This package is an RPC extension for Supervisor that allows Supervisor's configuration and state to be manipulated in ways that are not normally possible at runtime.
supervisor_twiddler packages are
available on PyPI.
You download them from there or you can use pip
to
automatically install or upgrade:
$ pip install -U supervisor_twiddler
After installing the package, add these lines to your supervisord.conf
file
to register the twiddler interface:
[rpcinterface:twiddler]
supervisor.rpcinterface_factory = supervisor_twiddler.rpcinterface:make_twiddler_rpcinterface
You must restart Supervisor for the twiddler interface to be loaded.
There are times when it is useful to be able to dynamically add and remove process configurations on a supervisord instance. This is the functionality that supervisor_twiddler provides. After restarting supervisord, the changes made by supervisor_twiddler will not persist.
The following Python interpreter session demonstrates the usage.
First, a ServerProxy object must be configured. If supervisord is listening on an inet socket, ServerProxy configuration is simple:
>>> import xmlrpclib
>>> s = xmlrpclib.ServerProxy('http://localhost:9001')
If supervisord is listening on a domain socket, ServerProxy can be configured with SupervisorTransport. The URL must still be supplied and be a valid HTTP URL to appease ServerProxy, but it is superfluous.
>>> import xmlrpclib
>>> from supervisor.xmlrpc import SupervisorTransport
>>> s = xmlrpclib.ServerProxy('http://127.0.0.1/whatever',
... SupervisorTransport('', '', 'unix:///path/to/supervisor.sock'))
Once ServerProxy has been configured appropriately, we can now exercise supervisor_twiddler:
>>> s.twiddler.getAPIVersion()
'0.1'
>>> s.twiddler.addProgramToGroup('group_name', 'ls', {'command':'ls -l',
... 'autostart':'false', 'autorestart':'false', 'startsecs':'0'})
True
>>> s.supervisor.startProcess('group_name:ls')
True
>>> s.supervisor.readProcessLog('group_name:ls', 0, 50)
'total 0
drwxr-xr-x 9 mnaberez mnaberez 306 Nov'
In the session above, a new program called ls was added to the existing group called group_name. You can create empty groups in supervisord.conf by adding an empty [group:x] section.
The process was configured to not start automatically (autostart), not restart automatically (autorestart), and startsecs was set to zero so Supervisor would not think its quick termination was an error.
The process was then started and its output read using the normal API commands provided by Supervisor.
All RPC extensions for Supervisor follow a convention where a method called getAPIVersion() is available. supervisor_twiddler provides this:
twiddler.getAPIVersion()
It is highly recommended that when you develop software that uses supervisor_twiddler, you test the API version before making method calls.
Process groups are defined in supervisord.conf as group sections. Assume supervisord.conf contained sections [group:foo] and [group:bar]:
twiddler.getGroupNames()
The return value would then return an array: ["foo", "bar"]. It is possible to use supervisor_twiddler to add new process groups at runtime, and these will also be included in the results returned by twiddler.getGroupNames().
In supervisord.conf, a [program:x] section will result in one or more processes, depending on numprocs and named by process_name.
The twiddler.addProgramToGroup() method makes it possible to add a new program to a group (resulting in one or more processes) and then control these processes as if they had existed originally in supervisord.conf.
twiddler.addProgramToGroup("group_name", "foo",
{"command": "/usr/bin/foo"})
The first parameter (group_name) is the group name where the new process will belong. While there is no restriction on what groups can be used, it is recommended that you keep your supervisord.conf groups static. You can add new process groups just for your dynamic processes, and this will help you track them easier.
The second parameter (foo) is the name of the new program to add to the group, as it would have been written in the [program:foo] section supervisord.conf.
The final parameter is a dict (XML-RPC "struct") containing the program options. These are the same options as in the supervisor.conf program section and follow the same rules. The only required key is command.
When you add a program in this way and do not specify the autostart option, the process will start on the next transition of Supervisor's state machine (almost immediately). You might want to set autostart to false and then start the process with supervisor.startProcess().
Similarly, you might want to set autorestart to false if you don't want Supervisor to restart it immediately after it exits.
If the process you are adding exits quickly, make sure that you set startsecs to 0. Otherwise, Supervisor will think the process failed to start and will give an abnormal termination error.
When processes are no longer needed in the supervisord runtime configuration, the twiddler.removeProcessFromGroup() method can be used:
twiddler.removeProcessFromGroup("group_name", "process_name")
To be removed, the process must not be running. It must have terminated on its own or have been stopped with supervisor.stopProcess().
The twiddler.log() method allows you to write arbitrary messages to Supervisor's main log. When you twiddle with Supervisor's configuration, this method is useful for logging messages about what was done.
twiddler.log("This is an informational message", "INFO")
The first argument is a string message to write to the log. The second argument is the log level and is optional (defaults to INFO). The log level may be a string or an integer.
Log levels are defined in the supervisor.loggers module and at the time of writing are: CRIT (50), ERRO (40), WARN (30), INFO (20), DEBG (10), TRAC (5), and BLAT (3).
Any changes to the supervisord runtime configuration will not be persisted after Supervisor is shut down.
Your Supervisor instance should never be exposed to the outside world. With supervisor_twiddler, anyone with access to the API has the ability to run arbitrary commands on the server.