Building better microservices with GRPC

Today, massive systems are running on microservices communicating with each other using REST APIs. HTTP is easy to get started, loosely structured and does good job in exchanging messages. But it's convenience comes with a performance trade-off, which takes us back to other optimal alternative: gRPC

Transcript

1. Building better microservices with
   GRPC
   Pycon India 2018
   

   View Slide

2. Narendran
   @DudeWhoCode
   

   www.dudewho.codes
   Software
   Consultant,
   Tarka Labs
   

   View Slide

3. Netflix microservices traffic diagram [https://www.honeycomb.io/microservices/]
   

   View Slide

4. REST APIs
   

   View Slide

5. What’s wrong with REST ?
   

   View Slide

6. REST is good…
   ✤ Text based protocol
   ✤ Human readable API contracts
   ✤ Great tooling for testing and inspection
   ✤ Postman, FTW.
   ✤ Well defined HTTP status codes
   

   View Slide

7. …but
   

   View Slide

8. …but
   ✤ Systems needs machine readable API contract
   ✤ Streaming is not easy
   ✤ Duplex communication is impossible
   ✤ Inconsistent patterns
   ✤ HTTP/1.1
   

   View Slide

9. Inconsistent patterns and payload ?
   

   View Slide

10. HTTP methods
    ✤ GET
    ✤ POST
    ✤ PUT
    ✤ DELETE
    ✤ PATCH
    ✤ HEAD
    ✤ OPTIONS
    

    View Slide

11. Q: Why isn’t the demo working ?
    

    View Slide

12. Q: Why isn’t the demo working ?
    A: The API broke
    

    View Slide

13. HTTP/1.1
    

    View Slide

14. 1990 1995 2005 2005 2010 2015 2018
    HTTP/0.9
    HTTP/1.0
    HTTP/1.1
    HTTP/2
    

    View Slide

15. HTTP/1.1 is slow. Why ?
    

    View Slide

16. Connection overload
    

    New TCP connection for
    every HTTP request
    HTTP/1.1 is slow. Why ?
    

    View Slide

17. Connection overload
    

    New TCP connection for
    every HTTP request
    Header overload
    

    Uncompressed, repeating
    plain text headers for every
    http request
    HTTP/1.1 is slow. Why ?
    

    View Slide

18. HTTP/1.1 is slow. Why ?
    Connection overload
    

    New TCP connection for
    every HTTP request
    Header overload
    

    Uncompressed, repeating
    plain text headers for every
    http request
    HOL blocking
    

    A line of packets is held up
    by the first packet
    

    View Slide

19. REST on HTTP/1.1
    

    View Slide

20. What is gRPC ?
    

    View Slide

21. gRPC is a protocol where a client application can directly call
    methods on a server application on a different machine as if it
    was a local object, making it easier for you to create
    distributed applications and services.
    www.grpc.io
    

    View Slide

22. Companies that uses GRPC in production
    

    View Slide

23. Such Innovashun
    Much Scale
    So New
    Wow
    So Distributed
    Developer Doge
    

    View Slide

24. gRPC
    The idea of treating
    network operations as
    remote procedure calls
    goes back at least to the
    1970s in early ARPANET
    documents.
    

    View Slide

25. gRPC
    The idea of treating
    network operations as
    remote procedure calls
    goes back at least to the
    1970s in early ARPANET
    documents.
    First practical
    implementation was called
    Lupine at Xerox PARC. One
    of the first business uses of
    RPC was by Xerox under
    the name "Courier" in 1981
    

    View Slide

26. gRPC
    The idea of treating
    network operations as
    remote procedure calls
    goes back at least to the
    1970s in early ARPANET
    documents.
    First practical
    implementation was called
    Lupine at Xerox PARC. One
    of the first business uses of
    RPC was by Xerox under
    the name "Courier" in 1981
    The first popular
    implementation of RPC on
    Unix was Sun's RPC (now
    called ONC RPC), used as
    the basis for Network File
    System (NFS).
    

