Under the Hood of Reactive Data Access

Slides to the presentation I gave at Spring I/O 2018

Conference page: https://2018.springio.net/sessions/under-the-hood-of-reactive-data-access
GitHub repository: https://github.com/mp911de/under-the-hood-reactive

Transcript

1. Mark Paluch
   @mp911de
   UNDER THE HOOD OF
   REACTIVE DATA ACCESS
   

   View Slide

2. Spring Data Modules
   JPA
   

   View Slide

3. Reactive Spring Data Modules
   JPA
   

   View Slide

4. What about…
   JDBC?
   JPA?
   
   

   View Slide

5. Relational Databases
   •Not a good fit for reactive access
   
   •Limited scalability
   
   •Resource synchronization
   
   •ACID Transactions
   

   View Slide

6. JDBC
   •Fully blocking API
   
   •Non-blocking API announced: ADBA 

   (JDK-8188051)
   http://mail.openjdk.java.net/pipermail/jdbc-spec-discuss/ • JDK-8188051: http://hg.openjdk.java.net/jdk/sandbox/shortlog/JDK-8188051-branch
   

   View Slide

7. Query Behavior
   

   View Slide

8. Query by Time
   Transport
   

   View Slide

9. Synchronous Data Access
   Data store
   App
   Data store
   App
   Data store
   

   View Slide

10. Reactive Data Access
    Data store
    App
    Data store Data store Data store
    App
    
    

    View Slide

11. Differences to imperative Data Access
    •Drivers require threading infrastructure
    
    •Timeout handling different
    

    View Slide

12. Synchronous Timeouts
    Socket.setSoTimeout(…)
    ⏱
    

    View Slide

13. Aynchronous Timeouts
    •Active timeout tracking required
    

    View Slide

14. Reactive Drivers
    •End-to-end nonblocking
    
    •Asynchronous request-response processing
    
    •Non-blocking pooling
    
    •Non-blocking connect
    

    View Slide

15. Application Driver Application
    Request
    Server selection
    Queueing
    Write Response decoding
    Response buffering
    Request completion Response
    

    View Slide

16. Reactive NoSQL drivers
    

    View Slide

17. MongoDB
    •Reactive-over-async driver
    
    •Cursor lifecycle
    
    •Pluggable I/O
    
    •Pooled connections
    Java.NIO
    

    View Slide

18. Query by Time
    Request
    Response
    (chunk)
    Response
    (complete)
    Prefetch
    (cont.d)
    Query
    Emit
    elements
    Result
    processing
    

    View Slide

19. MongoDB Driver Fetching
    Operator
    MongoDB Subscriber
    request(Long.MAX_VALUE)
    request(Long.MAX_VALUE)
    

    View Slide

20. MongoDB Driver Fetching
    Filter
    MongoDB Subscriber
    request(Long.MAX_VALUE)
    request(Long.MAX_VALUE)
    

    View Slide

21. MongoDB Driver Fetching
    Filter
    MongoDB Subscriber
    request(Long.MAX_VALUE)
    request(1)
    Dropped element
    

    View Slide

22. Specific Features
    •Cursor size derived from demand
    
    •Tailable Cursors (@Tailable)
    
    •MongoDB 3.6: Change Streams
    

    View Slide

23. Redis
    •Reactive driver (Lettuce 5)
    
    •Based on Netty and Project Reactor
    
    •Singleton connection
    
    •Optionally: Pooled connections
    

    View Slide

24. Command by Time
    Request
    Command
    Command
    (subscription)
    Response
    (chunk)
    Response
    (complete)
    

    View Slide

25. Specifics
    •Use ByteBuffer instead byte[]
    
    •null handling: Emission of empty buffers
    
    •Suspend reading if no demand
    

    View Slide

26. Apache Cassandra
    •Reactive bridge over async driver
    
    •Netty-based driver (Datastax)
    
    •Pooled connections
    

    View Slide

27. Query by Time
    Request
    Response
    (chunk)
    Response
    (complete)
    Query
    Emit
    elements
    Result processing
    Next page
    

    View Slide

28. Outlook
    •Cassandra driver 4.0 in the works
    
    •Likely support for reactive execution model
    
    •Emit items as they are received
    https://github.com/datastax/java-driver/tree/4.x • https://datastax-oss.atlassian.net/browse/JAVA-1505
    

    View Slide

29. Couchbase
    •Reactive driver driver
    
    •Based on Netty and RxJava 1
    
    •Multiple connection schemes
    

    View Slide

30. Key-Value, DCP
    •Request
    
    •Response
    
    •Emit results

    Search, Query, View
    •Request
    
    •Response streaming
    

    View Slide

31. Command by Time
    Request
    Command
    Command
    (subscription)
    Response
    (chunk)
    Response
    (complete)
    

    View Slide

32. Outlook
    •Couchbase driver likely to move to RxJava 2
    
    •No tickets/decisions yet
    

    View Slide

33. Imperative and Reactive Together
    

    View Slide

34. Blocking with Imperative Programming
    
    

    View Slide

35. Blocking with Reactive Programming
    
    

    View Slide

36. Imperative and Reactive
    @Component
    public class ApplicationInitializer {
    ReactiveMongoOperations ops;
    @PostConstruct
    public void postConstruct() {
    ops.remove(Person.class).all().block();
    }
    }
    

    View Slide

37. Imperative and Reactive
    @Component
    public class ApplicationInitializer {
    ReactiveMongoOperations ops;
    @PostConstruct
    public void postConstruct() {
    ops.remove(Person.class).all().block();
    }
    }
    
    

    View Slide

38. Event Loop
    Event Loop
    (single Thread)
    

    View Slide

39. Event Loop
    Event Loop
    (single Thread)
    

    View Slide

40. Event Loop
    Process next item
    Callback
    Register Callback
    Return
    Work
    Database
    Computation
    File
    I/O
    Event Loop
    (single Thread)
    Work Queue
    Work
    Request Callback
    

    View Slide

41. Event Loop
    Process next item
    Callback Work
    Event Loop
    (single Thread)
    Work Queue
    Request Callback
    

    View Slide

42. Event Loop
    Event Loop
    (single Thread)
    Work Queue Process next item
    Work
    Callback
    Work
    Wait for
    Request completion
    Return
    Request
    Request Callback
    

    View Slide

43. Event Loop
    Event Loop
    (single Thread)
    Work Queue Process next item
    Work
    Callback
    Work
    Wait for
    Request completion
    Return
    Request
    Request Callback
    

    View Slide

44. Thou shalt not block
    

    View Slide

45. Resources
    •Slides – mp911.de/uhd-springio18
    
    •Example Code – mp911.de/uhd-code
    
    •Spring Data Examples – Repository @ Github
    
    •Find us on Gitter!
    

    View Slide

46. Mark Paluch
    @mp911de
    THANKS!
    

    View Slide