Re: query cache

Robert Haas <robertmhaas@gmail.com> writes:
> Interestingly, Peter Geoghegan's blog post on the pg_stat_statements
> patch you just committed[1] claims that the overhead of fingerprinting
> queries was only 1-2.5%, which is less than I would have thought, so
> if we ever get to the point where we're fairly sure we've got problem
> three licked, it might make sense to revisit this due to problems one
> and two.

Maybe.

> It's also probably worth keeping in mind the next time we
> bump the protocol version: it would be nice to have a way of doing
> prepare-bind-execute in a single protocol message, which I believe to
> be not possible at present.

Huh?  That's the standard way of doing it, actually.  You send
Prepare/Bind/Execute/Sync in one packet, and wait for results.

This requires that you understand the query well enough to perform
Bind without seeing a Describe result, but that seems essential
to any one-round-trip case anyway.  It's not the protocol design
that is holding back anybody who wants to do that.
        regards, tom lane