Before getting started I want to thank Alexis Bellido at ventanazul.com for his patients with my Django questions, check out his post Django questions and answers with a Swedish guy for some more info on Django setup (if you didnt figure it out I am that Swedish guy).

Installation guides
When I set up Django locally I wanted to make my local development environment as similar to a production environment as possible, it will hopefully make production deployment easier down the line. To do this I am running Python 2.5, Django 1.0, PostgreSQL 8.3 (and thus the neccessary python driver psycopg2) and mod_python on Apache. There are some good installation guides out there, so setting up this stuff was mostly a walk in the park, check out the install guide in the Django Book, the quick install guide at Django Project or the install guide at WebMonkey. Of course there are some gotchas that took me quite some time to figure out…

Uninstall old versions of Django
Before getting started I already had Django 0.96 installed, and I  just installed Django 1.0 over it assuming that it was going to replace the old version. I was wrong and when I tried to run my new Django installation I got errors like “NameError: name ‘url’ is not defined” and “ImportError: cannot import name WEEKDAYS_ABBR” – neither which made much sense to me. It turns out that if you do not uninstall an old Django installation first you get a mix of new and old Django files, and that just do not work very well. So take care to acctually read and follow the instructions at Remove any old versions of Django.

Use the right version of the db driver (duh!)
To get the PostgreSQL working correctly with your Python install you have to use a version of the driver Psycopg2 that matches both your Python version and your Apache version, otherwise things just do not work and you do not really get a helpful error message. You can get the window versions of Psycopg2 here.

Configure Apache for mod_python
Once Django and Apache are installed you can create your new Django project (via django-admin.py startproject myNewProject) where Apache can find it (in htdocs for example). To be able to use Apache as your webserver you also need to install and configure mod_python. Installing is straight forward (see the install gudies above for more info), but configuring Apache took me some time. You need to edit httpd.conf and add the following to the end of the file:

<Location "/myNewProject">
SetHandler python-program
PythonHandler django.core.handlers.modpython
SetEnv DJANGO_SETTINGS_MODULE settings
PythonOption django.root /myNewProject
PythonDebug On
PythonPath "['C:\Program Files\Apache Group\Apache2\htdocs\myNewProject'] + sys.path"
</Location>

myNewProject is of course the name of your own Django project.

Hopefully this is of use to someone other than me, if not, then at least I have my notes organised for the next time. If I have missed anything or gotten anything wrong I would very much appreciate your feedback!

This post was originally published in swedish on Mashup.se, my blog about swedish mashups and APIs.

Only Python
If you don’t know Python you don’t have choice if you want to use Google App Engine, you just have to learn. It’s not a difficult programming language to learn if you already know how to program. Django, the Python framework that really speeds up writing applications for Google App Engine, is also quite easy to pick up (tip: Use the Google App Engine helper for Django). Before I started with Google App Engine I hadn’t written a single line pf Python and after a few intense days of concentration and coffee consumption I knew Python, Django and Google App Engine quite well.

Most 3rd party python libraries work perfectly on Google App Engine (Django is just one example), but there are limitations. Only libraries that are 100% Python can be used, so if the library has any code in C it can’t be used on Google App Engine. If a librarry has any code that makes a HTTP request or similar it can’t be used on Google App Engine either. All HTTP requests have to be done via Google App Engines URL Fetch API.

Differences between local dev environment and the production environment
One of the advantages with Google App Engine is that there is a local development server that simulates how Google App Engine works when an application is in production. This makes it easy to develop an application locally and then deploy it to the live servers on Google App Engine. As a developer it is important to pay attention to the differences between executing an application on the local server compared to the production environment. It’s no fun to spend time writing code that just don’t work once deployed.

The biggest difference betweent the two environments are that the requests made to 3rd parties work differently. If you have an application that uses the Delicous API it will work fine locally, but once deployed it wont work at all. The reason is that Delicous is blocking all requests from Google App Engine IP-addresses. Same thing is true for the Twitter API due to some HTTP headers that Google App Engine sets (Twitter claims to have fixed this now, haven’t had a chance to test yet). To avoid these problems you need to test your application often in the live environment, especially when you use APIs to call other services.

Datastore
The Google App Engine datastore has some limitations due to being a distributed database. The most obvious limitation is that there isn’t an “OR” operator in GQL (the Datastore version of SQL), but that is easily handled when coding. A more annoying limitations is that it is not possible to create a new entity (data object) in the Datastore via the Google App Engine dashboard unless there already exists an entity of this type, and that there is a real noticable delay between what one can see in the Datastorer in the Dashboard compared to what really is stored in the Datastore. This makes it very hard to really check what data that is stored in the Datastore in any given moment, which makes debugging more difficult.

No scheduled processes
One of the most restrictive limitations with Google App Engine is that there is no way to start a process, other than via a HTTP request. There is no type of recurring scheduled process (like a cron job) and no triggers or hooks to use to start a process when a special event occures. Almost all web applications need som kind of scheduled process to function correctly – to clean up old data, send emails, consolidate statistics or fetch data from an RSS feed once an hour.

The easiest way to get around this limitation is to create a cron job on another server that calls an URL in the Google App Engine application. If you have a lot of visitors you can also perform the background process as part of a user request. For example you could import data from an RSS feed when a user logs in to your application. This will of course make the user experience slow and there is no guarantee that users will perform the action to trigger the background process att the right times.

No matter which solution is implemented one will quickly bump into the next limitation of Google App Engine, that only short processes are permitted.

Only short processes
Something I have learned the hard way is that Google App Engine only is built to handle applications where a users makes a request and quickly gets an answer back. There is no support for a process that  executes for a longer time, and for the time being this limitations seems to be approximatly 9 secondes/process. After that an exception is thrown and the process killed. It does not end there, even if you are nowhere close to use the assigned CPU resources you can quickly use too many resources with long running processes since they have their own unique resource pool. If you use too many resources you application is shut down for 24 hours, and right now there is no way to buy extra resources.

This is a really serious limitation in Google App Engine that it is really difficult to get around. If you need heavy processes it is recommended that you use Amazon EC2 or something similar. To handle (not get around) the limitation you have to handle exceptions in a nice way and use transactions. More about this in William Vambenepe’s very informative post Emulating a long-running process (and a scheduler) in Google App Engine. He has some tips on how to get around this limitation, even if it is not recommended since you risk that your application is shut down.

More limitations
There are more limitations, read  Google App Engine: The good, the bad, and the ugly? for a longer list. Just keep in mind that some of the limitations mentioned in that post already have been addressed by Google.

To summerize you need to really know what limitations there are in Google App Engine before you spend time and energy on developing an application. If the limitations are not a problem then there is a lot to gain by using Google App Engine.