fix(platform): Re-enable middleware span sampling

[sentry-junior]

Summary

Re-enables middleware span ingestion on docs by removing the hard drop of Next.js Middleware.execute roots in tracesSampler.

Why

#18775 intentionally sampled middleware roots at 0 because they were unclassifiable noise at the time. We want middleware spans back for investigation / visibility.

SDK instrumentation still emits http.server.middleware on current @sentry/nextjs — only the app sampler was suppressing them. No SDK bump required.

Change

• Remove isMiddlewareRootSpan early-return (return 0) from src/tracesSampler.ts
• Middleware roots now use the same traffic-based rates as other roots (header classification when available, otherwise UA / unknown)
• Keep docs.request.classified in middleware.ts as the unsampled request counter

Verify after deploy

span.op:http.server.middleware release:<new sha>

Should go non-zero on sentry/docs.

Risks

• Middleware roots may still often classify as unknown when sampled before request headers are available, so volume can be higher than classifiable server spans
• Possible temporary duplicate child spans from the SDK wrapper until the unreleased JS SDK change that stops starting a wrapper span lands

Test plan

• Review sampler diff
• After merge/deploy, confirm middleware spans appear for the new release
• Confirm docs.request.classified still flows

Requested by Charly Gomez via Junior.

--

View Junior Session

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
develop-docs	Ready	Preview	Jul 30, 2026 2:59pm
sentry-docs	Ready	Preview	Jul 30, 2026 2:59pm

[sergical]

Superseded by #18887 — same goal, different sampling shape.

This PR removes isMiddlewareRootSpan entirely, so middleware roots fall through to the normal classification path. Worth noting what that resolves to in practice: there's no user-agent available on a middleware root — Next.js creates the Middleware.execute span in a detached sandbox before any request data reaches Sentry — so the UA fallback never has anything to match on and the rate is always SAMPLE_RATES.unknown (30%). At current traffic (~2.4M requests/day through the middleware, per docs.request.classified) that's ~720k spans/day, which is roughly 3x the ~264k/day that motivated #18775.

#18887 keeps the detection and returns 1% instead of 0 (~24k/day), and additionally names the spans from inside middleware.ts — middleware GET redirect / rewrite / passthrough with url.path, traffic_type, and device_type attributes. The naming is the part that makes them worth ingesting; without it you get 720k/day of bare middleware GET with no route detail, which is what the old beforeSendTransaction hack was trying to work around.

Also carried your duplicate-child-span note over to #18887 as an open question — if that SDK behavior is real, actual volume there could be ~2x the estimate. Happy to raise the 1% if 24k/day turns out too thin for the investigation you had in mind.