Frank has used his psychic powers to pretty much pre-empt my answer.
This twists the data semantics more than the âderived metricsâ support is willing to contemplate.
The âinstancesâ youâve added would need instance domain support (enumeration, internal index to external name mapping and vice versa, an indom number) and it would only work for component metrics that have the same dimension (in the pmUnits sense) and probably only then for metrics that are scalar (have no associated instance domain).
But with these restrictions, it would be technically feasible to define a set-value metric (as a derived metric) where the elements of the set are an ordered enumeration of the values of the metrics in the set). Unfortunately there is a lot of code, and little available effort, between âtechnically feasibleâ and âavailable in PCPâ.
From: Martin Spier [mailto:mspier@xxxxxxxxxxx]
Sent: Saturday, 26 July 2014 4:26 AM
To: Ken McDonell
Cc: pcp@xxxxxxxxxxx
Subject: Re: [pcp] Combined metrics with PCP_DERIVED_CONFIG
Sure. I'm building an UI that will consume PCP's API and besides other things, plot the live data. One of the charts I need to draw is composed of user and sys CPU. As of now, I need to fetch two metrics, kernel.all.cpu.sys and kernel.all.cpu.user, each returning a separate data structure and combine them. Eg.: