如何使用 ClickHouse 窗口漏斗模式？

Posted 2023-03-25

【中文标题】如何使用 ClickHouse 窗口漏斗模式？

【英文标题】：How to use ClickHouse windowFunnel mode?

【发布时间】：2021-10-26 10:54:34

【问题描述】：

我想在 ClickHouse 中使用windowFunnel() 函数。这是我的查询：

SELECT level,
       count() AS count
FROM (
      SELECT device_id,
             windowFunnel(6000000000000)(creation_time, event_id = 100,
                          event_id = 101) AS level
      FROM event
      WHERE ((creation_time >= '2021-08-01')
          AND (creation_time <= '2021-09-30'))
        AND (event_id in [100,101])
      GROUP BY device_id)
GROUP BY level
ORDER BY level ASC;

我希望 ClickHouse 返回不干预其他事件的事件序列。例如，如果序列是 100 => 102 => 101，它应该停止在 102 处找到 100 => 101。 ClickHouse 文档在 windowFunnel strict_order 模式下被认为是这个特性。但是当我以这种方式使用它时：

SELECT level,
       count() AS count
FROM (
      SELECT device_id,
             windowFunnel(6000000000000, ['strict_order'])(creation_time, event_id = 100,
                          event_id = 101) AS level
      FROM event
      WHERE ((creation_time >= '2021-08-01')
          AND (creation_time <= '2021-09-30'))
        AND (event_id in [100,101])
      GROUP BY device_id)
GROUP BY level
ORDER BY level ASC;

ClickHouse 抛出错误提示：

Code: 170, e.displayText() = DB::Exception: Bad get: has Array, requested String (version 21.7.2.7 (official build))

我不知道如何在 windowFunnel() 函数中使用模式？ 任何帮助将不胜感激。

【参考方案1】：

我终于可以找到如何使用windowsFunnel() 模式查询 CH。我应该删除数组：

SELECT level,
   count() AS count
FROM (
  SELECT device_id,
         windowFunnel(6000000000000, 'strict_order')
(creation_time, event_id = 100,
                      event_id = 101) AS level
  FROM event
  WHERE ((creation_time >= '2021-08-01')
      AND (creation_time <= '2021-09-30'))
    AND (event_id in [100,101])
  GROUP BY device_id)
GROUP BY level
ORDER BY level ASC;

我只是想不通为什么 CH 文档在 windowsFunnel() 语法中使用 []，我认为这里有一个错误：

windowFunnel(window, [mode, [mode, ... ]])(timestamp, cond1, cond2, ..., condN)

【讨论】：

没有错。 en.wikipedia.org/wiki/Extended_Backus%E2%80%93Naur_form 选项可以通过方括号 [ ... ] 表示。也就是说，在方括号中设置的所有内容可能只出现一次，或者根本不出现