Understanding the Messaging with Rabbit MQ
First, it is highly recommended to read Rabbit MQ’s tutorial.
It explains the basics and the concepts of this messaging server.
Now, let’s talk about the way Roboconf uses RabbitMQ.
This schema illustrates what is discussed below.
This organization applies to a given Roboconf domain.
Every domain follow this architecture, but all the queues, exchanges and clients are isolated from other domains.
Clients and Queues
The DM and all the agents have their own client for RabbitMQ.
Each of them has also its own queue. Unlike in former versions, this is not contextual within an application. The DM has a single queue and a single exchange, used for all the applications. Agents are by definition tied to a given application.
The DM’s queue is called
Queue names are prefixed with the domain name.
The queue for an agent is named according to the scoped instance it is associated with.
Within a Roboconf application, an instance path is unique. When an agent starts, it is associated with a scoped instance. The name of an agent’s queue is thus…
Notice that we replace the
/ character by
_ in queue names.
There are 3 exchanges.
Exchanges are important because no client sends directly a message to a queue. Queues are hidden behind exchanges. Exchanges act as filters for messages. They will be routed (or dropped) according to routing rules (routing keys).
The first exchange is for the Deployment Manager. It is called
The second exchange is used by agents within an application:
The third and last one is used by agents for inter-application exchanges:
Exchange names are prefixed with the domain name.
Routing keys are routing rules that determine what is done of a message when it reaches an exchange.
Routing keys are deduced from messaging contexts (it can Roboconf variables names, agent names, etc). As a reminder, variables are what component instances export or import (the famous runtime dependencies).
Routing keys are used for the 3 exchanges.
This corresponds to the topic keyword in Rabbit MQ.
Routing keys do not use domain names.
Routing Keys Example
Let’s consider the following instances.
We assume they are already associated with an agent.
MyWebApplication is an instance of WebApplication.
It exports WebApplication.ip and WebApplication.port.
It imports MySql.ip and MySql.port.
MyDatabase is an instance of MySql.
It exports MySql.ip and MySql.port. And it imports nothing.
We deploy MyWebApplication.
The agent creates its queue and starts listening to it. It also registers a routing key called those.that.exports.MySql. When a message reaches the agent exchange, it will be redirected to the agent’s queue.
We start MyWebApplication.
It publishes the variables it exports to the agents exchange with the routing key those.exports.WebApplication. Those that listen it will receive them.
We deploy MyDatabase.
There is no import, so the agent has routing key to register.
We now start MyDatabase.
It publishes the variables it exports to the agent exchange with the routing key those.exports.MySql. Those that listen it, including the MyWebApplication instance, will receive them.
When we say an agent publishes, it sends a message to the agent exchange, with the right routing key those.exports.(…).
When we say an agent listens, it actually binds its queue with the agent exchange AND the right routing key those.exports.(…).
By creating bindings between queues and the agents’ exchange, we determine routes for messages.
And we are sure the messages only reach the clients that actually need this information.
This works the same way when we send messages to the DM and for inter-application messages.