On Thu, May 10, 2012 at 3:42 PM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
Alban Hertroys <haramrae@gmail.com> writes: > On 10 May 2012 15:05, Radosław Smogura <rsmogura@softperience.eu> wrote: >> May I ask what kind of planning may occur during insert?
> Well, for example, if there's a unique constraint on the table then > the database will have to check that the newly inserted values don't > conflict with values that are already in the table. It needs to plan > an efficient strategy for that, which depends on the values being > inserted.
There is no planning associated with checking unique constraints; that's just a matter for the index mechanisms.
I think the real point here is that a simple INSERT/VALUES has such a trivial plan that there is hardly any gain to be had by avoiding the planning stage. Then the other overhead of a prepared statement (looking up the saved plan, checking it's not stale, etc) outweighs that. Or at least it could. 3x slower seems a bit fishy; I wonder whether there's some client-side inefficiency involved in that. Doing performance measurements with pgAdmin seems pretty questionable in the first place ...
