer ${{ secrets.BUF_TOKEN }}"
         CONTAINER_ID: test_${{ github.run_id }}_${{ github.run_attempt }}_rust_dynamo
     timeout-minutes: 30
     steps:

.github/workflows/nightly-ci.yml

@@ -965,6 +965,8 @@ jobs:
         dir: ['.', 'lib/bindings/python', 'lib/bindings/kvbm']
     permissions:
       contents: read
+    env:
+      CARGO_REGISTRIES_BUF_TOKEN: "Bearer ${{ secrets.BUF_TOKEN }}"
     steps:
     - uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd  # v6.0.2
       with:

.github/workflows/pre-merge.yml

@@ -237,6 +237,8 @@ jobs:
       matrix: { dir: ['.', 'lib/bindings/python', 'lib/runtime/examples', 'lib/bindings/kvbm'] }
     permissions:
       contents: read
+    env:
+      CARGO_REGISTRIES_BUF_TOKEN: "Bearer ${{ secrets.BUF_TOKEN }}"
     steps:
     - uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd  # v6.0.2
       with:
@@ -341,6 +343,8 @@ jobs:
       matrix: { dir: ['.', 'lib/bindings/python', 'lib/runtime/examples', 'lib/bindings/kvbm'] }
     permissions:
       contents: read
+    env:
+      CARGO_REGISTRIES_BUF_TOKEN: "Bearer ${{ secrets.BUF_TOKEN }}"
     steps:
     - uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd  # v6.0.2
       with:

.github/workflows/shared-build-image.yml

@@ -186,6 +186,8 @@ on:
         required: false
       HF_TOKEN:
         required: false
+      BUF_TOKEN:
+        required: true
     outputs:
       target_tag_plain:
         description: 'Plain runtime image tag prefix'
@@ -370,6 +372,7 @@ jobs:
           IMAGE_REPOSITORY: ${{ vars.ECR_REPOSITORY }}
           AWS_DEFAULT_REGION: ${{ vars.AWS_DEFAULT_REGION }}
           SCCACHE_S3_BUCKET: ${{ secrets.SCCACHE_S3_BUCKET }}
+          BUF_TOKEN: ${{ secrets.BUF_TOKEN }}
         timeout-minutes: 60
         run: |
           set -x
@@ -500,6 +503,7 @@ jobs:
           cuda_version: ${{ matrix.cuda_version }}
           aws_default_region: ${{ vars.AWS_DEFAULT_REGION }}
           sccache_s3_bucket: ${{ secrets.SCCACHE_S3_BUCKET }}
+          buf_token: ${{ secrets.BUF_TOKEN }}
           no_cache: ${{ inputs.no_cache }}
           extra_tags: ${{ steps.extra-tags.outputs.tags }}
           push_image: ${{ inputs.push_image }}
@@ -606,6 +610,7 @@ jobs:
           # wheel_builder/pre_runtime cache instead of a cold rebuild.
           sccache_bucket: ${{ secrets.SCCACHE_S3_BUCKET }}
           sccache_region: ${{ vars.AWS_DEFAULT_REGION }}
+          buf_token: ${{ secrets.BUF_TOKEN }}
           # Must match the build's render: EFA images attribute libfabric /
           # aws-ofi-nccl via --make-efa, and a mismatch would cold-miss the cache.
           make_efa: ${{ inputs.make_efa }}

README.md

@@ -204,6 +204,10 @@ sudo apt install -y build-essential libhwloc-dev libudev-dev pkg-config libclang
 # Install Rust
 curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh && source $HOME/.cargo/env
 
+# Authenticate Cargo to the Buf Schema Registry
+export BUF_TOKEN="your-buf-token"
+cargo login --registry buf "Bearer ${BUF_TOKEN}"
+
 # Create venv and build
 uv venv dynamo && source dynamo/bin/activate
 uv pip install pip 'maturin[patchelf]'

