High Performance Websites with Scalable Workers

There comes a point in time with a website when eventually need to do something in the background. There are always cron jobs, but eventually those either don't scale well, or are not responsive enough. Learn about how to help your website efficiently scale by using workers. We'll discuss the fundamental theory behind workers and how to easily implement them. We'll learn about several different technologies to help manage workers such as Beanstalkd, Gearman, Supervisord, Redis, and others.

Transcript

1. Text
   High Performance Websites
   with Scalable Workers
   @JustinCarmony
   

   View Slide

2. Introduction
   About Presenter
   • Director of Development
   for Deseret Digital Media
   • President of Utah PHP Usergroup
   • PHP Developer for 8+ years
   • Goofy Dad
   

   View Slide

3. Introduction
   About Presentation
   • There is a Vagrant demo project located at:
   https://github.com/JustinCarmonyDotCom/
   PHP-Workers-Tutorial
   aka http://bit.ly/10WYeUf
   • Feel free to clone it and follow along
   

   View Slide

4. Introduction
   Normally I Have Really
   Nice Slides...
   

   View Slide

5. Introduction
   ... but my work gave me
   this template (which is
   ugly) ...
   

   View Slide

6. Introduction
   ... and this guy came
   along ....
   

   View Slide

7. Introduction
   

   View Slide

8. Introduction
   So No Fancy Slides
   this Time
   ( Sorry :p )
   

   View Slide

9. Introduction
   So Lets Identify The
   Problem
   

   View Slide

10. The Problem
    Example: Submit
    New Blog Post
    Fill Out
    Form
    Display
    Success
    Page
    

    View Slide

11. The Problem
    Example: Submit
    New Blog Post
    Fill Out
    Form
    Display
    Success
    Page
    Save
    to DB
    

    View Slide

12. The Problem
    Example: Submit
    New Blog Post
    Fill Out
    Form
    Display
    Success
    Page
    Save
    to DB
    Post to
    Twitter
    & FB
    

    View Slide

13. The Problem
    Example: Submit
    New Blog Post
    Fill Out
    Form
    Display
    Success
    Page
    Save
    to DB
    Post to
    Twitter
    & FB
    Send
    Emails
    

    View Slide

14. The Problem
    Example: Submit
    New Blog Post
    Fill Out
    Form
    Display
    Success
    Page
    Save
    to DB
    Post to
    Twitter
    & FB
    Send
    Emails
    Warm
    Cache
    

    View Slide

15. The Problem
    Example: Submit
    New Blog Post
    Fill Out
    Form
    Display
    Success
    Page
    Save
    to DB
    Post to
    Twitter
    & FB
    Send
    Emails
    Warm
    Cache
    Update
    Queues
    

    View Slide

16. The Problem
    Example: Submit
    New Blog Post
    Fill Out
    Form
    Display
    Success
    Page
    Save
    to DB
    Post to
    Twitter
    & FB
    Send
    Emails
    Warm
    Cache
    Update
    Queues
    Create
    Images
    

    View Slide

17. The Problem
    Example: Submit
    New Blog Post
    Fill Out
    Form
    Display
    Success
    Page
    Save
    to DB
    Post to
    Twitter
    & FB
    Send
    Emails
    Warm
    Cache
    Update
    Queues
    Create
    Images
    Find
    Related
    Posts
    

    View Slide

18. The Problem
    Example: Submit
    New Blog Post
    Fill Out
    Form
    Display
    Success
    Page
    Save
    to DB
    Post to
    Twitter
    & FB
    Send
    Emails
    Warm
    Cache
    Update
    Queues
    Create
    Images
    Find
    Related
    Posts
    Another
    Thing
    Another
    Thing
    Another
    Thing
    Another
    Thing
    Another
    Thing
    Another
    Thing
    Another
    Thing
    

    View Slide

19. The Problem
    It’s SLOW!
    

    View Slide

20. The Problem
    Problems With
    Linear Approach
    • Time From “Save” to “Success” Longer &
    Longer
    • What happens if you get an error? Will it
    block all other items afterwards?
    • What if you have lots of authors? Everyone
    stuck in this linear approach?
    

