On Tue, Oct 11, 2022 at 9:27 AM Alvaro Herrera <alvherre@xxxxxxxxxxxxxx> wrote: > I remember having an hypothesis, upon getting a report of this exact > problem on a customer system once, that it could be due to killtuple not > propagating to standbys except by FPIs. I do not remember if we proved > that true or not. I do not remember observing that tables were being > read, however. That's true, but it doesn't matter whether or not there are LP_DEAD bits set on the standby, since in any case they cannot be trusted when in Hot Standby mode. IndexScanDescData.ignore_killed_tuples will be set to false on the standby. -- Peter Geoghegan