This is related to a design choice of hadoop: your clients need to have a correct server configuration for job submission to work. This is not (yet) implemented from communication with the server. so the client config files need to be right.

This is very brittle, and forces us to build up the production components through inheritance, placing in all the various attributes that are needed to work.

If we could build our conf not just through inheritance from a parent (and nesting), but by referencing a Cluster CD/Prim, then we woudn't need to patch every template to deploy work to different customers. something like

SubmitJob {
 cluster LAZY HplbCluster;
 job "...";
}

This could be done with a tweaked sfResolve derivative (that also tried to resolve via the cluster ), and with the code to build a ManagedConfiguration to read in Cluster content (first)