This page is meant as a template for writing a KIP. To create a KIP choose Tools->Copy on this page and modify with your content and replace the heading with the next KIP number and a description of your issue. Replace anything in italics with your own description.

Status

Current state: Under Discussion

Discussion thread: here

JIRA: here

Please keep the discussion on the mailing list rather than commenting on the wiki (wiki discussions get unwieldy fast).

Motivation

Though there does exist a converter in the connect-json library called "JsonConverter", there are limitations as to the domain of JSON payloads this converter is compatible with on the Sink Connector side when serializing them into Kafka Connect datatypes; When reading byte arrays from Kafka, the JsonConverter expects its inputs to be a JSON envelope that contains the fields "schema" and "payload", otherwise it'll throw a DataException reporting:

JsonConverter with schemas.enable requires "schema" and "payload" fields and may not contain additional fields. If you are trying to deserialize plain JSON data, set schemas.enable=false in your converter configuration.

(when schemas.enable is true) or

JSON value converted to Kafka Connect must be in envelope containing schema

(when schemas.enable is false)

For example, if your JSON payload looks something on the order of:

...

{"test1":"hi","test2":"there","test3":12,"test4":12.5,"test5":null,"test6":true,"test7":false,"test8":["element1","element2"],"test9":[],"test10":[1]}

The inferred schema should yield a STRUCT with the following fields:

This is achievable by the introduction of introducing a recursive method, inferSchema(JsonNode jsonValue), which is capable of both 1.) inferring the schema for simple JSON documents, and 2.) making a recursive call to break down JSON documents with complex data types into their constituent parts and Schemas:

 privateprivate Schema inferSchema(JsonNode jsonValue) {
        switch (jsonValue.getNodeType()) {
            case NULL:
                return Schema.OPTIONAL_STRING_SCHEMA;
            case BOOLEAN:
                return Schema.BOOLEAN_SCHEMA;
            case NUMBER:
                if (jsonValue.isIntegralNumber()) {
                    return Schema.INT64_SCHEMA;
                }
                else {
                    return Schema.FLOAT64_SCHEMA;
                }
            case ARRAY:
                SchemaBuilder arrayBuilder = SchemaBuilder.array(jsonValue.elements().hasNext() ? inferSchema(jsonValue.elements().next()) : Schema.OPTIONAL_STRING_SCHEMA);
                return arrayBuilder.build();
            case OBJECT:
                SchemaBuilder structBuilder = SchemaBuilder.struct();
                Iterator<Map.Entry<String, JsonNode>> it = jsonValue.fields();
                while (it.hasNext()) {
                    Map.Entry<String, JsonNode> entry = it.next();
                    structBuilder.field(entry.getKey(), inferSchema(entry.getValue()));
                }
                return structBuilder.build();
            case STRING:
                return Schema.STRING_SCHEMA;
        case BINARY:
        defaultcase MISSING:
        case POJO:
       return null;default:
         }   return null;
    }
}

In the current model, the assumption made in the event that a JSON value is not capable of being intelligibly inferenced (null, []) is that the actual schema is a STRING.

Compatibility, Deprecation, and Migration Plan

• What impact (if any) will there be on existing users?
• If we are changing behavior how will we phase out the older behavior?
• If we need special migration tools, describe them here.
• When will we remove the existing behavior?

Rejected Alternatives

Since the default value for schemas.infer.enable is false, existing instances of JsonConverters will remain unaffected. In order to migrate to using these new capabilities for existing connectors already configured to use the JsonConverter, users will have to send a PUT request to the /connectors/{connector name}/config endpoint and include this property:

value.converter.schemas.infer.enable: "true"

Once the connector properties are updated, users can then feel free to strip away schema data from the upstream producers as they are no longer necessary for unmarshalling the Kafka Connect Schema.

Rejected Alternatives

N/A

 If there are alternative ways of accomplishing the same thing, what were they? The purpose of this section is to motivate why the design is the way it is and not some other way.