PyConAu Brisbane August 2015

OpenStack: A vision for the future

Monty Taylor

• psychoanalytic thespia

• are we making a toolkit for people to use to develop clouds? Or to deploy clouds in various place?

• OS should provide computers and networks that work

• OS should not chase 12-factor apps; there are other, new tools that can do this (e.g. Kubernetes)

• OS should provide a happy home for app frameworks

• OS should, by default, give me a directly routable IP

Using PyCharm Educational Edition

Thea Koutsoukis

• PyCharm is an IDE

  • Syntax and error highlighting

  • code completion

  • version control integration

  • python console

  • free and open source

  • "fill in the missing code" projects (course creator plugin)

  • easy to pick up, but offers good IDE functionality as they progress

  • jetbrains.com/pycharm-educational

• suitable for school age kids

• courses

  • allows students to learn at their own pace

  • provides progress status and hints when the students make errors

  • warning: plan lessons before you start, as currently you can't change files around easily

  • no external file input, can't keep code, no reading and writing to files

  • good for very specific syntax and output

• can also function as a regular IDE that is student friendly

  • good error messages

  • syntax highlighting

  • code completion

• lesson plan idea

  • recap

  • teach class

  • do lessons

  • big project

• alternatives

  • online interpreters

  • useful because you don't have to install anything

  • may not have a great deal of features

  • rely on good internet connections

  • make not work on all browsers

Using Flask and Python Anywhere in the Classroom

Brad Nielsen

• easier to work online than worrying about configuration in a BYOD school

• prefers Python because it is easier to get into than things like Java

• online environments are good because it lets the students try things they are interested in while the teacher functions as a mentor

• was looking for a python web server kids could program on both inside and outside the school network

• Flask and Django solve the same problems but Django can be more complex and restrictive

• pythonprogramming.net has some good Flask, Django, SQL, and other python learning resources

• Flask

  • easy to handle server requests, responses, and posted form data

  • easy to render dynamic content using the Jinja template library

  • easy to interact with a SQLite database without having to build database objects like Django

  • simple session handling

Orchestration and CI/CD with Ansible and OpenStack

Simone Soldateschi

• Rackspace devops guy

• CI pipeline to build container

• full stack

• orchestration

• service discovery

• CI pipeline go live

• Desired state: write code to tell systems how to manage themselves

• Full stack: hardware -> OpenStack -> CoreOS -> Docker

Trials & tribulations teaching Python at CoderDojo

Caleb Hattingh

• Short, complete programs

• better to start with something that works and let the students make changes from there than to give them something broken and ask them to fix it

• use these to launch discussions into real world computer and programming topics

Writing Python For Kids

Brendan Scott

• wrote Python for Kids for Dummies, published by Wiley

• publisher saw Brendan's blog and asked him to write a book

• took about 12 months from initial contact to publication

• only about a month of that was drafting the content

Open edX: An edX partner perspective

John Zornig

• edX is a non-profit, open-source Massive Open Online Course (MOOC) set up by Harvard and MIT in 2012

• edX code is available on GitHub

• UQ was one of the first universities to get involved with edX (UQx)

• primarily written in Python and Django

• edX.org is a SaaS based on open edX which is free but must be self-managed and hosted

• due to the number of universities involved, it can take a while to get pull requests merged to edX

OpenStack Works, so now what?

Monty Taylor

• as an app dev, I want to deploy and run an app on the internet so that my customers all over the world can consume it

• as an app dev, I want to deploy the application across multiple clouds so that my service survives issues in any one of them

• spin up and tear down 10-20k VMs on 3 clouds a day using only OS API

• 2000 devs work on OS, and every patch is integration tested twice before merging; 1.7 million test jobs in the last six months; 18 Terabytes of log data in six months

• each test runs on a single use cloud slave

• they run no hardware servers for these tests; everything is run on Rackspace and HP clouds

• Gerrit

  • traditional enterprise Java application (i.e. not particularly cloud-native)

  • run as a single Nova VM with a Cinder volume (which works well enough)

  • scale out farm of git replicas

  • 2500 changes every week

  • 15,000 change revisions every week

  • 10,000 new changes merged every 42 days

• nodepool

  • cloud native

  • purpose build in Python

  • keeps a pool of nodes ready to go so tests can be run immediately

  • tears down used nodes and creates new ones to replace them

  • multi-cloud

  • fully elastic -> responds to demand automatically

• All of this works, so what next? Make it easier!

