Hi,
    Do you use the COUNT_DISTINCT(HLL) or the COUNT_DISTINCT(Bitmap) ? I 
recommend you to use COUNT_DISTINCT(HLL) if it is not your situation.
    On the another, please check log of receiver side and find if there are any 
error/exceptions ? 

--

Best wishes to you ! 
From :Xiaoxiang Yu




在 2020-10-21 10:59:04,"风起雨生" <526344...@qq.com> 写道:

各位老师好:
       以下是我们在使用kylin 3.1.0 版本中遇到的奇怪问题。
概要:
      当使用real-time 
实时指标时,在低频请求的情况下,去重指标数值是正常的,当并发请求时,例如多位运营人员打开大屏,去重指标会逐渐变大,甚至会超过count(*)指标,这个异常值的出现频率有一定的规律,似乎每五分钟出现一次,当查询频率降低,后续的去重指标会逐渐正常,但是之前的异常值会被持久化,不可恢复。
查询SQL:select minute_start, count(distinct device_id) as uv ,count(*) as pv from 
ONLINE_BZ_HEART_V3 where day_start='2020-10-20' group by minute_start order by 
minute_start desc ;
实时表结构:   


结果列表截图:


以上。
      

Reply via email to