Ignite write-through with Postgres backend data conversion

Question

I'm setting up Apache Ignite with Postgres as a persistent store. In postgres I have JSON and Array columns, so I implemented custom transformer to convert everything that is not primitive to JSON string. Here is the code:

public class PostgresTransformer extends JdbcTypesDefaultTransformer {

    @Override
    public Object getColumnValue(ResultSet rs, int colIdx, Class<?> type) throws SQLException {
        if (type == ArrayOfFloat.class) {
            Array raw = rs.getArray(colIdx);
            ArrayOfFloat obj = new ArrayOfFloat(raw);
            return obj.toString();
        }

        if (type == ArrayOfInt.class) {
            Array raw = rs.getArray(colIdx);
            ArrayOfInt obj = new ArrayOfInt(raw);
            return obj.toString();
        }

        if (type == ArrayOfString.class) {
            Array raw = rs.getArray(colIdx);
            ArrayOfString obj = new ArrayOfString(raw);
            return obj.toString();
        }

        if (type == JSON.class) {
            String raw = rs.getString(colIdx);
            JSON obj = new JSON(raw);
            return obj.toString();
        }

        return super.getColumnValue(rs, colIdx, type);
    }
}

In main-server.xml I also use this corresponding configuration:

<property name="transformer">
     <bean class="com.mycorp.ignite.transformers.PostgresTransformer"/>
</property>

<bean class="org.apache.ignite.cache.store.jdbc.JdbcTypeField">
      <constructor-arg>
            <util:constant static-field="java.sql.Types.OTHER"/>
      </constructor-arg>
      <constructor-arg value="neighbourhood_stats"/>
      <constructor-arg value="com.mycorp.ignite.types.JSON"/>
      <constructor-arg value="neighbourhoodStats"/>
</bean>

This works well for me for read-through. My question is: How to convert it back to valid postgres data type when i'm saving data back to DB via wright-through? There is no setColumnValue method in transformer. Only getColumnValue . Thanks!

Answer 1

I don't think you should use CacheJdbcPojoStore in this case. Just create your own implementation of the CacheStore interface and do all the required conversions there.