google-bigquery - 查询按城市获取每日负账户余额

Question

我有一份随时间推移的账户余额清单。架构如下所示：

+-------------+---------+---------+----------------------+
| customer_id | city_id |  value  |  timestamp           |
+-------------+---------+---------+----------------------+
| 1           | 1       |  -500   | 2019-02-12T00:00:00  |
| 2           | 1       |  -200   | 2019-02-12T00:00:00  |
| 3           | 2       |  200    | 2019-02-10T00:00:00  |
| 4           | 1       |  -10    | 2019-02-09T00:00:00  |
+-------------+ --------+---------+----------------------+

我想汇总这些数据，以便获得按城市划分并按时间排序的每日总负账户余额：

+---------+---------+--------------+
| city_id |  value  |   timestamp  |
+---------+---------+--------------+
| 1       | -500    |  2019-02-12  |
| 1       | -200    |  2019-02-10  |
| 1       | -10     |  2019-02-09  |
+ --------+---------+--------------+

我试过的：

SELECT city_id, FORMAT_TIMESTAMP("%Y-%m-%d", TIMESTAMP(timestamp)) as date,
  SUM(value) OVER (PARTITION BY city_id ORDER BY FORMAT_TIMESTAMP("%Y-%m-%d", TIMESTAMP(timestamp))) negative_account_balance 
FROM `account_balances`
WHERE value < 0

但是，这给了我奇怪的帐户余额值，例如-5.985856421224E10。任何想法为什么？除此之外，查询会多次为同一城市和同一天生成条目。我希望它在同一天只返回同一个城市一次。

Answer 1

以下是 BigQuery 标准 SQL

#standardSQL
SELECT city_id, account_balance, `date` FROM (
  SELECT city_id, `date`, 
    SUM(value) OVER(PARTITION BY city_id ORDER BY `date`) account_balance 
  FROM (
    SELECT city_id, DATE(TIMESTAMP(t.timestamp)) AS `date`, SUM(value) value
    FROM `project.dataset.account_balances` t
    GROUP BY city_id, `date` )
)
WHERE account_balance< 0   

您可以使用示例/虚拟数据进行测试，使用上面的示例，如下例所示

#standardSQL
WITH `project.dataset.account_balances` AS (
  SELECT 1 customer_id, 1 city_id, -500 value, '2019-02-12T00:00:00' `timestamp` UNION ALL
  SELECT 2, 1, -200, '2019-02-12T00:00:00' UNION ALL
  SELECT 5, 1, 100, '2019-02-13T00:00:00' UNION ALL
  SELECT 3, 2, 200, '2019-02-10T00:00:00' UNION ALL
  SELECT 4, 1, -10, '2019-02-09T00:00:00' 
)
SELECT city_id, account_balance, `date` FROM (
  SELECT city_id, `date`, 
    SUM(value) OVER(PARTITION BY city_id ORDER BY `date`) account_balance 
  FROM (
    SELECT city_id, DATE(TIMESTAMP(t.timestamp)) AS `date`, SUM(value) value
    FROM `project.dataset.account_balances` t
    GROUP BY city_id, `date` )
)
WHERE account_balance< 0   

产生以下结果

Row city_id account_balance date     
1   1       -10             2019-02-09   
2   1       -710            2019-02-12   
3   1       -610            2019-02-13   

Answer 2

我采取了一种更简单的方法并使用了这个 sql（顺便说一句，当我尝试你的原始查询时，我得到了一个看起来不错的结果）

SELECT city_id, FORMAT_TIMESTAMP("%Y-%m-%d", TIMESTAMP(timestamp)) as date,
  SUM(value) as value
FROM `account_balances`
GROUP BY city_id, timestamp
HAVING value < 0

我使用此数据进行了检查（注意：我更改了日期格式以匹配 BigQuery 格式，尽管结果都相同）

WITH account_balances as (
SELECT 1 AS customer_id, 1 as city_id, -500 as value, '2019-02-12 00:00:00' as timestamp UNION ALL
SELECT 2 AS customer_id, 1 as city_id, -200 as value, '2019-02-12 00:00:00' as timestamp UNION ALL
SELECT 3 AS customer_id, 2 as city_id, 200 as value, '2019-02-10 00:00:00' as timestamp UNION ALL
SELECT 4 AS customer_id, 1 as city_id, -10 as value, '2019-02-09 00:00:00' as timestamp
)

SELECT city_id, FORMAT_TIMESTAMP("%Y-%m-%d", TIMESTAMP(timestamp)) as date,
  SUM(value) as value
FROM `account_balances`
GROUP BY city_id, timestamp
HAVING value < 0

这是结果：