Django-supervisor combines the process-management awesomeness of supervisord with the convenience of Django's management scripts.
Running a Django project these days often entails much more than just starting up a webserver. You might need to have Django running under FCGI or CherryPy, with background tasks being managed by celeryd, periodic tasks scheduled by celerybeat, and any number of other processes all cooperating to keep the project up and running.
When you're just developing or debugging, it's a pain having to start and stop all these different processes by hand.
When you're deploying, it's a pain to make sure that each process is hooked into the system startup scripts with the correct configuration.
Django-supervisor provides a convenient bridge between your Django project and the supervisord process control system. It makes starting all the processes required by your project as simple as:
$ python myproject/manage.py supervisor
Django-supervisor is admittedly quite a thin layer on top of the wonderful functionality provided by supervisord. But by integrating tightly with Django's management scripts you gain several advantages:
- manage.py remains the single point of control for running your project.
- Running all those processes is just as easy in development as it is in production.
- You get auto-reloading for all processes when running in debug mode.
- Process configuration can depend on Django settings and environment variables, and have paths relative to your project and/or apps.
- Apps can provide default process configurations, which projects can then tweak or override as needed.
Django-supervisor is a wrapper around supervisord, so it uses the same configuration file format. Basically, you write an ini-style config file where each section defines a process to be launched. Some examples can be found below, but you'll want to refer to the supervisord docs for all the finer details:
http://www.supervisord.org
To get started, just include "djsupervisor" in your INSTALLED_APPS and drop a "supervisord.conf" file in your project directory, right next to the main manage.py script.
A simple example config might run both the Django development server and the Celery task daemon:
[program:webserver]
command={{ PYTHON }} {{ PROJECT_DIR }}/manage.py runserver --noreload
[program:celeryworker]
command={{ PYTHON }} {{ PROJECT_DIR }}/manage.py celery worker -l info
[program:celerybeat]
command={{ PYTHON }} {{ PROJECT_DIR }}/manage.py celery beat -l info
Now when you run the "supervisor" management command, it will detect this file and start the two processes for you.
Notice that the config file is interpreted using Django's templating engine. This lets you do fun things like locate files relative to the project root directory.
Better yet, you can make parts of the config conditional based on project settings or on the environment. For example, you might start the development server when debugging but run under FCGI in production:
[program:webserver]
{% if settings.DEBUG %}
command={{ PYTHON }} {{ PROJECT_DIR }}/manage.py runserver
{% else %}
command={{ PYTHON }} {{ PROJECT_DIR }}/manage.py runfcgi host=127.0.0.1 port=8025
{% endif %}
Django-supervisor provides a new Django management command named "supervisor" which allows you to control all of the processes belonging to your project.
When run without arguments, it will spawn supervisord to launch and monitor all the configured processs. Here's some example output using the config file shown in the previous section:
$ python myproject/manage.py supervisor 2011-06-07 23:46:45,253 INFO RPC interface 'supervisor' initialized 2011-06-07 23:46:45,253 INFO supervisord started with pid 4787 2011-06-07 23:46:46,258 INFO spawned: 'celeryd' with pid 4799 2011-06-07 23:46:46,275 INFO spawned: 'webserver' with pid 4801 2011-06-07 23:46:47,456 INFO success: webserver entered RUNNING state, process has stayed up for > than 1 seconds (startsecs) 2011-06-07 23:46:56,512 INFO success: celeryd entered RUNNING state, process has stayed up for > than 10 seconds (startsecs)
By default the "supervisor" command will stay in the foreground and print status updates to the console. Pass the --daemonize option to have it run in the background. You can also tweak its behaviour using all of supervisord's standard options in the config file.
