Hi， I would like to introduce the current requirements first. Imagine when the user executes an SQL:
Select last_value(*),max_time(*) from root.sg1.* The result is: +-------------+-------------+------------+---------------+-----------+-------------+ ｜last_value(s1) | max_time(s1)| last_time(s2) | max_value(s2) | last_time(N) | max_value(N) | +-------------+-------------+------------+---------------+-----------+-------------+ We all know that in the Internet of things, a device typically contains multiple sensors, and a set of requirements contains multiple devices. The result set above will have N * M columns, sometimes up to 50,000 columns, which is a very unfriendly way to return data. So the way we expect to return the data is： +-------+-------------+-------------+ ｜ path | max_time |. last_value. | +-------+-------------+-------------+ | s1. |. 1 | 1 | +-------+-------------+-------------+ |. s2 | 2 | 2 | +-------+-------------+-------------+ ……. Business processes max_time and last_value in many cases come in pairs, so we would like to see a simplified query function named last, so we have listed several candidates for SQL to support the new return format conversion. Which one do you prefer? 1. select last(s1), last(s2) from root.sg.d 2. select last(s1), last(s2) from root.sg.d align by sensor 3. select last(s1, s2) from root.sg.d 4. select last s1, s2 from root.sg.d 5. last s1, s2 from root.sg.d 6. select s1, s2 from root.sg.d last 7. select s1, s2 from root.sg.d onlylast Thanks --- Dawei Liu > 2020年2月21日 下午8:09，Jialin Qiao <[email protected]> 写道： > > Hi, > > Actually we do not have "last" syntax now. There are two ways to treat last: > > (1) Treat it as a special case of raw data query > (2)Treat it as a aggregation query > > Before deciding, let's consider two questions: > > (1) Is last an aggregation function or is the result format of last the same > with aggregation queries? > > I thinks it's not. One aggregation function should return one column. For > example, the result of "select count(s1) from xx" has one column: count(s1). > However, the last should return three columns: path, time, value > > (2) Could "last" be used with other aggregations at the same time? > > NO, if users issue a last query, he should query the last of all selected > series. "select last(s1), count(s2)" is not allowed. > i.e., it could only be this form: select last(s1), last(s2), last(s3), > last(s4) from root.sg.d1. And if so, why we duplicate last so many times... > > Therefor, I prefer treat last as a special case of raw data query. i.e., 4. > select last s1, s2 from root.sg.d1 > > Thanks, > -- > Jialin Qiao > School of Software, Tsinghua University > > 乔嘉林 > 清华大学 软件学院 > >> -----原始邮件----- >> 发件人: "Xiangdong Huang" <[email protected]> >> 发送时间: 2020-02-21 19:44:46 (星期五) >> 收件人: [email protected] >> 抄送: >> 主题: Re: Last Query syntax vote >> >> Hi, >> >> +1 for select last(s1), last(s2) from root.sg.d >> >> I think this vote description does not show why we need a new syntax and >> the data result format we will get... >> >> For example, the first SQL is a common query statement, but why it is not >> suitable and you want a new syntax? >> >> It is hard to let other guys joining the vote. >> >> Best, >> ----------------------------------- >> Xiangdong Huang >> School of Software, Tsinghua University >> >> 黄向东 >> 清华大学 软件学院 >> >> >> Dawei Liu <[email protected]> 于2020年2月21日周五 下午5:35写道： >> >>> Hi, >>> >>> I’d vote for >>> >>> [2] select last(s1), last(s2) from root.sg.d align by sensor >>> >>> It looks simple, it's easy to use, it's human logic >>> >>> >>> Thanks >>> --- >>> Dawei Liu >>> >>> >>> >>>> 2020年2月21日 下午5:26，Wei Shao <[email protected]> 写道： >>>> >>>> Hi, >>>> >>>> >>>> I'd vote for >>>> [4]. select last s1, s2 from root.sg.d >>>> >>>> >>>> Thanks, >>>> Wei >>>> >>>> >>>> >>>> At 2020-02-21 17:08:14, "Wei Shao" <[email protected]> wrote: >>>>> Hi, >>>>> >>>>> >>>>> We are to determine the exact syntax of Last Query for IoTDB. Last >>> query is to select the latest time-value pair result for a given timeseries. >>>>> You can choose from the following syntax options. >>>>> 1. select last(s1), last(s2) from root.sg.d >>>>> 2. select last(s1), last(s2) from root.sg.d align by sensor >>>>> 3. select last(s1, s2) from root.sg.d >>>>> 4. select last s1, s2 from root.sg.d >>>>> 5. last s1, s2 from root.sg.d >>>>> 6. select s1, s2 from root.sg.d last >>>>> 7. select s1, s2 from root.sg.d onlylast >>>>> >>>>> >>>>> Please vote for the option you are comfortable with. >>>>> >>>>> >>>>> Thanks, >>>>> Wei >>> >>>
