Backward Comaptibility issue and uncertainity in Schema Registry

问题

I have a use case where I have a JSON and I want to generate schema and record out of the JSON and publish a record. I have configured the value serializer and Schema setting is Backward compatible.

First JSON

String json = "{\n" +

     "    \"id\": 1,\n" +

     "    \"name\": \"Headphones\",\n" +

     "    \"price\": 1250.0,\n" +

     "    \"tags\": [\"home\", \"green\"]\n" +

     "}\n"
     ;

Version 1 schema registered.

Received message in avro console consumer.

Second JSON.

String json = "{\n" +

    "    \"id\": 1,\n" +

    "    \"price\": 1250.0,\n" +
    "    \"tags\": [\"home\", \"green\"]\n" +
    "}\n"
    ;

Registered schema Successfully. Sent message.

Now tried sending the JSON 1 sent successfully

Schema 3:

String json = "{\n" +

    "    \"id\": 1,\n" +
    "    \"name\": \"Headphones\",\n" +

    "    \"tags\": [\"home\", \"green\"]\n" +
    "}\n"
    ;

Got error for this case. Caused by: io.confluent.kafka.schemaregistry.client.rest.exceptions.RestClientException: Schema being registered is incompatible with an earlier schema; error code: 409

How is that schema generated from 2nd JSON was registered and the third one was rejected? Although I didn't have any Default key for the deleted field? Is it that the Schema Registry always accepts the 1st evolution? (2nd schema over 1st)

Schema in schema registry

Version 1 schema

{ "fields": [

{

  "doc": "Type inferred from '1'",

  "name": "id",

  "type": "int"

},

{

  "doc": "Type inferred from '\"Headphones\"'",

  "name": "name",

  "type": "string"

},

{

  "doc": "Type inferred from '1250.0'",

  "name": "price",

  "type": "double"

},

{

  "doc": "Type inferred from '[\"home\",\"green\"]'",

  "name": "tags",

  "type": {

    "items": "string",

    "type": "array"

  }

}
],
"name": "myschema",
"type": "record"   }

Version 2:

{ "fields": [

{

  "doc": "Type inferred from '1'",

  "name": "id",

  "type": "int"

},

{

  "doc": "Type inferred from '1250.0'",

  "name": "price",

  "type": "double"

},

{

  "doc": "Type inferred from '[\"home\",\"green\"]'",

  "name": "tags",

  "type": {

    "items": "string",

    "type": "array"

  }

}
],
"name": "myschema",
"type": "record"   }

回答1:

Let's go over the backwards compatibility rules... https://docs.confluent.io/current/schema-registry/avro.html#compatibility-types

First, the default isn't transitive, so version 3 only will look at version 2.

The backwards rule states you can delete fields or add optional fields (those with a default). I assume your schema generator tool doesn't know how to use optionals, so you're only allowed to delete, not add.

Between version 1 and 2, you've deleted the name field, which is valid.

Between version 2 and the incoming 3, it thinks you're trying to post a new schema which removes price (this is okay}, but adds a required name field, which is not allowed.

来源：https://stackoverflow.com/questions/59626557/backward-comaptibility-issue-and-uncertainity-in-schema-registry

标签

apache-kafka

avro

kafka-producer-api

confluent-schema-registry