    View Slide

21. The Solution - Workers
    The Solution:
    PHP Workers
    

    View Slide

22. The Solution
    What are Workers?
    • Separate PHP Processes that typically run
    from the CLI
    • Listen to a Queue for “jobs”
    • When it receives a job, it does it
    • When it finished, listens for more jobs
    

    View Slide

23. The Solution
    The Workflow
    

    View Slide

24. Web Application
    The Solution
    The Workflow
    

    View Slide

25. Job
    Web Application
    The Solution
    The Workflow
    

    View Slide

26. Job
    Web Application
    The Solution
    The Workflow
    Job
    Job
    Job
    Job
    Job
    Job
    Job
    Job
    Job
    Job
    Job
    Queue
    of Jobs
    

    View Slide

27. Job
    Web Application
    The Solution
    The Workflow
    Job
    Job
    Job
    Job
    Job
    Job
    Job
    Job
    Job
    Job
    Job
    Queue
    of Jobs
    

    View Slide

28. Job
    Web Application
    The Solution
    The Workflow
    Job
    Job
    Job
    Job
    Job
    Job
    Job
    Job
    Job
    Job
    Job
    Queue
    of Jobs
    Worker
    

    View Slide

29. Job
    Web Application
    The Solution
    The Workflow
    Listen
    Job
    Job
    Job
    Job
    Job
    Job
    Job
    Job
    Job
    Job
    Job
    Queue
    of Jobs
    Worker
    

    View Slide

30. Job
    Web Application
    The Solution
    The Workflow
    Listen
    Job
    Job
    Job
    Job
    Job
    Job
    Job
    Job
    Job
    Job
    Job
    Queue
    of Jobs
    Worker
    

    View Slide

31. Job
    Web Application
    The Solution
    The Workflow
    Listen
    Job
    Job
    Job
    Job
    Job
    Job
    Job
    Job
    Job
    Job
    Job
    Queue
    of Jobs
    Worker
    

    View Slide

32. Job
    Web Application
    The Solution
    The Workflow
    Listen
    Result
    Job
    Job
    Job
    Job
    Job
    Job
    Job
    Job
    Job
    Job
    Job
    Queue
    of Jobs
    Worker
    

    View Slide

33. Job
    Web Application
    The Solution
    The Workflow
    Listen
    Result
    Job
    Job
    Job
    Job
    Job
    Job
    Job
    Job
    Job
    Job
    Job
    Queue
    of Jobs
    Worker
    

    View Slide

34. The Solution
    Example: Submit
    New Blog Post
    Fill Out
    Form
    

    View Slide

35. The Solution
    Example: Submit
    New Blog Post
    Fill Out
    Form
    Queue Jobs
    

    View Slide

36. The Solution
    Example: Submit
    New Blog Post
    Fill Out
    Form
    Job
    Queue Jobs
    Job
    Job
    Job
    Job
    Job
    Job
    Job
    Job
    Job
    Job
    Job
    Job
    Job
    Job
    Job
    Job
    Job
    Job
    Job
    Job
    Job
    Job
    Job
    Job
    Job
    Job
    Job
    Job
    Job
    Job
    Job
    Job
    Job
    Job
    Job
    

    View Slide

37. The Solution
    Example: Submit
    New Blog Post
    Fill Out
    Form
    Display
    Success
    Page
    Job
    Queue Jobs
    Job
    Job
    Job
    Job
    Job
    Job
    Job
    Job
    Job
    Job
    Job
    Job
    Job
    Job
    Job
    Job
    Job
    Job
    Job
    Job
    Job
    Job
    Job
    Job
    Job
    Job
    Job
    Job
    Job
    Job
    Job
    Job
    Job
    Job
    Job
    

    View Slide

38. The Solution
    Okay... the Theory is
    Pretty Straight
    forward...
    

    View Slide

39. The Solution
    Lets see how you
    actually do it...
    

    View Slide

40. Getting Started
    What We’ll Need
    • PHP 5.3
    • Queue -- Beanstalkd
    • Manage Queue - Pheanstalk Library
    • Manage Workers - Solo / Supervisord
    • Store Worker Status - Redis
    • Talk with Redis - Predis Library
    

