El día martes, mayo 24, 2022 a las 12:11:49 -0400, Tom Lane escribió: > Laurenz Albe <laurenz.albe@xxxxxxxxxxx> writes: > > It may well be that somebody deleted or updated a few rows between the time > > the cursor was materialized and the time the 50000th row was fetched. > > Even without HOLD, a cursor will return a view of the data as it stood > when the cursor was opened, just as a plain SELECT does. There is > *plenty* of time for another session to get in there if you've been > groveling through 50K records one at a time. Tom, Thanks for pointing us in the right direction where to look for a solution. The CURSOR was opened around 23:11 pm and the CTID not found at 23:21 pm, i.e. ten minutes later. This piece of software does every night some housekeeping work in the circulation area of our LMS (Library Management System) and is meant to run as a standalone job (only one process after the other). We're trying to figure out with the customer if something else was started/running at this time between 23:11 and 23:21, to shut this off in the future. Is it possible that the PostgreSQL 13.1 server does something by its own to invalidate the rowid? matthias -- Matthias Apitz, ✉ guru@xxxxxxxxxxx, http://www.unixarea.de/ +49-176-38902045 Public GnuPG key: http://www.unixarea.de/key.pub
