Possible lock held too long issue in connection pool being threads +1

[morgo]

There is a possible race in acquiring locks the connection pool to threads +1:

• The checksum acquires a LOCKS TABLES .. READ to start the RR connections for checksumming (1 connection)
• The checkpoint thread starts, to update the checkpoint table (1 connection)
• The checksum RR threads are opened (N threads connections; the last one waits on the checkpoint before it can start).
• The checkpoint thread finishes.

I don't think this is much more than a very brief stall, but from a design perspective we should be acquiring and releasing locks as soon as possible without contention. So we may want to consider either:

• A separate pool that is unrestricted for admin commands.
• Using threads +2 for the pool size.

I'm yet to prove this in production. It was just a thought while inspecting the code.