We Ralph Wiggumed WebStreams to Make Them 10x Faster

The Problem

Native WebStreams (WHATWG Streams API) carry significant overhead for server-side JavaScript. Node.js pipeline() hits 7,900 MB/s while native WebStreams pipeThrough manages only 630 MB/s — a 12x gap. Every reader.read() call triggers four allocations and a microtask queue hop, even when data is already buffered:

1. Allocate a ReadableStreamDefaultReadRequest with callback slots
2. Enqueue the request into the stream's internal queue
3. Create a new Promise
4. Resolve through the microtask queue

The Solution: fast-webstreams

Available as experimental-fast-webstreams on npm. Passes 1,100 of 1,116 Web Platform Tests (Node.js passes 1,099).

Key Optimizations

Zero-Promise Pipeline Chains — When chaining pipeThrough and pipeTo between fast streams, upstream links are recorded without immediate piping. On pipeTo(), it traverses upstream, collects underlying Node.js stream objects, and executes a single pipeline() call. Result: ~6,200 MB/s (9.8x faster).

Synchronous Read Resolution — Attempts synchronous reads via nodeReadable.read(). When data is buffered, returns Promise.resolve({value, done}) without event loop delays. Result: ~12,400 MB/s (3.7x faster).

React Flight Optimization — For the RSC byte stream pattern (external enqueuing via captured controller), a custom LiteReadable buffer with direct callback dispatch replaces EventEmitter. Result: 1,600 MB/s vs 110 MB/s (14.6x faster).

Fetch Response Wrapping — Global ReadableStream patching gives native fetch response bodies fast stream shells. Result: 830 MB/s from 260 MB/s (3.2x faster).

Benchmarks

Surprising Findings

• "The spec is smarter than it looks" — attempted shortcuts frequently violated WPT for legitimate reasons. The ReadableStreamDefaultReadRequest pattern handles edge cases like cancellation during reads and error identity through locked streams.
• Promise.resolve(obj) always checks for thenables, affecting object creation in hot paths.
• Node.js pipeline() cannot replace WHATWG pipeTo due to fundamental differences in error propagation, stream locking, and cancellation semantics.

Upstream Contributions

Matteo Collina's PR (nodejs/node#61807) brings two optimizations directly into Node.js core:

1. read() fast path — Returns resolved Promises when data is buffered, avoiding request object allocation
2. pipeTo() batch reads — Processes multiple buffered reads without per-chunk request objects while respecting backpressure

Result: ~17-20% faster buffered reads and ~11% faster pipeTo for all Node.js users.

AI-Assisted Development

Most of fast-webstreams was built with AI assistance, enabled by two factors: 1,116 Web Platform Tests provided machine-checkable correctness validation, and an early benchmark suite measured throughput impact per optimization. The loop: implement → WPT validate → benchmark → revert if tests fail or no improvement.