Redis keyspace analysis: alerting

From #320 (closed):

Once we have several weeks worth of data from #361 (closed) and #362 (closed), it should be inspected to see if there are consistencies that might be amenable to alerting. It's nearly impossible to predict what this might be; we'll only know when we have the data.

If we can correlate any inconsistencies (spikes, drops, weirdness) with an outages, incidents, or degradations, then they should be prime targets for alerting, particularly if they precede the problem by any significant amount of time.

It is also acceptable if, after review, we cannot find any useful alerting. This is a followup step only, and the data itself has merit without alerting.