Kevin Grittner a écrit :
>> And each morning ... all data is INSERTed new
>
> I recommend VACUUM ANALYZE of the table(s) after this step. Without
> that, the first query to read each tuple sets its hint bits and
> rewrites it, causing a surprising delay at unpredictable times
> (although heavier near the start of the day).
Ok great, thanks for the advice, I added it at the end of the process...
Denis