Implicit trigger dispatch using ScheduledDataLoaderRegistry

[aisensiy]

Hi, I am trying to use dataloader without graphql. I know there is a ScheduledDataLoaderRegistry to automatically call the dispatch in every DataLoader but I still don't know how to use it. The code below seems not making the registry automatically trigger the dispatchAll as expected.

@SpringBootApplication
public class DataloaderAndMapstructApplication {

  public static void main(String[] args) {
    SpringApplication.run(DataloaderAndMapstructApplication.class, args);
  }

  @Bean
  public ScheduledDataLoaderRegistry registry(UserRepository userRepository) {
    DispatchPredicate depthOrTimePredicate = DispatchPredicate
        .dispatchIfDepthGreaterThan(10)
        .or(DispatchPredicate.dispatchIfLongerThan(Duration.ofMillis(200)));

    ScheduledDataLoaderRegistry registry = ScheduledDataLoaderRegistry.newScheduledRegistry()
        .dispatchPredicate(depthOrTimePredicate)
        .schedule(Duration.ofMillis(40))
        .build();

    registry.register("userLoader", DataLoaderFactory.newDataLoader((BatchLoader<String, User>) (userIds) -> CompletableFuture.supplyAsync(() -> {
      return userRepository.findAllById(userIds);
    })));
    return registry;
  }

  @Slf4j
  @Service
  @RequiredArgsConstructor
  public static class SchedulingRegistryRunner {
    private final ScheduledDataLoaderRegistry registry;

    @Scheduled(fixedDelay = 200)
    public void run() {
      log.info("run....");
      // When calling 10 times of the dataloader, I still got 10 repository.findByUsername and 2 seconds to finish it.
      registry.dispatchAll(); 
    }
  }
}

[bbakerman]

ScheduledDataLoaderRegistry registry = ScheduledDataLoaderRegistry.newScheduledRegistry()
        .dispatchPredicate(depthOrTimePredicate)
        .schedule(Duration.ofMillis(40))
        .build();

ScheduledDataLoaderRegistry requires a ScheduledExecutorService to run - this is the repeatable code pump that will periodically invoke the "predicate" to see if there is anything to dispatch.

Out of the box it has private ScheduledExecutorService scheduledExecutorService = Executors.newSingleThreadScheduledExecutor(); but really you should be deliberater in choosing one.

You should NOT your SchedulingRegistryRunner in this example code. You should not have another periodic code pump callinf dispatch. I mean it should not hurt but its not needed.

In this code example I cant see how you call the data loader. I guess its registry.getDataLoader("userLoader") somewhere.

Also note that re-using data loader instances between requests can lead to security problems - eg request A loads data X and then later request B will see that (potentially) cached data - which may be allowed or may not.

[aisensiy]

In this code example I cant see how you call the data loader.

Thanks for your reply. And I am trying to use mapstruct and dataloader together to make it easier to translate domain object to DTO.

// ============================ Domain ===================
@Entity
@Getter
@NoArgsConstructor(access = AccessLevel.PROTECTED)
public class Comment {
  @Id
  @GeneratedValue
  private Long id;

  private String content;
  private Long articleId;
  private String authorId;
  private Instant createdAt;
}

@Entity
@Getter
@NoArgsConstructor(access = AccessLevel.PROTECTED)
public class User {
  @Id
  private String username;
  private String email;
  private Instant createdAt;
}

// ============================ DTO ===================

@Data
public class CommentDTO {
  private Long id;
  private String content;
  private Instant createdAt;
  private UserDTO author;
}

@Data
public class UserDTO {
  String username;
  String email;
}

// ============================ Mapper ===================
@Mapper(componentModel = "spring")
public abstract class UserDTOMapper {
  private @Autowired DataLoaderRegistry dataLoaderRegistry;

  abstract UserDTO toDTO(User user);

  @SneakyThrows
  User toEntity(String id) {
    DataLoader<String, User> userDataLoader = dataLoaderRegistry.getDataLoader("userLoader");
    return userDataLoader.load(id).get();   // <-------- This is wrong, this will stuck the thread
  }
}

@Mapper(componentModel = "spring", uses = {UserDTOMapper.class})
public abstract class CommentDTOMapper {
  @Autowired private DataLoaderRegistry registry;

  @Mapping(source = "authorId", target = "author")
  abstract CommentDTO commentToCommentDTO(Comment comment);

  abstract List<CommentDTO> map(List<Comment> comments);
}

I know I am wrong to call dataloader.load(xx).get immediately. But I still do not have a good idea to make this work.

[bbakerman]

DataLoader is about 2 things - caching results and BATCHING a set of individual loads

In order to make a batch you need some time window to decide how long a batch is? eg do we wait 5 ms and then all outstanding dataloader calls are dispatched then?

or in grpahql do we dispatcher per field level say?

This is how batching works - combine N calls into one batch call on some boundary of choice.

userDataLoader.load(id).get();

This will never work if not one ever dispatches. But if should work if your dataLoaderRegistry is being dispatched periodically say. You need to make sure thats the case.