feat: implement Rust proxy gateway for vLLM Responses API#24
Open
leseb wants to merge 6 commits into
Open
Conversation
Replaces the Python gateway stub with a full Rust implementation using axum, reqwest, and tokio. Supports both streaming (SSE) and non-streaming proxy modes, hop-by-hop header filtering, API key injection, and comprehensive error mapping (502/504). Includes 11 tests covering passthrough, auth, streaming, error propagation, mid-stream failure, connection errors, and timeouts. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> Signed-off-by: Sébastien Han <seb@redhat.com>
leseb
requested review from
bbrowning,
franciscojavierarceo,
jiahuei,
maralbahari,
noobHappylife,
qandrew and
tjtanaa
as code owners
maralbahari
reviewed
May 12, 2026
Derive clap::Args on RuntimeConfig directly and use #[arg(skip)] for llm_api_base, eliminating the duplicate GatewayOpts struct and the manual build_config mapping function. Addresses review feedback from @maralbahari. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> Signed-off-by: Sébastien Han <seb@redhat.com>
Collaborator
|
@leseb I was wondering if we should validate that the library choices are optimal and verify stateless proxy throughput/latency against the Python baseline. When I was working on translating Python to Rust, I noticed some performance regression compared to the Python baseline, so I thought it would be best to try and avoid such regressions from the start. |
Introduces a proper Error enum via thiserror, replacing Box<dyn std::error::Error> across the library. ProxyState::new now returns Result instead of panicking on HTTP client build failure. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> Signed-off-by: Sébastien Han <seb@redhat.com>
Collaborator
Author
Can you share more? I'll add a criterion benchmarks |
Measures proxy latency overhead by comparing direct upstream requests against proxied requests for both non-streaming and streaming (SSE) paths. Establishes a baseline for regression detection. Results on Apple M-series: non_stream/direct: ~65 µs non_stream/proxied: ~133 µs (68 µs overhead) stream/direct: ~91 µs stream/proxied: ~462 µs (371 µs overhead) Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> Signed-off-by: Sébastien Han <seb@redhat.com>
Replaces generic "upstream" terminology with "vllm" to make it clear what the gateway is proxying to. Affects config fields, error variants, function names, variables, and test helpers. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> Signed-off-by: Sébastien Han <seb@redhat.com>
Signed-off-by: Sébastien Han <seb@redhat.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
--llm-api-base) and integrated (serve <model>spawning vLLM as a subprocess)config,proxy,app,servermodules with a clean separation of concernsTest Plan
cargo test(11/11 green)cargo clippy --all-targets -- -D warningscargo fmt -- --check🤖 Generated with Claude Code