Joel asked this question - I figured it is good to post to h-dev, too, since I'm sure others don't know this. The "big picture" is: First, remember the MiQ UI is connected to Hawkular-Services server via websocket. The agents all connect to some Hawkular-Services server via a websocket. There can be more than one Hawkular-Services server running and the UI may be talking to a server that is different than the server the agent is talking to: UI <---websocket---> Server A ^ | (message bus) | V Server B <---websocket---> agent 1) MiQ sends a JSON message over the websocket to a Hawkular-Services server (Server A in the diagram above). 2) Hawkular-Services server A looks at the ResourcePath in the message to determine what agent is responsible for managing that resource. 3) Hawkular-Services server A addresses the message to the proper agent and puts the JSON message on the agent's queue on the message bus. 3b) At this point the Hawkular-Services server A sends back a JSON message to the UI over the websocket that says "message forwarded to the bus". 4) The Hawkular-Services server that has a websocket connection to the targeted agent picks off that message from the bus (Server B in the diagram) 5) That Hawkular-Services server B forwards the message to the agent via its websocket connection. 6) Agent looks the the message's ResourcePath and the rest of the JSON to determine what it needs to do. It does the action. // now the process goes in reverse 7) Agent sends back a response to the server B over websocket (either a success or fail message) 8) Server B takes response, figures out which UI the response should go to, and puts it on message bus addressed to the correct UI 9) The Server A takes the response message off the bus 10) Server A sends the response message to the UI over its websocket connection. In the case of a "Deployment" JSON message, yes, the agent will trigger a full discovery scan so it can quickly discover the new deployment you just added to WildFly. Most of the time, its just using ResourcePath API (just that simple inventory POJO). HOWEVER, I believe Jay added some code that does interact with Inventory to do things like create relationships in inventory when new cluster entities are added. That is why (I think) he did this commit. Though I can't remember where this code is that builds inventory relationships - he'll have to point that out. But I'm pretty sure there is some inventory stuff going on under the covers for that stuff - more than just using the canonical ResourcePath stuff.