[wan9chi]

Motivation

remote_cache::upload_without_token failed on #969's macOS CI (job). The test expects the backend's 401 Invalid credentials for an upload without a token. Instead vp run got HTTP status 502, and remote-cache-server stop reported POST /store failed: write EPIPE. The test predates the dry-run stack; the race is in the e2e proxy that sits in front of the backend.

Changes

• The proxy streamed each request body to the backend as it arrived (request.pipe(upstream)). The Worker rejects an upload by its headers and answers 401 before reading the body. workerd then closes the connection, resetting it when body bytes are still unread.
• On a slow runner, a later piece of the body reached the proxy after that reset but before the proxy had read the 401. Writing it failed with EPIPE, Node destroyed the socket with the 401 still unread, and the proxy answered 502.
• The proxy now reads the whole request first and sends it with upstream.end(body). The body then goes out together with the headers, before the backend can answer, and an error after the answer is ignored, as before.

🤖 Generated with Claude Code

---

Note

Low Risk
Changes only the e2e remote-cache proxy’s request forwarding; production cache paths are unaffected.

Overview
Fixes a race in the e2e remote-cache proxy where uploads without a token were reported as 502 instead of the backend’s early 401.

The tap no longer streams the client body with request.pipe(upstream). It buffers the full request, then sends it with upstream.end(Buffer.concat(body)) so headers and body reach the Worker before it can reject and reset the connection. Comments now describe that reset-after-answer behavior and why late body writes could drop the unread 401.

^{Reviewed by Cursor Bugbot for commit e63c3be. Bugbot is set up for automated code reviews on this repo. Configure here.}

[github-actions]

fspy benchmark

linux

dynamic/launch             change  +0.72%  [ -3.89% ..  +5.74%]  overhead   +61.18%
dynamic/access             change  -0.78%  [ -5.30% ..  +3.43%]  overhead   +19.82%
dynamic/access-relative    change  +0.00%  [ -3.38% ..  +3.47%]  overhead   +68.54%
dynamic/access-contended   change  +1.06%  [ -7.34% .. +10.11%]  overhead   +28.35%
static/launch              change  -0.70%  [ -6.14% ..  +5.74%]  overhead  +195.78%
static/access              change  +0.50%  [ -9.58% .. +17.00%]  overhead +1458.78%
static/access-relative     change  -0.51%  [-12.03% ..  +2.90%]  overhead +1930.66%
static/access-contended    change  -0.56%  [ -2.26% ..  +0.67%]  overhead +1673.96%

macos

dynamic/launch             change  -0.14%  [ -7.89% ..  +6.88%]  overhead  +236.16%
dynamic/access             change  -0.38%  [ -9.08% ..  +7.97%]  overhead    +4.07%
dynamic/access-relative    change  -1.19%  [ -6.45% ..  +5.24%]  overhead  +234.56%
dynamic/access-contended   change  +0.00%  [ -6.09% .. +12.97%]  overhead    +3.08%

windows

dynamic/launch             change  -0.29%  [ -2.54% ..  +1.88%]  overhead   +33.92%
dynamic/access             change  +0.00%  [ -0.62% ..  +0.21%]  overhead    +1.27%
dynamic/access-relative    change  +0.00%  [ -0.62% ..  +0.62%]  overhead    +1.70%
dynamic/access-contended   change  +0.00%  [ -0.41% ..  +0.41%]  overhead    +1.46%