hyparquet-writer/src/write.js

95 lines
2.8 KiB
JavaScript
Raw Normal View History

import { writeColumn } from './column.js'
2025-03-26 04:06:43 +00:00
import { Writer } from './writer.js'
import { writeMetadata } from './metadata.js'
import { getSchemaElementForValues } from './schema.js'
2025-03-26 04:06:43 +00:00
/**
* Write data as parquet to an ArrayBuffer
*
2025-04-03 07:42:54 +00:00
* @import {ColumnChunk, DecodedArray, FileMetaData, RowGroup, SchemaElement, SchemaTree} from 'hyparquet'
2025-04-01 06:32:14 +00:00
* @import {KeyValue} from 'hyparquet/src/types.js'
* @import {ColumnData} from '../src/types.js'
2025-03-27 07:27:22 +00:00
* @param {object} options
* @param {ColumnData[]} options.columnData
* @param {boolean} [options.compressed]
2025-04-03 20:21:57 +00:00
* @param {boolean} [options.statistics]
2025-04-03 07:42:54 +00:00
* @param {number} [options.rowGroupSize]
2025-03-31 20:42:57 +00:00
* @param {KeyValue[]} [options.kvMetadata]
2025-03-26 04:06:43 +00:00
* @returns {ArrayBuffer}
*/
2025-04-03 20:21:57 +00:00
export function parquetWrite({ columnData, compressed = true, statistics = true, rowGroupSize = 100000, kvMetadata }) {
const num_rows = columnData.length ? BigInt(columnData[0].data.length) : 0n
2025-04-03 07:42:54 +00:00
const writer = new Writer()
2025-03-26 04:06:43 +00:00
2025-04-03 07:42:54 +00:00
// construct schema
2025-03-26 04:06:43 +00:00
/** @type {SchemaElement[]} */
const schema = [{
name: 'root',
num_children: columnData.length,
2025-03-26 04:06:43 +00:00
}]
2025-03-28 23:13:27 +00:00
for (const { name, data, type } of columnData) {
2025-04-03 07:42:54 +00:00
// check if all columns have the same length
if (BigInt(data.length) !== num_rows) {
throw new Error('columns must have the same length')
}
// auto-detect type
2025-03-28 23:13:27 +00:00
const schemaElement = getSchemaElementForValues(name, data, type)
if (!schemaElement.type) throw new Error(`column ${name} cannot determine type`)
schema.push(schemaElement)
2025-04-03 07:42:54 +00:00
}
// write header PAR1
writer.appendUint32(0x31524150)
/** @type {RowGroup[]} */
const row_groups = []
for (let i = 0; i < num_rows; i += rowGroupSize) {
const groupStart = writer.offset
// row group columns
/** @type {ColumnChunk[]} */
const columns = []
// write columns
for (let i = 0; i < columnData.length; i++) {
const { name, data } = columnData[i]
const file_offset = BigInt(writer.offset)
const schemaPath = [schema[0], schema[i + 1]]
2025-04-03 20:21:57 +00:00
const meta_data = writeColumn(writer, schemaPath, data, compressed, statistics)
2025-04-03 07:42:54 +00:00
// save metadata
columns.push({
file_path: name,
file_offset,
meta_data,
})
}
row_groups.push({
columns,
total_byte_size: BigInt(writer.offset - groupStart),
num_rows: BigInt(Math.min(rowGroupSize, Number(num_rows) - i)),
2025-03-26 04:06:43 +00:00
})
}
2025-04-03 07:42:54 +00:00
// write metadata
2025-03-26 04:06:43 +00:00
/** @type {FileMetaData} */
const metadata = {
version: 2,
created_by: 'hyparquet',
schema,
num_rows,
2025-04-03 07:42:54 +00:00
row_groups,
2025-03-26 04:06:43 +00:00
metadata_length: 0,
2025-03-31 20:42:57 +00:00
key_value_metadata: kvMetadata,
2025-03-26 04:06:43 +00:00
}
// @ts-ignore don't want to actually serialize metadata_length
delete metadata.metadata_length
writeMetadata(writer, metadata)
2025-04-03 07:42:54 +00:00
// write footer PAR1
2025-03-26 04:06:43 +00:00
writer.appendUint32(0x31524150)
return writer.getBuffer()
}