[pcp] Derived metrics vs metrics with instances

[Paul Smith]

> 
> 
> No, this is something for RFE-land (certainly doable if the need is
> sufficient) ... you could define
> 
> my.new.metric = network.interface.in.bytes + network.interface.out.bytes
> 
> and then pick the "eth1" instance of my.new.metric (the indom of the
> derived metric is propagated from the operands).
> 

yeah, except that's not what I need right now.  We have instance domains for each 'controller', so we have:

aconex.controllers.count[mel/registerDocumentServiceController]
aconex.controllers.count[mel/correspondenceServiceController]
…..

many many more, and we have an existing computed metric for a 'total' controller that sums up all controllers:

aconex.controllers.count[mel/total]

Now, there are some sub-grouping within all the controllers, there are a small handful, only 4 that are our public API controllers. I would like to have something like

aconex.controllers.count[mel/apis] = aconex.controllers.count[mel/total]-aconex.controllers.count[mel/registerDocumentServiceController]-aconex.controllers.count[mel/correspondenceServiceController] - …<other API specific controllers>

aconex.controllers.count[mel/nonapis]=aconex.controllers.count[mel/total]-aconex.controllers.count[mel/apis]

this would give me the overall ratio of controller rates of APIs vs the non-apis (main actual user traffic).

Paul