Python Web Summit March 8, 2012

Note

Taking notes on a panel is really challenging. Apologies on whatever or to whoever I miss. Any misquotes are my fault and not the fault of the speaker.

Introduction

by Michael Ryabushkin

How did this start?

  • People:

    Chris McDonough
    Mike Orr
    Phil Jenvey
    Mike Bayer
    Danny Greenfeld
    Audrey Roy
    

During the summit

  • Panels are important, but take people aside if need be

Creating a Better Deployment Story

Moderated by Jacob Kaplan-Moss (Django DBFL)

Panelists:

  • Tarek Ziadé (Distribute/Packaging)

    • New distutils lets you specify versions of third-party packages. But... redhat and other OS tools have their own package names. Ugh.
  • Nate Aune (DjangoZoom, Appsembler)

    • All tools (Plone, Django) makes it hard to do deployments. Hence his deployment startups (DjangoZoom and Appsembler).
    • We need to come up with a standard and insist on using it.
  • Kenneth Reitz (Heroku)

    • pip needs to be able to set versions of Python
    • PyPI needs more attention
  • Ian Bicking (Paste/WebOb/Silver Lining)

    • Let’s create a formal Application Package specification.
    • Problem to overcome is the difference between developers and sys admin
  • Jim Fulton (Buildout/Zope)

    • Company does development and maintains 500 applications
    • “Packaging needs to be better so we can deploy more easily.”

Questions

  1. How come Java .war files do it better than Python?
  2. How do we make it so that Python is as easy to deploy to the web as PHP.
  3. What is the status quo?
  4. What are people working on to make this better?

Formal Application Package Specification?

Ian Bicking’s thought on how to do it. Would have these things:

Formal path specification
Formal dependency listings
Project description
Python version labeling
Include non-Python components (database, libraries, fortran, etc)

Jacob’s comment: Is this going down the route of Chef/Puppet?

Namespacing Distributions

  • Tarek as to play namespacing games to make sure that we get to use DistUtils backporting across versions of Python.
  • Armin Ronacher commented: “The fact that having the same package name for distutils2/packaging would be a problem shows the root of the issue: no proper version deps.

Takeaways

  • Force deployed applications be a package.

  • Formal application specification?

  • Specify Python versions in virtualenv/buildout

  • The challenge of dealing with vendor named projects

  • Jacob is going to publish a specification that will hopefully get the community moving. And he invites others to participate on his work.

    • The others pledged to help out or work on best practices.

Porting to Python 3

Moderated by Barry Warsaw (Canonical, Python FLUFL)

  • Chris McDonough (Pyramid/WebOb)

  • Armin Ronacher (Flask, Jinja2)

  • Guido van Rossum (Python)

    • I don’t know much about how you guys do Python web development and want to know so I can make it easier for Python 3 conversions.
    • There was artificial ambiguity introduced in Python 2 in regards to strings.
  • Mike Bayer (SQLAlchemy/Mako)

    • The Python database API needs some love. I agree in that a huge, unmentioned hurdle for Python three are other libraries besides web frameworks and unicode. DBAPI, PIL, etc.
    • PEP 249 doesn’t mention unicode. http://www.python.org/dev/peps/pep-0249/
  • Robert Brewer (CherryPy)

Questions

Compelling arguments for/on Python 3

  • Armin: Python 3 has some powerful features for sockets and other components that Python 2.x lacks.
  • Barry Warsaw: Newer and more powerful libraries being written in Python 3.
  • Wayne Witzel: Give a ton options for porting to python3, they won’t choose any of them. Most people just want to be told what is right.
  • Barry Warsaw: 2-to-3 tool is useful for getting started, but once in the weeds I find that I dive into the code.

Final Thoughts

Note

my summary of their statements.

  • Chris McDonough: We need to make people more enthusiastic about Python 3.
  • Armin Ronacher: Improve the guides on porting.
  • Guido Van Rossum: This will be resolved. It’s going to be a while, but we can make it. We’ll remember how hard it was to move forward to Python 3.
  • Mike Bayer: This will be resolved when we think in Python 3 by default. And make Python 2.x a boring backport.
  • Robert Brewer:

Factoring Code for Reuse

Moderated by Danny Greenfeld (consumernotebook.com)

  • Tres Seaver (Zope/CMF/Pyramid)
  • Mariano Reingart (Web2Py)
  • Alex Gaynor (Django/PyPy)
  • Michael Foord (IronPython, Mock)
  • Carl Meyer (Virtualenv, Pip)

“State-Of” Multi-Talk Round 1

Each of these speakers, a leader in their field, gets time to talk about his subject.

