Skip to content

zstd decoding complete frame in read stream halts stream #64741

Description

@laverdet

Version

v26.5.0

Platform

Darwin Mac.lan 24.6.0 Darwin Kernel Version 24.6.0: Tue Apr 21 20:19:12 PDT 2026; root:xnu-11417.140.69.710.16~1/RELEASE_ARM64_T6041 arm64

Subsystem

No response

What steps will reproduce the bug?

import { Buffer } from 'node:buffer';
import * as consumers from 'node:stream/consumers';
import * as zlib from 'node:zlib';

function consume(codec, writes) {
	const inflate = codec.createDecompress();
	const result = consumers.buffer(inflate);
	writes.forEach(write => inflate.write(write));
	inflate.end();
	return result;
}

function describe(buffer) {
	const aa = buffer.includes(0x41);
	const bb = buffer.includes(0x42);
	const which = function() {
		if (aa && bb) {
			return 'frameA + frameB';
		} else if (aa) {
			return 'frameA only';
		} else if (bb) {
			return 'frameB only';
		} else {
			return 'nothing';
		}
	}();
	return `${buffer.length} bytes (${which})`;
}

async function attempt(codec, writes) {
	try {
		return describe(await consume(codec, writes));
	} catch (error) {
		return `threw: ${error.message}`;
	}
}

const tests = [ {
	name: 'zstd',
	compress: buffer => zlib.zstdCompressSync(buffer),
	createDecompress: () => zlib.createZstdDecompress(),
}, {
	name: 'gzip',
	compress: buffer => zlib.gzipSync(buffer),
	createDecompress: () => zlib.createGunzip(),
} ];

console.log(`node ${process.version}`);
for (const codec of tests) {
	// 'A' / 'B'
	const frameA = codec.compress(Buffer.alloc(2000, 0x41));
	const frameB = codec.compress(Buffer.alloc(2000, 0x42));
	const concatenated = Buffer.concat([ frameA, frameB ]);

	// BUG: both frames in a single write — the boundary between them falls inside one input buffer.
	const bug = await attempt(codec, [ concatenated ]);
	// COUNTER-EXAMPLE: the same two frames, one per write — the boundary lands at a write boundary.
	const counter = await attempt(codec, [ frameA, frameB ]);

	console.log(`${codec.name}:`);
	console.log(`  stream, both frames in ONE write   -> ${bug}`);
	console.log(`  stream, one frame per write        -> ${counter}`);
}

How often does it reproduce? Is there a required condition?

This reproduces every time

What is the expected behavior? Why is that the expected behavior?

Decoding concatenated zstd payloads should decode correctly as one stream. Indeed, zstdcat will decode this correctly (omitted from example but you can take my word for it).

What do you see instead?

marcel[1:42:21PM] [~/xx] ~/Downloads/node-v26.5.0-darwin-arm64/bin/node a.mjs
node v26.5.0
zstd:
  stream, both frames in ONE write   -> 2000 bytes (frameA only)
  stream, one frame per write        -> 4000 bytes (frameA + frameB)
gzip:
  stream, both frames in ONE write   -> 4000 bytes (frameA + frameB)
  stream, one frame per write        -> 4000 bytes (frameA + frameB)

Additional information

This will cause spurious failures while decoding zstd streams. The odds of it happening are a function of stream read size vs zstd chunk size. In my case I ran into this while decoding a stream after about 3GB and tracked it down to this bug.

Metadata

Metadata

Assignees

No one assigned

    Labels

    streamIssues and PRs related to the stream subsystem.zlibIssues and PRs related to the zlib subsystem.

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions