Upgrade to Pro — share decks privately, control downloads, hide ads and more …

Gearman: A Job Server to Scale

Gearman: A Job Server to Scale

Brief introduction to the gearman job server.

Mike Willbanks

April 07, 2012
Tweet

More Decks by Mike Willbanks

Other Decks in Programming

Transcript

  1. A Job Server to Scale By Mike Willbanks Software Engineering

    Manager CaringBridge MinneBar April 7, 2012
  2. 2 •Talk  Slides will be online later! •Me 

    Software Engineering Manager at CaringBridge  MNPHP Organizer  Open Source Contributor (Zend Framework and various others)  Where you can find me: • Twitter: mwillbanks G+: Mike Willbanks • IRC (freenode): mwillbanks Blog: http://blog.digitalstruct.com • GitHub: https://github.com/mwillbanks Housekeeping…
  3. 3 • What is Gearman  Yeah yeah… • Main

    Concepts  How it really works • Quick Start  Get it up and running and start playing. • The Details  How can it be a tech talk without details? • Some use cases  How you might use it. • Questions  Although you can bring them up at anytime! Agenda
  4. 5 “Gearman provides a generic application framework to farm out

    work to other machines or processes that are better suited to do the work. It allows you to do work in parallel, to load balance processing, and to call functions between languages.” Official Statement
  5. 6 •Gearman consists of a daemon, client and worker 

    At the core, they are simply small programs. •The daemon handles the negotiation of work  Workers and Clients •The worker does the work •The client requests work to be done What The Hell? Tell me!
  6. 8 •Gearman works on linux •API implementations available  PHP

     Perl  Java  Ruby  Python Platforms
  7. 13 •Head to gearman.org •Click Download •Click on the LaunchPad

    download •Download the Binary •Unpack the binary •./configure && make && make install •Bam! You’re off!  For more advanced configuration see ./configure –help •Starting  gearmand -d Installation
  8. 14 •Starting the Daemon  gearmand –d •Worker – command

    line style  gearman -w -f wc -- wc –l •Client – command line style  gearman -f wc < /etc/passwd •Check it! Simple Bash Example
  9. 16 •So, you know… we all like to talk about

    ourselves…  Yes, I wrote a layer on top of Zend Framework called Zend_Gearman; wow unique.  https://github.com/mwillbanks/Zend_Gearman PHP – Zend Framework
  10. 18 •Gearman by default is an in-memory queue  Leaving

    this as the default is ideal; however, does not work in all environments. •Persistent Queues  Libdrizzle  Libsqlite3  Libmemcached  Postgres  TokyoCabinet  MySQL  Redis Persistence
  11. 19 •Persistent queues require specific configuration during the compilation of

    gearman. •Additionally, arguments to the gearman daemon need to be passed to talk to the specific persistence layer. •Each persistence layer is actually built as a plugin to gearmand  http://bazaar.launchpad.net/~tangent- org/gearmand/trunk/files/head:/libgearman- server/plugins/queue/ Getting Up and Running with Persistence
  12. 21 •Clients send work to the gearmand server  This

    is called the workload; it can be anything that can become a string.  Utilize an open format; it will make life easier if you chose to use a different language for processing • XML, JSON, etc. • Yes, you can serialize objects if you wanted to… not recommended although. Clients
  13. 22 •Workers are the dudes in the factory doing all

    the work •Generally they will run as a daemon in the background •Workers register a function that they perform  They should ONLY be doing a single task.  This makes them far easier to manage. •The worker does the work and “can” return results  If you are doing the work asynchronously you generally do not return the result.  Synchronous work you will return the result. Workers
  14. 23 •Utilizing the Database  If you keep a database

    connection • Must have the ability to reconnect to the database. • Watch for connection timeouts •Handling Memory Leaks  Watch the amount of memory and detect leaks then kill the worker. •Request Languages  PHP for instance, sometimes slows down after hundreds of executions, kill it off if you know this will happen. Workers – special notes
  15. 24 •Workers sometimes have issues and die, or you need

    to boot them back up after a restart  Utilizing a service to watch your workers and ensure they are always running is a GOOD thing. •Supervisord  Can watch processes, restart them if they die or get killed  Can manage multiple processes of the same program  Can start and stop your workers. •When running workers, BE SURE to handle KILL signals such as SIGKILL. Keeping the Daemon Running
  16. 26 •Until recently you were writing something against the gearman

    socket interface…  telnet on port 4730  Write “STATUS” • Gives you the registered functions, number of workers and items in the queue. •Gearman Monitor – PHP Project  NOTE: I’ve never actually attempted this; BUT it is referenced on gearman.org so it must be doing something!  https://github.com/yugene/Gearman-Monitor Monitoring
  17. 28 •If you resize images on your web server: 

    Web servers should serve, not process images.  Images require a lot of memory AND processing power • They are best to be processed on their own! •Processing in the Background  Generally will require a change to your workflow and checking the status with XHR to see if the job has been completed. • This allows you to process them as you have resources available. • Have enough workers to process them “quickly enough” Images
  18. 30 •Sending email and/or generating templates and processing variables can

    take up time, time that is better spent getting the user to the next page. •The feedback on the mail doesn’t really make a difference so it is great to send it to the background. Email
  19. 32 •Get all of your logs to a single place

    •Process the logs to produce analytical data •Impression / Click Tracking •Why run a cron over your logs nightly?  Real-time data is where it is at! Log Analysis / Aggregation
  20. Questions? These slides will be posted to SlideShare & SpeakerDeck.

    Slideshare: http://www.slideshare.net/mwillbanks SpeakerDeck: http://speakerdeck.com/u/mwillbanks Twitter: mwillbanks G+: Mike Willbanks IRC (freenode): mwillbanks Blog: http://blog.digitalstruct.com GitHub: https://github.com/mwillbanks