查询性能较慢，仅适用于“简单”的Postgres查询

Question

我有一个简单的SQL查询，表现不太好。数据集也非常简单，只是一个任务表和一个user_tasks表。在DB目前只有50.000条记录

我在tasks字段和user_tasks字段上有索引。

查询完成时间> 300毫秒。

SELECT DISTINCT ON (t0."id") *
FROM "tasks" t0  INNER JOIN
     "user_tasks" u1
     ON TRUE 
WHERE (t0."id" = u1."task_id") AND
      (((u1."user_id" = 'f4325fd5-8862-4563-a534-957d76ac98b9') AND t0."delegate_id" IS NULL) 
OR ((t0."delegate_id" = 'f4325fd5-8862-4563-a534-957d76ac98b9') AND   
       (u1."user_id" != 'f4325fd5-8862-4563-a534-957d76ac98b9'))) 
ORDER BY t0."id", t0."start_at"

这是解释分析的输出

Unique  (cost=75397.69..75398.18 rows=98 width=2979) (actual time=479.995..480.041 rows=100 loops=1)
  ->  Sort  (cost=75397.69..75397.93 rows=98 width=2979) (actual time=479.994..480.004 rows=100 loops=1)
    Sort Key: t0.id, t0.start_at
      Sort Method: quicksort  Memory: 51kB
        ->  Gather  (cost=1000.85..75394.45 rows=98 width=2979) (actual time=17.529..479.734 rows=100 loops=1)
          Workers Planned: 2
            Workers Launched: 2
              ->  Merge Join  (cost=0.84..74384.65 rows=41 width=2979) (actual time=64.989..475.499 rows=33 loops=3)
                Merge Cond: (t0.id = u1.task_id)
                  Join Filter: (((u1.user_id = 'f4325fd5-8862-4563-a534-957d76ac98b9'::uuid) AND (t0.delegate_id IS NULL)) OR ((t0.delegate_id = 'f4325fd5-8862-4563-a534-957d76ac98b9'::uuid) AND (u1.user_id <> 'f4325fd5-8862-4563-a534-957d76ac98b9'::uuid)))
                    Rows Removed by Join Filter: 123267
                      ->  Parallel Index Scan using tasks_id_index on tasks t0  (cost=0.42..44772.95 rows=154209 width=2919) (actual time=0.023..125.719 rows=123300 loops=3)
                        ->  Index Scan using user_tasks_task_id_index on user_tasks u1  (cost=0.42..25604.24 rows=369900 width=36) (actual time=0.042..253.000 rows=369755 loops=3)
Planning time: 0.506 ms
Execution time: 488.518 ms

我将查询条带化，直到我发现它变得很快。

EXPLAIN ANALYZE SELECT DISTINCT ON (t0."id") * FROM "tasks" AS t0 
INNER JOIN "user_tasks" AS u1 ON (t0."id" = u1."task_id")
WHERE t0."delegate_id" = 'f4325fd5-8862-4563-a534-957d76ac98b9'  
OR  u1."user_id" = 'f4325fd5-8862-4563-a534-957d76ac98b9'

如果我删除OR和语句它变快〜0.7毫秒，如果我查询WHERE u1。“user_id”=并且如果我在WHERE t0查询它的速度很快。“delegate_id”=它的快速

该任务在delegate_id上​​有索引user_tasks在user_id和task_id＆user_task_id上有索引

Answer 1

搞定了

EXPLAIN ANALYZE 
    SELECT * 
    FROM "tasks" AS t0 INNER JOIN
       (select *
        from user_tasks ut
        where user_id = 'f4325fd5-8862-4563-a534-957d76ac98b9'
       ) u1
        ON (t0."id" = u1."task_id")
     WHERE t0."delegate_id" = 'f4325fd5-8862-4563-a534-957d76ac98b9'