事件发生时识别日期范围的T-SQL查询

Question

I am trying to identify the date ranges when an organization is on our "monitor" list.

My data looks like this:

OrgCode OrgName           ReviewDate    MonitorList
8000    Organization A    3/6/2014      1
8000    Organization A    6/4/2014      1
8000    Organization A    9/4/2014      1
8000    Organization A    12/4/2014     0
8000    Organization A    3/5/2015      1
8000    Organization A    6/4/2015      1
8000    Organization A    9/16/2015     1
8000    Organization A    12/16/2015    1
8000    Organization A    3/9/2016      1
8000    Organization A    6/2/2016      1
8000    Organization A    9/8/2016      1
8000    Organization A    12/8/2016     1
8000    Organization A    3/9/2017      0
8000    Organization A    6/14/2018     0

The query output I'm seeking looks like this:

OrgCode OrgName           MonitorStartDate  MonitorEndDate
8000    Organization A    3/6/2014          12/4/2014
8000    Organization A    3/5/2015          3/9/2017

This organization, Organization A, has appeared on our monitor list twice: from 3/6/2014 to 12/4/2014, and 3/5/2015 to 3/9/2017.

I've tried to accomplish this in a few ways, including,

• Varieties of LEAD() and LAG() ; and,
• GROUP BY OrgCode, OrgName, MonitorList and defining MonitorStartDate as MIN(ReviewDate) and MonitorEndDate as MAX(ReviewDate) .

The second method did not account for the fact that these organizations may be on/off the monitor list multiple times. I still think some combinations of LEAD() or LAG() might work; but, not by themselves.

Any guidance you folks can provide would be great and thanks for the help!

Answer 1

Use a running sum to classify rows into groups re-setting the value when 0 is encountered and lead to get the next row's date because the end date has to be from the first 0 encountered. Then use min and max on the corresponding columns with necessary groupings.

select orgcode,orgname
,min(case when monitorlist=1 then reviewdate end) as monitorstartdate
,max(next_dt) as monitorenddate
from (select t.*,
      sum(case when monitorlist=0 then 1 else 0 end) over(partition by orgcode order by reviewdate) as grp,
      lead(reviewdate) over(partition by orgcode order by reviewdate) as next_dt
      from tbl t
     ) t
group by orgcode,orgname,grp
having max(cast(monitorlist as int))=1

Answer 2

With this query

select orgcode,orgname,format(min(reviewdate),'M/d/yyyy') as monitorstartdate,format(max(next_dt),'M/d/yyyy') as monitorenddate
from (select t.*,
   sum(case when monitorlist=0 then 1 else 0 end) 
     over(partition by orgcode order by reviewdate) as grp,
     lead(reviewdate) over(partition by orgcode order by reviewdate) as next_dt
   from tbl t
   ) t
group by orgcode,orgname,grp,MonitorList
having MonitorList = 1

the result is as follows

orgcode     orgname             monitorstartdate    monitorenddate
8000        "Organization A"    3/6/2014            12/4/2014
8000        "Organization A"    3/5/2015            3/9/2017

The Fiddle link is here if people want to verify.

Answer 3

You can identify the groups by counting the number of 0's on or after each row. The rest is just aggregation:

select orgcode, orgname, min(ReviewDate) as MonitorStartDate,
       coalesce(min(case when monitorlist = 0 then ReviewDate end),
                max(ReviewDate)
               ) as MontiroEndDate
from (select t.*,
             sum(case when monitorlist = 0 then 1 else 0 end) over (partition by orgcode order by reviewdate desc) as grp             
      from t
     ) t
group by orgcode, orgname, grp
having max(monitorlist) = 1;

The logic for the end date is aa bit tricky:

• It is the ReviewDate of the "0" record.
• If there are none, the latest ReviewDate is used.

Here is a SQL Fiddle demonstrating it.