如何从 Clickhouse 中选择百分比数据？

Question

给定下表：

CREATE TABLE main
(
    `job_id` UUID,
    `request_time` DateTime,
    `host_id` UInt8,
    `status_code` LowCardinality(String),
)
ENGINE = MergeTree
ORDER BY request_time
SETTINGS index_granularity = 8192

我正在尝试按主机获取所有状态及其相对百分比。为此，我需要计算按主机和状态分组的结果，并将每个计数字段除以其总主机字段数。

例如，这个查询可以在 MySQL 中工作：

SELECT
    main.host_id,
    main.status_code,
    COUNT() AS status_count,
    COUNT() / sub.host_total * 100 AS percent
FROM
    main
INNER JOIN (
    SELECT host_id, COUNT() AS host_total
    FROM main
    GROUP BY host_id
) AS sub ON (sub.host_id = main.host_id)
GROUP BY
    main.host_id,
    main.status_code

但是 ClickHouse 抛出：

DB::Exception：未知标识符：host_total；有列：host_id、status_code、count()：处理时 host_id、status_code、count() AS status_count、(count() / host_total) * 100 AS 百分比。 (UNKNOWN_IDENTIFIER)

可能是因为correlated (dependent) subqueries are not supported。

有人建议我使用CTE，所以我尝试了这个：

WITH sub AS (
    SELECT host_id, COUNT() AS host_total
    FROM main
    GROUP BY host_id
)
SELECT
    main.host_id,
    main.status_code,
    COUNT() AS status_count,
    COUNT() / (SELECT host_total FROM sub WHERE sub.host_id = main.host_id) * 100 AS percent
FROM
    main
GROUP BY
    main.host_id,
    main.status_code

但还是没有运气：

DB::Exception: 处理查询时缺少列：'main.host_id'：'SELECT host_total FROM sub WHERE host_id = main.host_id'，必需列：'host_total''host_id''main.host_id''host_total' 'host_id' 'main.host_id': 处理时 (SELECT host_total FROM sub WHERE sub.host_id = main.host_id) AS _subquery20: 处理时 count() / ((SELECT host_total FROM sub WHERE sub.host_id = main.host_id) AS _subquery20): 处理时 (count() / ((SELECT host_total FROM sub WHERE sub.host_id = main.host_id) AS _subquery20)) * 100 AS 百分比。 (UNKNOWN_IDENTIFIER)

Answer 1

CH 抛出不正确的错误。 https://github.com/ClickHouse/ClickHouse/issues/4567

host_total 应该在 groupby 部分或在聚合函数下

insert into main(request_time, host_id,status_code) values ( now(), 1, 200);
insert into main(request_time, host_id,status_code) values ( now(), 1, 500);
insert into main(request_time, host_id,status_code) values ( now(), 1, 200);
insert into main(request_time, host_id,status_code) values ( now(), 2, 500);
insert into main(request_time, host_id,status_code) values ( now(), 2, 200);
insert into main(request_time, host_id,status_code) values ( now(), 3, 500);

SELECT
    main.host_id,
    main.status_code,
    COUNT() AS status_count,
    round((COUNT() / any(sub.host_total)) * 100, 2) AS percent
FROM main
INNER JOIN
(
    SELECT
        host_id,
        COUNT() AS host_total
    FROM main
    GROUP BY host_id
) AS sub ON sub.host_id = main.host_id
GROUP BY
    main.host_id,
    main.status_code
ORDER BY
    main.host_id ASC,
    main.status_code ASC

┌─host_id─┬─status_code─┬─status_count─┬─percent─┐
│       1 │ 200         │            2 │   66.67 │
│       1 │ 500         │            1 │   33.33 │
│       2 │ 200         │            1 │      50 │
│       2 │ 500         │            1 │      50 │
│       3 │ 500         │            1 │     100 │
└─────────┴─────────────┴──────────────┴─────────┘

但是有更好的方法来解决它：

窗口函数

SELECT
    host_id,
    status_code,
    status_count,
    round((status_count / host_total) * 100, 2) AS percent
FROM
(
    SELECT
        host_id,
        status_code,
        status_count,
        sum(status_count) OVER (PARTITION BY host_id) AS host_total
    FROM
    (
        SELECT
            host_id,
            status_code,
            COUNT() AS status_count
        FROM main
        GROUP BY
            host_id,
            status_code
    )
)
ORDER BY
    host_id ASC,
    status_code ASC

┌─host_id─┬─status_code─┬─status_count─┬─percent─┐
│       1 │ 200         │            2 │   66.67 │
│       1 │ 500         │            1 │   33.33 │
│       2 │ 200         │            1 │      50 │
│       2 │ 500         │            1 │      50 │
│       3 │ 500         │            1 │     100 │
└─────────┴─────────────┴──────────────┴─────────┘

数组

SELECT
    host_id,
    status_code,
    status_count,
    round((status_count / host_total) * 100, 2) AS percent
FROM
(
    SELECT
        host_id,
        sumMap([CAST(status_code, 'String')], [1]) AS ga,
        count() AS host_total
    FROM main
    GROUP BY host_id
)
ARRAY JOIN
    ga.1 AS status_code,
    ga.2 AS status_count

┌─host_id─┬─status_code─┬─status_count─┬─percent─┐
│       1 │ 200         │            2 │   66.67 │
│       1 │ 500         │            1 │   33.33 │
│       2 │ 200         │            1 │      50 │
│       2 │ 500         │            1 │      50 │
│       3 │ 500         │            1 │     100 │
└─────────┴─────────────┴──────────────┴─────────┘