在 Impala 中将聚合函数与重采样相结合

问题描述

我在 Hadoop 中有一个表，其中我有不同传感器单元的数据，采样时间 ts 为 1 毫秒。我可以使用 Impala 中的以下查询，使用不同聚合函数的组合重新采样单个单元的数据（假设我想使用 LAST_VALUE() 作为聚合函数每 5 分钟重新采样一次数据）：

SELECT DISTINCT * 
from ( select ts_resample,unit,last_value(Val1) over (partition by ts_resample order by ts rows between unbounded preceding and unbounded following) as Val1,last_value(Val2) over (partition by ts_resample order by ts rows between unbounded preceding and unbounded following) as Val2
from (
SELECT cast(cast(unix_timestamp(cast(ts/1000 as TIMESTAMP))/300 as bigint)*300 as TIMESTAMP) as ts_resample,ts as ts,unit as unit,Val1 as Val1,Val2 as Val2
FROM Sensor_Data.Table1 WHERE unit='Unit1') as t) as tt

如果我为单个单元运行此查询，那么我会得到正确的答案并且没有问题。

但是如果我想根据一些聚合函数为每个单元重新采样数据，例如LAST_VALUE() 然后我得到了错误的答案，每个单元的重新采样结果是相同的，尽管每个单元的数据不同。下面给出了我正在运行的查询，其中我没有在 WHERE 子句中指定任何单位：

SELECT DISTINCT * 
from(
select ts_resample,Val2 as Val2
FROM Sensor_Data.Table1) as t) as tt

对当前数据中的三个单元使用上述查询得到的结果如下：

ts_resample             unit    Val1    Val2
2020-12-01 00:00:00     unit1   0.8974  10.485
2020-12-01 00:00:00     unit2   0.8974  10.485
2020-12-01 00:00:00     unit3   0.8974  10.485
2020-12-01 00:05:00     unit1   0.9041  11.854
2020-12-01 00:05:00     unit2   0.9041  11.854
2020-12-01 00:05:00     unit3   0.9041  11.854

实际上，我想获得每个单位不同的每个单位的最后一个值。如下图：

ts_resample             unit    Val1    Val2
2020-12-01 00:00:00     unit1   0.8974  10.485
2020-12-01 00:00:00     unit2   0.9014  11.954
2020-12-01 00:00:00     unit3   0.7854  10.821
2020-12-01 00:05:00     unit1   0.9841  11.125
2020-12-01 00:05:00     unit2   0.8742  10.963
2020-12-01 00:05:00     unit3   0.9632  11.784

有人能告诉我我的查询有什么问题吗？

谢谢

解决方法

我通过使用 ts_resample 在分区中提供单元信息解决了这个问题。最终解决方案如下：

SELECT DISTINCT * 
from(
select ts_resample,unit,last_value(Val1) over (partition by ts_resample,unit order by ts rows between unbounded preceding and unbounded following) as Val1,last_value(Val2) over (partition by ts_resample,unit order by ts rows between unbounded preceding and unbounded following) as Val2
from (
SELECT cast(cast(unix_timestamp(cast(ts/1000 as TIMESTAMP))/300 as bigint)*300 as TIMESTAMP) as ts_resample,ts as ts,unit as unit,Val1 as Val1,Val2 as Val2
FROM Sensor_Data.Table1) as t) as tt

在此之后，我得到了我想要的结果并在我的问题中显示。