For each query, we should record some basic information: the winning plan chosen, the optimal plan, if there are plan flips, the variance in latency, and how random sampling (vs chunked sampling) impacts the latency and plan chosen.

At the end of this investigation, we should have an idea of whether or not we are choosing near-optimal join orders (and algorithms) for all of these queries and, if not, where we should focus our improvements.

The random sampling investigation is to understand SERVER-118980. We should also understand the interaction with SERVER-117085. Perhaps after that ticket, we don’t need random sampling for TPC-H – for now – since we are able to make use of unique fields. See also SERVER-119639.