Thursday, April 28, 2016

Lazy Evaluation and SQL Queries in the Django Shell

In Django terms, a QuerySet is an iterable of database records. What's nice about them is that they are evaluated only when you're ready for the results.

This means that even if it takes you a few lines of code to chain multiple queries, the Django ORM combines them into a single query. Less queries mean your database doesn't have to work as hard, and your website runs faster.


Evaluating a QuerySet Repeatedly

Imagine that we work for Häagen-Dazs and have access to their Django shell. We can use this to our advantage by hunting for free ice cream promotions.

Here, we get the active Promo objects. We evaluate the results just to see what promos are available. Then we filter them on the word free.

>>> results = Promo.objects.active()

>>> results
[<Promo: Free Flavors on Your Birthday>, <Promo: 10% Off All Cones>, 
<Promo: Buy 1, Get 1 Free>]

>>> results = results.filter(
>>>     Q(name__istartswith='free') |
>>>     Q(description__icontains='free')
>>> )

>>> results
[<Promo: Free Flavors on Your Birthday>]

The queries generated by the above are:

from django.db import connection

>>> connection.queries
[ {'sql': 'SELECT "flavors_promo"."id", "flavors_promo"."name", 
"flavors_promo"."description", "flavors_promo"."status" FROM 
"flavors_promo" WHERE "flavors_promo"."status" = \'active\' 
LIMIT 21',
  'time': '0.000'},
 {'sql': 'SELECT "flavors_promo"."id", "flavors_promo"."name", 
"flavors_promo"."description", "flavors_promo"."status" FROM 
"flavors_promo" WHERE ("flavors_promo"."status" = \'active\' 
AND ("flavors_promo"."name" LIKE \'free%\' ESCAPE \'\\\' OR 
"flavors_promo"."description" LIKE \'%free%\' ESCAPE \'\\\')) 
LIMIT 21',
  'time': '0.001'}]

There are 2 queries because we evaluated the results twice.

The first query was from the first time we retrieved all the active promos. It's pretty short. It just selects Promo records where promo.status is active.

The second query was from the second time we evaluated results, after we filtered for "free" in the promo names and descriptions.

As a side note, there is a bit of extra work in the second query as the second query still has that WHERE 'flavors_promo'.'status' = 'active' part. One might expect filter() to simply filter on the already-retrieved results rather than hitting the database again. But that's alright because the extra time is negligible.

Before we move on, let's reset the list of queries:

>>> from django.db import reset_queries
>>> reset_queries()

Evaluating a QuerySet Once

Now, let's look at what the queries would be if we only evaluated the results QuerySet once. Let's try building the same QuerySet again. Oh wait, just for fun, let's chain another operation so that we can be really sure that lazy evaluation is happening.

>>> results = Promo.objects.active()

>>> results = results.filter(
...     Q(name__istartswith=name) |
...     Q(description__icontains=name)
... )

>>> results = results.exclude(status='melted')

>>> results
[<Promo: Free Flavors on Your Birthday>]

As you can see, there's only one query:

>>> connection.queries
[{'sql': 'SELECT "flavors_promo"."id", "flavors_promo"."name", 
"flavors_promo"."description", "flavors_promo"."status" FROM 
"flavors_promo" WHERE ("flavors_promo"."status" = \'active\' AND
("flavors_promo"."name" LIKE \'free%\' ESCAPE \'\\\' OR 
"flavors_promo"."description" LIKE \'%free%\' ESCAPE \'\\\') AND 
NOT ("flavors_promo"."status" = \'melted\')) LIMIT 21',
  'time': '0.001'}]

Thanks to lazy evaluation, only one query was constructed, despite chaining multiple operations. That was nice.

Sure, the query could have been more optimal without the AND NOT melted part, but arguably that wasn't Django's fault, it was mine. But it gives me a clue about which operation I didn't need to chain in the Python code.

Next Steps

Try this on one of your projects. Open the Django shell, then try out some queries and see how they are evaluated. In particular, look at queries from one of your slower views.

