[Feature] Add process-local in-memory KV indexer and Router integration (#33370)

Co-authored-by: Wu, Yutong <yutong.wu@amd.com>
Co-authored-by: TianDi101 <ditian12@amd.com>
Co-authored-by: Zhangheng <hzh0425@apache.org>
This commit is contained in:
wuyl1
2026-08-20 10:45:35 +08:00
committed by GitHub
co-authored by Wu, Yutong TianDi101 Zhangheng
parent 238ba40c27
commit 360d10d6bc
42 changed files with 6603 additions and 174 deletions
+11 -5
View File
@@ -108,6 +108,9 @@ jobs:
steps:
- uses: actions/checkout@v4
- name: Install protoc
run: sudo bash scripts/ci/utils/install_protoc.sh
- name: Verify dynamo-* SHA pin
run: |
set -euo pipefail
@@ -160,17 +163,17 @@ jobs:
- name: cargo check
working-directory: experimental/sgl-router
run: cargo check --all-targets
run: cargo check --workspace --all-targets
- name: cargo clippy
working-directory: experimental/sgl-router
run: cargo clippy --all-targets -- -D warnings
run: cargo clippy --workspace --all-targets -- -D warnings
- name: cargo fmt
working-directory: experimental/sgl-router
run: |
rustup toolchain install nightly --profile minimal --component rustfmt
cargo +nightly fmt -- --check
cargo +nightly fmt --all -- --check
# cargo-deny 0.19.6+ required for CVSS 4.0 parsing.
- name: Install cargo-deny
@@ -196,6 +199,9 @@ jobs:
steps:
- uses: actions/checkout@v4
- name: Install protoc
run: sudo bash scripts/ci/utils/install_protoc.sh
- name: Install OS deps (libssl, pkg-config)
run: |
sudo apt-get update
@@ -246,7 +252,7 @@ jobs:
# the Qwen3-0.6B tokenizer.json. Filter is matched against the
# full test path `tokenizer::parity::parity_matrix`; substring
# `parity_matrix` is unique to that test.
run: cargo test --release -- --skip parity_matrix
run: cargo test --release --workspace -- --skip parity_matrix
# Regenerate the cross-impl block-hash parity fixture and fail if it
# differs from the committed file. The Python script replicates
@@ -366,7 +372,7 @@ jobs:
working-directory: experimental/sgl-router
run: |
source "$HOME/.cargo/env"
cargo build --release
cargo build --release --workspace
# Install SGLang from the local checkout in editable mode (no PyPI
# version pin) — mirrors `pr-test-rust.yml` so the router e2e runs