Once the supervisor is up and running, you can interact with it to control the running processes. Running "manage.py supervisor shell" will launch the interactive supervisorctl command shell. From here you can view process status and start/stop/restart individual processes:
$ python myproject/manage.py supervisor shell celeryd RUNNING pid 4799, uptime 0:03:17 webserver RUNNING pid 4801, uptime 0:03:17 supervisor> supervisor> help default commands (type help <topic>): ===================================== add clear fg open quit remove restart start stop update avail exit maintail pid reload reread shutdown status tail version supervisor> supervisor> stop celeryd celeryd: stopped supervisor> supervisor> status celeryd STOPPED Jun 07 11:51 PM webserver RUNNING pid 4801, uptime 0:04:45 supervisor>
You can also issue individual process-manangement commands directly on the command-line:
$ python myproject/manage.py supervisor start celeryd celeryd: started $ $ python myproject/manage.py supervisor status celeryd RUNNING pid 4937, uptime 0:00:55 webserver RUNNING pid 4801, uptime 0:09:05 $ $ python myproject/manage.py supervisor shutdown Shut down $
For details of all the available management commands, consult the supervisord documentation.
The "supervisor" command accepts the following options:
--daemonize run the supervisord process in the background --pidfile store PID of supervisord process in this file --logfile write supervisord logs to this file --project-dir use this as the django project directory --launch=program launch program automatically at supervisor startup --nolaunch=program don't launch program automatically at startup --exclude=program remove program from the supervisord config --include=program include program in the supervisord config --autoreload=program restart program when code files change --noreload don't restart programs when code files change
Django-supervisor provides some extra niceties on top of the configuration language of supervisord.
All supervisord.conf files are rendered through Django's templating system. This allows you to interpolate values from the settings or environment, and conditionally switch processes on or off. The template context for each configuration file contains the following variables:
PROJECT_DIR the top-level directory of your project (i.e. the
directory containing your manage.py script).
APP_DIR for app-provided config files, the top-level
directory containing the application code.
PYTHON full path to the current python interpreter.
SUPERVISOR_OPTIONS the command-line options passed to manage.py.
settings the Django settings module, as seen by your code.
environ the os.environ dict, as seen by your code.
Django-supervisor recognises some special config-file options that are useful when merging multiple app-specific and project-specific configuration files.
The [program:__defaults__] section can be used to provide default options for all other [program] sections. These options will only be used if none of the config files found by django-supervisor provide that option for a specific program.
The [program:__overrides__] section can be used to override options for all configured programs. These options will be applied to all processes regardless of what any other config file has to say.
Finally, you can completely disable a [program] section by setting the option "exclude" to true. This is mostly useful for disabling process definitions provided by a third-party application.
Here's an example config file that shows them all in action:
; We want all programs to redirect stderr by default, ; unless specifically configured otherwise. [program:__defaults__] redirect_stderr=true ; We force all programs to run as user "nobody" [program:__overrides__] user=nobody ; Disable auto-reloading on code changes by excluding that program. [program:autoreload] exclude=true
The supervisord and supervisorctl programs interact with each other via an XML-RPC control socket. This provides a great deal of flexibility and control over security, but you have to configure it just so or things won't work.
For convenience during development, django-supervisor provides automatic control socket configuration. By default it binds the server to localhost on a fixed-but-randomish port, and sets up a username and password based on settings.SECRET_KEY.
For production deployment, you might like to reconfigure this by setting up the [inet_http_server] or [unix_http_server] sections. Django-supervisor will honour any such settings you provide.
When running in debug mode, django-supervisor automatically defines a process named "autoreload". This is very similar to the auto-reloading feature of the Django development server, but works across all configured processes. For example, this will let you automatically restart both the dev server and celeryd whenever your code changes.
To prevent an individual program from being auto-reloaded, set its "autoreload" option to false:
[program:non-python-related] autoreload=false
To switch off the autoreload process entirely, you can pass the --noreload option to supervisor or just exclude it in your project config file like so:
[program:autoreload] exclude=true
Optionally, the file patterns on which autoreload listens for changes can be set in your project's settings.py:
SUPERVISOR_AUTORELOAD_PATTERNS = [".py", ".pyc", ".pyo"] SUPERVISOR_AUTORELOAD_IGNORE_PATTERNS = [".", "#*", "*~"]
There aren't any more docs online yet. Sorry. I'm working on a little tutorial and some examples, but I need to actually use the project a little more first to make sure it all fits together the way I want...