parquet-json

Apache Parquet JSON integration

Description

This project is a spin-off of the parquet-mr project. We propose to implement a converter to write JsonNode objects to parquet directly without intermediately format. To do so, this project implements the WriteSupport interface for Jackson JsonNode objects, and relies on a OpenAPI based schema definition.

This project is mostly based on the ProtocolBuffer and Avro converters implementations.

Data mapping

OpenAPI Type	OpenAPI format	Parquet	Comment
integer	int16	int16	not a valid OAPI type
integer	int32	int32
integer	int64	int64
integer	-	int32	default format int32
number	float	float
number	double	double
number	-	float	default format float
string	-	String	logical type
string	password	String	logical type
string	email	String	logical type
string	UUID	String	to be improved
string	byte	String	base64 encoded bytes string
string	binary	binary	not supported
string	date	date	logical type
string	date-time	timestamp	MILLIS precision
boolean	-	boolean
arrays	-	list	logical type, array of maps not implemented
object	-	GroupType
oneOf	-	Union	not implemented
allOf	-		not supported
map	-	map	keys as string only, "free form" objects and "Fixed Keys" not supported
enum	-	enum	only string type supported

How to use the converter

Given for example a schema definition in a file openapi.yaml as:

openapi: 3.0.1
info:
  title: Some schemas
  description: Some schemas for parquet-json usage example
  version: 1.0.0
servers:
  - url: 'https://getyourguide.com'
paths: {}
components:
  schemas:
    MyObject:
      title: MyObject
      type: object
      properties:
        key_string:
          type: string
          nullable: false
          default: 'a string'
        key_int32:
          type: integer
          format: int32
          nullable: true
          default: 1
        is_true:
          type: boolean
          nullable: true
          default: true

the converter can be used to write a parquet file on the local FS with:

    Configuration conf = new Configuration();
    conf.set("fs.file.impl", org.apache.hadoop.fs.LocalFileSystem.class.getName());

    OpenAPI openAPI = new OpenAPIV3Parser().read("openapi.yaml");

    ObjectSchema schema = (ObjectSchema) openAPI.getComponents().getSchemas().get("MyObject");

    ObjectMapper mapper = new ObjectMapper();

    String output = "./example.parquet";
    Path path = new Path(output);

    ParquetWriter<JsonNode> writer =
        JsonParquetWriter.Builder(path)
            .withSchema(schema)
            .withConf(conf)
            .withCompressionCodec(CompressionCodecName.SNAPPY)
            .withDictionaryEncoding(true)
            .withPageSize(1024 * 1024)
            .build();

    String json =
        "{\"key_string\":\"hello\",\"key_int32\":32,\"is_true\":true}";
    JsonNode payload = mapper.readTree(json);
    
    writer.write(payload);
    
    writer.close();

Known limitations

• Currently works only with schemas of type OpenAPI (https://github.com/swagger-api/swagger-parser/) and data payload of type JsonNode (Jackson library).
• The schema must be fully resolved (no internal or external ref)
• Union types (oneOf) not implemented yet
• Readers (from Parquet to JsonNode/OpenAPI) are not implemented (we don't need this part here at GetYourGuide)

Contributing

We welcome pull requests; if you are planning to perform bigger changes then it makes sense to file an issue first.

Security

For sensitive security matters please contact [email protected].

Legal

Copyright 2020 GetYourGuide GmbH.

parquet-json is licensed under the Apache License, Version 2.0. See LICENSE for the full text.