parquet-wasm
TypeScript icon, indicating that this package has built-in type declarations

0.6.0-beta.2 • Public • Published

WASM Parquet npm version

WebAssembly bindings to read and write the Apache Parquet format to and from Apache Arrow.

This is designed to be used alongside a JavaScript Arrow implementation, such as the canonical JS Arrow library.

Including all compression codecs, the brotli-encoded WASM bundle is 907KB.

Install

parquet-wasm is published to NPM. Install with

yarn add parquet-wasm
# or
npm install parquet-wasm

API

Two APIs?

Important Note!: the maintainer of arrow2 and parquet2 has stepped back, and therefore I plan to deprecate the parquet2-based API in the future. See issue #308.

These bindings expose two APIs to users because there are two separate implementations of Parquet and Arrow in Rust.

  • parquet and arrow: These are the "official" Rust implementations of Arrow and Parquet. These projects started earlier and may be more feature complete.
  • parquet2 and arrow2: These are safer (in terms of memory access) and claim to be faster, though I haven't written my own benchmarks yet.

Since these parallel projects exist, why not give the user the choice of which to use? In general the reading API is identical in both APIs, however the write options differ between the two projects.

Choice of bundles

Presumably no one wants to use both parquet and parquet2 at once, so the default bundles separate parquet and parquet2 into separate entry points to keep bundle size as small as possible. The following describe the six bundles available:

Entry point Rust crates used Description Documentation
parquet-wasm/bundler/arrow1 parquet and arrow "Bundler" build, to be used in bundlers such as Webpack Link
parquet-wasm/node/arrow1 parquet and arrow Node build, to be used with require in NodeJS Link
parquet-wasm/esm/arrow1 parquet and arrow ESM, to be used directly from the Web as an ES Module Link
parquet-wasm or parquet-wasm/bundler/arrow2 parquet2 and arrow2 "Bundler" build, to be used in bundlers such as Webpack Link
parquet-wasm/node/arrow2 parquet2 and arrow2 Node build, to be used with require in NodeJS Link
parquet-wasm/esm/arrow2 parquet2 and arrow2 ESM, to be used directly from the Web as an ES Module Link

Note that when using the esm bundles, the default export must be awaited. See here for an example.

arrow2 API

This implementation uses the arrow2 and parquet2 Rust crates.

This is the default implementation and is more full-featured, including metadata handling and async reading. Refer to the API documentation for more details and examples.

arrow API

This implementation uses the arrow and parquet Rust crates.

Refer to the API documentation for more details and examples.

Debug functions

These functions are not present in normal builds to cut down on bundle size. To create a custom build, see Custom Builds below.

setPanicHook

setPanicHook(): void

Sets console_error_panic_hook in Rust, which provides better debugging of panics by having more informative console.error messages. Initialize this first if you're getting errors such as RuntimeError: Unreachable executed.

The WASM bundle must be compiled with the console_error_panic_hook for this function to exist.

Example

import * as arrow from "apache-arrow";
import * as parquet from "parquet-wasm";

// Create Arrow Table in JS
const LENGTH = 2000;
const rainAmounts = Float32Array.from({ length: LENGTH }, () =>
  Number((Math.random() * 20).toFixed(1))
);

const rainDates = Array.from(
  { length: LENGTH },
  (_, i) => new Date(Date.now() - 1000 * 60 * 60 * 24 * i)
);

const rainfall = arrow.tableFromArrays({
  precipitation: rainAmounts,
  date: rainDates,
});

// Write Arrow Table to Parquet

// wasmTable is an Arrow table in WebAssembly memory
const wasmTable = parquet.Table.fromIPCStream(arrow.tableToIPC(rainfall, "stream"));
const writerProperties = new parquet.WriterPropertiesBuilder()
  .setCompression(parquet.Compression.ZSTD)
  .build();
const parquetUint8Array = parquet.writeParquet(wasmTable, writerProperties);

// Read Parquet buffer back to Arrow Table
// arrowWasmTable is an Arrow table in WebAssembly memory
const arrowWasmTable = parquet.readParquet(parquetUint8Array);

// table is now an Arrow table in JS memory
const table = arrow.tableFromIPC(arrowWasmTable.intoIPCStream());
console.log(table.schema.toString());
// Schema<{ 0: precipitation: Float32, 1: date: Date64<MILLISECOND> }>

Published examples

(These may use older versions of the library with a different API).

Performance considerations

Tl;dr: When you have a Table object (resulting from readParquet), try the new Table.intoFFI API to move it to JavaScript memory. This API is less well tested than the Table.intoIPCStream API, but should be faster and have much less memory overhead (by a factor of 2). If you hit any bugs, please create a reproducible issue.

Under the hood, parquet-wasm first decodes a Parquet file into Arrow in WebAssembly memory. But then that WebAssembly memory needs to be copied into JavaScript for use by Arrow JS. The "normal" conversion APIs (e.g. Table.intoIPCStream) use the Arrow IPC format to get the data back to JavaScript. But this requires another memory copy inside WebAssembly to assemble the various arrays into a single buffer to be copied back to JS.

Instead, the new Table.intoFFI API uses Arrow's C Data Interface to be able to copy or view Arrow arrays from within WebAssembly memory without any serialization.

Note that this approach uses the arrow-js-ffi library to parse the Arrow C Data Interface definitions. This library has not yet been tested in production, so it may have bugs!

