如何在SQL中使用时间范围计算最大行数？

Question

This is the table in Postgresql:

mydb=# \d login_log 
              Table "public.login_log"
   Column    |           Type           | Modifiers 
-------------+--------------------------+-----------
 id          | integer                  | 
 login_start | timestamp with time zone | 
 login_end   | timestamp with time zone | 

some rows:

1 | 2015-03-19 10:00:00  | 2015-03-19 13:30:00
2 | 2015-03-19 10:20:00  | 2015-03-19 13:20:00
3 | 2015-03-19 13:00:00  | 2015-03-19 16:00:00
4 | 2015-03-19 13:10:00  | 2015-03-19 16:00:00
5 | 2015-03-19 14:30:00  | 2015-03-19 15:30:00
6 | 2015-03-19 15:00:00  | 2015-03-19 15:30:00
7 | 2015-03-19 12:00:00  | 2015-03-19 18:00:00

I need a SQL to count out in which time range there is the max logged users.

with the example above, the result is:

in time range: 2015-03-19 13:10:00 ~ 2015-03-19 13:20:00 , 5 users logged in. (1, 2, 3, 4, 7)

Answer 1

Use range types (construct them "on the fly"). They offer quite a few helpful functions and operators . You would only need to define a custom aggregate , which will provide you the overall intersection. So - you would end up with something like this:

with common as (
  select (intersection(tsrange(login_start, login_end))) as period 
  from login_log 
)
select 
  -- common.period, 
  -- array_agg(id) 
  *
from common, login_log
WHERE tsrange(login_start, login_end) && common.period
-- GROUP BY common.period
/* 
for some reason, when uncommenting the "^--..." lines, 
and commenting the "*" one - sqlfiddle shows an empty result. 
Nevertheless it works on my local posgres...
*/ 

See the working example: http://sqlfiddle.com/#!15/0c9c6/10

Answer 2

Find different timestamps of interest using UNION ALL, count number of active users at these timestamps:

select ts,
       (select count(*) from login_log t2
        where timestamps.ts between t2.login_start and t2.login_end) as count
from (select login_start as ts
      from login_log
      union all
      select login_end
      from login_log) as timestamps
order by count desc
fetch first 1 row only

Finally order descending and pick the highest value only!

(From a non Postgresql user, so some details may be wrong... Please comment if that's the case and I'll edit!)