From Generator to Fiber the Road to Coroutine in PHP

COSCUP 2022

Transcript

1. From Generator to Fiber
   The Road to Coroutine
   in PHP
   @COSCUP 2022
   Albert Chen
   

   View Slide

2. 02 Blocking and Async I/O in PHP
   Outline
   01 I/O Models in Linux
   03 What is Generator?
   04 What is Fiber in PHP 8.1?
   05 How to use Async I/O with Coroutine?
   07 Q&A
   06 Coroutines in Swoole
   

   View Slide

3. Read/Write
   Read/Write
   (O_NONBLOCK)
   I/O Multiplexing
   (select/poll/epoll)
   AIO
   I/O Models in Linux
   Simplified Matrix of Linux I/O Models
   Blocking Non-Blocking
   Synchronous
   Asynchronous
   (https://developer.ibm.com/articles/l-async)
   

   View Slide

4. I/O Models in Linux
   Synchronous Blocking I/O
   (https://developer.ibm.com/articles/l-async)
   

   View Slide

5. I/O Models in Linux
   Synchronous Non-Blocking I/O
   (https://developer.ibm.com/articles/l-async)
   

   View Slide

6. I/O Models in Linux
   Asynchronous Blocking I/O
   (https://developer.ibm.com/articles/l-async)
   

   View Slide

7. I/O Models in Linux
   Asynchronous Non-Blocking I/O
   (https://developer.ibm.com/articles/l-async)
   

   View Slide

8. I/O Models in Linux
   Comparison of the I/O Models
   (https://rickhw.github.io/2019/02/27/ComputerScience/IO-Models)
   

   View Slide

9. 
   select poll epoll
   Data
   Structue
   Array Link List Hash Table
   Big O O(n) O(n) O(1)
   FD Limit
   1024 (x86)
   2048 (x64)
   Unlimited Unlimited
   I/O Multiplexing
   I/O Multiplexing in Linux
   

   View Slide

10. I/O Multiplexing
    I/O Multiplexing in Linux
    (http://daemonforums.org/showthread.php?t=2124)
    

    View Slide

11. Blocking I/O in PHP
    All I/O functions in PHP are born to be synchronous
    Easy for developers to understand
    Less-efficient for utilization of CPU usage
    

    View Slide

12. sleep file_get_contents shell_exec end
    Blocking I/O in PHP
    1s 3s 3s
    process idle
    + +
    process idle process idle
    7s
    =
    

    View Slide

13. 1s worker
    1s worker
    1s worker
    Blocking I/O in PHP
    If one request takes 1s in I/O waiting
    Each worker can handle only one request and the same time
    Concurrency in PHP depends on worker numbers
    Context switch between processes are expensive!
    request
    request
    request
    

    View Slide

14. Asynchronous I/O in PHP
    The simplest but worst-performant solution in PHP
    spatie/async package (needs pcntl and posix extensions)
    

    View Slide

15. Asynchronous I/O in PHP
    The simplest but worst-performant solution in PHP
    spatie/async package (needs pcntl and posix extensions)
    worker
    process process
    fork
    fork
    

    View Slide

16. Asynchronous I/O in PHP
    ReactPHP
    

    View Slide

17. Asynchronous I/O in PHP
    ReactPHP
    

    View Slide

18. Asynchronous I/O in PHP
    Amp
    

    View Slide

19. Asynchronous I/O in PHP
    Amp
    

    View Slide

20. Asynchronous I/O in PHP
    Async I/O Extensions in PHP
    ext-event: libevent wrapper
    ext-ev: libev wrapper
    ext-uv: libuv wrapper
    Both ReactPHP and Amp support these extensions as
    
    event-loop drivers
    

    View Slide

21. Asynchronous I/O in PHP
    Event Loop
    Event Queue
    Stream
    Stream
    Stream
    Stream
    Stream
    Event Loop
    callbacks
    

