我可以查询clickhouse中累积列的每小时增量吗?

xie*_*ong 8 clickhouse

我想节省事件时间和每 30 秒产生的总电量。总金额不会每次都重置为零。这只是从仪表开始到现在的总数,而不是30秒内产生的总数。

有没有办法查询生成的电柱总量的每日、每周或每月聚合(也许不仅仅是 sum 或 avg)?

还是通过设计一张AggregatingMergeTree桌子?

我不需要保留每条记录,只需要每天、每周和每月的聚合。

例如 :

create table meter_record (
   event_time Datetime,
   generated_total Int64
)
Run Code Online (Sandbox Code Playgroud)

vla*_*mir 6

更新

喜欢使用SimpleAggregateFunction代替的聚合函数为简单的功能,如中值平均最小最大加快聚集体的计算。


让我们建议您需要计算该表的中值平均值离散聚合:

CREATE TABLE meter_record (
   event_time Datetime,
   generated_total Int64  
)
ENGINE = MergeTree
PARTITION BY (toYYYYMM(event_time))
ORDER BY (event_time);
Run Code Online (Sandbox Code Playgroud)

使用AggregatingMergeTree计算所需的聚合:

CREATE MATERIALIZED VIEW meter_aggregates_mv
ENGINE = AggregatingMergeTree()
PARTITION BY toYYYYMM(day)
ORDER BY (day) 
AS
SELECT  
  toDate(toStartOfDay(event_time)) AS day, 
  /* aggregates to calculate the day's section left and right endpoints */
  minState(generated_total) min_generated_total,
  maxState(generated_total) max_generated_total,
  /* specific aggregates */
  medianState(generated_total) AS totalMedian,
  avgState(generated_total) AS totalAvg,
  varPopState(generated_total) AS totalDispersion
  /* ... */
FROM meter_record
GROUP BY day;
Run Code Online (Sandbox Code Playgroud)

要获得所需的每日/每周/每月(以及任何基于日的聚合,如季度年度)聚合,请使用以下查询:

/* daily report */
SELECT 
  day,
  minMerge(min_generated_total) min_generated_total,
  maxMerge(max_generated_total) max_generated_total,
  medianMerge(totalMedian) AS totalMedian,
  avgMerge(totalAvg) AS totalAvg,
  varPopMerge(totalDispersion) AS totalDispersion
FROM meter_aggregates_mv
/*WHERE day >= '2019-02-05' and day < '2019-07-01'*/
GROUP BY day;

/* weekly report */
SELECT 
  toStartOfWeek(day, 1) monday,
  minMerge(min_generated_total) min_generated_total,
  maxMerge(max_generated_total) max_generated_total,
  medianMerge(totalMedian) AS totalMedian,
  avgMerge(totalAvg) AS totalAvg,
  varPopMerge(totalDispersion) AS totalDispersion
FROM meter_aggregates_mv
/*WHERE day >= '2019-02-05' and day < '2019-07-01'*/
GROUP BY monday;

/* monthly report */
SELECT 
  toStartOfMonth(day) month,
  minMerge(min_generated_total) min_generated_total,
  maxMerge(max_generated_total) max_generated_total,
  medianMerge(totalMedian) AS totalMedian,
  avgMerge(totalAvg) AS totalAvg,
  varPopMerge(totalDispersion) AS totalDispersion
FROM meter_aggregates_mv
/*WHERE day >= '2019-02-05' and day < '2019-07-01'*/
GROUP BY month;

/* get daily / weekly / monthly reports in one query (thanks @Denis Zhuravlev for advise) */
SELECT
  day,
  toStartOfWeek(day, 1) AS week,
  toStartOfMonth(day) AS month,
  minMerge(min_generated_total) min_generated_total,
  maxMerge(max_generated_total) max_generated_total,
  medianMerge(totalMedian) AS totalMedian,
  avgMerge(totalAvg) AS totalAvg,
  varPopMerge(totalDispersion) AS totalDispersion
FROM meter_aggregates_mv
/*WHERE (day >= '2019-05-01') AND (day < '2019-06-01')*/
GROUP BY month, week, day WITH ROLLUP
ORDER BY day, week, month;
Run Code Online (Sandbox Code Playgroud)

评论:

  • 您指出原始数据不需要您只是聚合,因此您可以将meter_record -table 的引擎设置为Null,手动清理meter_record(请参阅DROP PARTITION)或定义TTL以自动执行此操作

  • 删除原始数据是不好的做法,因为它无法计算历史数据的新聚合或恢复现有聚合等

  • 物化视图meter_aggregates_mv将只包含在创建视图后插入到表meter_record 中的数据。要更改此行为,在视图定义中使用POPULATE