    View Slide

27. How RPC works ?
    

    View Slide

28. www.grpc.io
    

    View Slide

29. Why gRPC(is faster) ?
    

    View Slide

30. gRPC uses HTTP/2
    ✤ Minimizes header overhead
    ✤ Minimizes HOL blocking
    ✤ Minimizes number of connections
    ✤ Thereby, minimizing API latency
    

    View Slide

31. Headers in HTTP/1.1 [plain text]
    :method GET
    :scheme Https
    :host tarkalabs.com
    :path /team
    user-agent Chrome
    auth_token 3286354f34s3546
    :method GET
    :scheme Https
    :host tarkalabs.com
    :path /projects
    user-agent Chrome
    auth_token 3286354f34s3546
    Request #1 Request #2
    

    View Slide

32. Headers in HTTP/2 [HPAC]
    :method GET
    :scheme Https
    :host tarkalabs.com
    :path /team
    user-agent Chrome
    auth_token 3286354f34s3546
    Index of repeating fields +
    :path /projects
    Request #1 Request #2
    

    View Slide

33. Optimized connections
    https://docs.google.com/presentation/d/1r7QXGYOLCh4fcUq0jDdDwKJWNqWK1o4xMtYpKZCJYjM/present?slide=id.p19
    ✤ All the data is sent as frames over single TCP connection
    ✤ Duplex communication happened between server and client in a
    connection
    

    View Slide

34. gRPC uses ProtocolBuffers
    ✤ gRPC uses protocol buffers for seriailzation
    ✤ Protocol buffers: A structured and faster way for data
    serialization
    

    View Slide

35. Protobufs are Schema Oriented
    message Person {
    required string name = 1;
    required int32 id = 2;
    optional string email = 3;
    }
    

    View Slide

36. Protobufs are backward compatible
    if (version == 3) {
    ...
    } else if (version > 4) {
    if (version == 5) {
    ...
    }
    ...
    }
    // No ugly versioning
    

    View Slide

37. Protobufs are faster
    https://auth0.com/blog/beating-json-performance-with-protobuf/
    

    View Slide

38. Even more faster between
    microservices
    https://auth0.com/blog/beating-json-performance-with-protobuf/
    

    View Slide

39. gRPC basics
    ✤ Define your function
    ✤ Service Definition using protobuf
    ✤ Generate server stub, extend it and start the server
    ✤ Generate client stub
    

    View Slide

40. Connections
    ✤ Unary
    ✤ Response stream
    ✤ Request stream
    ✤ Bi-directional stream
    Server Client
    

    View Slide

41. Supported Languages
    ✤ Python
    ✤ C++
    ✤ C#
    ✤ Dart
    ✤ Go
    ✤ Node.js
    ✤ Objective-C
    ✤ PHP
    ✤ Java
    ✤ Ruby
    

    View Slide

42. Demo
    

    View Slide

43. Drawbacks
    ✤ No browser support
    ✤ Not enough documentation
    ✤ No standardisation across languages
    ✤ Smaller community
    ✤ No out of the box Load balancers from cloud providers
    

    View Slide

44. gRPC or REST …?
    

    View Slide

45. …depends
    ✤ Choose gRPC if for server to server communication
    and services that follows same development cycle
    (mircoservices)
    ✤ Choose REST for public facing services or the service
    with interacts with your browser
    

    View Slide

46. There is no such thing as best
    solution. There can only be
    systems that are more appropriate
    in a particular set of
    circumstances.
    – Ancient Chinese Saying
    https://en.wikipedia.org/wiki/Wise_old_man
    

    View Slide

47. There is no such thing as best
    solution. There can only be
    systems that are more appropriate
    in a particular set of
    circumstances.
    – The Pragmatic Programmer
    

    View Slide

48. gRPC Resources
    ✤ gRPC python Greeter Quickstart [https://grpc.io/docs/
    quickstart/python.html]
    ✤ gRPC python docs [https://grpc.io/docs/tutorials/basic/
    python.html]
    ✤ Protocol Buffers Introduction [https://developers.google.com/
    protocol-buffers/docs/overview]
    ✤ Protocol Buffers python reference [https://
    developers.google.com/protocol-buffers/docs/reference/
    python-generated]
    

    View Slide

49. Naren

    @DudeWhoCode
    www.dudewho.codes/talks
    

    View Slide