Can I split an Apache Avro schema across multiple files?

Question

I can do,

{
    \"type\": \"record\",
    \"name\": \"Foo\",
    \"fields\": [
        {\"name\": \"bar\", \"type\": {
            \"type\": \"record\",
             


        

Answer 1

I assume, your motivation is (as my own) structuring your schema definition and avoiding copy&paste-errors.

To achieve that, you can also use Avro IDL. It allows to define avro schemas on a higher level. Reusing types is possible within the same file and also across multiple files.

To generate the .avsc-files run

$ java -jar avro-tools-1.7.7.jar idl2schemata my-protocol.avdl

The resulting .avsc-files will look pretty much the same as your initial example, but as they are generated from the .avdl you'll not get lost in the verbose json-format.

Answer 2

Yes, it's possible.

I've done that in my java project by defining common schema files in avro-maven-plugin Example:

search_result.avro:

{"namespace": "com.myorg.other",
 "type": "record",
 "name": "SearchResult",
 "fields": [
     {"name": "type", "type": "SearchResultType"},
     {"name": "keyWord",  "type": "string"},
     {"name": "searchEngine", "type": "string"},
     {"name": "position", "type": "int"},
     {"name": "userAction", "type": "UserAction"}
 ]
}

search_suggest.avro:

{"namespace": "com.myorg.other",
 "type": "record",
 "name": "SearchSuggest",
 "fields": [
     {"name": "suggest", "type": "string"},
     {"name": "request",  "type": "string"},
     {"name": "searchEngine", "type": "string"},
     {"name": "position", "type": "int"},
     {"name": "userAction", "type": "UserAction"},
     {"name": "timestamp", "type": "long"}
 ]
}

user_action.avro:

{"namespace": "com.myorg.other",
 "type": "enum",
 "name": "UserAction",
 "symbols": ["S", "V", "C"]
}

search_result_type.avro

{"namespace": "com.myorg.other",
 "type": "enum",
 "name": "SearchResultType",
 "symbols": ["O", "S", "A"]
}

avro-maven-plugin configuration:

<plugin>
    <groupId>org.apache.avro</groupId>
<artifactId>avro-maven-plugin</artifactId>
<version>1.7.4</version>
    <executions>
    <execution>
        <phase>generate-sources</phase>
        <goals>
        <goal>schema</goal>
        </goals>
    <configuration>
     <sourceDirectory>${project.basedir}/src/main/resources/avro</sourceDirectory>
         <outputDirectory>${project.basedir}/src/main/java/</outputDirectory>
     <includes>
         <include>**/*.avro</include>
     </includes>
     <imports>
              <import>${project.basedir}/src/main/resources/avro/user_action.avro</import>
              <import>${project.basedir}/src/main/resources/avro/search_result_type.avro</import>
     </imports>
       </configuration>
     </execution>
</executions>
</plugin>

Answer 3

The order of imports in the pom.xml matters. You must import the subtypes first before processing the rest.

<imports>
    <import>${project.basedir}/src/main/resources/avro/Bar.avro</import>
    <import>${project.basedir}/src/main/resources/avro/Foo.avro</import>
</imports>

That would unblock the codegen from emitting undefined name: Bar.avro error.

Answer 4

You need to import the avsc file in avro-maven plugin where you have first written the object schema that you want to reuse

<plugin>
<groupId>org.apache.avro</groupId>
<artifactId>avro-maven-plugin</artifactId>
<version>${avro.maven.plugin.version}</version>
<configuration>
    <stringType>String</stringType>
</configuration>
<executions>
    <execution>
        <phase>generate-sources</phase>
        <goals>
            <goal>schema</goal>
        </goals>
        <configuration>
            <sourceDirectory>src/main/java/com/xyz/avro</sourceDirectory> // Avro directory
            <imports>
                <import>src/main/java/com/xyz/avro/file.avsc</import> // Import here
            </imports>
        </configuration>
    </execution>
</executions>

Answer 5

You can also define multiple schemas inside of one file:

schemas.avsc:

[
{
    "type": "record",
    "name": "Bar",
    "fields": [ ]
},
{
    "type": "record",
    "name": "Foo",
    "fields": [
        {"name": "bar", "type": "Bar"}
    ]
}
]

If you want to reuse the schemas in multiple places this is not super nice but it improves readability and maintainability a lot in my opinion.

Answer 6

From what I have been able to figure out so far, no.

There is a good write up about someone who coded their own method for doing this here:

http://www.infoq.com/articles/ApacheAvro