Marc Cousin wrote:
> all the databases of the cluster are regularly vacuumed (at least once a day),
> all the stats are up to date.
If stats say 0 ("estimated rows") rows but 40M rows are present stats
are clearly not up-to-date. We had interpretation problems of
pg_class.reltuples because it was read as int, not as float, but that
was fare earlier than 1.4.
If SELECT reltuples FROM pg_class WHERE relname='<foo>' returns
non-zero, but estimated rows is 0, your platform's strtod might have a
locale problem, but I doubt that because from my observations pgsql will
always return [1-9].[0-9](n)e[1-9](n), i.e. if the decimal point was the
problem est. rowcount would be between 1 and 9.
Regards,
Andreas