MySQL - Peak visit counts by datetime period

Question

I have a visits table (id int, start datetime, end datetime), and I you wish to track peak visit counts.

Example data:

+------+---------------------+----         


        

Answer 1

This is not very efficient, but it'll give you your result:

select U.dt1 as date-time-1, DATE_ADD(U.dt2,INTERVAL -1 SECOND) as date-time-2, 
    (select count(id) from Visits where 
    (dt1 >= u.dt1 and dt1<U.dt2)  --(dt1)dt2
    or (dt1<u.dt1 and dt2>=u.dt2)   -- dt1()dt2
    --or (dt2 >= u.dt1 and dt2<U.dt2) -- dt1(dt2) (comment this line to get your result which I believe is incorrect)
    ) as count 
from (
    select A.dt1 as dt1, (
        select min(M.dt) from ( select min(dt2) as dt from Visits where dt2 > A.dt1 union select min(dt1) as dt from Visits where dt1 > A.dt1) M
    ) as dt2 from Visits A
union 
    select B.dt2 as dt1, (
        select min(M.dt) from ( select min(dt2) as dt from Visits where dt2 > b.dt2 union select min(dt1) as dt from Visits where dt1 > b.dt2) M 
    ) as dt2 from Visits b where B.dt2 <> (select max(dt2) from Visits)
) U 

I've commented the condition checking to see if a visit starts before a range and ends within it to get the same result set as you but I believe you should consider this.

Answer 2

So to make this work you need to understand your periods and the overlapping between then. We agreed in comments that to make it work in the right way you should have from the second row on, at least one second added to the prior end. In order to understand that I will add a graph of how your periods will be and right bellow the periods from the visits table so you will see in the end that the times (since all periods are same day and hour, I will leave just minutes and seconds on the graph)

13:00        13:30         14:26      14:39
^            ^             ^          ^           
|------------||-----------||----------||-----------|
              |_ 13:31    |_ 14:25     |_ 14:40    |_ 20:05

--and in your table
13:00                                              20:05
^                                                  ^
|--------------------------------------------------|
             |------------|           14:39        20:05
             |_ 13:30     |_ 14:25    ^            ^
                                      |------------|

To achieve such periods table I've created a VIEW to facilitate the query, here is the code to it:

create or replace view vw_times as
  select dtstart as dt from visits
   UNION
  select dtend as dt from visits;

The purpose of this view is to identify all dates starts and ends of your given periods.

And here is the query that will produce such periods scenarios:

SELECT case when cnt>1 
           then date_add(dtstart,interval 1 second) 
           else dtstart 
           end as dtstart,
         dtend
  from (SELECT dtstart, 
               dtend, 
               @ct:=@ct+1 as cnt
          FROM ( SELECT t1.dt as dtstart,
                        (select min(x.dt) 
                           from vw_times as x
                          where x.dt > t1.dt
                         ) as dtend
                   FROM vw_times t1,
                        (select @ct := 0) as cttab
                  ORDER BY t1.dt
                ) t2
          WHERE dtend is not null
        ) as t3

And from it you can LEFT JOIN your table to find the overlapping periods like this:

SELECT times.dtstart, times.dtend, count(*)
  FROM (SELECT case when cnt>1 
                 then date_add(dtstart,interval 1 second) 
                 else dtstart 
                 end as dtstart,
               dtend
        from (SELECT dtstart, 
                     dtend, 
                     @ct:=@ct+1 as cnt
                FROM ( SELECT t1.dt as dtstart,
                              (select min(x.dt) 
                                 from vw_times as x
                                where x.dt > t1.dt
                               ) as dtend
                         FROM vw_times t1,
                              (select @ct := 0) as cttab
                        ORDER BY t1.dt
                      ) t2
                WHERE dtend is not null
              ) as t3
       ) as times 
       LEFT JOIN visits v 
              ON (    times.dtstart >= v.dtstart
                  AND times.dtend <= v.dtend)
 GROUP BY times.dtstart, times.dtend

This will result in:

dtstart                      dtend                   count(*)
July, 04 2016 19:13:00       July, 04 2016 19:13:30     1
July, 04 2016 19:13:31       July, 04 2016 19:14:25     2
July, 04 2016 19:14:26       July, 04 2016 19:14:39     1
July, 04 2016 19:14:40       July, 04 2016 19:20:05     2

See it working here: http://sqlfiddle.com/#!9/3509ff/10

EDIT

Since you added a comment with the final result, it would make the final query even smaller:

SELECT times.dtstart, 
       case when times.dtend = vmax.maxend
            then date_add(times.dtend, interval 1 second)
            else times.dtend
            end as dtend, 
       count(*)
  FROM  (SELECT dtstart, 
               dtend
          FROM ( SELECT t1.dt as dtstart,
                        (select min(date_sub(x.dt, interval 1 second)) 
                           from vw_times as x
                          where x.dt > t1.dt
                        ) as dtend
                   FROM vw_times t1
                  ORDER BY t1.dt
               ) t2
        WHERE t2.dtend is not null
       ) as times 
       LEFT JOIN visits as v
              ON (    times.dtstart >= v.dtstart
                  AND times.dtend <= v.dtend)
       LEFT JOIN (select max(date_sub(v.dtend, interval 1 second)) as maxend
                    from visits v) vmax
              ON ( times.dtend = vmax.maxend )
 GROUP BY times.dtstart, 
          case when times.dtend = vmax.maxend
            then date_add(times.dtend, interval 1 second)
            else times.dtend
            end

That will result in:

dtstart                   dtend                 count(*)
July, 04 2016 19:13:00    2016-07-04 19:13:29    1
July, 04 2016 19:13:30    2016-07-04 19:14:24    2
July, 04 2016 19:14:25    2016-07-04 19:14:38    1
July, 04 2016 19:14:39    2016-07-04 19:20:05    2

See it here: http://sqlfiddle.com/#!9/3509ff/24