avro-kedavro

A library to parse raw json to avro with magic!

Why avro-kedavro?

• We want to store information about wizards in S3 to query with Athena later using this schema:

{
  "name": "Wizard",
  "type": "record",
  "namespace": "com.avro.kedavro",
  "fields": [
      {
		"name": "name",
		"type": [
		  "null",
		  "string"
		],
		"default": null
	  },
      {
		"name": "id",
		"type": "long"
	  },
      {
		"name": "timestamp",
		"type": "long",
        "logicalType": "timestamp-millis"
	  }
  ]
}

• We have multiple sources reporting wizards and leaving the data in JSON format in a stream... the problem is that each source uses a different format, so we have reports like:

  • {"name": "Harry", "id": 12345, "timestamp": 1571128870}
  • {"name": "Ron", "id": "98765", "timestamp": 1571128870}
  • {"name": "Hermione", "id": "56784", "timestamp": "1571128870000"}

  None of these reports are valid for the schema we have:

  • All of them will fail just with the name field, since the union in JSON-avro should be: "name": {"string": "..."}
  • Only the first record has the id as long
  • The schema expects the timestamp as a long with milliseconds, but none of the reports is correct: or they don't have milliseconds, or it's a string instead of a long

• We could try to implement an specific solution for each record, but what happens when we start dealing with 10 different types? And with 100? And even more, what if we want to change some schemas? Changing a schema would mean to go through all the parsers we built for specific "events". So we need some kind of magic where we have:

  • avro schema
  • JSON record
  • Some rules like: switch strings to numbers, or switch timestamps to timestamps with milliseconds, ...

  Well... that magic is avro-kedavro!

How to use it

avro-kedavro is design to work with goavro. The idea is avro-kedavro will parse your raw JSON to avro-JSON supported by your schema, so you can use goavro to generate your avro OCF files

Example:

import (
	"encoding/json"
	"fmt"

	"github.com/linkedin/goavro"
	"github.com/ouzi-dev/avro-kedavro/pkg/kedavro"
)

const schema = `{
	"name": "Wizard",
	"type": "record",
	"namespace": "com.avro.kedavro",
	"fields": [
		{
		  "name": "name",
		  "type": [
			"null",
			"string"
		  ],
		  "default": null
		},
		{
		  "name": "id",
		  "type": "long"
		},
		{
		  "name": "timestamp",
		  "type": "long",
		  "logicalType": "timestamp-millis"
		}
	]
  }`

const JSONrecord = `{"name": "Voldemort", "id": "66666", "timestamp": "1571128870"}`

func ParseToJSONAvro() error {
	p, err := kedavro.NewParser(schema, kedavro.WithStringToNumber(), kedavro.WithTimestampToMillis())
	if err != nil {
		// Error parsing schema
		return err
	}

	avroJSON, err := p.Parse([]byte(JSONrecord))
	if err != nil {
		// Error parsing record with schema
		return err
	}

	// Marshal the map to show the result from avro-kedavro
	kedavroJSONResult, err := json.Marshal(avroJSON)
	if err != nil {
		// Error marshaling kedavro result
		return err
	}

	fmt.Println(string(kedavroJSONResult))
	// this will print: {"name": {"string": "Voldemort"}, "id": 66666, "timestamp": 1571128870000}

	// use goavro to test the generated avroJSON is valid for the schema
	codec, err := goavro.NewCodec(schema)
	if err != nil {
		// Error parsing schema
		return err
	}

	textual, err := codec.TextualFromNative(nil, avroJSON)
	if err != nil {
		// Error avroJSON
		return err
	}

	fmt.Println(string(textual))
	// this will print: {"name": {"string": "Voldemort"}, "id": 66666, "timestamp": 1571128870000}
	return nil
}

Options

avro-kedavro supports 4 different options for now:

• WithStringToNumber() will try to parse strings as numbers: {"test": "1234.56"} => {"test": 1234.56}
• WithStringToBool() will try to parse strings as booleans: {"test": "False"} => {"test": false}
• WithTimestampToMillis() will add milliseconds to timestamps, only works for logicalType="timestamp-millis" fields: {"test": 1571128870} => {"test": time.Time(1571128870000)}
• WithTimestampToMicros() will add microseconds to timestamps, only works for logicalType="timestamp-micros" fields: {"test": 1571128870} => {"test": time.Time(1571128870000000)}
• WithDateTimeFormat(format string) will try to parse a string to a timestamp using the format specified as param, only works for logicalType="timestamp-millis" or logicalType="timestamp-micros" fields: {"test": "2019-10-14T12:45:18Z"} => (using time.RFC3339 as format and type logicalType="timestamp-millis) => {"test": time.Time(15710571180000)}
• WithNowForNullTimestamp will set time.Now() if the field is null, only works for logicalType="timestamp-millis" or logicalType="timestamp-micros" fields.

Supported types

Not all the avro types are supported by avro-kedavro yet! The current supported types are:

Avro	Go
null	nil
boolean	bool
bytes	[]byte
float	float32
double	float64
long	int64
int	int32
string	string
union	see below
record	map[string]interface{}

Unsupported types:

Avro
enum
fixed
map
array

Supported Unions

Only unions with two elements where the first one is null and the second is a supported type different than record are currently supported by avro-kedavro:

First field	Second field
null	boolean
null	bytes
null	float
null	double
null	long
null	int
null	string

Supported Logical Types

For now only two logical types are supported:

Avro	Go
timestamp-millis	time.Time
timestamp-micros	time.Time

About timestamps

For logical types of type timestamp, the schema has to be defined always as a long.

Accepted values in json for timestamps are:

• Numeric values: for example 1586502702 will be accepted as a timestamp, if a numeric value has decimals, those decimals will be ignored when parsing to time.Time
• Strings: only if WithStringToNumber() option is provided, the string will be parsed like:
  • If the string is a number without decimals: it will be treated as a timestamp (in seconds, milliseconds, or microseconds depending on the provided options to the parser)
  • If the string is a number with decimals: it will be treated as a timestamp where the decimals will be consider fractions of seconds.
    • If the selected type is timestamp-millis the parser will keep the first three decimals.
    • If the selected type is timestamp-micros the parser will keep the first six decimals.
  • If the string has non-numeric characters: the parser will try to parse the string to time.Time using the provided format with the option WithDateTimeFormat(format string)
• null: only if WithNowForNullTimestamp() option is provided. When the option is provided, if a null is found for a timestamp-millis or timestamp-micros field, time.Now() will be used as value.