Marko Tiikkaja <marko(at)joh(dot)to> writes:
> The failure is a number of levels down a call stack of PL/PgSQL
> functions, but I can reproduce it at will by calling the function. I
> unfortunately can't narrow it down to a smaller test case, but attached
> is an xlogdump of the session. The query that finally breaks the
> elephant's back is a SELECT .. FOR UPDATE from relid=54511.
> Any ideas on how to debug this?
If this test case used to pass, it'd be pretty useful to git bisect
to find where it started to fail.
regards, tom lane