在 Big Query 中使用“Lead”窗口函数时出现时间戳问题

Question

我正在尝试获取客户的第一个订单、他们的下一个订单以及两个订单之间的天数差异。看起来很简单。我遵循的步骤如下：

1. 使用 MIN() 和 LEAD() 函数拉客户的第一个和第二个订单
2. 使用这 2 个字段运行 DATEDIFF 以获得天数差异。

简短的代码如下所示：

SELECT cust, MIN(ord_time) first_ord, LEAD(ord_time, 1) 
                                      OVER 
                                      (PARTITION BY customer_id
                                      ORDER BY ord_time) next_ord
FROM
(SELECT cust, ord_time
FROM df.orders
GROUP EACH BY cust, ord_time)

其中还有一些其他过滤连接和分组，但这是基本块。

输出应该是一个包含客户 ID 的字段和两个时间戳字段。两个时间戳字段如下所示：

Timestamps in Output

所以一切看起来都很棒。但是，当我尝试使用这两个字段运行 DATEDIFF() 函数时，一切都返回 Null。

此外，当我将鼠标悬停在任一时间戳字段上时，它会告诉我数据类型是 TIMESTAMP，但是当我尝试将任何类型的时间戳转换为秒或其他任何内容时，next_ord 字段会导致它失败并出现“类型未知”。

只是寻找我做错了什么或任何解决此问题的方法。

感谢您的帮助。

Answer 1

我认为这与 wendow 函数如何处理时间戳有关

这是我目前看到的：

1.

当源数据点是字符串时 - 一切都按预期：

SELECT 
  customer_id,
  first_ord,
  next_ord,
  DATEDIFF(next_ord, first_ord) AS diff
FROM (
  SELECT 
    customer_id, 
    LEAD(ord_time, 0) OVER (PARTITION BY customer_id ORDER BY ord_time) first_ord, 
    LEAD(ord_time, 1) OVER (PARTITION BY customer_id ORDER BY ord_time) next_ord,
    ROW_NUMBER() OVER (PARTITION BY customer_id ORDER BY ord_time) num
  FROM 
    (SELECT 1 AS customer_id, '2014-04-08 09:51:24 UTC' AS ord_time),
    (SELECT 1 AS customer_id, '2014-04-08 09:53:31 UTC' AS ord_time),
    (SELECT 1 AS customer_id, '2014-05-08 09:53:31 UTC' AS ord_time),
    (SELECT 2 AS customer_id, '2014-09-12 17:20:43 UTC' AS ord_time),
    (SELECT 2 AS customer_id, '2015-04-16 21:44:18 UTC' AS ord_time),
)
WHERE num = 1

结果：

customer_id       first_ord             next_ord    diff     
1   2014-04-08 09:51:24 UTC 2014-04-08 09:53:31 UTC 0    
2   2014-09-12 17:20:43 UTC 2015-04-16 21:44:18 UTC 216  

2.

当源数据点是时间戳时 - 结果为 null 如您在问题中所述：

SELECT 
  customer_id,
  first_ord,
  next_ord,
  DATEDIFF(next_ord, first_ord) AS diff
FROM (
  SELECT 
    customer_id, 
    LEAD(ord_time, 0) OVER (PARTITION BY customer_id ORDER BY ord_time) first_ord, 
    LEAD(ord_time, 1) OVER (PARTITION BY customer_id ORDER BY ord_time) next_ord,
    ROW_NUMBER() OVER (PARTITION BY customer_id ORDER BY ord_time) num
  FROM 
    (SELECT 1 AS customer_id, TIMESTAMP('2014-04-08 09:51:24 UTC') AS ord_time),
    (SELECT 1 AS customer_id, TIMESTAMP('2014-04-08 09:53:31 UTC') AS ord_time),
    (SELECT 1 AS customer_id, TIMESTAMP('2014-05-08 09:53:31 UTC') AS ord_time),
    (SELECT 2 AS customer_id, TIMESTAMP('2014-09-12 17:20:43 UTC') AS ord_time),
    (SELECT 2 AS customer_id, TIMESTAMP('2015-04-16 21:44:18 UTC') AS ord_time),
)
WHERE num = 1

结果：

customer_id       first_ord             next_ord    diff     
1   2014-04-08 09:51:24 UTC 2014-04-08 09:53:31 UTC null     
2   2014-09-12 17:20:43 UTC 2015-04-16 21:44:18 UTC null     

3.

为了“修复”，我必须做一些如下的转换：

SELECT 
  customer_id,
  TIMESTAMP(first_ord) as first_ord,
  TIMESTAMP(next_ord) as next_ord,
  DATEDIFF(next_ord, first_ord) AS diff
FROM (
  SELECT 
    customer_id, 
    LEAD(STRING(ord_time), 0) OVER (PARTITION BY customer_id ORDER BY ord_time) first_ord, 
    LEAD(STRING(ord_time), 1) OVER (PARTITION BY customer_id ORDER BY ord_time) next_ord,
    ROW_NUMBER() OVER (PARTITION BY customer_id ORDER BY ord_time) num
  FROM 
    (SELECT 1 AS customer_id, TIMESTAMP('2014-04-08 09:51:24 UTC') AS ord_time),
    (SELECT 1 AS customer_id, TIMESTAMP('2014-04-08 09:53:31 UTC') AS ord_time),
    (SELECT 1 AS customer_id, TIMESTAMP('2014-05-08 09:53:31 UTC') AS ord_time),
    (SELECT 2 AS customer_id, TIMESTAMP('2014-09-12 17:20:43 UTC') AS ord_time),
    (SELECT 2 AS customer_id, TIMESTAMP('2015-04-16 21:44:18 UTC') AS ord_time)
)
WHERE num = 1

结果是：

customer_id       first_ord             next_ord    diff     
1   2014-04-08 09:51:24 UTC 2014-04-08 09:53:31 UTC 0    
2   2014-09-12 17:20:43 UTC 2015-04-16 21:44:18 UTC 216