Understanding service internals¶
Registering synchronous services¶
What happens when a method is decorated with @service
?
The @service
decorator simply marks the method as a service by
setting adding an attribute called _morse_service
of type bool
and with value True
to the method.
Before actually registering the service, a mapping between the component
and one or several middlewares that will manage incoming requests must
be defined (for instance, we may want the
morse.actuators.waypoint.Waypoint
’s goto
service to be
managed by the YARP middleware for the component MainRobot
).
These mappings are defined by your builder script (so they are simulation-dependent).
At start up, morse.blender.main.init()
does the following
- Reads the configuration;
- Instantiates classes associated with each component;
- Registers the mappings (with
morse.core.services.MorseServices.register_request_manager_mapping()
); - Calls
morse.core.abstractobject.AbstractObject.register_services()
on each component instance.
The morse.core.abstractobject.AbstractObject.register_services()
method iterates over
every method marked as a MORSE service within the class, and calls
morse.core.services.do_service_registration()
on it.
This method finds the middlewares responsible for managing this
component’s services, and calls
morse.core.request_manager.RequestManager.register_service()
.
It is up to each middleware to manage the registration of new services. They may have to, for instance, expose the new service into a shared directory (yellow pages), etc.
Upon incoming request…¶
When a new request comes in, the middleware-specific part receives it,
deserializes it calls
morse.core.request_manager.RequestManager.on_incoming_request()
with it.
This method dispatches the request to the correct component, and executes
it (for synchronous services) or starts the execution and returns an
internal request ID (for asynchronous services).
This internal request ID can be used by the middleware to track the status of a request.
On completion, the
morse.core.request_manager.RequestManager.on_service_completion()
callback is invoked, and the final result can be sent back to the
client.
Asynchronous services¶
Registration of asynchronous services is almost the same as for synchronous
services. The @async_service
decorator simply calls the @service
decorator with the asynchronous
parameter set to True
, which results in
the original method being wrapped in a new method that takes an extra parameter
(a callback) and calls
morse.core.abstractobject.AbstractObject.set_service_callback()
.
Simplified version of the @service
decorator:
def service(fn, asynchronous=False):
dfn = fn
if asynchonous:
def decorated_fn(self, callback, *param):
self._set_service_callback(callback)
fn(self, *param)
dfn = decorated_fn
dfn.__name__ = fn.__name__
dfn._morse_service = True
dfn._morse_service_is_async = async
return dfn
However, asynchronous services behaviour differs when a request comes in:
- The
morse.core.request_manager.RequestManager.on_incoming_request()
method creates a new callback function for this service; - It invokes the original method with this callback.
- When
morse.core.abstractobject.AbstractObject.completed()
is invoked (i.e., when the service is completed), the callback is executed. - This in turn causes the
morse.core.request_manager.RequestManager.on_service_completion()
method to be invoked, to notify middleware-specific request managers that the task is complete.
Manually registering services¶
While usually unnecessary, it is possible to to manually register services (i.e., without using decorators).
This first code snippet shows how to register a synchronous service that uses sockets as the communication interface:
from morse.middleware.socket_request_manager import SocketRequestManager
def add(a, b):
return a + b
req_manager = SocketRequestManager()
req_manager.register_service("test_component", add)
while True:
# This calls the middleware specific part, responsible for reading
# incoming requests and writing back pending results.
req_manager.process()
# In a real case, you don't want to block on such a loop, and MORSE
# calls req_manager.process() on your behalf
If you run this sample code, you can test it with a simple Telnet session:
> telnet localhost 4000
Connected to localhost.
> req1 test_component add (1,2)
req1 OK 3
Note
The socket-based protocol is fairly simple: you provide a request id, the name of the component that offers the service, the name of the service, and (optionally) parameters. Parameters must be contained in a valid Python iterable (a tuple, like in the example, or an array).
Here, req1
is the custom request id, chosen by the client.
For asynchronous services, a callback function is passed to the service. It allows the service to notify when it is complete.
This second code snippet shows an example of asynchronous service:
import types
from morse.core import status
from morse.middleware.socket_request_manager import SocketRequestManager
State = types.SimpleNamespace()
State.result_cb = None
State.run_computation = False
State.value = None
# Here is where we start our asynchronous service.
# Arbitrary parameters may be passed, but the first one
# must be the callback to set the service result upon completion.
def start_computation(result_setter, start_value):
State.result_cb = result_setter
State.value = start_value
State.run_computation = True
# the service must return true if the task was successfully started
return True
# This is the actual code called at each simulation step in our component
def complex_computation(a):
if a == 0:
# At the end of the computation, we set the result.
# the result can be any valid Python object
State.result_cb((status.SUCCESS, "done!"))
State.run_computation = False
morse.sleep(1)
return a - 1
req_manager = SocketRequestManager()
# here we explicitely register an asynchronous service.
# the optional 'service_name' argument allows to define a custom service
# name.
req_manager.register_async_service("test_component", start_computation, service_name = "compute")
while True:
req_manager.process()
if State.run_computation:
State.value = complex_computation(State.value)
print("Value is now %i" % State.value)
If you test the code with Telnet:
> telnet localhost 4000
Connected to localhost.
> req2 test_component compute (5,)
[after 5 seconds]
req2 OK done!
Note
When passing a single parameter, you still need to pass a valid Python iterable,
even if it has only one element. Hence the (5,)
.