Hi. I am working with Flink DataStream API. We have a custom window with custom Window state. To not write custom serizalization code we just reused Flink POJO serializer like that:

@Override
public TypeSerializer<SessionTimeWindow> getWindowSerializer(ExecutionConfig executionConfig) {
    return Types.POJO(SessionTimeWindow.class).createSerializer(new ExecutionConfig());
}

Unfortunately there is an issue with schema evolution of this state.

org.apache.flink.util.StateMigrationException: The new namespace serializer (org.apache.flink.api.java.typeutils.runtime.PojoSerializer@e18e3041) must be compatible with the old namespace serializer (org.apache.flink.api.java.typeutils.runtime.PojoSerializer@97bf27ff).

after debugging that I found out that in

org.apache.flink.contrib.streaming.state.RocksDBKeyedStateBackend#updateRestoredStateMetaInfo

TypeSerializerSchemaCompatibility<N> s =
                restoredKvStateMetaInfo.updateNamespaceSerializer(namespaceSerializer);
        if (s.isCompatibleAfterMigration() || s.isIncompatible()) {
            throw new StateMigrationException(

the

updateNamespaceSerializer

does return

Type.COMPATIBLE_AFTER_MIGRATION

which is like I would expect but then why it is written in a way that compatible migratable state still throws a

StateMigrationException

? What am I missing?