Question

我已经读过How do I speed up counting rows in a PostgreSQL table?和https://wiki.postgresql.org/wiki/Slow_Counting，但是我并没有接近更好的结果。另一个问题中大多数答案的问题都没有任何类型的筛选，并且依赖于表范围的统计信息。

我有一个约有1000万行的表，当前数据以页为单位提取（我知道这种分页策略并不理想），并且我必须向用户显示总数（业务需求）。查询速度很快，通常小于200毫秒，如下所示：

explain analyze
SELECT  DISTINCT ON (ID) table.*
FROM "table"
WHERE "table"."deleted_at" IS NULL
GROUP BY "table"."id"
ORDER BY "table"."id" DESC
LIMIT 25 OFFSET 200;


    QUERY PLAN
---------------------------------------------------------------------------------------------------------------------------------------------------------------------
 Limit  (cost=530.48..585.91 rows=25 width=252) (actual time=0.870..0.942 rows=25 loops=1)
   ->  Unique  (cost=87.00..19878232.36 rows=8964709 width=252) (actual time=0.328..0.899 rows=225 loops=1)
         ->  Group  (cost=87.00..15395877.86 rows=8964709 width=252) (actual time=0.327..0.747 rows=225 loops=1)
               Group Key: id
               ->  Index Scan Backward using table_pkey on table  (cost=87.00..10913523.36 rows=8964709 width=252) (actual time=0.324..0.535 rows=225 loops=1)
                     Filter: (deleted_at IS NULL)
                     Rows Removed by Filter: 397
 Planning time: 0.174 ms
 Execution time: 0.986 ms
(9 rows)

Time: 77.437 ms

问题是当我尝试通过以下方式显示计数时：

explain analyze
SELECT COUNT(*) AS count_all, "table"."id" AS id
FROM "table"
WHERE "table"."deleted_at" IS NULL
GROUP BY "table"."id";
                                                                      QUERY PLAN
-------------------------------------------------------------------------------------------------------------------------------------------------------
 GroupAggregate  (cost=87.00..21194868.36 rows=10282202 width=4) (actual time=0.016..16984.904 rows=10343557 loops=1)
   Group Key: id
   ->  Index Scan using table_pkey on table  (cost=87.00..5771565.36 rows=10282202 width=4) (actual time=0.012..11435.350 rows=10343557 loops=1)
         Filter: (deleted_at IS NULL)
         Rows Removed by Filter: 2170
 Planning time: 0.098 ms
 Execution time: 18638.381 ms
(7 rows)

此刻我不能使用概率计数，但是我也无法忍受10到50秒来返回计数。我还有其他方法可以加快速度吗？

加快Postgresql行计数

0 个答案: