On 10/17/14, 11:47 PM, Tom Lane wrote:
> Marko Tiikkaja <marko@joh.to> writes:
>> This week we had one of the most annoying problems I've ever encountered
>> with postgres. We had a big index on multiple columns, say, foo(a, b,
>> c). According to pg_stat_all_indexes the index was being used *all the
>> time*. However, after looking into our queries more closely, it turns
>> out that it was only being used to look up statistics for the foo.a
>> column to estimate merge scan viability during planning. But this took
>> hours for two people to track down.
>
>> So what I'd like to have is a way to be able to distinguish between
>> indexes being used to answer queries, and ones being only used for stats
>> lookups during planning.
>
> Why? Used is used.
Because I don't need a 30GB index on foo(a,b,c) to look up statistics.
If I ever have a problem, I can replace it with a 5GB one on foo(a).
.marko