Object Serialization

GridGain 9 provides a way to serialize your java objects and types and send the data between servers and clients.

Native Types

GridGain handles native type serialization automatically. For example, the following Compute job accepts an Integer and returns an Integer:

class IntegerComputeJob implements ComputeJob<Integer, Integer> {
    @Override
    public @Nullable CompletableFuture<Integer> executeAsync(
        JobExecutionContext context, @Nullable Integer arg
    ) {
        return completedFuture(arg - 1);
    }
}

Since both the argument and the result are native types and are serialized automatically, you do not need any additional code to handle the serialization:

try (IgniteClient client = IgniteClient.builder().addresses("address/to/cluster:port").build()) {
    Integer result = client.compute().execute(
        JobTarget.anyNode(client.clusterNodes()),
        JobDescriptor.builder(IntegerComputeJob.class).build(),
        1
    );
}

Tuples

GridGain is designed around working with Tuples, and handles tuple serialization automatically. For example, the following Job accepts Tuple and returns Tuple:

class TupleComputeJob implements ComputeJob<Tuple, Tuple> {
    @Override
    public @Nullable CompletableFuture<Tuple> executeAsync(JobExecutionContext context, @Nullable Tuple arg) {
        Tuple resultTuple = Tuple.copy(arg);
        resultTuple.set("col", "new value");

        return completedFuture(resultTuple);
    }
}

Since both the argument and the result are tuples, they are serialized automatically, and you do not need to handle serialization:

try (IgniteClient client = IgniteClient.builder().addresses("address/to/cluster:port").build()) {
    Tuple resultTuple = client.compute().execute(
        JobTarget.anyNode(client.clusterNodes()),
        JobDescriptor.builder(TupleComputeJob.class).build(),
        Tuple.create().set("col", "value")
    );
}

User Objects

User objects are not marshalled automatically, and require user code to handle.

Here is an example of custom objects marshalled with custom logic (using ObjectMapper, but you can do whatever you think is good for your use case). Let’s start with Compute job definition that should be a part of the same deployment unit.

Server-Side

The code below shows how to handle marshalling on a server, so that it can properly send the data to the clients and receive their responses:

This is the custom object that we will be using as an argument the job:

class ArgumentCustomServerObject {
    int arg1;
    String arg2;
}

We need to define a marshaller for it using the ObjectMapper object:

final ObjectMapper MAPPER = new ObjectMapper();

class ArgumentCustomServerObjectMarshaller implements Marshaller<ArgumentCustomServerObject, byte[]> {
    @Override
    public byte @Nullable [] marshal(@Nullable ArgumentCustomServerObject object) throws UnsupportedObjectTypeMarshallingException {
        try {
            return MAPPER.writeValueAsBytes(object);
        } catch (JsonProcessingException e) {
            throw new RuntimeException(e);
        }
    }

    @Override
    public @Nullable ArgumentCustomServerObject unmarshal(byte @Nullable [] raw) throws UnsupportedObjectTypeMarshallingException {
        try {
            return MAPPER.readValue(raw, ArgumentCustomServerObject.class);
        } catch (IOException e) {
            throw new RuntimeException(e);
        }
    }
}

Let’s also create another object that will be used to store Compute job results, and the corresponding marshaller:

class ResultCustomServerObject {
    int res1;
    String res2;
    long res3;
}

class ResultCustomServerObjectMarshaller implements Marshaller<ResultCustomServerObject, byte[]> {
    @Override
    public byte @Nullable [] marshal(@Nullable ResultCustomServerObject object) throws UnsupportedObjectTypeMarshallingException {
        try {
            return MAPPER.writeValueAsBytes(object);
        } catch (JsonProcessingException e) {
            throw new RuntimeException(e);
        }
    }

    @Override
    public @Nullable ResultCustomServerObject unmarshal(byte @Nullable [] raw) throws UnsupportedObjectTypeMarshallingException {
        try {
            return MAPPER.readValue(raw, ResultCustomServerObject.class);
        } catch (IOException e) {
            throw new RuntimeException(e);
        }
    }
}

