用 udagg 应该能比较完美的解决你的问题 ^.^

On 12/6/19, Djeng Lee <lee.ro...@gmail.com> wrote:
> 存在查询
>
> 在 2019/12/5 下午4:06,“Yuan,Youjun”<yuanyou...@baidu.com> 写入:
>
>     Count=0的窗口如何能得到呢?没有数据就没有产出。
>     然而可以同rows
> over窗口,将两个前后窗口的sum-当前的count,可以间接得到两个窗口的count是否相等。同时辅以前后窗口时间的差,来辅助判断。
>
> 最终在自定义函数last_value_str/first_value_str的帮助下,勉强得以实现(尚不完美,可能出现连续的ONLINE的输出)
>     下面是我的SQL,仅供参考:
>
>     INSERT INTO mysink
>     SELECT userid, lastts, case when preCnt <= 0 OR tsdiff > 10 THEN
> 'ONLINE' ELSE 'offline' END AS status
>     FROM (
>       SELECT curCnt, preCnt, lastts, firstts, userid, yeardiff * 31536000 +
> monthdiff * 2678400 + daydiff * 86400 + hourdiff * 3600 + mindiff * 60 +
> seconddiff AS tsdiff
>       FROM (
>               SELECT curCnt, preCnt,
>                       cast(substring(lastts, 1, 4) as bigint) - 
> cast(substring(firstts, 1,
> 4) as bigint) as yeardiff,
>                       cast(substring(lastts, 6, 2) as bigint) - 
> cast(substring(firstts, 6,
> 2) as bigint) as monthdiff,
>                       cast(substring(lastts, 9, 2) as bigint) - 
> cast(substring(firstts, 9,
> 2) as bigint) as daydiff,
>                       cast(substring(lastts, 12, 2) as bigint) - 
> cast(substring(firstts,
> 12, 2) as bigint) as hourdiff,
>                       cast(substring(lastts, 15, 2) as bigint) - 
> cast(substring(firstts,
> 15, 2) as bigint) as mindiff,
>                       cast(substring(lastts, 18, 2) as bigint) - 
> cast(substring(firstts,
> 18, 2) as bigint) as seconddiff,
>                       lastts, firstts, userid
>               FROM (
>                       SELECT userid, cnt AS curCnt, sum(cnt) OVER w - cnt as 
> preCnt,
> last_value_str(ts0) OVER w as lastts, first_value_str(ts0) OVER w as firstts
>
>                       FROM (
>                               SELECT HOP_PROCTIME(rowtime, interval '5' 
> second, interval '10'
> second) AS rowtime, count(*) as cnt, userid, last_value_str(cast(rowtime AS
> varchar)) AS ts0
>                               FROM mysrc
>                               GROUP BY userid, hop(rowtime, interval '5' 
> second, interval '10'
> second)
>                       )
>                       WINDOW w as (PARTITION BY userid ORDER BY rowtime ROWS 
> BETWEEN 1
> PRECEDING AND CURRENT ROW)
>               )
>       )
>       WHERE (preCnt <= 0 OR yeardiff * 31536000 + monthdiff * 2678400 +
> daydiff * 86400 + hourdiff * 3600 + mindiff * 60 + seconddiff > 10) OR
> (curCnt = preCnt AND lastts = lastts)
>     )
>
>     -----邮件原件-----
>     发件人: 1193216154 <1193216...@qq.com>
>     发送时间: Thursday, December 5, 2019 2:43 PM
>     收件人: user-zh <user-zh@flink.apache.org>
>     主题: 回复: 回复: 回复:如何用SQL表达对设备离在线监控
>
>     可以考虑用flink cep,应该可以解决你的问题。
>
>
>     ------------------&nbsp;原始邮件&nbsp;------------------
>     发件人:&nbsp;"Djeng Lee"<lee.ro...@gmail.com&gt;;
>     发送时间:&nbsp;2019年12月5日(星期四) 下午2:40
>     收件人:&nbsp;"user-zh@flink.apache.org"<user-zh@flink.apache.org&gt;;
>
>     主题:&nbsp;Re: 回复: 回复:如何用SQL表达对设备离在线监控
>
>
>
>     上线时间,前n窗口count == 0 , 后n窗口count &gt; 1。说明是上线。由此得出上线时间.
