Re: speed up a logical replica setup

[Thomas Munro]

040_pg_createsubscriber.pl seems to be failing occasionally on
culicidae near "--dry-run on node S".  I couldn't immediately see why.
That animal is using EXEC_BACKEND and I also saw a one-off failure a
bit like that on my own local Linux + EXEC_BACKEND test run
(sorry I didn't keep the details around).  Coincidence?