Hi,
On 2023-02-14 16:04:16 +0900, Michael Paquier wrote:
> On Mon, Feb 13, 2023 at 09:45:12AM -0800, Andres Freund wrote:
> > Shouldn't there at least be some basic verification of pg_stat_statements
> > output being sane after running the test? Even if that's perhaps just actually
> > printing the statements.
>
> There is a total of 20k entries in pg_stat_statements if the max is
> high enough to store everything. Only dumping the first 100
> characters of each query generates at least 1MB worth of logs, which
> would bloat a lot of the buildfarm in each run. So I would not do
> that. One thing may be perhaps to show a count of the queries in five
> categories: select, insert, delete, update and the rest?
I didn't mean printing in the sense of outputting the statements to the tap
log. Maybe creating a temp table or such for all the queries. And yes, then
doing some top-level analysis on it like you describe sounds like a good idea.
Greetings,
Andres Freund