Coroutine Based Concurrency with PHP and Swoole

Slide 1

Slide 1 text

Coroutine Based Concurrency with PHP and Swoole Bruce Dou | @doubaokun [email protected] 20th Feb 2020 PHPUK2020

Slide 2

Slide 2 text

About me • Founder at Transfon • Twitter: @doubaokun • Email: [email protected] • Linkedin: linkedin.com/in/brucedou • Github: github.com/doubaokun Twitter: @doubaokun

Slide 3

Slide 3 text

Agenda • Rethinking about software development • Concurrency, I/O, Scheduling and OS • Coroutine: Cooperative multi-tasking • Swoole Coroutine Internal • Swoole • Coroutine and Swoole code samples • Demo Twitter: @doubaokun

Slide 4

Slide 4 text

Rethinking about software development Twitter: @doubaokun

Slide 5

Slide 5 text

Twitter: @doubaokun Mapping real world problems onto machine

Slide 6

Slide 6 text

Execution Unit Twitter: @doubaokun

Slide 7

Slide 7 text

I/O and Task Processing Twitter: @doubaokun

Slide 8

Slide 8 text

Twitter: @doubaokun

Slide 9

Slide 9 text

Twitter: @doubaokun Mapping real world problems onto machine with several levels abstraction

Slide 10

Slide 10 text

Twitter: @doubaokun

Slide 11

Slide 11 text

Twitter: @doubaokun Mapping real world problems onto machine with several levels abstraction with real world limitations in mind

Slide 12

Slide 12 text

Twitter: @doubaokun 1. 1 CPU cycle 0.3 ns 2. Main memory access 120 ns 3. Goroutine 170 ns 4. Swoole Coroutine 190 ns 5. Linux OS context switch 1,000 ns 6. Linux thread launch 5,000 ns 7. Linux process launch 20,000 ns 8. Per-thread stack is 8MB (ulimit -s)

Slide 13

Slide 13 text

• Unpredictable external performance • Client side I/O • Server side I/O • Queue theory Twitter: @doubaokun

Slide 14

Slide 14 text

Twitter: @doubaokun wait time utilization

Slide 15

Slide 15 text

Twitter: @doubaokun Mapping real world problems onto machine with several levels abstraction with real world limitations in mind about management and structure

Slide 16

Slide 16 text

Scaling up the power Twitter: @doubaokun

Slide 17

Slide 17 text

Scaling up the power Twitter: @doubaokun

Slide 18

Slide 18 text

= Scaling the concurrency Twitter: @doubaokun

Slide 19

Slide 19 text

What is Concurrency? Twitter: @doubaokun

Slide 20

Slide 20 text

Concurrency is managing multi-tasking Twitter: @doubaokun

Slide 21

Slide 21 text

Concurrency is managing multi-tasking in both real world and machine world Twitter: @doubaokun

Slide 22

Slide 22 text

Twitter: @doubaokun

Slide 23

Slide 23 text

What is blocking Twitter: @doubaokun

Slide 24

Slide 24 text

Twitter: @doubaokun

Slide 25

Slide 25 text

Blocking is sequential Twitter: @doubaokun

Slide 26

Slide 26 text

Blocking and sequential are the nature of human mind and machine Twitter: @doubaokun

Slide 27

Slide 27 text

Blocking and sequential are the nature of a single Execution Unit Twitter: @doubaokun

Slide 28

Slide 28 text

What is blocking I/O Twitter: @doubaokun

Slide 29

Slide 29 text

Twitter: @doubaokun

Slide 30

Slide 30 text

Blocking I/O is the nature of machine Twitter: @doubaokun

Slide 31

Slide 31 text

What is async I/O Twitter: @doubaokun

Slide 32

Slide 32 text

Twitter: @doubaokun

Slide 33

Slide 33 text

The nature of real world I/O Twitter: @doubaokun

Slide 34

Slide 34 text

The nature of real world is better resources utilisation Twitter: @doubaokun

Slide 35

Slide 35 text

Mapping real world to machine Twitter: @doubaokun

Slide 36

Slide 36 text

Mapping real world to machines: Unlimited Execution Unit and I/O vs Limited resources and machines Twitter: @doubaokun

