On Mon, Mar 2, 2020 at 2:40 PM PG Bug reporting form
<noreply@postgresql.org> wrote:
> Looking at the sources of both extensions, I can see, that pgstatindex() is
> using psprintf(INT64_FORMAT) for page counters and psprintf("%u") for root
> page, while bt_metap() is using only psprintf("%d");
>
> I assume psprintf("%u") should be used at least for metad->btm_root and
> metad->btm_fastroot in the bt_metap(PG_FUNCTION_ARGS) function.
While this has been wrong forever, the immediate problem here is
probably not metad->btm_root. In practice, the root block number is
usually fairly low. Very large indexes tend to have a root page that
only has a small number of tuples, because the root page was created
relatively early in the lifetime of the index. Past a certain point,
the root page receives new tuples so infrequently that it almost never
happens.
I think it's more likely that the problem here is the relatively new
column returned by bt_metap(), oldest_xact. That has only been around
since Postgres v11.
I'm not sure how we should handle this in the backbranches, since only
a change in the CREATE FUNCTION declaration of bt_metap() can truly
fix the problem. I suppose that we could work around the problem with
some kind of kludge, but come up with a real fix for v13.
--
Peter Geoghegan
