Re: Extreme bloating of intarray GiST indexes

Josh Berkus <josh@agliodbs.com> writes:
> So, some data:

> corp=# select * from public.pgstattuple('idx__listings_features');
> -[ RECORD 1 ]------+---------
> table_len          | 54190080
> tuple_count        | 7786
> tuple_len          | 2117792
> tuple_percent      | 3.91
> dead_tuple_count   | 0
> dead_tuple_len     | 0
> dead_tuple_percent | 0
> free_space         | 49297536
> free_percent       | 90.97
>                      ^^^^^^^^^
> Well, that explains the bloating.  Why all that free space, though?

> Maybe autovac isn't running?

No, because you have under 10% dead tuples in the main table.
I think this is sufficient proof of the crummy-page-splits theory.
Can you provide the data in the column that's indexed?
        regards, tom lane