Bug: Current algorithm for checking activity can miss some

[krancour]

Currently, the zeroscaler monitors activity only for deployments that are already scaled to their minimum and it does this on a periodic basis. For the sake of illustration, let's say imagine a one pod minimum that we check on every two minutes.

It's possible that a pod dies and is replaced with a new one in the two minutes between checks. This means whatever new activity might have occurred on the first pod isn't accounted for. In a scenario such as this, we should be assuming there was activity on the pod that died to avoid a premature scale-to-zero.

A similar case involves a pod that dies in the two minutes between checks and is replaced by a new pod that also dies in the same interval, and is replaced with a third pod. When the next check occurs, only the third pod will be checked for new metrics. Our current algorithm wouldn't even be aware of the short-lived pod that still might have had some activity.

We already have an open issue for improving our metrics-gathering approach (#17) in support of solutions to other issues (#15 and #16), so this issue isn't its own action item so much as it supplements #17 by documenting a specific deficiency with the current approach.

[krancour]

This was fixed in #27.