    View Slide

41. Getting Started
    Wait, What About
    Gearman?
    • It’s great to use!
    • Requires to install modules.
    • If you understand how to use Beanstalkd,
    its easy to use Gearman
    • Lots of talks about Gearman, like to give
    some exposure to Beanstalkd
    

    View Slide

42. Getting Started
    Install / Setup
    Beanstalkd
    • Debian/Ubuntu:
    apt-get install beanstalkd
    • Fedora/CentOS/RHEL:
    su -c 'yum install beanstalkd'
    • Mac Homebrew:
    brew install beanstalkd
    

    View Slide

43. Getting Started
    Beanstalkd on
    Windows
    • ... use a VM. Checkout Vagrant!
    

    View Slide

44. Getting Started
    Pheanstalk for PHP
    • Most Tested PHP Beanstalkd Library
    • Installable via Composer
    • Located at:
    https://github.com/pda/pheanstalk
    

    View Slide

45. Workers
    The Trick: Managing
    Your Workers
    • Putting stuff in a queue is easy!
    • Keeping workers up & running consistently
    is the hard part!
    • We’ll talk about PHP workers, but you can
    use any language to write your workers.
    

    View Slide

46. Workers
    Features for our
    PHP Workers
    • Run for a long time
    • Self-restarting
    • Can be restarted
    • Reports it’s current status
    • Blocking, Not Polling
    • Monitor It’s Own Performance
    

    View Slide

47. Workers
    Timing For your
    Workers
    • Allowing them to run indefinitely:
    set_timeout_limit(0);
    • Set a max runtime for them to gracefully
    restart. Example:
    $time_limit = 60 * 60 * 1; // Minimum of 1 hour
    $time_limit += rand(0, 60 * 10); // Adding additional time
    

    View Slide

48. The Queue
    Anatomy of a
    Beanstalkd Job
    • What makes up a job
    • Priority
    • Delay
    • Time-To-Run
    • Data
    

    View Slide

49. Text
    Beanstalkd Actions
    • put (creates job)
    • put with delay
    • reserve (worker selects
    job)
    • release (puts job back
    into queue)
    • release with delay
    • delete (removes job)
    • bury (bury job at end of
    queue to wait)
    • kick (kick a buried job
    back into the queue)
    • touch (request more
    time on a job)
    

    View Slide

50. The Queue
    Typical Job Life Cycle
    Client
    put
    Job Worker
    retrieve delete
    Poof!
    

    View Slide

51. Poof!
    The Queue
    Possible Job Flow
    DELAYED
    put with
    delay
    (time passes)
    READY RESERVED
    reserve
    put
    release with delay
    release
    delete
    BURIED
    bury
    kick
    Poof! delete
    

    View Slide

52. Demo Time
    Lets look at the code!
    

    View Slide

53. High Performance
    Okay, so... what about
    the High Performance
    Part?
    

    View Slide

54. High Performance
    Using Workers to
    Increase Performance
    • Make Writing to Your “Queue” Extremely
    Fast (Beanstalkd, Redis, In-Memory)
    • Do only what you must in your request,
    Queue Everything Else
    • Find IO Operations & Queue Them Instead
    of Perform Them
    

    View Slide

55. High Peformance
    Caching is for Reads
    Queueing is for Writes
    

    View Slide

56. High Performance
    Possible Things to
    Queue: Your Writes
    • Logging
    • Analytics
    • Some Queries
    • Communication with other APIs
    • Writing to your Data Sources
    

    View Slide

57. High Performance
    Identifying What To
    Queue
    • Use XHGui/XHProf to find slow parts of
    your application
    • Use StatsD/Graphite to Gather Stats
    Across Requests
    • Does it communicate with something else?
    

    View Slide

58. Questions
    Questions?
    

    View Slide

59. The End
    Thank You!
    Twitter: @JustinCarmony
    Website: http://www.justincarmony.com/
    Email: justin@justincarmony.com
    

    View Slide