Horton

Re: 64-bit queryId?

Robert Haas <robertmhaas@gmail.com>

View thread View on postgr.es
From: Robert Haas <robertmhaas@gmail.com>
To: Michael Paquier <michael.paquier@gmail.com>
Cc: Alexander Korotkov <a.korotkov@postgrespro.ru>, Tom Lane <tgl@sss.pgh.pa.us>, Peter Geoghegan <pg@bowt.ie>, "Joshua D. Drake" <jd@commandprompt.com>, Greg Stark <stark@mit.edu>, "pgsql-hackers@postgresql.org" <pgsql-hackers@postgresql.org>
Date: 2017-10-04T14:04:11Z
Lists: pgsql-hackers 
On Wed, Oct 4, 2017 at 9:49 AM, Michael Paquier
<michael.paquier@gmail.com> wrote:
> I am still on the learning curve with pg_stat_statements... This still
> does not look complete to me. pgss_hash_fn only makes use of the last
> four bytes of the query ID. What about computing the hash using as
> also the first four bytes? With the current code, if the last four
> bytes of two queries match then they would be counted together looking
> at pgss_store().

Not really; dynahash won't merge two keys just because their hash
codes come out the same.  But you're right; that's probably not the
best way to do it.   TBH, why do we even have pgss_hash_fn?  It seems
like using tag_hash would be superior.

> I have spotted as well this comment in pg_stat_statements.c:
>     /* Increment the counts, except when jstate is not NULL */
>     if (!jstate)
> I think that this should be "when jstate is NULL".

I think that you're right, but that's unrelated to this patch.

-- 
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company

Commits

  1. pg_stat_statements: Add a comment about the dangers of padding bytes.

  2. pg_stat_statements: Widen query IDs from 32 bits to 64 bits.