Efficient MySQL Table Setup and Queries - mysql

Question

Suppose I have the following database setup (a simplified version from what I actually have):
Table: news_posting (500,000+ entries)
| --------------------------------------------------------------|
| posting_id | name | is_active | released_date | token |
| 1 | posting_1 | 1 | 2013-01-10 | 123 |
| 2 | posting_2 | 1 | 2013-01-11 | 124 |
| 3 | posting_3 | 0 | 2013-01-12 | 125 |
| --------------------------------------------------------------|
PRIMARY posting_id
INDEX sorting ON (is_active, released_date, token)
Table: news_category (500 entries)
| ------------------------------|
| category_id | name |
| 1 | category_1 |
| 2 | category_2 |
| 3 | category_3 |
| ------------------------------|
PRIMARY category_id
Table: news_cat_match (1,000,000+ entries)
| ------------------------------|
| category_id | posting_id |
| 1 | 1 |
| 2 | 1 |
| 3 | 1 |
| 2 | 2 |
| 3 | 2 |
| 1 | 3 |
| 2 | 3 |
| ------------------------------|
UNIQUE idx (category_id, posting_id)
My task is as follows. I must get a list of 50 latest news postings (at some offset) that are active, that are before today's date, and that are in one of the 20 or so categories that are specified in the request. Before I choose the 50 news postings to return, I must sort the appropriate news postings by token in descending order. My query is currently similar to the following:
SELECT DISTINCT posting_id
FROM news_posting np
INNER JOIN news_cat_match ncm ON (ncm.posting_id = np.posting_id AND ncm.category_id IN (1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20))
WHERE np.is_active = 1
AND np.released_date < '2013-01-28'
ORDER BY np.token DESC LIMIT 50
With just one specified category_id the query does not involve a filesort and is reasonably fast because it does not have to process removal of duplicate results. However, calling EXPLAIN on the above query that has multiple category_id's returns a table that says that there is filesort to be done. And, the query is extremely slow on my data set.
Is there any way to optimize the table setup and/or the query?

Answer 1

I was able to get the above query to run even faster than with a single-value category list version by rewriting it as follows:
SELECT posting_id
FROM news_posting np
WHERE np.is_active = 1
AND np.released_date < '2013-01-28'
AND EXISTS (
SELECT ncm.posting_id
FROM news_cat_match ncm
WHERE ncm.posting_id = np.posting_id
AND ncm.category_id IN (1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20)
LIMIT 1
)
ORDER BY np.token DESC LIMIT 50
This now takes under a second on my data set.
The sad part is that this is even faster than if there is just one category_id specified. That's because the subset of news items is bigger than with just one category_id, so it finds the results more quickly.
Now my next question is whether this can be optimized for cases when a category has only few news that are spread in time?
The following is still pretty slow on my development machine. Although it's fast enough on the production server, I would like to optimize this if possible.
SELECT DISTINCT posting_id
FROM news_posting np
INNER JOIN news_cat_match ncm ON (ncm.posting_id = np.posting_id AND ncm.category_id = 1)
WHERE np.is_active = 1
AND np.released_date < '2013-01-28'
ORDER BY np.token DESC LIMIT 50
Does anyone have any further suggestions?