    View Slide

22. Asynchronous I/O in PHP
    Event Loop in ReactPHP
    

    View Slide

23. Asynchronous I/O in PHP
    Event Loop in ReactPHP
    

    View Slide

24. Generator
    Supported from PHP 5.4
    Generator provides an easy way to implement simple iterators
    Generator allows you to write code that uses foreach to iterate over
    
    a set of data without needing to build an array in memory
    Generator can yield as many times as it needs to provide the values
    
    to be iterated over
    (From official PHP manual)
    

    View Slide

25. Generator
    Reading large file with Generator
    Regular Version Generator Version
    (https://clouding.city/php/generator)
    

    View Slide

26. Generator
    Generators, also known as semi-coroutines, are a subset of
    
    coroutines
    Coroutines can control where execution continues immediately
    
    after they yield, while Generators cannot, instead transferring
    
    control back to the generator's caller
    The yield statement in a generator does not specify a coroutine
    
    to jump to, but rather passes a value back to a parent routine
    (From Wikipedia)
    

    View Slide

27. Generators
    yield iteration
    Generator
    Producer Consumer
    Pub/Sub Pattern
    two way data transfer
    

    View Slide

28. Generator
    Two Way Data Transfer in Generator
    ?
    (https://www.npopov.com/2012/12/22/Cooperative-multitasking-using-coroutines-in-PHP.html)
    

    View Slide

29. Generator
    Two Way Data Transfer in Generator
    yield1
    ret1
    yield2
    ret2
    NULL
    the first var_dump in gen
    the var_dump of the ->send()
    again from within gen
    the return value of ->send()
    (https://www.npopov.com/2012/12/22/Cooperative-multitasking-using-coroutines-in-PHP.html)
    

    View Slide

30. Generator
    Generator in ReactPHP's Coroutine
    (https://github.com/reactphp/async)
    

    View Slide

31. Generator
    Generator in ReactPHP's Coroutine
    (https://github.com/reactphp/async/blob/4.x/src/functions.php)
    

    View Slide

32. Generator
    Generator in ReactPHP's Coroutine
    (https://github.com/reactphp/async/blob/4.x/src/functions.php)
    

    View Slide

33. Generator
    Generator in Amp's Coroutine
    (https://amphp.org/http-client)
    

    View Slide

34. Generator
    Generator in Amp's Coroutine
    (https://github.com/amphp/amp/blob/master/lib/Coroutine.php)
    

    View Slide

35. Generator
    Generator in Amp's Coroutine
    (https://github.com/amphp/amp/blob/master/lib/Coroutine.php)
    

    View Slide

36. Fiber
    New feature in PHP 8.1
    Submitted and developed by Amp
    Unlike stack-less Generators, each Fiber has its own call stack,
    
    allowing them to be paused within deeply nested function calls
    Execution may be interrupted anywhere in the call stack using
    
    Fiber::suspend()
    Once suspended, execution of the fiber may be resumed with any
    
    value using Fiber::resume()
    (From official PHP manual)
    

    View Slide

37. Fiber
    Execution in Fiber
    (https://php.watch/versions/8.1/fibers)
    Sequential Fiber
    

    View Slide

38. Fiber
    Fiber Class
    

    View Slide

39. Fiber
    Basic Usage
    ?
    

    View Slide

40. Fiber
    Basic Usage
    fiber start
    fiber suspend
    fiber continue
    done
    

    View Slide

41. Fiber
    Basic Usage
    ?
    

    View Slide

42. Fiber
    Basic Usage
    start fiber 1
    end fiber 1
    start fiber 2
    end fiber 2
    done
    

    View Slide

43. Fiber
    Fibers don't make blocking code
    become non-blocking magically!
    

    View Slide

44. Fiber
    Basic Usage
    ?
    

    View Slide

45. Fiber
    Basic Usage
    fiber start
    0
    1
    2
    3
    .........
    

    View Slide

46. Fiber
    Fibers only support Cooperative Scheduling
    for I/O-bound tasks
    

    View Slide

47. Fiber
    Fiber in ReactPHP
    (https://github.com/reactphp/async#async)
    

    View Slide

48. Fiber
    Fiber in ReactPHP
    (https://github.com/reactphp/async/blob/4.x/src/functions.php)
    

    View Slide

49. Fiber
    Fiber in ReactPHP
    (https://github.com/reactphp/async/blob/4.x/src/functions.php)
    

    View Slide

50. Fiber
    Fiber in Amp
    (https://github.com/amphp/http-client/blob/v5/examples/concurrency/1-concurrent-fetch.php)
    