Slide 37

Slide 37 text

Concurrency is hard Twitter: @doubaokun

Slide 38

Slide 38 text

Concurrency models • Process • Process pool • Thread • Thread pool • Single thread async I/O • Light weight thread: Fiber / Coroutine / … Twitter: @doubaokun

Slide 39

Slide 39 text

Concurrency is hard • Inefficient scaling • Race condition Twitter: @doubaokun

Slide 40

Slide 40 text

https://unixism.net/2019/04/linux-applications-performance-introduction/ Twitter: @doubaokun Performance of Concurrent I/O models

Slide 41

Slide 41 text

Machine I/O Concurrency Twitter: @doubaokun

Slide 42

Slide 42 text

Decoupling I/O and CPU Twitter: @doubaokun

Slide 43

Slide 43 text

Async machine I/O Twitter: @doubaokun

Slide 44

Slide 44 text

Twitter: @doubaokun

Slide 45

Slide 45 text

Event based callbacks ? Twitter: @doubaokun

Slide 46

Slide 46 text

not the nature of human mind Twitter: @doubaokun

Slide 47

Slide 47 text

Async machine I/O should be abstracted Twitter: @doubaokun

Slide 48

Slide 48 text

Understand the machine better Twitter: @doubaokun

Slide 49

Slide 49 text

Concurrency and Operating System Twitter: @doubaokun

Slide 50

Slide 50 text

What is a function? Twitter: @doubaokun

Slide 51

Slide 51 text

a sequence of instructions that takes inputs and returns outputs Twitter: @doubaokun

Slide 52

Slide 52 text

function add($a, $b) { return $a + $b; } Twitter: @doubaokun

Slide 53

Slide 53 text

What is Context? Twitter: @doubaokun

Slide 54

Slide 54 text

Twitter: @doubaokun

Slide 55

Slide 55 text

Call Stack and Stack Frame Twitter: @doubaokun

Slide 56

Slide 56 text

function add($a, $b) { return $a + $b; } function plus5($c, 5) { return add($c, 5); } Twitter: @doubaokun

Slide 57

Slide 57 text

Twitter: @doubaokun

Slide 58

Slide 58 text

What is context switch? Twitter: @doubaokun

Slide 59

Slide 59 text

Twitter: @doubaokun

Slide 60

Slide 60 text

History of multi-tasking in OS Twitter: @doubaokun

Slide 61

Slide 61 text

1. No scheduler Twitter: @doubaokun

Slide 62

Slide 62 text

2. Cooperative Scheduler Non-preemptive Twitter: @doubaokun

Slide 63

Slide 63 text

3. Preemptive timer interrupt scheduler Twitter: @doubaokun

Slide 64

Slide 64 text

Twitter: @doubaokun

Slide 65

Slide 65 text

Twitter: @doubaokun

Slide 66

Slide 66 text

Thread context switch can be expensive Twitter: @doubaokun

Slide 67

Slide 67 text

Not scalable Twitter: @doubaokun

Slide 68

Slide 68 text

Light weight thread Twitter: @doubaokun

Slide 69

Slide 69 text

Light weight Execution Unit Twitter: @doubaokun

Slide 70

Slide 70 text

Light weight Execution Unit provided by machine/library Twitter: @doubaokun

Slide 71

Slide 71 text

Light weight Execution Unit provided by machine to map real world problems Twitter: @doubaokun

Slide 72

Slide 72 text

Light weight Execution Unit provided by machine to map real world problems to execute real world tasks Twitter: @doubaokun

Slide 73

Slide 73 text

What is Coroutine? Twitter: @doubaokun

Slide 74

Slide 74 text

“Coroutines are computer components that generalise subroutines for non-preemptive multitasking, by allowing execution to be suspended and resumed.” - Wikipedia Twitter: @doubaokun

Slide 75

Slide 75 text

“a type of functions that enables concurrency via cooperative multitasking” Twitter: @doubaokun

