Scaling Python

A talk I gave in November, 2015 at Pycon CA in Toronto, Canada

Transcript

1. SCALING PYTHON
   James King
   @agentultra
   

   View Slide

2. • CPython
   • Openstack
   • Firefox
   • WebGL
   • Many, many
   more…
   

   View Slide

3. View Slide

4. –Johnny Know-It-All
   “Dynamic languages do not
   scale to programming-in-the-
   large”
   

   View Slide

5. –Super Enterprise Hacker
   “Scripting languages are great
   for toys, one-off jobs, and
   prototypes but do not scale for
   serious projects”
   

   View Slide

6. –Armchair Academic
   “You cannot possibly ship
   software without a formal
   verification of correctness”
   

   View Slide

7. POPPYCOCK!
   • 4+ million LOC
   • Python (2.7 for the most part,
   sigh)
   • 2,351 contributors to the
   Liberty release
   • 28,647 accepted contributions
   • HP, Redhat, Mirantis
   contributed over 13k commits
   to Liberty as the top 3 out of
   165 companies
   Source: http://stackalytics.com
   http://activity.openstack.org/dash/browser/?release=liberty
   

   View Slide

8. POPPYCOCK!
   • RackSpace
   • HP
   • DreamCompute
   • Internap
   • … and many more
   

   View Slide

9. DREAMCOMPUTE
   • 2721.31 VMs running on average
   • 1.38 VCPU per VM on average
   • 1.65 GB of RAM per VM on average
   • 3755 VCPUs active
   • 4501 GB of RAM active
   

   View Slide

10. IT WORKS PRETTY WELL
    

    View Slide

11. IT WORKS PRETTY WELL
    … and it’s all powered by a wimpy toy-language with
    no static types or SAT-solving compilers…
    

    View Slide

12. View Slide

13. BLUEPRINTS
    https://blueprints.launchpad.net/nova
    

    View Slide

14. DEVSTACK
    http://docs.openstack.org/developer/devstack/
    

    View Slide

15. CODE REVIEW
    https://review.openstack.org
    

    View Slide

16. CODE REVIEW
    https://review.openstack.org
    

    View Slide

17. CONTINUOUS INTEGRATION
    

    View Slide

18. View Slide

19. THANK YOU
    

    View Slide