On Mon, Oct 11, 2021 at 11:12 AM Peter Geoghegan <p...@bowt.ie> wrote: > Sure, the user might not be happy with --parent-check throwing an > error on a replica. But in practice most users won't want to do that > anyway. Even on a primary it's usually not possible as a practical > matter, because the locking implications are *bad* -- it's just too > disruptive, for too little extra coverage. And so when --parent-check > fails on a replica, it really is very likely that the user should just > not do that. Which is easy: just remove --parent-check, and try again.
We should have a warning box about this in the pg_amcheck docs. Users should think carefully about ever using --parent-check, since it alone totally changes the locking requirements (actually --rootdescend will do that too, but only because that option also implies --parent-check). -- Peter Geoghegan