MySQL Left Outer Join & Count - 'Where'?

Question

I have a relatively simple query that returns a user profile, together with 2 counts related to that user (stream events & inventory);

SELECT u.*, r.regionName, COUNT(i.fmid) AS invcount, COUNT(s.fmid) AS streamcount
FROM fm_users u
JOIN fm_regions r
ON u.region=r.regionid
LEFT OUTER JOIN fm_inventory i
ON u.fmid=i.fmid
LEFT OUTER JOIN fm_stream s
ON u.fmid=s.fmid
WHERE u.username='sampleuser'

Both the inventory & stream values could be zero, hence using a left outer join.

However, the values for both currently return numbers for all users, not the specific user (always identified as integer 'fmid' in each table). This is obviously because the query doesn't specify a 'where' for either count - but I'm not sure where. If I change the last part to this;

WHERE u.username='sampleuser' AND s.fmid=u.fmid AND i.fmid=u.fmid
GROUP BY u.fmid

It still returns incorrect numbers, albeit 'different' numbers depending on the search criteria - and the name number for both invcount and streamcount.

score 1 · Accepted Answer · answered Mar 12 '18 at 12:59

Your query cross joins the fm_inventory and fm_stream tables for each user. For example if a specific user has 2 matching fm_inventory rows, and fm_stream also has two, the result will have 4 (=2 x 2) rows and each will be counted in both COUNT() functions.

The way to accomplish what you are doing is to use correlated subselects for the counts.

SELECT u.*, r.regionName, 
   (select COUNT(*) from fm_inventory i
      where i.fmid = u.fmid) AS invcount, 
   (select COUNT(*) from fm_stream s
      where s.fmid = u.fmid) AS streamcount
FROM fm_users u
JOIN fm_regions r
ON u.region=r.regionid
WHERE u.username='sampleuser'

This allows the counts to be independent of each other. This works even if more than just one user is selected. In that case you need a group by clause, but otherwise it is the same syntax.

Perfect, thank you! Your example perfectly matched what I was seeing too. — Lee, Mar 12 '18 at 13:01

score 0 · Answer 2 · answered Mar 12 '18 at 12:47

0

When you mix normal columns with aggregates, include all normal columns in the GROUP BY. In this particular case, you are missing the r.regionName in the GROUP BY.

See longer explanation.

answered Mar 12 '18 at 12:47

slaakso

8,331
2
16
27

Thanks for the comment - I wasn't aware of that, however even adding that I'm seeing the same results returned – Lee Mar 12 '18 at 12:51
What is it that you are counting with COUNT(i.fmid)? Number of fm_inventory rows for the user? If so, you can count with COUNT(distinct i.id) assuming the fm_inventory has id column? – slaakso Mar 12 '18 at 13:01

MySQL Left Outer Join & Count - 'Where'?

2 Answers2