Counting instances in table - mysql

Question

I've got this tag system for tagging blog entries and such. The tags are in one table, containing only a tag name and a primary key. Then I have another table with objects that are using the tags.
It could look something like this:
_________________________________
| tags |
--------------------------------|
| id | name |
|-------------------------------|
| 1 | Scuba diving |
| 2 | Dancing |
---------------------------------
_________________________________
| tag_objects |
--------------------------------|
| id | tag | object |
|-------------------------------|
| 1 | 2 | 13 |
| 2 | 2 | 18 |
| 3 | 1 | 24 |
---------------------------------
Now, what I need to accomplish is to to add a column to the tags table, called "occurrences" or something. For each tag in tags, occurrences should be set to the number of times that tag is used in tag_objects.
So basically something like (obviously pseudo-code):
foreach(tags):
UPDATE tags
SET occurrences = (SELECT COUNT(id)
FROM tag_objects
WHERE tag = tags.id);
When people create new posts and stuff in the future, I'll just have a trigger to update the count, but I have a couple of thousand rows already that I need to count first. I don't know how to do this, so any assistance would be appreciated.

Answer 1

The easiest way to do this, without any extra tables, would be:
First add the extra field:
mysql> alter table tags add occurs int
default 0;
Then just update this new field with the number of occurences.
mysql> update tags left join (select tag,
count(id) as cnt from tag_objects
group by tag) as subq on
tags.id=subq.tag set
occurs=coalesce(subq.cnt,0);
Note the use of the left join to ensure all tags are counted, even the unused ones. The coalesce-function will convert NULL to 0.

Answer 2

You have done a good work, your query must work.
But, this will result in awful performance. I advise you to recreate a table :
CREATE TABLE newTags AS
SELECT t.id, t.name, COUNT(*) AS occurrences
FROM tags t
INNER JOIN tag_objects to
ON to.tag = tags.id
GROUP BY t.id, t.name
This will be very fast.

Answer 3

Unless you really need to denormalize your data, you should stay away from that. Counting on indexed columns is usually very fast. I am a big fan of clean and normalized data ;-)

Answer 4

I would generally not want to store computed values in columns on the database - it's messy, can easily get out of sync, and offends the deities of normalization.
However, if you really must have a database entity with the count, rather than calculating on the fly, I'd create a view (http://dev.mysql.com/doc/refman/5.0/en/create-view.html) which stores the pre-computed value, using the SQL provided by Scorpio
CREATE view tag_occurences AS
SELECT t.id, t.name,
COUNT(*) AS occurrences
FROM tags t
INNER JOIN tag_objects to
ON to.tag = tags.id
GROUP BY t.id, t.name

Answer 5

I think you will gain better performance if you will be incrementing and decrementing the value of occurrences on table tag_objects insert/delete trigger.

Answer 6

Your psuedeo code will work exactly as written (without the foreach loop). At least it would in oracle, I assume MySQL lets you use a correlated subquery as the value too.

Answer 7

For the inserting of new rows you could use a query like:
INSERT INTO tags VALUES(x,y,z,1) ON DUPLICATE KEY UPDATE occurrences = occurrences+1;
I didn't check the syntax, but something like that.