You can also do similar things with Django Debug Toolbar. However, in the shell you can dissect your Python code line by line, which can be very helpful.

Monday, November 9, 2015

Solving UnicodeDecodeErrors Due to Opening Binary Files


Common Scenario: Walking Directory Tree and Opening Files

A common thing to do in Python is to go through a directory tree, opening each file and doing something with the file's text.

for path in paths:
    for line in open(path, 'r'):
        # Do something with each line of the file here.
        # Go ahead, right inside the for loop.
        # It's a text file, so imagine the possibilities.

Here, we iterate over all the paths in the directory tree. For each path, we open the file for reading. Then we go through each line of the file and do something with it.

The Problem

This works well enough for many situations, but at some point you end up running into a UnicodeDecodeError when you try to open a particular file. Usually, it's because that file isn't a text file: for example, it might be a JPEG or a font file.

Those errors are scary! They look like this:

for line in open(path, 'r'):
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 

self = <encodings.utf_8.IncrementalDecoder object at 0x10349a320>
input = b"\x00\x00\x01\x00\x02\x00  \x00\x00\x01\x00 \x00(\x10\x00\x00&\x00\x00\x00\x10\x10\x00\x00\x01\x00 \x00(\x04\x00\x00N...00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00"
final = False

def decode(self, input, final=False):
    # decode input (taking the buffer into account)
    data = self.buffer + input
>       (result, consumed) = self._buffer_decode(data, self.errors, final)
E       UnicodeDecodeError: 'utf-8' codec can't decode byte 0xad in position 89: invalid start byte

Before you go into a UnicodeDecodePanic trying out all the variants of open, io.open, unicode_open, etc., think about whether the file you're trying to open is even a text file.

The Solution

To solve the problem of accidentally opening non-text files, you can use BinaryOrNot's is_binary function. Just check to make sure the file isn't a binary before attempting to open it, like this:

from binaryornot.check import is_binary

for path in paths:
    if not is_binary(path):
        for line in open(path, 'r'):
            # Do something with each line of the file here.
            # Go ahead, right inside the for loop.
            # It's a text file, so imagine the possibilities.

This is a real-life code example. In fact, it comes from a fix to cookiecutter-django's tests that I just committed this weekend, which comes from Cookiecutter core code.

BinaryOrNot is a package that guesses whether a file is binary or text. I put it together a couple of years ago in order to use it in Cookiecutter. Since then, I've found uses for it over and over in various projects.

More Info

BinaryOrNot on GitHub: https://github.com/audreyr/binaryornot
Project documentation: http://binaryornot.readthedocs.org/

Tuesday, November 3, 2015

Intensive Django Training With the 91st Cyberspace Operations Squadron

Daniel and I just returned from a trip to San Antonio, Texas, where we taught one of our intensive Django training workshops at Lackland Air Force Base.

We prepared a customized version of our curriculum to meet the needs of the 91st Cyberspace Operations Squadron of the US Air Force.

Our intensive Django training at Lackland Air Force Base in San Antonio, Texas.

Teaching such a sharp, enthusiastic group and seeing everyone grasp difficult concepts so rapidly was a huge thrill. As instructors who like challenges, we tend to err on the side of assuming that our students can handle anything, so we threw a lot of very advanced topics at the group, wondering how much would click. On the last day as they were putting their knowledge into practice during hands-on project time, it was apparent that even the hardest parts had made an imprint.

For more info, see Intensive Django Training with the US Air Force, Daniel's detailed blog post about the training experience.

Special thanks to Capt. Jonathan D. Miller for making this possible. It was an honor to work with you and your team.

Tuesday, May 26, 2015

Our Trip to DjangoGirls Ensenada, Mexico

This weekend, Daniel and I drove down to Ensenada, Mexico to speak and coach at DjangoGirls Ensenada. It was a 2-day workshop for women of any level of experience to get a taste of web application development.