I wrote an interactive blog post on this approach and the Arrow C Data Interface if you want to read more!

Example

import * as arrow from "apache-arrow";
import { parseRecordBatch } from "arrow-js-ffi";
// Edit the `parquet-wasm` import as necessary
import { readParquet, wasmMemory } from "parquet-wasm/node2";

// A reference to the WebAssembly memory object.
const WASM_MEMORY = wasmMemory();

const resp = await fetch("https://example.com/file.parquet");
const parquetUint8Array = new Uint8Array(await resp.arrayBuffer());
const wasmArrowTable = readParquet(parquetUint8Array).intoFFI();

const recordBatches: arrow.RecordBatch[] = [];
for (let i = 0; i < wasmArrowTable.numBatches(); i++) {
  // Note: Unless you know what you're doing, setting `true` below is recommended to _copy_
  // table data from WebAssembly into JavaScript memory. This may become the default in the
  // future.
  const recordBatch = parseRecordBatch(
    WASM_MEMORY.buffer,
    wasmArrowTable.arrayAddr(i),
    wasmArrowTable.schemaAddr(),
    true
  );
  recordBatches.push(recordBatch);
}

const table = new arrow.Table(recordBatches);

// VERY IMPORTANT! You must call `drop` on the Wasm table object when you're done using it
// to release the Wasm memory.
// Note that any access to the pointers in this table is undefined behavior after this call.
// Calling any `wasmArrowTable` method will error.
wasmArrowTable.drop();

Compression support

The Parquet specification permits several compression codecs. This library currently supports:

  • [x] Uncompressed
  • [x] Snappy
  • [x] Gzip
  • [x] Brotli
  • [x] ZSTD
  • [ ] LZ4 (deprecated)
  • [x] LZ4_RAW. Supported in arrow2 only.

LZ4 support in Parquet is a bit messy. As described here, there are two LZ4 compression options in Parquet (as of version 2.9.0). The original version LZ4 is now deprecated; it used an undocumented framing scheme which made interoperability difficult. The specification now reads:

It is strongly suggested that implementors of Parquet writers deprecate this compression codec in their user-facing APIs, and advise users to switch to the newer, interoperable LZ4_RAW codec.

It's currently unknown how widespread the ecosystem support is for LZ4_RAW. As of pyarrow v7, it now writes LZ4_RAW by default and presumably has read support for it as well.

Custom builds

In some cases, you may know ahead of time that your Parquet files will only include a single compression codec, say Snappy, or even no compression at all. In these cases, you may want to create a custom build of parquet-wasm to keep bundle size at a minimum. If you install the Rust toolchain and wasm-pack (see Development), you can create a custom build with only the compression codecs you require.

Note that this project uses Cargo syntax newly released in version 1.60. So you need version 1.60 or higher to compile this project. To upgrade your toolchain, use rustup update stable.

Example custom builds

Reader-only bundle with Snappy compression using the arrow and parquet crates:

wasm-pack build --no-default-features --features arrow1 --features snappy --features reader

Writer-only bundle with no compression support using the arrow2 and parquet2 crates, targeting Node:

wasm-pack build --target nodejs --no-default-features --features arrow2 --features writer

Debug bundle with reader and writer support, targeting Node, using arrow and parquet crates with all their supported compressions, with console_error_panic_hook enabled:

wasm-pack build \
  --dev \
  --target nodejs \
  --no-default-features \
  --features arrow1 \
  --features reader \
  --features writer \
  --features all_compressions \
  --features debug
# Or, given the fact that the default feature includes several of these features, a shorter version:
wasm-pack build --dev --target nodejs --features debug

Refer to the wasm-pack documentation for more info on flags such as --release, --dev, target, and to the Cargo documentation for more info on how to use features.

Available features

By default, arrow, all_compressions, reader, and writer features are enabled. Use --no-default-features to remove these defaults.

  • arrow1: Use the arrow and parquet crates
  • arrow2: Use the arrow2 and parquet2 crates
  • reader: Activate read support.
  • writer: Activate write support.
  • async: Activate asynchronous read support (only applies to the arrow2 endpoints).
  • all_compressions: Activate all supported compressions for the crate(s) in use.
  • brotli: Activate Brotli compression.
  • gzip: Activate Gzip compression.
  • snappy: Activate Snappy compression.
  • zstd: Activate ZSTD compression.
  • lz4: Activate LZ4_RAW compression (only applies to the arrow2 endpoints).
  • debug: Expose the setPanicHook function for better error messages for Rust panics.

Node <20

On Node versions before 20, you'll have to polyfill the Web Cryptography API.

Future work

  • [ ] Example of pushdown predicate filtering, to download only chunks that match a specific condition
  • [ ] Column filtering, to download only certain columns
  • [ ] More tests

Acknowledgements

A starting point of my work came from @my-liminal-space's read-parquet-browser (which is also dual licensed MIT and Apache 2).

@domoritz's arrow-wasm was a very helpful reference for bootstrapping Rust-WASM bindings.

Package Sidebar

Install

npm i parquet-wasm

Weekly Downloads

7,338

Version

0.6.0-beta.2

License

MIT OR Apache-2.0

Unpacked Size

16.4 MB

Total Files

18

Last publish

Collaborators

  • kylebarron