>     离线时间,前n 窗口count&gt;=1, 后n窗口count==0,说明下线,由此可得下线时间。
>     前n后n都&gt;1 作为心跳维持。
>
>
>
>     在 2019/12/5 下午2:06,“Yuan,Youjun”<yuanyou...@baidu.com&gt; 写入:
>
>     &nbsp;&nbsp;&nbsp; 谢谢你的回复。
>     &nbsp;&nbsp;&nbsp;
> 这种方案比较有意思,只是还不能区分设备第一次心跳产生的count=1的消息(上线),和设备最后一次心跳产生的count=1的消息(下线)
>     &nbsp;&nbsp;&nbsp;
>     &nbsp;&nbsp;&nbsp; -----邮件原件-----
>     &nbsp;&nbsp;&nbsp; 发件人: 1193216154 <1193216...@qq.com&gt;
> &nbsp;&nbsp;&nbsp; 发送时间: Wednesday, December 4, 2019 9:39 PM
> &nbsp;&nbsp;&nbsp; 收件人: user-zh <user-zh@flink.apache.org&gt;
> &nbsp;&nbsp;&nbsp; 主题: 回复:如何用SQL表达对设备离在线监控
>     &nbsp;&nbsp;&nbsp;
>     &nbsp;&nbsp;&nbsp; 设定一个滑动窗口,窗口大小大于等于2n,滑动间隔大于等于n,若一次窗口结算,count
> 大于等于2,则在线,否则下线
>     &nbsp;&nbsp;&nbsp;
>     &nbsp;&nbsp;&nbsp; ---原始邮件---
>     &nbsp;&nbsp;&nbsp; 发件人: "Yuan,Youjun"<yuanyou...@baidu.com&amp;gt;
>     &nbsp;&nbsp;&nbsp; 发送时间: 2019年12月4日(周三) 晚上6:49
>     &nbsp;&nbsp;&nbsp; 收件人:
> "user-zh@flink.apache.org"<user-zh@flink.apache.org&amp;gt;;
>     &nbsp;&nbsp;&nbsp; 主题: 如何用SQL表达对设备离在线监控
>     &nbsp;&nbsp;&nbsp;
>     &nbsp;&nbsp;&nbsp;
>     &nbsp;&nbsp;&nbsp; Hi all,
>     &nbsp;&nbsp;&nbsp;
>     &nbsp;&nbsp;&nbsp;
> 假设我们有很多设备,设备正常工作期间会定时发送心跳到服务器。如果某个设备在超过N分钟的时间内,没有发送任何心跳到服务器,服务器会认为设备已经离线。直到下一次心跳,才判定设备为在线。
>     &nbsp;&nbsp;&nbsp;
> 需求:在判定设备离线时,产出一条设备离线消息;在设备经过一次离线后,第一次心跳时,产出一条设备上线的消息;
>     &nbsp;&nbsp;&nbsp; 假设设备上报的消息包含当前时间(ts)和设备id(deviceid):
>     &nbsp;&nbsp;&nbsp; 1575456144,dev1
>     &nbsp;&nbsp;&nbsp; 1575456146,dev2
>     &nbsp;&nbsp;&nbsp; 1575456147,dev1
>     &nbsp;&nbsp;&nbsp; ….
>     &nbsp;&nbsp;&nbsp;
>     &nbsp;&nbsp;&nbsp; 产出的离在线消息分别格式如下(第一列为设备离在线时间):
>     &nbsp;&nbsp;&nbsp; 1575456158,dev1,offline &nbsp;&nbsp;&nbsp;
> 1575456169,dev2,online &nbsp;&nbsp;&nbsp; &nbsp;&nbsp;&nbsp;
> 能否用一条SQL来定义这个作业呢?
>     &nbsp;&nbsp;&nbsp;
>     &nbsp;&nbsp;&nbsp; 谢谢!
>     &nbsp;&nbsp;&nbsp; 袁尤军
>     &nbsp;&nbsp;&nbsp;
>
>

回复