Question

This question has been asked before but I am facing a slightly different problem.

I have a table which logs events and stores their timestamps (as datetime). I need to be able to break up time into chunks and get number of events that occurred in that interval. The interval can be custom (Say from 5 minutes to 1 hour and even beyond).

The obvious solution is to convert the datetime to unix_timestamp divide it by number of seconds in the interval, take its floor function and multiply it back by the number of seconds. Finally convert the unix_timestamp back to the datetime format.

This works fine for small intervals.

select 
from_unixtime(floor(unix_timestamp(event.timestamp)/300)*300) as start_time,
count(*) as total 
from event 
where timestamp>='2012-08-03 00:00:00' 
group by start_time;

This gives the correct output

+---------------------+-------+
| start_time          | total |
+---------------------+-------+
| 2012-08-03 00:00:00 |    11 |
| 2012-08-03 00:05:00 |     4 |
| 2012-08-03 00:10:00 |     4 |
| 2012-08-03 00:15:00 |     7 |
| 2012-08-03 00:20:00 |     8 |
| 2012-08-03 00:25:00 |     1 |
| 2012-08-03 00:30:00 |     1 |
| 2012-08-03 00:35:00 |     3 |
| 2012-08-03 00:40:00 |     3 |
| 2012-08-03 00:45:00 |     5 |
~~~~~OUTPUT SNIPPED~~~~~~~~~~~~

But if I increase the interval to say 1 hour (3600 sec)

mysql> select from_unixtime(floor(unix_timestamp(event.timestamp)/3600)*3600) as start_time, count(*) as total from event where timestamp>='2012-08-03 00:00:00' group by start_time;
+---------------------+-------+
| start_time          | total |
+---------------------+-------+
| 2012-08-02 23:30:00 |    35 |
| 2012-08-03 00:30:00 |    30 |
| 2012-08-03 01:30:00 |    12 |
| 2012-08-03 02:30:00 |    18 |
| 2012-08-03 03:30:00 |    12 |
| 2012-08-03 04:30:00 |     4 |
| 2012-08-03 05:30:00 |     3 |
| 2012-08-03 06:30:00 |    13 |
| 2012-08-03 07:30:00 |   269 |
| 2012-08-03 08:30:00 |   681 |
| 2012-08-03 09:30:00 |  1523 |
| 2012-08-03 10:30:00 |   911 |
+---------------------+-------+

The reason, as far as I could gauge, for the boundaries not being set properly is that unix_timestamp will convert time from my local timezone (GMT + 0530) to UTC and then output the numerical value.

So a value like 2012-08-03 00:00:00 will actually be 2012-08-02 18:30:00. Dividing and using floor will set the minutes part to 00. But when I use from_unixtime, it will convert it back to GMT + 0530 and hence give me intervals that begin at 30 mins.

How do I ensure the query works correctly irrespective of the timezone? I use MySQL 5.1.52 so to_seconds() is not available

EDIT:
The query should also fire correctly irrespective of the interval (can be hours, minutes, days). A generic solution would be appreciated

Answer 1

You can use TIMESTAMPDIFF to group by intervals of time:

For a specified interval of hours, you can use:

SELECT   '2012-08-03 00:00:00' + 
         INTERVAL FLOOR(TIMESTAMPDIFF(HOUR, '2012-08-03 00:00:00', timestamp) / <n>) * <n> HOUR AS start_time,
         COUNT(*) AS total 
FROM     event 
WHERE    timestamp >= '2012-08-03 00:00:00'
GROUP BY start_time

Replace the occurances of 2012-08-03 00:00:00 with your minimum input date.

<n> is your specified interval in hours (every 2 hours, 3 hours, etc.), and you can do the same for minutes:

SELECT   '2012-08-03 00:00:00' + 
         INTERVAL FLOOR(TIMESTAMPDIFF(MINUTE, '2012-08-03 00:00:00', timestamp) / <n>) * <n> MINUTE AS start_time,
         COUNT(*) AS total 
FROM     event 
WHERE    timestamp >= '2012-08-03 00:00:00'
GROUP BY start_time

Where <n> is your specified interval in minutes (every 45 minutes, 90 minutes, etc).

Be sure you’re passing in your minimum input date (in this example 2012-08-03 00:00:00) as the second parameter to TIMESTAMPDIFF.

---

EDIT: If you don’t want to worry about which interval unit to pick in the TIMESTAMPDIFF function, then of course just do the interval by seconds (300 = 5 minutes, 3600 = 1 hour, 7200 = 2 hours, etc.)

SELECT   '2012-08-03 00:00:00' + 
         INTERVAL FLOOR(TIMESTAMPDIFF(SECOND, '2012-08-03 00:00:00', timestamp) / <n>) * <n> SECOND AS start_time,
         COUNT(*) AS total 
FROM     event 
WHERE    timestamp >= '2012-08-03 00:00:00'
GROUP BY start_time

EDIT2: To address your comment pertaining to reducing the number of areas in the statement where you have to pass in your minimum parameter date, you can use:

SELECT   b.mindate + 
         INTERVAL FLOOR(TIMESTAMPDIFF(SECOND, b.mindate, timestamp) / <n>) * <n> SECOND AS start_time,
         COUNT(*) AS total 
FROM     event 
JOIN     (SELECT '2012-08-03 00:00:00' AS mindate) b ON timestamp >= b.mindate
GROUP BY start_time

And simply pass in your minimum datetime parameter once into the join subselect.

You can even make a second column in the join subselect for your seconds interval (e.g. 3600) and name the column something like secinterval… then change the <n>‘s to b.secinterval, so you only have to pass in your minimum date parameter AND interval one time each.

---

SQLFiddle Demo