> Mark, Stephen, etc:
>
>> > I can see your point, however I wonder if the issue is that the
>> default
>> > stats settings of '10' (3000 rows, 10 histogram buckets) is too low,
>> and
>> > maybe we should consider making a higher value (say '100') the
>> default.
>>
>> Personally, I think that'd be reasonable.
>
> I don't really think that we want to increase sampling for *all* columns.
> Each column that doesn't get queried but has increased stats adds the size
> of
> the pg_stats table and the amount of vacuuming it needs, decreasing
> performance of stats lookups. Especially when very large text and BYTEA
> columns are involved.
>
> Neil and I talked about setting a seperate GUC,
> default_index_stats_target,
> for 8.0 that would dramatically increase the stats sampling for indexed
> columns only (say, to 250). Unfortunately, I was unable to come up with
> a
> good test case for setting this value.
>
> Perhaps the Tiger database would allow us to set up a good test case.
> Does
> anyone have a well-indexed PG Tiger I could play with?
Hey, I can give you a copy of RT1 which is fine, but it is 1.1G
compressed. I'd have to mail you a DVD.
---------------------------(end of broadcast)---------------------------
TIP 9: the planner will ignore your desire to choose an index scan if your
joining column's datatypes do not match
