I implmented a custom cache store for Ignite backed by DynamoDB:

public class DynamoDBCacheStore extends CacheStoreAdapter<Integer, String> {
    private final DynamoDbTable<Sensor> table;

    public DynamoDBCacheStore() {
       // initialize the DynamoDB client
    }

    @Override
    public Sensor load(Integer key) throws CacheLoaderException {
        // impl here
    }


    @Override
    public void write(Cache.Entry<? extends Integer, ? extends String> entry) throws CacheWriterException {
        // impl here
    }

    @Override
    public void delete(Object key) throws CacheWriterException {
        // impl here
    }
}

I would like to enable SQL support for this so I can use the cache with JDBC driver:

        CacheConfiguration<Integer, String> cacheCfg = new CacheConfiguration<Integer, String>()
                .setCacheStoreFactory(FactoryBuilder.factoryOf(DynamoDBCacheStore.class))
                .setCacheMode(CacheMode.PARTITIONED)
                .setAtomicityMode(CacheAtomicityMode.ATOMIC)
                .setName(cacheName)
                .setReadThrough(true)
                .setWriteThrough(true)
                .setIndexedTypes(Integer.class, String.class)

        IgniteConfiguration cfg = new IgniteConfiguration()
                .setIgniteInstanceName("embedded-ignite")
                .setClientMode(false)
                .setCacheConfiguration(cacheCfg);

However, when the cache is empty (cold start) without calling the cache.loadCache(), SQL API doesn’t trigger read through behvaior:

            // Running SQL query
            String sql = "SELECT * FROM LookUpTable WHERE id = ?";
            try (QueryCursor<List<?>> cursor = personCache.query(new SqlFieldsQuery(sql).setArgs(25))) {
                for (List<?> row : cursor) {
                    System.out.println(row);
                }
            }

Using the Key/Value API, read through behavior is enforced:

IgniteCache<Integer, String> cache = ignite.getOrCreateCache(cacheName);
cache.get(24);

Does SQL API only operates on dataset that is already residing in-memory? What would be the approach in case underlying storage is bigger than Ignite capacity?