SPARK SQL 按选择结果从组中选择

Question

我有一张名为table_new的表。在第一步中，我想按 id、kmstand、vacationname 和 vacationvalue 对结果进行分组，其中每个分组只有一个计数。对于此步骤，我已经创建了一个查询：

SELECT id, kmstand, vacationame, vacationvalue 
 FROM `db_1`.`table_new` 
 WHERE (vacationame='vacation1' 
    OR vacationame='vacation2' 
    OR vacationame='vacation3' 
    OR vacationame='vacation4') 
 GROUP BY id, kmstand, vacationame, vacationvalue 
 HAVING COUNT(*) = 1 ORDER BY id, kmstand DESC

结果是：

    id  kmstand vacationame vacationvalue
1   1   4000    vacation1   munich
2   1   4000    vacation1   stuttgart
3   1   5500    vacation4   koln
4   1   5500    vacation2   frankfurt
5   1   5500    vacation3   berlin
6   1   5500    vacation1   potsdam
7   2   6000    vacation2   new york
8   2   6000    vacation1   bangladesh
9   2   3000    vacation1   washington
10  2   3000    vacation3   chicago

现在

，我想选择kmstand和假期名称组合现在不同的id。这意味着结果应该是：

        id  kmstand vacationame vacationvalue
1   1   5500    vacation4   koln
2   1   5500    vacation2   frankfurt
3   1   5500    vacation3   berlin
4   1   5500    vacation1   potsdam
5   2   6000    vacation2   new york
6   2   6000    vacation1   bangladesh
7   2   3000    vacation1   washington
8   2   3000    vacation3   chicago

为此，我创建了以下嵌套 sql 查询：

    SELECT id, kmstand, count(*) as cnt 
FROM `db_1`.`table_new`
WHERE (SELECT id, kmstand, vacationame, vacationvalue 
     FROM `db_1`.`table_new` 
     WHERE (vacationame='vacation1' 
        OR vacationame='vacation2' 
        OR vacationame='vacation3' 
        OR vacationame='vacation4') 
     GROUP BY id, kmstand, vacationame, vacationvalue 
     HAVING COUNT(*) = 1 ORDER BY id, kmstand DESC) 
GROUP BY id, kmstand 
HAVING cnt = 1 
ORDER BY id, kmstand DESC

我也尝试了没有 where 子句或没有 from 的，但没有找到解决方案。对于此SQL查询，我收到以下错误消息：org.apache.spark.sql.AnalysisException: cannot recognize input near 'SELECT' 'id' ',' in expression specification; line 3 pos 7我很确定，语法不正确。你有什么建议吗？

Answer 1

这是问题的答案。现在在这里，我能够获得组合 id、kmstand 和 vacationame 不同的 id。

SELECT id, sumcnt, cnt2
    FROM(
        SELECT id, count(*) as cnt2, sum(cnt) as sumcnt
        FROM (
            SELECT id, kmstand, vacationame, count(*) as cnt
            FROM `db_1`.`table_new` 
            WHERE (vacationame='vacation1'
                OR vacationame='vacation2'
                OR vacationame='vacation3'
                OR vacationame='vacation4')
            GROUP BY id, kmstand, vacationame)T
        GROUP BY id)T
    WHERE (sumcnt/cnt2 = 1)

Answer 2

不熟悉 SPARK，但您可能希望：

SELECT id, kmstand, count(*) as cnt 
FROM (SELECT id, kmstand, vacationame, vacationvalue 
     FROM `db_1`.`table_new` 
     WHERE (vacationame='vacation1' 
        OR vacationame='vacation2' 
        OR vacationame='vacation3' 
        OR vacationame='vacation4') 
     GROUP BY id, kmstand, vacationame, vacationvalue 
     HAVING COUNT(*) = 1) T
GROUP BY id, kmstand 
HAVING cnt = 1 
ORDER BY id, kmstand DESC

请注意，我在 FROM 子句中为表添加了一个别名 ( T (。 这可能是必需的，也可能不是必需的，具体取决于您的 RDBMS。

另请注意，您通常不能在子查询中 ORDER BY。