Bertrand Roos <bertrand.roos@areal.fr> writes:
> In fact, I was telling the question because I have read on some forums
> that the auto vacuum deamon only count dead tuple so only update and
> delete operations can cause the scheduling of auto-analyse.
That might have been true years ago, don't remember at the moment.
But in modern PG versions auto-vacuum is driven by the estimated number
of dead tuples while auto-analyze is driven by the total count of
inserts+updates+deletes. It's easy to show in a standalone experiment
that auto-analyze *will* run against a table that's only had inserts.
With the numbers you're showing, auto-analyze should trigger once the
table gets to 20% new tuples. It would be interesting to see the
pg_stat_all_tables values for one of your problematic tables.
regards, tom lane