BigQuery 日期和时间函数在时间戳列上返回 NULL

Question

我正在使用 LAG() 和 LEAD() 从表中的一个时间戳列中提取 3 个时间戳列：timestamp、prev_timestamp、next_timestamp。我需要做一些简单的日期和时间格式化，但是当我在 prev_timestamp 或 next_timestamp 上使用 MONTH() 之类的函数时，它会返回 NULL。

结果列的架构类型是正确的 (TIMESTAMP)，并且由于某种原因，常规时间戳日期和时间格式有效。如何使它正确返回所有 3 列的月份？

示例代码为时间戳列返回月份，为上一个和下一个时间戳列返回 NULL：

SELECT 
  MONTH(timestamp) AS month,
  MONTH(prev_timestamp) AS prev_month,
  MONTH(next_timestamp) AS next_month
FROM (
  SELECT
    timestamp,
    LAG(timestamp,1) OVER (PARTITION BY id ORDER BY timestamp) prev_timestamp,
    LEAD(timestamp,1) OVER (PARTITION BY id ORDER BY timestamp) next_timestamp
  FROM timestamp_table
  )

Answer 1

所以在测试和检查了几件事后，我实际上从Mikhail's answer 中得到启发，并意识到他的答案是不正确的，因为滞后/领先不会返回毫秒，而是微秒（为什么？这是任何人的猜测）。

SELECT 
  MONTH(timestamp) AS month,
  MONTH(MSEC_TO_TIMESTAMP((prev_timestamp/1000))) AS prev_month,
  MONTH(MSEC_TO_TIMESTAMP((next_timestamp/1000))) AS next_month
FROM (
  SELECT
    timestamp,
    LAG(timestamp,1) OVER (PARTITION BY id ORDER BY timestamp) prev_timestamp,
    LEAD(timestamp,1) OVER (PARTITION BY id ORDER BY timestamp) next_timestamp
  FROM timestamp_table
  )

应该可以。我刚刚测试了创建一个包含三行时间戳的表。在没有 /1000 的情况下使用它，我的滞后/领先版本给出了不同的月份。我测试过，如果你不做除法，你最终会在第 47 个千年的某个地方结束。

Answer 2

试试下面

SELECT 
  MONTH(MSEC_TO_TIMESTAMP(timestamp)) AS month,
  MONTH(MSEC_TO_TIMESTAMP(prev_timestamp)) AS prev_month,
  MONTH(MSEC_TO_TIMESTAMP(next_timestamp)) AS next_month
FROM (
  SELECT
    timestamp,
    LAG(timestamp,1) OVER (PARTITION BY id ORDER BY timestamp) prev_timestamp,
    LEAD(timestamp,1) OVER (PARTITION BY id ORDER BY timestamp) next_timestamp
  FROM timestamp_table
  )