On Tue, Apr 16, 2019 at 12:00 PM Peter Geoghegan <p...@bowt.ie> wrote: > Can you be more specific? What was the cause of the corruption? I'm > always very interested in hearing about cases that amcheck could have > detected, but didn't.
FWIW, v4 indexes in Postgres 12 will support the new "rootdescend" verification option, which isn't lossy, and would certainly have detected your customer issue in practice. Admittedly the new check is quite expensive, even compared to the other bt_index_parent_check() checks, but it is nice that we now have a verification option that is *extremely* thorough, and uses _bt_search() directly. -- Peter Geoghegan