There is already a ThreadPool class, but it support only the scheduling of Future objects without a few features that Query Parallelism would rely on:
- Renameable (so that logs triggered by a parallel stream for the thread bound to conn56 are printed as conn56-1, conn56-2, etc)
- Server parameter to control maximum number (given the 127 tickets granted to Query by the WT engine, default size would be 64)
- Single acquire() API that returns N threads in one shot, we don't want to schedule work that would be not served immediately by
- not enough idle threads in the pool or
- not enough tickets in the ticket granting pool
- FTDC metric to report current number, in use, idle