Mastering Swoole PHP

讀書會#1

12/23

19:30~21:00

@17Live 大樓三樓 銀河灰會議室

About This Book

Zeev Suraski
Co-founder of Zend, Co-architect of PHP

Enrico Zimuel
Member of PHP-FIG and co-author of Expressive and Zend Framework

About This Book

• Background

• General concepts

• The practical world on Linux OS

• Connect with protocols

• Swoole runtime and internal

• Hands on Swoole PHP

• Use cases and patterns

What is Swoole?

• Since 2012

• Open source based on Apache-2.0 License

• 16.3k start on Dec. 2020

• More than 100 contributors

• Adopted by many enterprises in China

• Swoole Plus for enterprise support

What is Swoole?

• Not a new language, but an extension for PHP

• Provides many new features for traditional PHP

• New lifecycle in Swoole

• Server patterns, coroutine, async I/O, Process Management, etc.

• It's still under active development

• Stable for production environment

• High performance

Stateless PHP

• How do we serve PHP today?

  • ​PHP-FPM

  • mod_php for Apache

• Both patterns are all stateless

Stateless PHP

• Pros

  • Easy scaling

  • Simple, less risk causing memory leaks

• Cons

  • States can't be shared between requests

  • States must rely on external storages

  • Resources can't be reused efficiently

  • Connection cost is expensive (like database)

  • Not good for high performance

PHP FPM

• FPM (FastCGI Process Manager)

PHP FPM

PHP Architecture

• SAPI: Adapters like PHP-Cli, PHP-Fpm or embeded

• ZendVM: Composed with Compiler and Executor, like JVM in Java

• Extension: Including PHP Extension and Zend Extension

PHP Lifecycle

Blocking I/O in PHP

blocking I/O model

Blocking I/O in PHP

• PHP is originally created as glue layer to call C functions
• The I/O model is blocking by default
• Almost all client-side libraries involving I/O are blocking
• Multiple process for handling blocking I/O
• I/O bound concurrency depends on process number
• Cost for context switch in processes is expensive

Performance Problem

Identify the bottoleneck first!

Performance Problem

• Are we wasting resources by not using the reusable variables?
• Do we request the same data not changed from the database for each request?
• Are we requesting the data not used by the current request?
• Can we cache the remote data locally?
• Can we use async I/O instead of blockingI/O to save processes?
• Can we defer the heavy task by sending them into a queue and executing them separately?
• Etc.

Performance Problem

• Resource can't be reused

• I/O requests are blocking

• 80% time cost on blocking I/O

Concurrency Problem

• 100 requests at the same time need 100 processes

• 100 connections will be established as well

• Scale to increase concurrency

Concurrency Problem

• Synchronous Process

Concurrency Problem

• Asynchronous Process

Concurrency Problem

• Asynchronous Process

  • Save data into a queue then ACK back to the client
  • Check process status periodically
  • Waste resource for checking requests
  • Main process lost the control about these requests
  • Not applicable for dependent requests

Concurrency Problem

• Fast and Slow Components

Concurrency Models

Concurrency Models

• Actor

  • Erlang

• CSP (Communicating Sequential Processes)

  • Golang, Swoole

• STM (Software Transactional Memory)

  • ​Clojure

• Guilds

  • Ruby 3

Don't communicate by sharing memory; share memory by communicating. (Rob Pike)

Concurrency Models

• Actor

  • Multiple actors can run at the same

  • Each actor has its own identifier and internal states

  • Actors communicate with each other by sending asynchronous messages

  • Every actor has an mailbox (queue)

  • Actors are isolated

https://www.brianstorti.com/the-actor-model

Concurrency Models

• CSP

  • Decoupled by channel

  • Pub/Sub Pattern

  • Shared memory and lock are implemented in channel

  • Only one active consumer can access data

  • Processes are anonymous

Concurrency Models

• STM

  • Use shared memory

  • No need for lock

  • All changes of data are encapsulated in transactions

  • Every thread has a copy of its working data and can change its value

  • Like transactions in database

Concurrency Models

• Guild

  • Threads from different guilds can run in parallel

  • Threads that belong to the same Guild can’t execute in parallel (GGL)

  • Immutable objects can be shared (read) across Guilds

  • Guilds can communicate with each other using channel (deep copy)

https://www.slideshare.net/KoichiSasada/guild-prototype

Reference

• Mastering Swoole PHP
  • https://swoolebook.com

Discussion