On Sat, Jan 25, 2014 at 1:11 PM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
> I've chosen to handle failures to load query
> text data by just returning NULL for that query text, which seems
> reasonable given the new mindset that the query text is auxiliary data
> less important than the actual counts.
I guess that's okay.
> I've not done much more than smoke-testing on this; I'm thinking about
> hot-wiring the code to sometimes force it through the error paths,
> just to make sure they act as expected. I've also not done any
> real performance testing.
I was never too worried about the impact on performance, because any
price that must be paid is paid only when new entries are created,
which is typically a very rare event that was already expensive due to
requiring an exclusive lock. This patch will make it much rarer in
practice by increasing the pg_stat_statements.max default, and thus
reducing the impact on such exclusive locks on *all* sessions, not
just those executing less popular queries. I don't know about you, but
the reason that I didn't performance test this is that it's really
hard to think of a representative scenario where it could possibly
matter, even though it seems like there might be one.
--
Peter Geoghegan