Router Prefetch Exhausts HTTP/1.1 Connection Pool, Stalling Server Actions In Self-Hosted Next Start

The App Router's prefetch logic issues concurrent _rsc requests without a concurrency cap and does not prioritize them relative to Server Action mutations. On HTTP/1.1, the browser's six-connection limit is exhausted by prefetches; the action POST cannot be sent until a socket frees, and aborted prefetches may leave sockets in a broken state, causing the action to hang or the client to miss the revalidation response.

1. Clone https://github.com/SignaHub285/next-app-router-action-abort-repro 2. Run npm i && npm run repro 3. The script builds, starts next start behind a slow RSC proxy, drives Chromium through control (no prefetch pressure) and treatment (prefetch burst); treatment shows UI updated after click ~3876ms vs control ~278ms, with multiple aborted RSC requests.

This is a userland monkey-patch, not an official fix. It reduces prefetch concurrency on all HTTP versions; on HTTP/2 or HTTP/3 where concurrency is not limited by connections, it may slightly throttle prefetch speed. If a Server Action itself requires multiple concurrent requests, two free sockets may still be insufficient—adjust MAX_PREFETCH_CONCURRENCY downward. Remove the component and restore original fetch to roll back. Test thoroughly with your application's prefetch patterns.