Re: Horrific time for getting 1 record from an index?

On 11/11/13 3:51 PM, Daniel Farina wrote:
> On Mon, Nov 11, 2013 at 1:48 PM, Jim Nasby <jnasby@enova.com> wrote:
>> Postgres 9.1.9.
>>
>> explain analyze select min(insert_time) from cnu_stats.page_hits_raw ;
>> I checked and there were no un-granted locks... but I have a hard time
>> believing it actually too 257 seconds to get 2 pages (one index, one heap)
>> back from our SAN.
>
> Try adding EXPLAIN (ANALYZE, BUFFERS).  I am wondering if you are
> reading through a lot of pages addressing tuples not visible to the
> transaction.

explain (analyze,buffers) select min(insert_time) from cnu_stats.page_hits_raw ;
                                                                            QUERY PLAN

-----------------------------------------------------------------------------------------------------------------------------------------------------------------
  Result  (cost=0.12..0.13 rows=1 width=0) (actual time=119.347..119.347 rows=1 loops=1)
    Buffers: shared hit=1 read=9476
    InitPlan 1 (returns $0)
      ->  Limit  (cost=0.00..0.12 rows=1 width=8) (actual time=119.335..119.336 rows=1 loops=1)
            Buffers: shared hit=1 read=9476
            ->  Index Scan using page_hits_raw_pkey on page_hits_raw  (cost=0.00..5445004.65 rows=47165480 width=8)
(actualtime=119.333..119.333 rows=1 loops=1) 
                  Index Cond: (insert_time IS NOT NULL)
                  Buffers: shared hit=1 read=9476
  Total runtime: 119.382 ms
(9 rows)

We do run a regular process to remove older rows... I thought we were vacuuming after that process but maybe not.
--
Jim Nasby, Lead Data Architect   (512) 569-9461