将嵌套数组加载到 bigquery

Question

我有 JSON 格式的数据，其中包含嵌套数组。这是一个例子：

"data": {"events": [[1, 1271, 518, 945], [1, 1287, 495, 963],...

子数组的长度可以是 3 或 4，第一个数字是数据类型（大约有 30 种不同的）。有什么方法可以在不转换成字典“记录”的情况下将这些数据加载到 bigQuery？

谢谢， 亚龙

--编辑--

有this 的问题，有一个解决方法，但是有一个固定长度的子数组，所以我猜不适用..

Answer 1

这可能是错误的方向，因为尚不完全清楚您的最终目标是什么，但让我尝试帮助您
不知何故，我觉得您的目标表应如下所示

type    metric1 metric2 metric3  
   1       1271     518     945  
   1       1287     495     963  

所以，我的建议是分两步完成

第 1 步 - 将数据加载为 CSV，仅包含一个字段 - 假设表格 theTable 包含字段 data

                              data
 {"data": {"events": [[1, 1271, 518, 945], [1, 1287, 495, 963]]}}
 {"data": {"events": [[2, 111, 222, 333], [3, 444, 555, 666], [4, 777, 888, 999]]}} 

第 2 步 - 处理 theTable 以生成预期的架构（请参阅答案顶部）并保存到最终表格中。您可以为此使用以下查询

SELECT
  NTH(1, SPLIT(y)) AS type,
  NTH(2, SPLIT(y)) AS metric1,
  NTH(3, SPLIT(y)) AS metric2,
  NTH(4, SPLIT(y)) AS metric3,
FROM (
  SELECT 
    REPLACE(REPLACE(COALESCE(y0, y1, y2, y3, y4, y5, y6), '[', ''), ']', '') AS y  
  FROM (
    SELECT
      IF(k=0, JSON_EXTRACT(data, '$.data.events[0]'), NULL) AS y0,
      IF(k=1, JSON_EXTRACT(data, '$.data.events[1]'), NULL) AS y1,
      IF(k=2, JSON_EXTRACT(data, '$.data.events[2]'), NULL) AS y2,
      IF(k=3, JSON_EXTRACT(data, '$.data.events[3]'), NULL) AS y3,
      IF(k=4, JSON_EXTRACT(data, '$.data.events[4]'), NULL) AS y4,
      IF(k=5, JSON_EXTRACT(data, '$.data.events[5]'), NULL) AS y5,
      IF(k=6, JSON_EXTRACT(data, '$.data.events[6]'), NULL) AS y6,
    FROM theTable AS a
    CROSS JOIN (
      SELECT k FROM (SELECT 0 AS k), (SELECT 1 AS k), (SELECT 2 AS k), 
      (SELECT 3 AS k), (SELECT 4 AS k), (SELECT 5 AS k), (SELECT 6 AS k)
    ) AS b
  )
  HAVING NOT y IS NULL
)

结果将是

type    metric1 metric2 metric3  
   1       1271     518     945  
   1       1287     495     963  
   2        111     222     333  
   3        444     555     666  
   4        777     888     999  

如您所见 - 此特定查询最多支持 7 个子数组，但您可以通过更改三个位置的代码来减少或增加它

#1

REPLACE(REPLACE(COALESCE(y0, y1, y2, y3, y4, y5, y6), '[', ''), ']', '') AS y    

#2

IF(k=0, JSON_EXTRACT(data, '$.data.events[0]'), NULL) AS y0,
IF(k=1, JSON_EXTRACT(data, '$.data.events[1]'), NULL) AS y1,
IF(k=2, JSON_EXTRACT(data, '$.data.events[2]'), NULL) AS y2,
IF(k=3, JSON_EXTRACT(data, '$.data.events[3]'), NULL) AS y3,
IF(k=4, JSON_EXTRACT(data, '$.data.events[4]'), NULL) AS y4,
IF(k=5, JSON_EXTRACT(data, '$.data.events[5]'), NULL) AS y5,
IF(k=6, JSON_EXTRACT(data, '$.data.events[6]'), NULL) AS y6,

#3

SELECT k FROM (SELECT 0 AS k), (SELECT 1 AS k), (SELECT 2 AS k), 
(SELECT 3 AS k), (SELECT 4 AS k), (SELECT 5 AS k), (SELECT 6 AS k)

最后，只测试转换逻辑，不加载实际数据 - 你可以使用下面的脚本

SELECT
  NTH(1, SPLIT(y)) AS type,
  NTH(2, SPLIT(y)) AS metric1,
  NTH(3, SPLIT(y)) AS metric2,
  NTH(4, SPLIT(y)) AS metric3,
FROM (
  SELECT 
    REPLACE(REPLACE(COALESCE(y0, y1, y2, y3, y4, y5, y6), '[', ''), ']', '') AS y  
  FROM (
    SELECT
      IF(k=0, JSON_EXTRACT(data, '$.data.events[0]'), NULL) AS y0,
      IF(k=1, JSON_EXTRACT(data, '$.data.events[1]'), NULL) AS y1,
      IF(k=2, JSON_EXTRACT(data, '$.data.events[2]'), NULL) AS y2,
      IF(k=3, JSON_EXTRACT(data, '$.data.events[3]'), NULL) AS y3,
      IF(k=4, JSON_EXTRACT(data, '$.data.events[4]'), NULL) AS y4,
      IF(k=5, JSON_EXTRACT(data, '$.data.events[5]'), NULL) AS y5,
      IF(k=6, JSON_EXTRACT(data, '$.data.events[6]'), NULL) AS y6,
    FROM (
    SELECT data FROM 
      (SELECT '{"data": {"events": [[1, 1271, 518, 945], [1, 1287, 495, 963]]}}' AS data),
      (SELECT '{"data": {"events": [[2, 111, 222, 333], [3, 444, 555, 666], [4, 777, 888, 999]]}}' AS data) 
    ) AS a
    CROSS JOIN (
      SELECT k FROM (SELECT 0 AS k), (SELECT 1 AS k), (SELECT 2 AS k), 
      (SELECT 3 AS k), (SELECT 4 AS k), (SELECT 5 AS k), (SELECT 6 AS k)
    ) AS b
  )
  HAVING NOT y IS NULL
)

希望这对您有所帮助！

Answer 2

无法直接加载数组数组；您需要使用记录来包装数组的内部级别。标准 SQL 的参考涉及到这一点（尽管就语言本身而言，而不是加载数据）：https://cloud.google.com/bigquery/sql-reference/arrays#building_arrays_of_arrays。