Messages


Information in a simulation can propagate in one of two ways - either through neighbors, which we cover in Behaviors and Topology, or through message passing.

Messages are a simple yet powerful way of moving data around a simulation and can be used to add, remove, and collect data from agents that are not necessarily near each other.

Agents send messages by adding them to the state.messages array, and they receive them by accessing context.messages().

Agents can create a message to send to either another agent or the simulation engine itself. Here, we send a message to an agent with an agent_name of "schelling":

JavaScript
Python
const behavior = (state, context) => {
  state.messages.push({
    to: "schelling",
    type: "data_point",
    data: {
      num_agents: 50,
    },
  });
};

If you want to jump right into code you can take a look at our Message Passing Colors simulation, which demos message passing among many agents.

You'll notice that each message is comprised of three fields: "to", "type", and "data."

  • to: the agent_name or agent_id of the agent that the message will be delivered to (this can also be an array of agents)
  • type: the type of message being sent, to help the message handler select the correct course of action
  • data: any data you want to send along with the message

You can use the helper function state.addMessage(to<String>, type<String>, data<Dict>) to add messages directly to the state.messages field (the helper function can only take one agent as a "to" argument).

JavaScript
Python
const behavior = (state, context) => {
  state.addMessage("foo", "bar", { msg: "hello" });
};

Messages are produced during a step, but are not delivered and processed until the next step.

Data flow for a single simulation step in HASH

Previous

Join our community of HASH developers