Trickster style query results caching in query-frontend #977
Description
Repeated large queries are can quickly "gum up" a Cortex cluster, as we limit the number of concurrent queries. The idea here is to cache query results and when the same query comes in with an overlapping time range, only query for the delta and merge the results.
We probably want to keep the cache externally, as the results can be long lived.
Challenges are start/end and step alignment - Grafana 5.2(?) does this now, but I suspect a bunch of users aren't running an up to date version yet. We could align start & end to the step in the frontend, but this runs into discussions about how we shouldn't mutate the query...
Finally, lets see if we can vendor / re use trickster code to achieve this - I've got no appetite for reimplementation.