Skip to content
Background workers' library for Google App Engine
Find file
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Failed to load latest commit information.



Warning: This project is under heavy development to bring it into some usable state first. If you want to use it, proceed with caution. The actual functionality provided by library may hardly resemble assumptions stated in this documentation. In other words: stuff is being made here. Help, if you can, but don't cry if something breaks.

Note: This library is to be used with Python 2.7 runtime on GAE.


gae-workers is a library that enables to execute long-running processes ("workers") on Google App Engine without the use of backends. For that, it uses a combination of task queue (for actual code execution) and memcache (for preserving the state of workers).


Once downloaded, include the library with your App Engine application. You should also add an entry to app.yaml that will make workers' tasks go to the script intended to handle them:

- url: /_ah/worker
  script: gaeworkers/

If you want, you can change the URL (among other parameters) by editing the file. (and, of course, changing in app.yaml as well).


Note: Examples below assume that 'gaeworkers' directory is in your app's root directory.

Using a worker is quite similar to working with the Python standard threading.Thread class. In general, you define a class inheriting from gaeworkers.Worker and implement its run() method to do your logic:

from gaeworkers import Worker
# ...
class MyWorker(Worker):
    def setup():
        self.query = Model.all()
    def run():
        for model in self.query:

There are few things to bear in mind though:

  • For best results, the run() method shall be a generator function that uses yield frequently. This allows gaeworkers to control the execution of the worker, measure the time it takes and estimate whether a deadline is looming and work shall be delegated to next task. Without the yielding, all code in run() has to be executed in one go; Python does not allow preempting.
  • State of worker object is preserved between queued tasks that are used for executing worker's code. Therefore any non-volatile data shall be stored in self's attributes.
  • run() is invoked "from the beginning" for every task spawned to handle the worker. Hence it is not a good place to have any sort of initialization. For that, implement the setup() method - it is ran only once per worker.

Starting a worker is straightforward:

worker = MyWorker(name='Model worker')

Assigning a name allows for easily distinguishing tasks belonging to different workers in App Engine logs and/or Appstats. The name is included in the query string worker's task URL, and is used as a name for the task.

Something went wrong with that request. Please try again.