如何按日期分组，但在选择语句中使用日期作为条件？

Question

I'm really not sure how to title this question, so sorry about that..

I have a site where users can "watch" products. When they start watching a product, a row is inserted into product_tracking , which looks like this:

+-----------------------+------------+------+-----+---------+----------------+
| Field                 | Type       | Null | Key | Default | Extra          |
+-----------------------+------------+------+-----+---------+----------------+
| id                    | bigint(11) | NO   | PRI | NULL    | auto_increment |
| u_id                  | int(11)    | NO   | MUL | NULL    |                |
| p_id                  | bigint(11) | NO   | MUL | NULL    |                |
| date_started_tracking | datetime   | NO   |     | NULL    |                |
| date_stopped_tracking | datetime   | YES  |     | NULL    |                |
+-----------------------+------------+------+-----+---------+----------------+

The datetime the user starts watching the product is inserted into date_started_tracking when the row is inserted, and date_stopped_tracking is null by default. When the user stops watching a product, date_stopped_tracking is updated with the current datetime.

I want to have a chart on my site showing the number of products being tracked over time. Something like this:

# tracks
|                       __/
|               ___    /      
|              /   \__/     
|             /       
|            /
|       ____/
|      /
|  ___/
| /
|/___________________________ date (grouped by day, week, or month)

I don't know how to write a query to retrieve the information required to make the chart.

What makes it difficult is that I want to group rows by day, week, or month, but only include rows in a group if date_started_tracking is <= the grouping date and date_stopped_tracking is null or > the grouping date (ie select rows showing products being tracked at that date).

In other words, for each day, week, or month (depending on the desired granularity), the number of products being watched by users at that time should be returned.

---

My attempt at a solution

The only idea I've come up with is creating a table of dates and then writing a query such as:

SELECT WEEK(d.date) as week, 
       (SELECT COUNT(pt.id) 
        FROM product_tracking pt
        WHERE pt.date_started_tracking <= d.date AND (pt.date_stopped_tracking IS NULL OR pt.date_stopped_tracking > d.date)
        ) as numWatches
FROM dates d
GROUP BY week
ORDER BY week ASC

I think this will work, but it requires creating a table of dates, which seems kind of retarded.

Is there a better way?

Answer 1

Your method works, but it won't scale very well. An alternative approach is to keep track of the tracking day-by-day using variables for the cumulative sums. This looks like:

select dte, sum(inc) as net,
       (@sum := @sum + sum(inc)) as total_for_day
from ((select date_start_tracking as dte, count(*) as inc
       from product_tracking
       group by date_start_tracking
      ) union all
      (select date_stopped_tracking, - count(*)
       from product_tracking
       where date_stopped_tracking is not null
       group by date_stopped_tracking
      )
     ) pt cross join
     (select @sum := 0) vars
group by dte
order by dte;

This is an outline of a solution. For your data, you might want one more than the stop date -- depending on whether or not the product is counted on that day.

You can then use this query as a subquery to aggregate for whatever period you like.