SQL查询性能优化

Question

UPDATE nas_backup
SET fiber_serviceability_class = '0', 
last_updated_ts = CURRENT_TIMESTAMP 
WHERE location_id IN ( 
SELECT location_id 
FROM ( 
WITH distinct_locs AS ( 
    SELECT location_id, boundary_type 
    FROM ( 
        SELECT location_id, boundary_type 
        FROM nc
        WHERE technology_type = 'Fibre' 
    ) 
    GROUP BY location_id, boundary_type 
    HAVING COUNT( * ) = 1
)
SELECT nas.location_id
FROM distinct_locs, nas_backup nas
WHERE distinct_locs.location_id = nas.location_id
AND distinct_locs.boundary_type = 'FSA'
GROUP BY nas.location_id
)
);

谁能建议一个优化查询的方法?现在需要5分钟以上。

nc表有1600万条记录，nas_backup表有200万条记录

Answer 1

EXISTS可以帮你一点忙。试试吧:

UPDATE nas_backup
SET fiber_serviceability_class = '0', 
last_updated_ts = CURRENT_TIMESTAMP 
-- Remove the IN condition and check only that at least one row exists
WHERE EXISTS ( 
SELECT location_id 
FROM ( 
    WITH distinct_locs AS ( 
        SELECT location_id, boundary_type 
        FROM ( 
            SELECT location_id, boundary_type 
            FROM nc
            WHERE technology_type = 'Fibre'
        ) 
        GROUP BY location_id, boundary_type 
        HAVING COUNT( * ) = 1
    )
    SELECT
        nas.location_id
    FROM
        distinct_locs
    WHERE
        distinct_locs.boundary_type = 'FSA'
        -- This is the new line
        AND distinct_locs.location_id = nas_backup.location_id
    GROUP BY
        nas.location_id
    )
);

然而，如果你能分享你的数据库结构和你的目标，这将更容易帮助我们。

下次请告诉我们你正在使用的DBMS的供应商和版本。

Answer 2

我在这里说的大部分事情都应该由一个好的优化器来完成，所以这主要是为了更好的可读性。

所有过滤器都应用于位置部分，因此将其带到子查询以减少结果集:

SELECT location_id, boundary_type 
FROM nc 
WHERE technology_type = 'Fibre' AND nc.boundary_type='FSA'
GROUP BY location_id, boundary_type 
HAVING COUNT(*) = 1

接下来，使用JOIN语法代替隐式JOIN:

UPDATE nas fiber_serviceability_class = '0', last_updated_ts = CURRENT_TIMESTAMP 
FROM nas_backup nas
    JOIN (SELECT location_id, boundary_type 
            FROM nc 
            WHERE technology_type = 'Fibre' AND nc.boundary_type='FSA'
            GROUP BY location_id, boundary_type 
            HAVING COUNT(*) = 1) loc ON loc.location_id=nas.location_id

我不知道是否有一个特殊的原因与HAVING子句的子查询。如果不知道你的结构和数据，这只是一个猜测，但也许你不需要。这是一个非常简单的查询，然后翻译为"更新每个nas_backup，其中技术类型是光纤和边界类型是FSA在同一时间"。

UPDATE nas fiber_serviceability_class = '0', last_updated_ts = CURRENT_TIMESTAMP 
FROM nas_backup nas
    JOIN nc ON nas.location_id=nc.location_id
WHERE nc.technology_type='Fibre'
    AND nc.boundary_type='FSA'