container/templates/wheel_builder.Dockerfile

@@ -571,11 +571,13 @@ ARG USE_SCCACHE
 {% if framework != "sglang" %}
 ARG ENABLE_MEDIA_FFMPEG
 {% endif %}
-RUN --mount=type=secret,id=aws-web-identity-token,target=/run/secrets/aws-token \
+RUN --mount=type=secret,id=buf_token,required=true \
+    --mount=type=secret,id=aws-web-identity-token,target=/run/secrets/aws-token \
     --mount=type=secret,id=aws-role-arn,env=AWS_ROLE_ARN \
     --mount=type=cache,target=/root/.cargo/registry,sharing=shared \
     --mount=type=cache,target=/root/.cargo/git,sharing=shared \
     --mount=type=cache,id=uv-root-{{ context.dynamo.uv_version }},target=/root/.cache/uv,sharing=shared \
+    export CARGO_REGISTRIES_BUF_TOKEN="Bearer $(cat /run/secrets/buf_token)" && \
     export AWS_WEB_IDENTITY_TOKEN_FILE=/run/secrets/aws-token && \
     export UV_CACHE_DIR=/root/.cache/uv && \
     export SCCACHE_S3_KEY_PREFIX=${SCCACHE_S3_KEY_PREFIX:-${TARGETARCH}} && \
@@ -655,8 +657,10 @@ ARG ENABLE_SOURCE_ARCHIVAL=false
 # Mount cargo registry + git caches so re-runs don't re-download the
 # ~750 crates from crates.io every build. `sharing=shared` lets parallel
 # builds (e.g. multiple frameworks in CI) read the same cache concurrently.
-RUN --mount=type=cache,target=/root/.cargo/registry,sharing=shared \
+RUN --mount=type=secret,id=buf_token,required=true \
+    --mount=type=cache,target=/root/.cargo/registry,sharing=shared \
     --mount=type=cache,target=/root/.cargo/git,sharing=shared \
+    export CARGO_REGISTRIES_BUF_TOKEN="Bearer $(cat /run/secrets/buf_token)" && \
     if [ "$ENABLE_SOURCE_ARCHIVAL" = "true" ]; then \
         mkdir -p /tmp/dynamo-vendor-full && \
         cd /opt/dynamo && \
@@ -796,11 +800,13 @@ COPY components/ /opt/dynamo/components/
 
 # Build kvbm wheel (with nixl linkage via auditwheel repair)
 ARG ENABLE_KVBM
-RUN --mount=type=secret,id=aws-web-identity-token,target=/run/secrets/aws-token \
+RUN --mount=type=secret,id=buf_token,required=true \
+    --mount=type=secret,id=aws-web-identity-token,target=/run/secrets/aws-token \
     --mount=type=secret,id=aws-role-arn,env=AWS_ROLE_ARN \
     --mount=type=cache,target=/root/.cargo/registry,sharing=shared \
     --mount=type=cache,target=/root/.cargo/git,sharing=shared \
     --mount=type=cache,id=uv-root-{{ context.dynamo.uv_version }},target=/root/.cache/uv,sharing=shared \
+    export CARGO_REGISTRIES_BUF_TOKEN="Bearer $(cat /run/secrets/buf_token)" && \
     export AWS_WEB_IDENTITY_TOKEN_FILE=/run/secrets/aws-token && \
     export UV_CACHE_DIR=/root/.cache/uv && \
     export SCCACHE_S3_KEY_PREFIX=${SCCACHE_S3_KEY_PREFIX:-${TARGETARCH}} && \

deploy/inference-gateway/epp/Dockerfile

@@ -87,10 +87,12 @@ COPY --from=dynamo deploy/inference-gateway/ext-proc/ deploy/inference-gateway/e
 # git caches are content-addressed (safe to persist); no target/ mount --
 # sccache caches compilations in S3 where stale artifacts can't be linked
 # against newer source.
-RUN --mount=type=cache,target=/usr/local/cargo/registry,id=cargo-registry-${TARGETARCH},sharing=shared \
+RUN --mount=type=secret,id=buf_token,required=true \
+    --mount=type=cache,target=/usr/local/cargo/registry,id=cargo-registry-${TARGETARCH},sharing=shared \
     --mount=type=cache,target=/usr/local/cargo/git,id=cargo-git-${TARGETARCH},sharing=shared \
     --mount=type=secret,id=aws-web-identity-token,target=/run/secrets/aws-token \
     --mount=type=secret,id=aws-role-arn,env=AWS_ROLE_ARN \
+    export CARGO_REGISTRIES_BUF_TOKEN="Bearer $(cat /run/secrets/buf_token)" && \
     export AWS_WEB_IDENTITY_TOKEN_FILE=/run/secrets/aws-token && \
     export SCCACHE_S3_KEY_PREFIX="${SCCACHE_S3_KEY_PREFIX:-epp-${TARGETARCH}}" && \
     if [ "$USE_SCCACHE" = "true" ]; then \