Sql 如何获取相对于BigQuery中的前一行已更改的列列表？ 为了更好的理解，考虑下面的例子和附图：

假设在ID相同的情况下，BigQuery中表中的任何列都不应该从第1天更改为第2天，那么如何获得行为不符合预期的列列表

换句话说，我想列出ID相同时从第1天更改到第2天的所有列

在示例的最后一列中，我将介绍所需的输出：

| ID | Day | Column1 | Column2 | Column3 | Column4 | Column5 | Column6 | Column7 | Desired outputs
| 1  | 1   |    x    |    x    |    x    |    x    |    x    |    x    |    x    |     Column3
| 1  | 2   |    x    |    x    |    y    |    x    |    x    |    x    |    x    |     Column3
| 2  | 1   |    x    |    x    |    x    |    x    |    x    |    x    |    x    |     Column2
| 2  | 2   |    x    |    y    |    x    |    x    |    x    |    x    |    x    |     Column2
| 3  | 1   |    x    |    x    |    x    |    x    |    x    |    x    |    x    | Column4,Column6
| 3  | 2   |    x    |    x    |    x    |    y    |    x    |    y    |    x    | Column4,Column6
| 4  | 1   |    x    |    x    |    x    |    x    |    x    |    x    |    y    |     Column7
| 4  | 2   |    x    |    x    |    x    |    x    |    x    |    x    |    x    |     Column7

问候,， 布鲁诺

---

这里有一个可能的解决方案（忽略WITH语句，这是为了复制您的表）：

计算每个列和每个ID的前一天值：

聚合每个ID和日期的修改列名

整个查询将为您提供

注意：unpivot操作很容易推广，您应该能够使用

SELECT COLUMN_NAME 
FROM yourdataset.INFORMATION_SCHEMA.COLUMNS
WHERE TABLE_NAME = yourtablename
AND COLUMN_NAME NOT IN ("ID", "Day")

使用下面的方法

select * except(prev_row, cur_row),
  (
    select ifnull(string_agg(col_name),'no changes')
    from (
      select * from unnest(split(prev_row)) p,
      unnest([struct(split(p, ':')[offset(0)] as col_name, split(p, ':')[safe_offset(1)] as value_previous)])
    )
    join (
      select * from unnest(split(cur_row)) c,
      unnest([struct(split(c, ':')[offset(0)] as col_name, split(c, ':')[safe_offset(1)] as value_current)]) 
    )
    using(col_name)
    where value_previous != value_current
    and not col_name in ('id', 'day')
  ) as changes_vs_previous_day
from (
  select *,
    translate(to_json_string(t), '{}"', '') cur_row,
    translate(to_json_string(lag(t) over(partition by id order by day)), '{}"', '') prev_row,
  from `project.dataset.table` t
)       

如果应用于问题中的样本数据，则输出为

更新：下面是上面的稍微重构版本，只是为了让它不那么冗长

#standardSQL
create temp function extract_name_value_pairs (row string) as (
  array(
    select as struct * from unnest(split(translate(row, '{}"', ''))) t,
    unnest([struct(split(t, ':')[offset(0)] as col_name, split(t, ':')[safe_offset(1)] as value)])
  )  
);
select * except(prev_row, cur_row),
  (
    select ifnull(string_agg(col_name),'no changes')
    from unnest(extract_name_value_pairs(prev_row)) p
    join unnest(extract_name_value_pairs(cur_row)) c
    using(col_name)
    where p.value != c.value
    and not col_name in ('id', 'day')
  ) as changes_vs_previous_day
from (
  select *,
    to_json_string(t) as cur_row,
    to_json_string(lag(t) over(partition by id order by day)) as prev_row,
  from `project.dataset.table` t
)

---

你有固定的列要检查吗？或者你正在寻找一个更通用的解决方案来允许任意数量的列？一个更通用的解决方案会更好，但我认为一个更简单的解决方案考虑20列会解决我的问题。检查和伟大的想法！非常感谢。没问题。不要忘记接受答案以结束问题（如果答案令人满意）

SELECT COLUMN_NAME 
FROM yourdataset.INFORMATION_SCHEMA.COLUMNS
WHERE TABLE_NAME = yourtablename
AND COLUMN_NAME NOT IN ("ID", "Day")

select * except(prev_row, cur_row),
  (
    select ifnull(string_agg(col_name),'no changes')
    from (
      select * from unnest(split(prev_row)) p,
      unnest([struct(split(p, ':')[offset(0)] as col_name, split(p, ':')[safe_offset(1)] as value_previous)])
    )
    join (
      select * from unnest(split(cur_row)) c,
      unnest([struct(split(c, ':')[offset(0)] as col_name, split(c, ':')[safe_offset(1)] as value_current)]) 
    )
    using(col_name)
    where value_previous != value_current
    and not col_name in ('id', 'day')
  ) as changes_vs_previous_day
from (
  select *,
    translate(to_json_string(t), '{}"', '') cur_row,
    translate(to_json_string(lag(t) over(partition by id order by day)), '{}"', '') prev_row,
  from `project.dataset.table` t
)       

#standardSQL
create temp function extract_name_value_pairs (row string) as (
  array(
    select as struct * from unnest(split(translate(row, '{}"', ''))) t,
    unnest([struct(split(t, ':')[offset(0)] as col_name, split(t, ':')[safe_offset(1)] as value)])
  )  
);
select * except(prev_row, cur_row),
  (
    select ifnull(string_agg(col_name),'no changes')
    from unnest(extract_name_value_pairs(prev_row)) p
    join unnest(extract_name_value_pairs(cur_row)) c
    using(col_name)
    where p.value != c.value
    and not col_name in ('id', 'day')
  ) as changes_vs_previous_day
from (
  select *,
    to_json_string(t) as cur_row,
    to_json_string(lag(t) over(partition by id order by day)) as prev_row,
  from `project.dataset.table` t
)