Query performance with min and filter - Mailing list pgsql-general

Hi Postgres community,

We seem to have found a situation where a query run using explain analyse or pgbench is incredibly fast, but run via Java under load performs very poorly – we’ve checked query performance metrics for our Postgres instance and can confirm that it’s the query running slowly, and not a Java problem.

The table has ~7 million rows, with total size of 44GB. As shown below, explain analyse gives a sub 1ms execution time, yet in practice the query takes an average of 2 seconds – it’s worth mentioning that a small percentage of requests do take ~1ms, so something non-deterministic is happening here.

The query in question is below, with explain analyse:

explain analyse SELECT coalesce (

    (SELECT min(msg_offset) - 1 FROM events WHERE created_utc >= CURRENT_TIMESTAMP - INTERVAL '60 SECONDS' ),

    (SELECT max(msg_offset) FROM events),

    0

);

                                                                                 QUERY PLAN                                                                                

---------------------------------------------------------------------------------------------------------------------------------------------------------------------------

-

 Result  (cost=2.03..2.04 rows=1 width=8) (actual time=0.183..0.183 rows=1 loops=1)

   InitPlan 1 (returns $1)

     ->  Aggregate  (cost=1.56..1.57 rows=1 width=8) (actual time=0.028..0.028 rows=1 loops=1)

           ->  Index Only Scan using test_4 on events  (cost=0.44..1.56 rows=1 width=8) (actual time=0.023..0.023 rows=0 loops=1)

                 Index Cond: (created_utc >= (now() - '00:01:00'::interval))

                 Heap Fetches: 0

   InitPlan 3 (returns $3)

     ->  Result  (cost=0.45..0.46 rows=1 width=8) (actual time=0.153..0.153 rows=1 loops=1)

           InitPlan 2 (returns $2)

             ->  Limit  (cost=0.43..0.45 rows=1 width=8) (actual time=0.152..0.152 rows=1 loops=1)

                   ->  Index Only Scan Backward using events_pkey on events events_1  (cost=0.43..156842.57 rows=7643099 width=8) (actual time=0.151..0.151 rows=1 loops=1)

                         Index Cond: (msg_offset IS NOT NULL)

                         Heap Fetches: 0

 Planning time: 2.480 ms

 Execution time: 0.242 ms

(15 rows)

Output of \d events, with a few test indexes added to try and improve performance:

                                             Table "public.events"

    Column    |            Type             | Collation | Nullable |                  Default                   

--------------+-----------------------------+-----------+----------+--------------------------------------------

 msg_offset   | bigint                      |           | not null | nextval('events_msg_offset_seq'::regclass)

 msg_key      | character varying           |           | not null | 

 content_type | character varying           |           | not null | 

 type         | character varying           |           | not null | 

 created_utc  | timestamp without time zone |           | not null | 

 data         | text                        |           |          | 

 event_size   | integer                     |           | not null | 1048576

 cluster_id   | bigint                      |           | not null | 1

Indexes:

    "events_pkey" PRIMARY KEY, btree (msg_offset)

    "cluster_idx" btree (msg_key, cluster_id)

    "cluster_type_filter_idx" btree (type, cluster_id)

    "created_idx" btree (created_utc, msg_key)

    "key_idx" btree (msg_key)

    "test_1" btree (created_utc, msg_offset)

    "test_2" btree (msg_offset, created_utc)

    "test_3" btree (created_utc, msg_offset)

    "test_4" btree (created_utc, msg_offset DESC)

    "type_idx" btree (type)

Any help understanding what is happening here, or guidance to improve performance would be much appreciated.

Regards,

Adam