MySQL数据库设计-捕获视频使用情况指标

Question

I recently got handed over this existing application. There is a mysql database table which is being used for tracking the time till where the user has last watched the video.

The simplistic bare version of the table is

id -> Primary_key
user_id
video_id
last_watched_time
last_viewed_time (DateTime)

In the last_watched_time column, they store the time in seconds from where to start playing the video again.

The way it is done is, when the user starts playing the video, a new record is inserted in the table (if record for that user and video does not exist) and then while the user is watching the video - every 20 seconds - an update is done on that record to capture the last watched time. This is so the if the user clicks on the next video without pressing the stop button - we know the time where he stopped.

So if a user watches a new video for 1 min and closes the browser, the last_watched_time column gets updated 3 times 20,40,60. The last value stored is 60. So when he comes back - the video starts playing from the 1 min mark.

id  |  user_id  |  video_id  |  last_watched_time  |  last_viewed_time
-------------------------------------------------------------------------
1   |     10    |     6      |       60            | 2018-10-01 10:10:10

So this is an existing table with live data.

Now they want to start measuring detailed metrics of the users usage like - In last 7 days , how many hours of video the user has watched broken down by day - In last 6 hours, how many hours of video the user has watched broken down by hour

So my first thought was to do the following - add another column called view_time to this table - change every 20 sec updates to insert statements

So for the same scenario above, the data in the table would be

id  |  user_id  |  video_id  |  last_watched_time  |  view_time  |  last_viewed_time
-----------------------------------------------------------------------------------------
1   |     10    |     6      |         0           |      0      |   2018-10-01 13:10:10
2   |     10    |     6      |        20           |     20      |   2018-10-01 13:10:30
3   |     10    |     6      |        40           |     20      |   2018-10-01 13:10:50
4   |     10    |     6      |        60           |     20      |   2018-10-01 13:11:10

Now if the same user comes back after 2 hours, forwards the video by 10 minutes and watches for 25 seconds

5   |     10    |     6      |       660           |    600      |   2018-10-01 15:11:10
6   |     10    |     6      |       680           |     20      |   2018-10-01 15:11:30

With this, I am still tracking the last_watched_time and if they want daily or hourly metrics, I can group by day or hour and sum up the view_time to know how many minutes the user watched that day or hour.

The obvious issue I see with this approach is if the user watches video for 4 hours - going by the 20 second insert statements to capture the last_watched_time - approximately 720 rows will be inserted into this table. And if we are talking about 100 users - the number just multiplies.

Is this approach even right. How should I go about it?

Answer 1

Do the work on INSERT instead of on SELECT . This way much less data needs to be stored, and the SELECTs are much faster.

If all you want is the aggregate watch times, then store only that. That is, when a new record comes in, augment the existing record if it was 20 seconds ago.

Since you want daily and hourly watch times, whenever it is a new hour, don't add to the existing record, but start a new record. At the extreme, this will shrink the number of rows by 180x (60 minutes @ 20-second intervals). For 'surfing', there may be no shrinkage.