Postgresql 9.3 Update Join 影响的行数与 Select 不同

Question

我需要从我的数据库中选择一些特定的位置，然后更新它们。

尴尬的是，在单独执行Select 语句时，它使用了 90k 行。 但是在对同一个 Select 执行 Update 语句时，它影响 80k 行。

这会产生 90k 行。

SELECT
    "Extent2"."AppId",
    "Extent2"."DateCreated"
FROM
    "dbo"."Route" AS "Extent1"
    INNER JOIN
    "dbo"."Positions" AS "Extent2" ON
        "Extent1"."BoundAppId" = "Extent2"."AppId" AND
        "Extent2"."DateCreated" > "Extent1"."TimeStarted" AND
        "Extent2"."DateCreated" < "Extent1"."TimeFinished"
WHERE
    "Extent1"."TimeStarted" IS NOT NULL AND
    "Extent1"."TimeFinished" IS NOT NULL AND
    FALSE = "Extent1"."IsDeleted" AND
    "Extent1"."TimeFinished" < '2016-02-06'

这会影响 80k 行 （嵌套的 Select 是相同的）

UPDATE "dbo"."Positions" AS j0
SET "ArchiveUntil" = '2020-02-16'
FROM (
    SELECT
        "Extent2"."AppId",
        "Extent2"."DateCreated"
    FROM
        "dbo"."Route" AS "Extent1"
        INNER JOIN
        "dbo"."Positions" AS "Extent2" ON
            "Extent1"."BoundAppId" = "Extent2"."AppId" AND
            "Extent2"."DateCreated" > "Extent1"."TimeStarted" AND
            "Extent2"."DateCreated" < "Extent1"."TimeFinished"
    WHERE
        "Extent1"."TimeStarted" IS NOT NULL AND
        "Extent1"."TimeFinished" IS NOT NULL AND
        FALSE = "Extent1"."IsDeleted" AND
        "Extent1"."TimeFinished" < '2016-02-06'
) as j1
WHERE (j0."AppId" = j1."AppId" AND j0."DateCreated" = j1."DateCreated")

此 SQL 是从 EntityFramework.Extended Library Source Commit 分支（未与父级合并）生成的，但对我来说似乎很正常......

也许不是？

Answer 1

您的查询没问题。我想您只是发现(AppId, DateCreated) 在Positions 中不是唯一的，或者其中一个值是NULL。

您可以运行此查询来验证计数：

select appid, DateCreated, count(*)
from dbo.Positions
group by appid, DateCreated
having count(*) > 1;

也有可能某些值是NULL，您可以使用以下方法找到：

select p.*
from dbo.Positions p
where appid is null or DateCreated is null;

编辑：

哎呀。我怀疑以上是正确的想法，只是错误的列。尝试运行这个版本（你原来的聚合查询）：

SELECT "Extent2"."AppId", "Extent2"."DateCreated", COUNT(*) as cnt
FROM
    "dbo"."Route" AS "Extent1"
    INNER JOIN
    "dbo"."Positions" AS "Extent2" ON
        "Extent1"."BoundAppId" = "Extent2"."AppId" AND
        "Extent2"."DateCreated" > "Extent1"."TimeStarted" AND
        "Extent2"."DateCreated" < "Extent1"."TimeFinished"
WHERE
    "Extent1"."TimeStarted" IS NOT NULL AND
    "Extent1"."TimeFinished" IS NOT NULL AND
    FALSE = "Extent1"."IsDeleted" AND
    "Extent1"."TimeFinished" < '2016-02-06'
GROUP BY "Extent2"."AppId", "Extent2"."DateCreated"
HAVING COUNT(*) >= 2;

我猜这是返回重复项（因此一对值被计算两次）。但是，一行在update 语句中只更新一次（因此同一对只更新一次）。

Answer 2

您没有使用子查询中的任何值来进行更新，您只是在检查跨越“dateCreated”的记录是否存在存在并且（也许）使用常量/文字值更新它。

UPDATE "dbo"."Positions" dst
SET "ArchiveUntil" = '2020-02-16'
WHERE EXISTS (
    SELECT 1
    FROM "dbo"."Route" src
    WHERE dst."BoundAppId" = src."AppId" 
      AND dst."DateCreated" > src."TimeStarted"
      AND dst."DateCreated" < src."TimeFinished"

      -- the two conditions below are redundant
      -- AND src."TimeStarted" IS NOT NULL
      -- AND src."TimeFinished" IS NOT NULL
      AND src."IsDeleted" <> True
      AND src."TimeFinished" < '2016-02-06'
        ) ;