Plugin package for supervisor that can perform fast restarts of supervisor process groups via an xmlrpc call

Due to inherent architecture design python-supervisor waits (currently, unpactched) one second after restarting a process. While this can be patched to a lower time value it represents a scalability issue given sufficient processes in a process group.

supervisor-restarter provides an xmlrpc call which can rapidly restart a process group by adding a new module to the supervisor.plugins namespace. Note that while supervisor is already a namespace, plugins will be added as such via the setuptools pkg_resources mechanism.

supervisor-restarter should be compatible with Python 2.4 through 2.7. It is untested (and probably incompatible) with Python 3 and higher.

To configure the restarter XML RPC interface you will need a main config section or drop file for supervisor containing the following:

[rpcinterface:restarter]
supervisor.rpcinterface_factory = supervisor.plugins.restarter:make_rpcinterface
;delay = 0.1
;timeout = 5.0
;stagger_factor = 1

The delay option can be used to configure the delay, in seconds, between internal "callback" iterations of the restartProcessGroup rpc call. A default of 0.2 seconds is used if none is otherwise configured.

The timeout option will configure the maximum amount of time, in seconds, that the xmlrpc method restartProcessGroup will be allowed to run; although "run" is slightly deceptive because, as indicated above, the underlying method is actually non-blocking and is called every delay seconds until it indicates the operation is complete. timeout simply bounds this to a maximum time. Once this time limit has been reached the xmlrpc call will return a Fault or a list containing faults (as dictionaries) if prior errors have accumulated. The timeout is not exact as it can only be checked every delay seconds. A default timeout of 5 seconds is used if not otherwise configured.

The stagger_factor option configures how many iterations (call of the internal handler every delay seconds) it takes to stop all processes. This is useful because the restarter will first try to start any processes that have completed stopping before it goes on to stop the next set, thus staggering can cause less overall disruption by allowing new processes to become available while the old ones are still being killed off. The default value of 1 indicates no staggering will be used.

$ python
Python 2.7.1 (...)
>>> import xmlrpclib
>>> server = xmlrpclib.Server('http://localhost:9001/RPC2')
>>> server.restarter.restartProcessGroup('someconfiguredgroup')
True

Eventually the plan is to extend supervisorctl in a similar fashion. However, for the moment, it's not possible to extend the ctl interface via a supervisor.d/ *.conf file, only by editing the main config. Thus, the following tool is now included in this package:

Restarts the process group passed on the command line. Takes all the same arguments as supervisorctl.

Example:

supervisorctl_restart_group -s http://myserver foobar

Restarts the foobar process group on myserver.