Slide 76

Slide 76 text

Cooperative multi-tasking Twitter: @doubaokun

Slide 77

Slide 77 text

Multi-tasking management strategy Twitter: @doubaokun

Slide 78

Slide 78 text

On machine: User space yield and resume Twitter: @doubaokun

Slide 79

Slide 79 text

1 3 start 2 4 Twitter: @doubaokun

Slide 80

Slide 80 text

Twitter: @doubaokun

Slide 81

Slide 81 text

Swoole Twitter: @doubaokun

Slide 82

Slide 82 text

https://www.swoole.co.uk/ https://github.com/swoole/swoole-src @php_swoole Coroutine-based concurrency library with PHP syntax Twitter: @doubaokun

Slide 83

Slide 83 text

Twitter: @doubaokun

Slide 84

Slide 84 text

Super fast Twitter: @doubaokun

Slide 85

Slide 85 text

Super fast https://tsh.io/blog/swoole-is-it-node-in-php-or-am-i-wrong/ Twitter: @doubaokun

Slide 86

Slide 86 text

Super fast Twitter: @doubaokun

Slide 87

Slide 87 text

Twitter: @doubaokun • Async I/O • Reuse states • Coroutine • Coroutine scheduler based on I/O

Slide 88

Slide 88 text

Twitter: @doubaokun

Slide 89

Slide 89 text

Swoole server Twitter: @doubaokun

Slide 90

Slide 90 text

HTTP server with PHP Swoole Twitter: @doubaokun

Slide 91

Slide 91 text

Server structure Twitter: @doubaokun

Slide 92

Slide 92 text

Server structure Twitter: @doubaokun

Slide 93

Slide 93 text

Swoole Coroutine Internal Twitter: @doubaokun

Slide 94

Slide 94 text

Twitter: @doubaokun

Slide 95

Slide 95 text

Use cases Twitter: @doubaokun

Slide 96

Slide 96 text

Twitter: @doubaokun

Slide 97

Slide 97 text

Common pitfalls Twitter: @doubaokun

Slide 98

Slide 98 text

Twitter: @doubaokun Common pitfalls • Looking to run it on M$ Windows • Trying to run it with PHP-FPM • Stateless mind • Variable scope: super global, global, static, local variables • Blocking within Coroutine context • Exception handling • Exit

Slide 99

Slide 99 text

Code samples Twitter: @doubaokun

Slide 100

Slide 100 text

Coroutine Twitter: @doubaokun

Slide 101

Slide 101 text

No content

Slide 102

Slide 102 text

1 3 start 2 4 Coroutine yield and resume

Slide 103

Slide 103 text

2 1 3 2 Coroutine structure

Slide 104

Slide 104 text

Server Twitter: @doubaokun

Slide 105

Slide 105 text

Server response timeout

Slide 106

Slide 106 text

Timer Twitter: @doubaokun

Slide 107

Slide 107 text

Timer

Slide 108

Slide 108 text

Client Twitter: @doubaokun

Slide 109

Slide 109 text

HTTP Client

Slide 110

Slide 110 text

Channel Twitter: @doubaokun

Slide 111

Slide 111 text

No content

Slide 112

Slide 112 text

Coroutine Pooling Twitter: @doubaokun

Slide 113

Slide 113 text

No content

Slide 114

Slide 114 text

No content

Slide 115

Slide 115 text

Linux process Twitter: @doubaokun

Slide 116

Slide 116 text

No content

Slide 117

Slide 117 text

HashTable Twitter: @doubaokun

Slide 118

Slide 118 text

No content

Slide 119

Slide 119 text

Task worker Twitter: @doubaokun

Slide 120

Slide 120 text

No content

Slide 121

Slide 121 text

No content

Slide 122

Slide 122 text

Coroutine Server Twitter: @doubaokun

Slide 123

Slide 123 text

Twitter: @doubaokun

Slide 124

Slide 124 text

Demo Twitter: @doubaokun

Slide 125

Slide 125 text

https://www.swoole.co.uk/ https://twitter.com/php_swoole https://github.com/swoole/swoole-src https://github.com/21days-dev/phpuk2020 Thank you