Dicovery mechanism

[ahilloffis]

For our application, we want to dynamically connect different devices that are represented by mango agents.
We tested a discovery mechanism via MQTT Container, which works quite well but an internal solution similar to something like Simple Service Discovery Protocol or similar for MQTT communication would be quite nice.

[rcschrg]

Just so you know, we currently have no plans to implement any version of agent discovery, but we generally think it can be a valuable feature for mango-python. @ahilloffis, if you want to come up with some sort of agent discovery, go for it and create PR.

[maurerle]

@ahilloffis what is your current discovery mechanism?

In my distributed simulations using MQTT, I have one container which has the DistributedClockManager and each other container has a DistributedClockAgent running.

The actual simulation agents only require to know the market, at which they register (so kind of a "registry", where agents announce interest to participate and from then on receive relevant messages).
But due to the requirement to have a synchronized simulation clock between the containers, the DistributedClockManager should wait for all other agents before starting the simulation.
It therefore requires the list of DistributedClockAgent to wait for in wait_all_online()..

This is of course solved when simulating in realtime using the AsyncioClock, do you have a different approach?

[ahilloffis]

@maurerle We have currently implemented our version, where we send special device/agent announcement messages via MQTT. We leveraged the MQTTContainer.subscribe_for_agent method to subscribe to non-Mango queues we control (by sending messages to these queues via Mango messages).

In our development, we are currently still centralised for the simulation. When we distribute the simulation, we will probably also use the DistributedClockAgent or a role version. Otherwise, we may implement a version of a fully distributed clock agent, which doesn't require a DistributionClockManager