Rudder

Devlopper note: there is twi hard problems,

1/ semantic to describe access to inventory values (it could be linked to #9299 if we actually model all inventory information has json path, and we will need to do that to let the user know what to query)
2/ performance and consistency

For this one, the problem is to load the correct "context" for the node in an atomic way (i.e in a way that the node context matches the directive for the same generation trigger).

To be clearer, today we have a very simple "atomic" way of doing that thing: at start of generation, we load all datas (rules, groupes, directives, parameters, and node properties) and we only rely on these loaded, immutable data for the remaining of the process.

Now, if we want to do the same with inventories, we are going to load FAR TO MUCH data, and the performance will be horrible.
So we need to somehow know before loading node information, for all node, the bounding context (aka the list of free variable for that node's applied directives), and only query for that (open question: node by node, or for all node ? Depends on node by node generation ?).

So we need to change our model, to have for each node the list of its free variable. But we can't have that today before quite far in the generation process, and we are bounding them with the node context in the same step (not clever). Quite a big change to forsee.

It generates a JSON

{
  "hostname":"server.rudder.local",
  "policyServerId":"root",
  "localAdministratorAccountName":"root",
  "archDescription":"x86_64",
  "ram":"1572864000",
  "timezone":"Europe/Paris",
  "os":{
    "name":"Debian",
    "fullName":"Debian GNU/Linux 9.1 (stretch)",
    "version":"9.1",
    "kernelVersion":"4.9.0-3-amd64",
    "servicePack":"" 
  },
  "machine":{
    "machineType":"VirtualMachineType(VirtualBox)",
    "manufacturer":"innotek GmbH" 
  }
}

in Rudder System Variable RUDDER_INVENTORY_VARS