Question

数据库是postgres。

我在同一张桌子上有两个查询。这两个查询完全相同，只是一个在With记录中使用文本列siteid，另一个在With记录中使用smallint列sitegid。但是解释分析对使用smallint sitegid的查询返回了更长的估计。这对我来说没有意义。有谁知道为什么？

我在这里为这两个查询添加了解释分析输出的一部分。它们完全相同，除了实际用于groupAggregate的时间。不顾一切地寻求帮助。谢谢！

查询就像：

With records as (...)
SELECT source, siteid,  observed_contingency_row, forecast_contingency_col,
                  count(*),sum(diff) as sum_error 
FROM records 
GROUP BY source, siteid,  observed_contingency_row, forecast_contingency_col 
ORDER BY source, siteid,  observed_contingency_row, forecast_contingency_col

这是一个非常长的查询，无法在此处发布所有内容。

解释分析的输出是：

->  WindowAgg  (cost=0.08..0.11 rows=1 width=74) (actual time=14.880..14.894 rows=15 loops=1)
     ->  Sort  (cost=0.08..0.09 rows=1 width=74) (actual time=14.876..14.877 rows=15 loops=1)
           Sort Key: records.observed_contingency_row, records.forecast_contingency_col
           Sort Method: quicksort  Memory: 27kB
           ->  GroupAggregate  (cost=0.03..0.07 rows=1 width=74) (actual time=14.775..14.869 rows=15 loops=1)
                 Group Key: records.source, records.siteid, records.observed_contingency_row, records.forecast_contingency_col
                 ->  Sort  (cost=0.03..0.04 rows=1 width=74) (actual time=14.762..14.773 rows=176 loops=1)
                       Sort Key: records.source, records.siteid, records.observed_contingency_row, records.forecast_contingency_col
                       Sort Method: quicksort  Memory: 38kB
                       ->  CTE Scan on records  (cost=0.00..0.02 rows=1 width=74) (actual time=10.664..14.306 rows=176 loops=1)

和

->  WindowAgg  (cost=0.08..0.11 rows=1 width=74) (actual time=1558.508..1558.516 rows=15 loops=1)
     ->  Sort  (cost=0.08..0.09 rows=1 width=74) (actual time=1558.503..1558.503 rows=15 loops=1)
           Sort Key: records.observed_contingency_row, records.forecast_contingency_col
           Sort Method: quicksort  Memory: 27kB
           ->  GroupAggregate  (cost=0.03..0.07 rows=1 width=74) (actual time=1558.403..1558.495 rows=15 loops=1)
                 Group Key: records.source, records.siteid, records.observed_contingency_row, records.forecast_contingency_col
                 ->  Sort  (cost=0.03..0.04 rows=1 width=74) (actual time=1558.388..1558.407 rows=176 loops=1)
                       Sort Key: records.source, records.siteid, records.observed_contingency_row, records.forecast_contingency_col
                       Sort Method: quicksort  Memory: 38kB
                       ->  CTE Scan on records  (cost=0.00..0.02 rows=1 width=74) (actual time=7.106..1557.779 rows=176 loops=1)

整数列的排序时间更长？

0 个答案: