perf(transcode): remove per-request route allocations on the hot path

[polaz]

Summary

Removes two avoidable allocations paid on every proxied request (the hottest path in the proxy).

Changes

1. Arc<RouteEntry> instead of deep clone. The handler closure captured RouteEntry by value, so axum's per-request clone of the handler deep-cloned its String fields (http_path, grpc_path, response_body). Wrapping the entry in Arc makes the per-request clone a single refcount bump.
2. Parse the gRPC path once. grpc_path was re-parsed into a PathAndQuery on every request in both the unary and streaming handlers. It's now parsed once at route-build time and stored as PathAndQuery (whose clone is a cheap Bytes refcount bump). An invalid path now skips the route at startup with a log, instead of a per-request 500 branch.

Not done (deliberately)

• SIMD / wildcopy: not applicable. There's no hot scalar byte loop in our code; byte-level work lives in serde_json / prost / hyper (already optimized), and the proxy is I/O-bound on the upstream gRPC RTT. Adding SIMD to the glue would be premature and wouldn't move P99.
• The serde_json::Value intermediate in request building could be skipped by constructing the DynamicMessage directly via reflection, but that's a larger refactor with real correctness surface (coercion, oneofs, nested fields) and little payoff on an I/O-bound path. Left as a possible future optimization.

Testing

No behavior change on the success path. cargo nextest run --features redis: 119 passed. clippy (all-features) + fmt clean.

Closes #41

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Organization UI

Review profile: ASSERTIVE

Plan: Pro Plus

Run ID: 0c0aa285-e7bc-45ad-8e61-d81a10d81667

📥 Commits

Reviewing files that changed from the base of the PR and between 8489371 and 1aa09ab.

📒 Files selected for processing (1)

• src/transcode/mod.rs

---

📝 Walkthrough

Summary by CodeRabbit

Release Notes

• Refactor
  • Improved request handling performance through optimized route resolution.
  • Route configuration errors are now detected at startup rather than during request processing.

Walkthrough

RouteEntry.grpc_path changes from String to a pre-parsed PathAndQuery. Route extraction (extract_routes, extract_streaming_routes) now parses the path once and skips routes with invalid paths. Each RouteEntry is wrapped in Arc before being captured by handler closures, and both transcode_handler and streaming_handler now accept Arc<RouteEntry> and clone the stored PathAndQuery directly.

Changes

Per-request allocation removal on the hot path

Layer / File(s)	Summary
RouteEntry struct change and parse-time path validation src/transcode/mod.rs	grpc_path field type changes from String to PathAndQuery. extract_routes and extract_streaming_routes parse the formatted path into PathAndQuery at route-build time; parse failures log an error and skip the route instead of deferring to a per-request 500.
Arc wiring in routes() and handler signature updates src/transcode/mod.rs	routes() wraps each primary, alias, and streaming RouteEntry in Arc before closure capture. streaming_handler and transcode_handler parameter types change to Arc<RouteEntry>, and both clone the pre-stored PathAndQuery instead of re-parsing a string.

Estimated code review effort

🎯 2 (Simple) | ⏱️ ~10 minutes

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title clearly and concisely describes the main optimization: removing per-request route allocations on the critical hot path.
Description check	✅ Passed	The description provides a detailed summary of changes including the two main optimizations (Arc wrapping and path parsing), testing results, and deliberate non-changes with justifications.
Linked Issues check	✅ Passed	The PR fully addresses both objectives from issue #41: wrapping RouteEntry in Arc to eliminate deep cloning, and parsing grpc_path once at route-build time instead of per-request.
Out of Scope Changes check	✅ Passed	All changes are directly related to the two stated optimizations in issue #41; no out-of-scope modifications are present.
Docstring Coverage	✅ Passed	Docstring coverage is 100.00% which is sufficient. The required threshold is 80.00%.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing Touches

📝 Generate docstrings

• Create stacked PR
• Commit on current branch

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch perf/#41-hot-path-allocs

---

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

[greptile-apps]

Greptile Summary

• Optimizes the REST-to-gRPC transcoding hot path in src/transcode/mod.rs.
• Stores route entries behind Arc so handler cloning avoids deep-copying route strings per request.
• Parses generated gRPC paths once during route extraction and stores them as reusable PathAndQuery values.
• Skips routes with invalid generated gRPC paths during startup instead of handling that parse failure per request.

Confidence Score: 5/5

The change appears safe to merge based on the limited, performance-focused scope and absence of identified correctness issues.

Only src/transcode/mod.rs changed, and the update preserves the success path while moving reusable route data into cheaper shared representations.

T-Rex Logs

What T-Rex did

• Ran the unary proxied-route route-allocations tests before and after, and observed that both runs completed with exit code 0 and reported success_behavior_unchanged=true.
• Encountered a runtime blocker when attempting to run cargo test transcode, because cargo was not found in the environment, blocking both before/after test attempts.

_{Ran code and verified through T-Rex}

_{Reviews (1): Last reviewed commit: "perf(transcode): remove per-request rout..." | Re-trigger Greptile}