Scaffold + Tools for creating/developing pyramid applications on Google App Engine
A scaffold to help you get started writing a pyramid aplication that will run on Google App Engine.
pyramid_appengine can be installed via pip or easy_install
$ pip install pyramid_appengine
Once installation has completed, an appengine_starter template will be made avaialable to use to create projects.
$ pcreate --l Available templates: appengine_starter: Pyramid scaffold for appengine ...
The project generated from the scaffold utilizes buildout 2 which no longer attempts to provide full or partial isolation from system python installations. As a result it is a good idea to use a virtualenv to provide that isolation.
This can be accomplished by simply creating a new virtualenv and using it’s interpreter to run bootstrap.py. Using virtualenv is out of scope but there is plenty of information on the internet on how to do it.
To get started, first create your project skeleton.
$ pcreate -t appengine_starter mynewproject
A buildout environment for your project will be created. once complete, run the buildout as usual
$ cd meynewproject $ virtualenv -p /usr/bin/python2.7 .env $ .env/bin/python2.7 bootstrap.py $ ./bin/buildout
The buildout will take care of downloading and installing the App Engine SDK (currently 1.6.3). it will be located in “./parts/google_appengine” all utils for deploying and running the development server will be located in “./bin”
Your source code for your project will be located at “./src/mynewproject”, a bundle of your source and it’s dependencies will be located at “./parts/mynewproject”
Running your project for development
your pyramid site will be running on port 8000 so point your browser at
The app engine admin console for your app is on port 8010
You will find the supervisor service on port 9999 at
From there you can check to see if the service is running, and you can start/stop/restart and tail the log.
Deploying your application to App Engine
Assuming you have created an application id “mynewproject” on app engine, the application can be deployed like so.
$ ./bin/appcfg update parts/mynewproject -A mynewproject -V dev
Then your application will be running at…
What It Does And Why
Most pyramid scaffolds create a project directory structure that is an installable through the pip/easy_install . However, App Engine applications do not support that format. Instead App Engine assumes that everything is contained in one directory including all of the projects dependencies not provided by the App Engine run time.
So a directory structure for an application deployable to App Engine looks like this…
/myproject/ /myproject/app.yaml /myproject/index.yaml /myproject/queue.yaml /myproject/pyramid /myproject/verlruse /myproject/jinja2 /myproject/newfangledlib
Because of this directory structure, which is vastly different from what is expected by other tools, we need a way to develop in your typical python egg format, but deploy in an App Engine format.
Buildout is a tool that can be used to support the kind of setup where you develop your application as an egg but deploy what App Engine expects. If you aren’t familiar with buildout you may want to read up on it. It has some of the same goals as virtualenv, but has more features via recipes to help with deployment.
For running the buildout you typically do …
$ /path/to/python bootstrap.py $ ./bin/buildout
The buildout.cfg file distributed with python_appengine does the following.
- creates a buildout environment where the source for your project is located at ./src/nameofproject
When buildout is run …
- all the dependencies for your project are downloaded and setup in the buildout environment
- the appengine sdk is downloaded and installed in the buildout environment under ./parts/google_appengine .
- tools such as devappserver, appcfg which are tools distributed with the app engine sdk are put in the buildouts bin directory
- a supervisor script to run the dev_appserver.py is generated
Buildout and Virtualenv
In order for Google App Engine’s Dev Server and upload script to function correctly all files which are being used by the project must be collected together into a flat hierarchy, as described above.
By default, however, buildout will not create directories for any packages already present in the system’s site-packages directory.
Since buildout 2.0 has been released, the suggested way to provide package isolation is to create a virtualenv and then use that interpreter to bootstrap your buildout.
Managing dependencies for deployment
As mentioned earlier, all dependencies must be contained in the applications deployment directory under parts or provided by the app engine runtime environment. As your application gets bigger and bigger you will likely edit the buildout.cfg from time to time to add more dependencies so that they are deployed with your application.
To update the dependencies for your application edit the packages attribute under the stanza for your project in the buildout.cfg and then run ./bin/buildout again to have the dependencies symlinked or copied to parts/mynewproject
As a general rule, having a thorough unit test suite is good. But in the authors opinion it is essential for app engine applications. The main reason being that app engine requires you to specify the datastore indexes you need to support the application at deployment time via the index.yaml.
The app engine sdk will update your index.yaml for you when you are running your application on the development server. But it requires you use something that generates an HTTP request in order to trigger the behavior. So in theory, you would have to make sure you hit every page of your application before you deploy to insure any new index needs caused by new or updated queries are recorded.
This method is error prone and time consuming. A better way is to have your unit tests generate it.
The project generated by the scaffold includes everything you need to do this. By using py.test and hooks specified in conftest.py, a couple things are guaranteed.
- a clean appengine environment is initialized before each test
- any changes to index.yaml are written after each test
Tests can be run from the root of the project directory like so.
$ ../../bin/python setup.py test
upgrade sdk to 1.8.7
fix setuptools craziness with rod.recipe.appengine by pinning setuptools in versions.cfg
add flag in supervisor config for devappserver to avoid checking for sdk updates when it starts up so that the app either comes up or errors out, not just hangs there waiting for user input that it will never get.
Slight modification to docs to give an actual opinion on whether to use virtualenv or not. As in yes, use a virtualenv for less python packaging pain.
rebuild with all the artifacts (reademe.rst, CHANGES.txt etc…..)
added supervisord scripts to run dev_appserver
App Engine SDK version bumped to 1.8.0
include versions.cfg template for scaffold
bump sdk to 1.7.5
bump pyramid version to 1.4
upgrade to buildout 2.0
added versions.cfg to make managing versions of dependencies easier.
bump sdk version to 1.7.2
adding “unzip=true” to the buildout.cfg template to help out with eggs that are distributed that way
pinning to pyramid version 1.3
bump SDK version to 1.6.6
changed buildout.cfg to ignore site-packages which should fix a lot of issues with various namespaces packages that pyramid potentially depends on.
upgrading buildout to appengine sdk 1.6.5 and pyramid 1.3
changed application template to not use an ini file because it doesn’t really add any value on appengine, but handling the paste* dependencies is troublesome for the recipe rod.recipe.appengine under some circumstances.
This is just the default behavior of course, there’s nothing stopping you from deploying with an ini file if you wish.
upgrading buildout to appengine sdk 1.6.4, the first SDK which works with python2.7.
fixes to address github issue #6
usage of project vs package used incorrectly.
minor changes to readme.rst