• Basic things you want to do which are currently harder than they should be:

  • get (make/fetch/find) a base image

  • upload it to each cloud

  • boot a VM on one or more clouds

  • available on the internet

• Get a base image

  • OS diskimage-builder

  • packer

  • images from Ubuntu/RH/SUSE

  • problem: hypervisor image file format

    • Rackspace uses VHD

    • HP uses qcow2

    • DreamHost uses RAW

  • problem: image API version

    • HP uses v1

    • vechost uses v2

  • problem: image task vs. PUT

  • problem: ensure it's on the internet

    • there are a whole bunch of different combinations between vendors

• solutions:

  • os-client-config

  • shade

    • a library to wrap business logic around client libraries

    • in use in infra nodepool and ansible (v2.0, multi-cloud support)

• Future directions:

  • get back to basics

  • shade existence is a bug

  • make some decision about divergences in the basic levels

  • take a stand even if one of our product managers disagrees

  • ensure that simple things are simple

Keynote: Consequences of an Insightful Algorithm

Carina C. Zona

• algorithms impose consequences on people all the time

  • an algorithm is a step-by-step set of operations for predictably arriving at a given outcome

  • in CS = patterns of instructions articulated in code

  • in the world generally = patterns articulated in a variety of ways

• deep learning is a way of training artificial neural networks

  • relies on an automated discovery of patterns

• a false positive or false negative when dealing with human data must be dealt with very carefully

  • people cannot be treated simply as numbers

• standards for ethical research involving humans and human data tend to be only considered carefully in academia, but should be observed in other areas too.

• we cannot intuit inner state, emotions, private subjectivity

• consent is permission granted != informed consent is permission freely granted

  • 'no' is the consequence-free alternative and the default value with informed appreciation and understanding (ahead of time) of the consequences

• we need to learn from professional/academic ethics

  • avoid harm to others

  • consider the potential impact on others

  • be honest and trustworthy

  • actively counter bias and inequality

  • audit outcomes

• insightful algorithms are growing in precision but the risk of damage is also rising

• anticipate diverse ways to screw up

  • this requires truly diverse teams having decision authority

• cultivate informed consent

• audit constantly

• recognize bias is inherent

• visionary about countering bias

• aim mining tools at public benefit consequences

  • currently these are mostly aimed at corporate uses

• commit to transparency

Learn You a Flask

Lachlan Blackhall

• example of Flask app running on Heroku

• flask at the core doesn't provide much before serving requests

• further functionality is provided by plugins

• use an ORM, such as SQLAlCHEMY to access SQL database; don't use raw SQL

  • we're Python programmers, work high up the stack!

Easy wins with Cython: fast and multi-core

Caleb Hattingh

• Cython lets you specify object/variable types to save time as Python doesn't have to determine type

• Cython is a super set of Python; it can compile anything written in Python into a C file

• easycython is a useful wrapper to help with cython code

• it speeds up Python code significantly when crunching large amounts of data

  • can be applied just to the bits of the code that need speeding up, the rest can just be standard python

• PyPy and Numba are both options for speeding up Python code as well

• Cython can be run multi-core (with no global interpreter lock)

• but limit your Cython to only the bit that really need it, as it adds some complexity and debugging difficulties

Rapid prototyping with teenagers

Katie Bell

• NCSS program

• also run a summer school (has been running for 20 years)

  • 2.5 days of teaching python

  • 3 days of project time, with an all-nighter on the last night

• three things excite kids for learning:

  • How stuff works

  • Building cool stuff

  • Problem solving

• goal of the project was to build a completely functioning social networking site

• build a site quickly with minimal magic

• used tornado for building the web page, but tornado can look a bit complicated so they built a wrapper and let the kids use that

• sqlite for persistent data as it is relatively simple and built into Python

• use simple tools

• group brainstorm only at the beginning

• use version control

• communication is paramount

• continual integration -> always have something working, have students demo often

• explain new procedures to the whole class at the same time

• share successes

On Being Pythonic

Peter Lovett