    View Slide

51. Fiber
    Fiber in Amp
    (https://github.com/amphp/amp/blob/v3/src/functions.php)
    

    View Slide

52. Fiber
    Fiber in Amp
    (https://github.com/revoltphp/event-loop/blob/main/src/EventLoop/FiberLocal.php)
    

    View Slide

53. Fiber
    Both ReactPHP and amp(3.0 beta) support Fiber now
    A comprehensive coroutine rather than semi-coroutine
    Eliminates yield statement for coroutines
    Fiber doesn't turn your code into non-blocking magically, traditional
    
    blocking code has to be rewritten for asynchronous I/O
    Help not too much for most of developers
    Fiber provides only bare minimum required to allow user code to
    
    implement full-stack coroutines in PHP
    Fiber doesn't support Preemptive Scheduling yet
    

    View Slide

54. Questions
    What's the difference between Asynchronous and Coroutine?
    What's the difference between Generator and Fiber?
    How to turn Blocking I/O stream turning into Non-Blocking?
    What is Event Loop? Is it required for Asynchronous?
    What's the benefits of Coroutine?
    Is Asynchronous a must for Coroutine?
    How to make built-in Blocking I/O functions Asynchronous?
    Can Asynchronous I/O be integrated with traditional Blocking I/O
    
    environment? (e.g. PHP-FPM)
    

    View Slide

55. Coroutine in Swoole
    Basic Usage
    ?
    

    View Slide

56. Coroutine in Swoole
    Basic Usage
    start coro 1
    start to resume 1 @1
    resume coro 1 @1
    start to resume 1 @2
    resume coro 1 @2
    main
    

    View Slide

57. Coroutine in Swoole
    Coroutine for Blocking I/O
    ?
    

    View Slide

58. Coroutine in Swoole
    Coroutine for Blocking I/O
    coroutine start
    114.45.154.86
    coroutine end
    main
    

    View Slide

59. Coroutine in Swoole
    Coroutine for Blocking I/O
    ?
    

    View Slide

60. Coroutine in Swoole
    Coroutine for Blocking I/O
    coroutine start
    main
    xxx.xxx.xxx.xxx
    coroutine end
    

    View Slide

61. Coroutine in Swoole
    Preemptive Scheduling
    ?
    

    View Slide

62. Coroutine in Swoole
    Preemptive Scheduling
    

    View Slide

63. Coroutine in Swoole
    Preemptive Scheduling
    ?
    

    View Slide

64. Coroutine in Swoole
    Fatal error: Uncaught Exception:
    12.5ms end in xxx
    Stack trace:
    #0 {main}
    thrown in xxx
    Preemptive Scheduling
    

    View Slide

65. Coroutine in Swoole
    Channel
    ?
    

    View Slide

66. Coroutine in Swoole
    Channel
    [coroutine 3]
    int(0)
    [coroutine 2] - 0
    [coroutine 3]
    int(1)
    [coroutine 2] - 1
    [coroutine 3]
    ......
    

    View Slide

67. Coroutine in Swoole
    Swoole implements its full coroutine features
    Built-in Blocking I/O functions can be replaced to Non-Blocking I/O
    
    automatically if hook flags are set
    Swoole supports Preemptive Scheduling for CPU-bound tasks
    Swoole Provides CSP Model for Coroutine Communications like in
    
    GoLang
    Yields and resumes take place behind the scenes for I/O functions
    
    switching
    

    View Slide

68. Q&A
    

    View Slide