Re: compute_query_id and pg_stat_statements

On Fri, May 14, 2021 at 12:26:23AM -0400, Tom Lane wrote:
> I then tried a really dumb xor'ing implementation, and
> that gives me a slowdown of 2.2%.  This could undoubtedly
> be improved on further, say by unrolling the loop or
> processing multiple bytes at once.  One problem with it
> is that I suspect it will tend to concentrate the entropy
> into the third/fourth and seventh/eighth bytes of the
> accumulator, since so many of the fields being jumbled
> are 4-byte or 8-byte fields with most of the entropy in
> their low-order bits.  Probably that could be improved
> with a bit more thought -- say, an extra bump of the
> nextbyte pointer after each field.
> 
> Anyway, I think that what we have here is quite an inferior
> implementation, and we can do better with some more thought.

Considering what even a simple query has to do, I am still baffled why
such a computation takes ~2%, though it obviously does since you have
confirmed that figure.

-- 
  Bruce Momjian          https://momjian.us
  EDB                                      https://enterprisedb.com

  If only the physical world exists, free will is an illusion.