Question

Background

My program is storing a series of objects, a set of tags, and the many-to-many associations between tags and objects, in a MySQL database. To give you an idea of the structure:

CREATE TABLE objects (
    object_id INT PRIMARY KEY,
    ...
);
CREATE TABLE tags (
    tag_name VARCHAR(32) NOT NULL
);
CREATE TABLE object_tags (
    object_id INT NOT NULL,
    tag_name VARCHAR(32) NOT NULL,
    PRIMARY KEY (object_id, tag_name)
);

Problem

I want to be able to query for all objects that are tagged with all of the tags in a given set. As an example, let’s say I have a live tree, a dead flower, an orangutan, and a ship as my objects, and I want to query all of those tagged living and plant. I expect to receive a list containing only the tree, assuming the tags match the characteristics of the objects.

Current Solution

Presently, given a list of tags T1, T2, …, Tn, I am solving the problem as follows:

1. Select all object_id columns from the object_tags table where tag_name is T1.
2. Join the result of (1) with the object_tags table, and select all object_id columns where tag_name is T2.
3. Join the result of (2) with the object_tags table again, and select all object_id columns where tag_name is T3.
4. Repeat as necessary for T4, …, Tn.
5. Join the result of (4) with the objects table and select the additional columns of the objects that are needed.

In practice (using Java), I start with the query string for the first tag, then prepend/append the string parts for the second tag, and so on in a loop, before finally prepending/appending the string parts that make up the overall query. Only then does the string actually get passed into a PreparedStatement and get executed on the server.

Edit: Expanding on my example from above, using this solution I would issue the following query:

SELECT object_id FROM object_tags JOIN (
    SELECT object_id FROM object_tags WHERE tag_name='living'
) AS _temp USING (object_id) WHERE tag_name='plant';

Question

Is there a better solution to this problem? Although the number of tags is not likely to be large, I am concerned about the performance of this solution, especially as the database grows in size. Furthermore, it is very difficult to read and maintain the code, especially when the additional concerns/constraints of the application are thrown in.

I am open to suggestions at any level, although the languages (MySQL and Java) are not variables at this point.

Answer 1

I don’t know about the performance of this solution, but you can simplify by using pattern matching in MySql to match a set of pipe-delimited tags (or any delimiter). This is a solution I’ve used before for similar applications with tag tables (@match would be a variable passed in by your Java code, I’ve harded coded a value for demonstration):

set @match = 'living|plant';
set @numtags =
    length(@match) - length(replace(@match, '|', '')) + 1;

select * from objects o
where @numtags =
(
    select count(*) from object_tags ot
    where concat('|',@match,'|')
        like concat('%|',ot.tag_name,'|%')
    and ot.object_id = o.object_id
)

Here is a working demo: http://sqlize.com/0vP6DgQh0j