msgpack-nodejs

msgpack-nodejs

Yet another javascript/nodejs implementation of MsgPack Spec.
The purpose behind is learning by doing, which focuses on modern tools/techniques of nodejs/typescript ecosystem.

Contents

• Usage
  • Examples
  • API
  • Options
• Project status
  • Compability
  • Limitation
  • TODO
  • Benchmark
• Implementation detail
  • Encode
  • Decode
  • Lesson learned
• References

Usage

npm i msgpack-nodejs
npm test

Example

Please check example.md

API

• encode(): any => Uint8Array
• decode(): Uint8Array => Exclude<any, Map>
• EncodeStream class: Exclude<any, null> => Buffer
• DecodeStream class: Buffer => Exclude<any, null> ¹
• registerExtension(): Register your own extension
• stringBufferStat(): Show string buffer copied count and size
• lruCacheStat(): Show cache hit/miss count
• bufferAllocatorStat(): Show how byte-array allocate new buffer
• prefixTrieStat(): Show prefix-trie hit/miss count
• applyOptions(): Manually control caching

Options

You can apply options like this

Key	type	default	Description
encoder.mapKeyCache.enabled	boolean	true	Cache map-key or not
encoder.mapKeyCache.size	number	30	How big is the mapKeyCache
encoder.stringCache.enabled	boolean	true	Cache any string except map-key or not
encoder.stringCache.size	number	100	How big is the stringCache
encoder.byteArray.base	number	1024	How many bytes will be allocated for every execution. Setting this would increase performance when handling many big JSON data
decoder.shortStringCache.enabled	boolean	true	Use prefix-trie or not
decoder.shortStringCache.lessThan	number	10	Only cache if string is shorter than this value
decoder.jsUtf8Decode.enabled	boolean	true	Use JS utf8-decode or not
decoder.jsUtf8Decode.lessThan	number	200	Only use JS utf8-decode if string is shorter than this value

Project status

Compability

Env	Executable?
Node.js 18	✅
Node.js 16	✅
Node.js 14	✅
Node.js 12	❌

Limitation

• Does not support float 32 encoding, because Javascript float is always 64-bit.

TODO

• Ext tests
• Map 16/32 tests

Benchmark

By utlizing the great benchmark tool by msgpack-lite, I thought the performance of this project would not be disappointing.

Runs on node.js 16 & laptop with R5-5625U.

operation	op	ms	op/s
buf = Buffer(JSON.stringify(obj));	1021200	5000	204240
obj = JSON.parse(buf);	1279500	5000	255900
buf = require("msgpack-lite").encode(obj);	685800	5000	137160
obj = require("msgpack-lite").decode(buf);	389800	5001	77944
buf = Buffer(require("msgpack.codec").msgpack.pack(obj));	713600	5000	142720
obj = require("msgpack.codec").msgpack.unpack(buf);	401300	5001	80243
buf = require("msgpack-js-v5").encode(obj);	284400	5000	56880
obj = require("msgpack-js-v5").decode(buf);	544600	5000	108920
buf = require("msgpack-js").encode(obj);	277100	5001	55408
obj = require("msgpack-js").decode(buf);	559800	5000	111960
buf = require("msgpack5")().encode(obj);	147700	5001	29534
obj = require("msgpack5")().decode(buf);	239500	5000	47900
buf = require("notepack").encode(obj);	1041500	5000	208300
obj = require("notepack").decode(buf);	671300	5000	134260
obj = require("msgpack-unpack").decode(buf);	163400	5001	32673
buf = require("msgpack-nodejs").encode(obj); (Run in sequence)	1148900	5000	229780
obj = require("msgpack-nodejs").decode(buf); (Run in sequence)	777500	5000	155500
buf = require("msgpack-nodejs").encode(obj); (Run exclusively)	1321900	5000	264380
obj = require("msgpack-nodejs").decode(buf); (Run exclusively)	805400	5000	161080

Implementation detail

Encode

Encoder uses a recursive function match() to match JSON structure (primitive value, object, array or nested), and pushes anything encoded into ByteArray that responsible for allocating buffer. Encoded string will be written in StringBuffer first and cached in LruCache.

Decode

Decoder uses parseBuffer() to read every value out, and push them into StructBuilder to rebuild whole JSON object. For string less than 200 bytes, use pure JS utf8Decode(), then cache in prefix trie.

Optimization strategies:

Cache

• To improve encoding performance, LruCache was used for caching encoded string and its header.
• To improve decoding peformance, prefix trie was deployed for Uint8Array caching.
• To avoid evicting, map-key caching and string caching were separated.

ArrayBuffer / TypedArray

• To efficiently allocate new buffer, every ByteArray begins with small buffer (1K). ²
• To efficiently handle unpredictable large JSON, ByteArray allocates exponentially.
• To avoid overhead on writing, ByteArray uses DataView calls as much as possible.

Node.js

• To maximize performance of array, use pre-allocated array. ³
• To maximize performance of string encoding, string are encoded in StringBuffer with encodeInto() to prevent unnecessary copying. Then these encoded content will be referenced by subarray() for writing and caching. ⁴
• To avoid overhead of TextDecoder(), decode UTF-8 bytes with pure JS when less than 200 bytes. ³
• To avoid GC overhead in decoder, every parsed typed value will be passed into builder.insertValue() directly. ⁵
• To avoid syntax penalty of private class fields under node.js 18, use TypeScript's syntax instead.

Lessons learned

• Javascript
  • The difference between ArrayBuffer, TypedArray and the node.js API Buffer ⁶
  • BigInt operators ⁷
  • left shift and right shift ^{8 9}
  • Private class features ¹⁰
  • Pre-allocated array ³
  • UTF-8 encoding/decoding ¹¹
  • Node.js transform stream ¹²
• Node.js
  • Profiler ¹³
• Typescript
  • Testing - ts-jest ¹⁴
  • Linter - typescript-eslint & lint-staged ¹⁵
  • Packaging for ESModule & CommonJS ¹⁶
• CI & CD
  • Github Actions ¹⁷

Footnotes

• After attaching another stream that does not expect a object as its input, you may encounter error ↩

• Thanks to kriszyp/msgpackr for inspiration of better buffer allocation strategy. ↩

• Thanks to AppSpector, this article gives very practical advices including pre-allocated array and manual decoding under 200 characters. ↩ ↩² ↩³

• Thanks to msgpack/msgpack-javascript for technique including UTF-8 bytes calculation and usage of encodeInto(), which led me to the ultra optimization strategy. ↩

• Memory Diagnostics - Using GC Trace ↩

• [JS] TypedArray, ArrayBuffer 和 DataView ↩

• BigInt operators ↩

• left shift ↩

• right shift ↩

• Private class features ↩

• UTF-8 encoding/decoding ↩

• samerbuna/efficient-node ↩

• Profiler ↩

• ts-jest ↩

• 使用 ESLint, Prettier, Husky, Lint-staged 以及 Commitizen 提升專案品質及一致性 ↩

• Best practices for creating a modern npm package ↩

• Github Actions ↩