multiple counting of pod memory

[rptaylor]

I did a bit of testing and confirmed that the self.memory query multi-counts pod memory: https://github.com/rptaylor/kuantifier/blob/master/python/KAPEL.py#L60

Each time the KSM pod restarts, or for a deployment of > 1 KSM pods for redundancy, (at least in our configuration) you get a new instance label, causing the reported memory to increase by 100% for each.

[rptaylor]

This shows the reported memory is 3x too high because the KSM pod was restarted twice, so the same metric is reported (and summed) 3 times.

[rptaylor]

Changing sum by (pod, uid) to max by (pod, uid) should fix it I think but that might adversely affect the ability to account multi-container pods.

[rptaylor]

FYI @mwestphall something to keep in mind for later.