A photo posted by Audrey Roy Greenfeld (@pyaudrey) on

The event was organized by DjangoGirlsMX with the help of the US Consulate General of Tijuana and the non-profit Hala Ken.

We asked the US Consulate and Hala Ken about why they decided to get involved. They answered that Django Girls workshops fit perfectly into two of their major areas of interest: new technology and women's empowerment.

We were honored to be invited as guest speakers and appreciative of the opportunity, knowing that we could make a big difference showing women new to Django that we cared.

At the end of the morning session, we gave a talk to inspire attendees to keep going with their programming journey. It was called "Programming Gives You Superpowers." Here are the slides.

Note: for fun we made the cover image a little fancier after the talk, otherwise it's the same :)

It was a fantastic experience getting to spend time with the web development community of Tijuana/Ensenada. So many of the Python Tijuana and Django Girls Tijuana organizers and members drove out to Ensenada and spent the night in hotels to help make this happen. We had fun coaching alongside them after our talk.

My co-author, co-presenter, co-everything husband PyDanny also blogged his account of it: My First Django Girls Event

Finally, we had such a great time that we're now working on planning an upcoming Inland Empire DjangoGirls/RailsGirls event. All are invited to help: RSVP here or here for the May 30 planning session.

Sunday, May 3, 2015

Two Scoops of Django 1.8 is out!

Daniel Roy Greenfeld and I have updated Two Scoops of Django to 1.8, since Django 1.8 is a Long Term Support version.

The book is now available as a PDF. I know this will make a lot of folks happy! The print paperback is coming soon (US and India editions to start).

More info: http://twoscoopspress.org/products/two-scoops-of-django-1-8


I wrote half the book, including some of the rather difficult parts :) I also did the illustrations. The book is filled with a ton of weird cartoons and silly humor.

Enjoy, and hope it's helpful!

Sunday, April 12, 2015

Spring Cleaning for Python Programmers

It's spring again, which means that for Python programmers, it's time to clean out your hard drive.

Instructions:

1. Add these lines to your .bashrc (or other shell rc) file:

alias rmpyc='find . -type f -name "*.pyc" -print -delete'

export PYTHONDONTWRITEBYTECODE=true

The first part gives you a handy rmpyc command to recursively delete .pyc files.

The second part tells Python not to write .pyc files anymore.

2. Source your rc file and run rmpyc from your home directory (on UNIX, from ~). This will delete all the Python bytecode from your home dir onward. You don't need to keep it around because it'll just get rewritten as needed anyway.

3. Delete the virtualenvs that you're not using. (e.g. if you use virtualenvwrapper, delete the directories in ~/.virtualenvs/ that you don't need).

4. If you use VirtualBox, delete the virtual machines that you don't need.

5. Delete the repos that you don't need around anymore.

In my case I freed up 3 GB by removing the .pyc files and 25 GB by removing the virtual machines. I forgot to check how much space my unused virtualenvs took up, but it was probably a non-trivial amount.

My numbers are probably higher than most because my laptop's almost 5 years old and I mess around with random Python packages a lot, but you should still be able to save some space. At the very least, it'll be like squeezing the last paste out of a toothpaste tube.


Note: originally the instructions said the following, but I updated them after advice from Dan Crosta, Glyph, and Kit. Thank you all so much for the tips!

alias rmpyc='find . -type f -name "*.pyc" -print0 | xargs -0 rm -v'

Tuesday, March 24, 2015

Pillow Flowers

Lately, I've been playing around with drawing flowers with Python and Pillow.

The trick to drawing flowers is to iterate around the petals in polar coordinates, and then convert polar to cartesian for drawing purposes.

Flower Experiment 1 Flower Experiment 1 Flower Experiment 1 Flower Experiment 1 Flower Experiment 1

I demoed some of this at the meetup that I hosted last night, Inland Empire Pyladies' Coding for Artists. There, Danny and I taught participants how to use Pillow, ellipses, rectangles, random number generation, and trigonometry to make basic 2D generative art.