On 02-10-2012 10:15, Peter Geoghegan wrote:
> There are other, similar tools that exist in proprietary databases.
> They expose a hash value, which is subject to exactly the same caveats
> as our own. They explicitly encourage the type of aggregation by
> third-party tools that I anticipate will happen with
> pg_stat_statements. I want to facilitate this, but I'm confident that
> the use of (dbid, userid, querytext) as a "candidate key" by these
> tools is going to cause confusion, in subtle ways. By using the hash
> value, those tools are exactly as well off as pg_stat_statements is,
> which is to say, as well off as possible.
>
It depends on how you will use this information. If it is for a rapid
analysis, it is fine to use a hash because the object internals won't change
(I hope not). However, if you want to analyze query statistics for a long
period of time (say a few months) or your environment is distributed, you
can't use the hash. There isn't a perfect solution but I see both cases being
useful for analysis tools.
-- Euler Taveira de Oliveira - Timbira http://www.timbira.com.br/ PostgreSQL: Consultoria, Desenvolvimento,
Suporte24x7 e Treinamento
