Query for Highest Rated Video (Likes / Dislikes) SLOW when using ORDER - mysql

Question

(Using MySQL) I have a table of videos (simplified):
+---------+-------------+-------------+
| videoID | videoName | videoAuthor |
+---------+-------------+-------------+
| 1 | cool_video1 | rocky |
| 2 | mixingTest2 | sensable |
+---------+-------------+-------------+
and a table of video ratings, that is, every time a user likes or dislikes a video, a row gets added/updated in the videoRatings table:
for example, this would mean the video with videoID 1 has two likes and one dislike. a "1" is a like in the videoRatings table. a "2" is a dislike (simplified)
+---------------+---------+-------------------+
| videoRatingID | videoID | videoRatingTypeID |
+---------------+---------+-------------------+
| 121 | 1 | 1 |
| 234 | 1 | 1 |
| 290 | 1 | 2 |
+---------------+---------+-------------------+
now, simple enough, all I want to do is grab the highest scoring videos of about 100,000 videos.
naturally, I do something like this:
SELECT Videos.videoID,
COUNT(CASE WHEN videoRatingTypeID =1 THEN 1 ELSE NULL END) AS likes,
COUNT(CASE WHEN videoRatingTypeID =2 THEN 1 ELSE NULL END) AS dislikes
FROM Videos
LEFT JOIN VideoRatings ON VideoRatings.videoID = Videos.videoID
GROUP BY Videos.videoID
ORDER BY likes DESC
but this query runs in about a half of a second. That worries me that when the videos table gets to >1mil, this will be even longer. The videoRatings table is very small (~40 rows), and the videos table is ~100,000 rows.
I have indexes on the videoID obviously in the Videos table and indexes in my videoRatings table on videoID, videoRatingID, and a composite index on videoID+videoRatingID
I don't see a better way to do this. I've read several posts about moving the order by outside. But when I do that:
SELECT * FROM (
SELECT Videos.videoID,
COUNT(CASE WHEN videoRatingTypeID =1 THEN 1 ELSE NULL END) AS likes,
COUNT(CASE WHEN videoRatingTypeID =2 THEN 1 ELSE NULL END) AS dislikes
FROM Videos
LEFT JOIN VideoRatings
GROUP BY Videos.videoID
) tmp
ORDER BY tmp.likes DESC
there is zero improvement.
What is a better way to approach this layout, or this query? Thank you!

Answer 1

For real scalability, I think you will need a solution that maintains a summary table. In the meantime, this might be faster:
select v.videoID,
(select count(*)
from VideoRatings vr
where vr.videoID = v.videoID and
videoRatingTypeId = 1
) as likes,
(select count(*)
from VideoRatings vr
where vr.videoID = v.videoID
videoRatingTypeId = 2
) as dislikes
from Videos v;
Be sure that you have an index on VideoRatings(videoId, videoRatingTypeId) (actually, the type id is not so important in the index, but it can help).
This replaces the group by on the whole set of videos and ratings, with index scans and small aggregations. This will scale up, as long as the index for videoRatings fits into memory.
EDIT:
Your video ratings table is very spartan, containing little more information than the summary number of likes and dislikes. Such a table might have the date/time of the rating and the person who did the rating, for instance.
BUT. You are adding new ratings by inserting rows into this table. Well, it is almost the same operation to update another table (which could be videos) with the information. Then think of your current table as the historical log.
The advantage to using updates is that you can truncate the log when it gets big. Right now, you have to keep every rating since the beginning of time for every video.