Re: Postgresql 'eats' all mi data partition

Tomas Szepe wrote:
>>[tgl(at)sss(dot)pgh(dot)pa(dot)us]
>>
>>
>>>indexes:
>>>stats_min_pkey primary key btree (ip, "start")
>>>stats_min_start btree ("start")
>>>stats_hr_pkey primary key btree (ip, "start")
>>>stats_hr_start btree ("start")
>>
>>>ip is of type "inet" in all tables.
>>>start is of type "timestamp without time zone" in all tables.
>>
>>Okay, so a pkey index entry will take 32 bytes counting overhead ...
>>you've got about 10:1 bloat on the stats_min indexes and 2:1 in stats_hr.
>>Definitely bad :-(
>
>
> The only difference between the way stats_min and stats_hr are updated
> stems from the fact that stats_min only holds records for the last 1440
> minutes (because of its killer time granularity), whereas stats_hr
> holds its data until we decide some of it is obsolete enough and
> issue a "delete from" by hand.

Are you sure that all indexes are needed and that a partial index could
not help ? What about the statistics on these indexes ? Are they really
used ?

Regards
Gaetano Mendola.