chore: serve static file benchmarks from memory

[dinwwwh]

The static file handler benchmarks read real files, so CodSpeed's single measured call depended on thread pool scheduling and on whatever else the event loop picked up, rather than on plugin code. They now serve the same files from an in-memory fs that still runs the whole plugin path: path resolution, containment checks, content type, ETag, conditional and range handling, and body streaming. Every measured call is now the same microtask chain, so these benches should stop flagging false regressions on unrelated PRs. The new numbers reset the baseline for all five benches.

Evidence

• False flags. Across the CodSpeed comments on fix(rpc): reject non-Blob values in serialized blobs #2060–fix(openapi): fix regex pattern for numeric path validation #2186, "serve file" was flagged 19 times, almost always at −11.3% to −12.9% (3.5 ms ↔ 4.0 ms). "range request" was flagged 4 times, "serve deeply nested encoded path" 3 times and "not modified (304)" twice. None of those PRs touched the plugin, and 25 of the 28 comments warned about different runtime environments.
• Unrelated code moves the numbers. chore: remove unreliable octet stream benchmarks #2188 only removes the octet stream benches and their helpers from __shared__/payloads.ts, yet CodSpeed reports range request +60%, nested path +51% and serve file +48%. This bench imported that module just for drainBody.
• Most of the measured call isn't the plugin. CodSpeed records 17–183 syscalls inside the single measured call of these benches (serve file: 161–183 across the last 43 runs), against 0 for almost every other bench. Under the CodSpeed analysis runner with strace, each serve call makes 6 fs requests through the libuv thread pool, with ~10 event-loop turns and ~18 futex handoffs. The per-call syscall count also varies from call to call (53–149) with identical code.
• Unrelated callbacks run inside calls. While a call waits on fs, the worker handles vitest's IPC traffic, such as the reply to the previous bench's [CodSpeed] … done log. Locally that lands in a warm-up call, but which call it lands in depends only on timing.

Results

Measured locally with the CodSpeed analysis runner (CODSPEED_ENV=1 CODSPEED_RUNNER_MODE=simulation) under strace. Valgrind isn't available here, so this is not a CodSpeed run.

• Syscalls inside the measured call drop from 14–149 (fs, futex, epoll) to none for four benches. Serve file is left with only V8's JIT code-map writes.
• Three runs, one with chore: remove unreliable octet stream benchmarks #2188's payloads.ts change applied, give identical syscall counts for all 40 calls. Before this change they differed between runs.
• All five scenarios still return the expected responses: 200, 200, 206 bytes 0-1023/10240, 304, and fall through.
• "serve file" still costs more than the others. As the first bench in the file, its measured call includes V8's baseline compilation of the shared request path, which is fixed for a given code state.

[cloudflare-workers-and-pages]

Deploying with    Cloudflare Workers

The latest updates on your project. Learn more about integrating Git with Workers.

Status	Name	Latest Commit	Preview URL	Updated (UTC)
✅ Deployment successful! View logs	orpc	08e13f5	Commit Preview URL Branch Preview URL	Oct 06 2026, 04:09 AM

[pullfrog]

✅ No new issues found.

Reviewed changes

• Serves bench files from memory instead of disk — benches/static-file-handler.bench.ts now injects an in-memory fs via vi.mock('node:fs') / vi.mock('node:fs/promises') (createReadStream, stat, realpath) so the single call CodSpeed measures no longer waits on the libuv thread pool or absorbs unrelated callbacks, which was the source of the false regressions across unrelated PRs.
• Drops the ./__shared__/payloads import — the file only used it for drainBody; a local copy is inlined so building those payloads no longer shifts V8 compilation into the first measured call.
• Deterministic representation metadata — fixed mtime and fixed 10KB buffers give a stable ETag, so all five scenarios still exercise the full plugin path (path resolution, containment, content type, ETag, conditional, range, streaming) and return 200, 200, 206 bytes 0-1023/10240, 304, and fall through.

