Basic optimisation with an index for mysql

Question

Hi, i have a question about basic mysql database optimisation. I have 3 tables, Articles, Tags and Taggings (which is a join table).

Articles         Taggings             Tags
id               id                   id
name             article_id           name
                 tag_id

I am retrieving the articles that exactly match the tags specified, with the following query

SELECT *, COUNT(*) AS c
FROM articles AS a
JOIN taggings AS tng ON a.id = tng.article_id
JOIN tags AS t ON t.id = tng.tag_id
WHERE t.name IN ("Red","Green")
GROUP BY a.id
HAVING c = 2

This query is slow, so I did an EXPLAIN, and got the following results:

Now, I don't really understand what I am doing here, but i believe that "type: ALL" is not good, so thought i would add indexes(BTREE) to both article_id and tag_id in the taggings table, and run the query again. Well that didn't look any better to my uneducated eye, same number of rows as previous one, and the type is still ALL in two of the cases.

So could someone tell me where I am going wrong please? will indexes not help me with this problem?

My Tag table will remain relatively small, so I thought the query should scan the Tag table for the tags I have specified, and then (through the indexes) be able to instantly retrieve the associated properties, and it should all be very quick, obviously something wrong in my thinking.

Thanks

[EDIT] - for Jay's comments

I added 10k articles, 30k taggings, and 6 tags, also added 2 indexs on tag.name and taggings.tag_id, the query still took a long time to run, 0.5-1 second, the EXPLAIN is below.

Answer 1

You could also try using joining to the tables twice instead of a GROUP BY. This sometimes produces a faster query:

SELECT a.*
FROM articles AS a
JOIN taggings AS tng1 ON a.id = tng1.article_id
JOIN tags AS t1 ON t1.id = tng1.tag_id AND t1.name = "Red"
JOIN taggings AS tng2 ON a.id = tng2.article_id
JOIN tags AS t2 ON t2.id = tng2.tag_id AND t2.name = "Green"

Answer 2

Because tags.name is the only column that really reduces a number of rows in result set, it must be indexed to make any tag-based search query faster.

Update: try to run this query

SELECT a.*
FROM articles AS a
JOIN taggings AS tng ON a.id = tng.article_id
JOIN tags AS t ON t.id = tng.tag_id
WHERE t.name IN ("Red","Green")
GROUP BY a.id
HAVING COUNT(DISTINCT t.id) = 2

Answer 3

There are a couple of things going on here.

First, your tables presently are apparently very small. When the table is small, the DBMS often finds it faster to read the whole thing rather than to use any index. To get meaningful EXPLAIN results, you need to get realistic numbers of records in the tables.

It also looks like you have the "id" fields declared as primary keys. Primary keys are a subclass of indexes, so those should be available. Note the explain plan indicates it used the primary key to find the Tag record.

The obvious starting point of this query is Tags. So if this is an important query, I'd create an indexon Tags(name). Then it wouldn't need to sequentially search the Tags table.

From there it should look up Taggings by tag_id. So you should have an index on that.

Then it could look up Article by article_id. That's the primary key so it should already be there.

So I think you'd get the most effective plan with two indexes: Tags(name) and Taggings(tag_id).