Formatter

Serializes buffered BulkWriter columns into one or more import files through the JSON or Parquet formatter implementation.

interface Formatter

IMPLEMENTATIONS:

• class JsonFormatter implements Formatter

  Writes buffered rows to a JSON file whose top-level rows property contains the serialized entities, including dynamic fields when enabled.

  • extension -

    [REQUIRED]

    Returns .json for generated files.

  • persist -

    [REQUIRED]

    Serializes the buffered columns and returns the generated JSON file paths.

• class ParquetFormatter implements Formatter

  Writes buffered rows to Parquet files using @shanghaikid/parquetjs and converts Milvus scalar, vector, array, and dynamic-field values to Parquet-compatible representations.

  • extension -

    [REQUIRED]

    Returns .parquet for generated files.

  • persist -

    [REQUIRED]

    Serializes the buffered columns and returns the generated Parquet file paths.

Example

Choose a formatter implementation

Creates the JSON and Parquet implementations exposed by the SDK.

import { JsonFormatter, ParquetFormatter } from '@zilliz/milvus2-sdk-node';

const jsonFormatter = new JsonFormatter();
const parquetFormatter = new ParquetFormatter();

console.log(jsonFormatter.extension); // .json
console.log(parquetFormatter.extension); // .parquet

Notes

• The Formatter interface exposes the readonly extension field and the persist(columns, dynamicRows, rowCount, dir, schema) method, which returns the generated local file paths.

• BulkWriter selects JsonFormatter when format is json and ParquetFormatter when format is parquet.

• Parquet output requires the SDK dependency @shanghaikid/parquetjs.