I verified the bench runs to completion, and the file passes eslint and the root tsc. The mock replaces only the fs surface the plugin actually reads, and the hoisted function declarations are safe to reference from the lazy vi.mock factories.

  ^{｜ View workflow run ｜ Using deepseek-v4.1-flash (free via Pullfrog for OSS) ｜ 𝕏}

[pkg-pr-new]

More templates

• @orpc/bun-playground
• @orpc/cloudflare-playground
• @orpc/nest-playground
• @orpc/next-playground

@orpc/ai-sdk

npm i https://pkg.pr.new/@orpc/ai-sdk@2189

@orpc/arktype

npm i https://pkg.pr.new/@orpc/arktype@2189

@orpc/bun

npm i https://pkg.pr.new/@orpc/bun@2189

@orpc/client

npm i https://pkg.pr.new/@orpc/client@2189

@orpc/cloudflare

npm i https://pkg.pr.new/@orpc/cloudflare@2189

@orpc/contract

npm i https://pkg.pr.new/@orpc/contract@2189

@orpc/experimental-effect

npm i https://pkg.pr.new/@orpc/experimental-effect@2189

@orpc/evlog

npm i https://pkg.pr.new/@orpc/evlog@2189

@orpc/hibernation

npm i https://pkg.pr.new/@orpc/hibernation@2189

@orpc/json-schema

npm i https://pkg.pr.new/@orpc/json-schema@2189

@orpc/experimental-lock

npm i https://pkg.pr.new/@orpc/experimental-lock@2189

@orpc/experimental-msw

npm i https://pkg.pr.new/@orpc/experimental-msw@2189

@orpc/nest

npm i https://pkg.pr.new/@orpc/nest@2189

@orpc/next

npm i https://pkg.pr.new/@orpc/next@2189

@orpc/node

npm i https://pkg.pr.new/@orpc/node@2189

@orpc/openapi

npm i https://pkg.pr.new/@orpc/openapi@2189

@orpc/opentelemetry

npm i https://pkg.pr.new/@orpc/opentelemetry@2189

@orpc/pinia-colada

npm i https://pkg.pr.new/@orpc/pinia-colada@2189

@orpc/pino

npm i https://pkg.pr.new/@orpc/pino@2189

@orpc/publisher

npm i https://pkg.pr.new/@orpc/publisher@2189

@orpc/ratelimit

npm i https://pkg.pr.new/@orpc/ratelimit@2189

@orpc/server

npm i https://pkg.pr.new/@orpc/server@2189

@orpc/shared

npm i https://pkg.pr.new/@orpc/shared@2189

@orpc/swr

npm i https://pkg.pr.new/@orpc/swr@2189

@orpc/tanstack-query

npm i https://pkg.pr.new/@orpc/tanstack-query@2189

@orpc/trpc

npm i https://pkg.pr.new/@orpc/trpc@2189

@orpc/valibot

npm i https://pkg.pr.new/@orpc/valibot@2189

@orpc/zod

npm i https://pkg.pr.new/@orpc/zod@2189

commit: 08e13f5

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[codspeed]

Merging this PR will improve performance by 52.74%

⚠️ Different runtime environments detected

Some benchmarks with significant performance changes were compared across different runtime environments,
which may affect the accuracy of the results.

Open the report in CodSpeed to investigate

⚡ 4 improved benchmarks
✅ 26 untouched benchmarks

Performance Changes

	Benchmark	BASE	HEAD	Efficiency
⚡	not modified (304)	361.2 µs	212.4 µs	+70.05%
⚡	range request	1,016.6 µs	667.8 µs	+52.24%
⚡	serve deeply nested encoded path	1,020.7 µs	672.8 µs	+51.72%
⚡	not found fall through	329.9 µs	238.1 µs	+38.58%

Tip

Curious why performance improved? Comment @codspeedbot explain why performance improved on this PR, or directly use the CodSpeed MCP with your agent.

---

_{Comparing claude/static-file-bench-determinism (08e13f5) with main (86c31c3)}