Graham Dumpleton (WSGI 2 ideas)

  • PEP 333 was created back in 2003

  • PEP 3333 was created back in 2010

  • Wanted something better:

    • Make it simpler

    • standardized high level request/response objects

    • Async support (not possible because so different)

    • Resource management

    • Unknown request content length

      • no compressed request content
      • No chunked requests
      • no full duplex HTTP
  • Has the boat sailed?

    • Too much legacy code relying on WSGI 1.0
    • Missed opportunity with Python 3
  • Graham’s ideas:

    • use context managers to improve resource management

      • need to override close() of the iterable
    • Implement wsgi.input as an iterable

    • TODO add what I missed

Benoît Chesneau (gunicorn)

The state of gunicorn:

  • First commit was November 30, 2009, Three users at first

  • preform model

  • Thread-safe

  • Automatic worker process management

  • Manage using signals

  • Natively support WSGI, Django, Paster

  • HTTP streams: decode on the fly http chunks

  • Supporting sendfile & FileWrapper

  • Simple Python configuration

  • Multiple workers (sync and async)

  • Various server hooks

  • use your own logger

  • Some issues:

    • Reload - graceful (HUP) or reexec (USR2)
    • The Django case: python-manage.py
    • Performance issues due to WSGI
    • CGI compatibility: headers
    • CONTENT_TYPE,CONTENT_LENGTH,SCRIPT_NAME
    • The WSGI spec needs to be completed
    • Async workers & blocking issues

      • IO access like the filesystem are not greened
      • C drivers
    • Web app configurations & deployment: we need a standard

  • Challenges:

    • Python 3

      • the case of sync workers
      • Handle extensions/plugins
    • New needs on the web: websockets, SPDY

    • modular HTTP & WSGI server in Python

Ben Bangert (Pylons Project)

State of the Pylons Project:

  • Lots of community plugins developing
  • Larger frameworks on top taking off along with bootstraps
  • Ploneconf Pyramid track
  • 62 reports, 31 devs with commit
  • lots of pyramid_* packages

Challenges

  • Porting to Python 3 (is doing very well)

  • organizing and simplifying sometimes overly pedantic documentation

  • TurboGears

    • Still exists
    • Some progress on community migration
  • Standardizing deployment

  • Foundation efforts

  • Pylons conf (189 members of the SF Bay Pylons meetup)

  • Increasing presence at conferences (not just Python ones)

  • More awareness of professional support

  • Moar books?

Robert Brewer (CherryPy)

State of CherryPy

  • It works in Python 2.7x
  • It works in Python 3.x
  • Being broken up into modular components so the WSGI HTTP server can be used in things like Pyramid

Promoting Python for Web Use

Moderated by Paul Everitt (Pyramid)

“State-Of” Multi-Talk - Round 2

Glyph Lefkowitz (Twisted)

  • It works for the web!
  • Lots of cool features
  • Works more as a container rather than a platform.
  • Has excellent support for Windows. Has an MSI, Executable, etc
  • Needs to fight the impression of being a giant library. It is actually small.

Jannis Leidel (Django Project)

  • Used in a lot of places around the world in small and gigantic projects

  • 21,700 user list

  • 7,000 developer list

  • 33 committers

  • 2,5000 downloads/day

  • 2,100 projects on PyPI

  • Django 1.4 almost out:

    • New project layout
    • Custom project templates
    • Standard WSGI entrypoint
    • Full timezone support
    • In browser testing
    • Cookie session backend
    • Clickjacking protection
    • New form wizard
    • i18n URLs
    • No exception wrapping templates anymore
    • *args, **kwargs

New orm functions:

select_for_update()
bulk_create()
distinct('filename')

New template stuff:

@assignment_tag
{% elif %}
{% static %}
{% truncatechars %}

Django 1.5 news:

  • Cool stuff is coming

Quotes

“Django isn’t a functional unit. You include it and it just sits there.”