The marshallers above define how to represent corresponding objects as byte[], and how to read these objects from byte[]. However, defining these classes does not enable custom serialization, as you need to specify the marshaller to use when serializing objects. In GridGain, this is done by overriding two methods in Compute job definition to use them as factory methods for marshallers:

The code below provides an example of implementing marshallers in a compute job:

class PojoComputeJob implements ComputeJob<ArgumentCustomServerObject, ResultCustomServerObject> {

    @Override
    public @Nullable CompletableFuture<ResultCustomServerObject> executeAsync(
        JobExecutionContext context,
        @Nullable ArgumentCustomServerObject arg
    ) {
        ResultCustomServerObject res = new ResultCustomServerObject();
        res.res1 = arg.arg1;
        res.res2 = arg.arg2;
        res.res3 = 1;

        return completedFuture(res);
    }

    @Override
    public Marshaller<ArgumentCustomServerObject, byte[]> inputMarshaller() {
        return new ArgumentCustomServerObjectMarshaller();
    }

    @Override
    public Marshaller<ResultCustomServerObject, byte[]> resultMarshaller() {
        return new ResultCustomServerObjectMarshaller();
    }
}

With this, the GridGain server will be able to handle marshalling the required objects to sending them to clients, and unmarshalling the client responses.

Client-Side

On the client side, largely the same code is required to handle the incoming objects and to marshal the response:

Define the custom object that is used for compute job:

class ArgumentCustomClientObject {
    int arg1;
    String arg2;
}

Define the marshaller for the object:

final ObjectMapper MAPPER = new ObjectMapper();

class ArgumentCustomClientObjectMarshaller implements Marshaller<ArgumentCustomClientObject, byte[]> {
    @Override
    public byte @Nullable [] marshal(@Nullable ArgumentCustomClientObject object) throws UnsupportedObjectTypeMarshallingException {
        try {
            return MAPPER.writeValueAsBytes(object);
        } catch (JsonProcessingException e) {
            throw new RuntimeException(e);
        }
    }

    @Override
    public @Nullable ArgumentCustomClientObject unmarshal(byte @Nullable [] raw) throws UnsupportedObjectTypeMarshallingException {
        try {
            return MAPPER.readValue(raw, ArgumentCustomClientObject.class);
        } catch (IOException e) {
            throw new RuntimeException(e);
        }
    }
}

Do the same for the result object:

class ResultCustomClientObject {
    int res1;
    String res2;
    long res3;
}


class ResultCustomClientObjectMarshaller implements Marshaller<ResultCustomClientObject, byte[]> {
    @Override
    public byte @Nullable [] marshal(@Nullable ResultCustomClientObject object) throws UnsupportedObjectTypeMarshallingException {
        try {
            return MAPPER.writeValueAsBytes(object);
        } catch (JsonProcessingException e) {
            throw new RuntimeException(e);
        }
    }

    @Override
    public @Nullable ResultCustomClientObject unmarshal(byte @Nullable [] raw) throws UnsupportedObjectTypeMarshallingException {
        try {
            return MAPPER.readValue(raw, ResultCustomClientObject.class);
        } catch (IOException e) {
            throw new RuntimeException(e);
        }
    }
}

// ....

Now that all marshallers are defined, you can start working with the custom objects and handle marshalling of arguments and results in your compute jobs:

try (IgniteClient client = IgniteClient.builder().addresses("address/to/cluster:port").build()) {

    // Marshalling example of pojo.
    ResultCustomClientObject resultPojo = client.compute().execute(
        JobTarget.anyNode(client.clusterNodes()),
        JobDescriptor.<ArgumentCustomClientObject, ResultCustomClientObject>builder(PojoComputeJob.class.getName())
            .argumentMarshaller(new ArgumentCustomClientObjectMarshaller())
            .resultMarshaller(new ResultCustomClientObjectMarshaller())
            .build(),
        new ArgumentCustomClientObject()
    );
}

© 2024 GridGain Systems, Inc. All Rights Reserved. Privacy Policy | Legal Notices. GridGain® is a registered trademark of GridGain Systems, Inc.
Apache, Apache Ignite, the Apache feather and the Apache Ignite logo are either registered trademarks or trademarks of The Apache Software Foundation.

Last updated on Aug 30, 2024