Counting number of joined rows in left join

Question

I'm trying to write an aggregate query in SQL which returns the count of all records joined to a given record in a table; If no records were joined to the given record, then the result for that record should be 0:

Data

My database looks like this (I'm not able to change the structure, unfortunately):

MESSAGE
----------------------------------------------
MESSAGEID   SENDER        SUBJECT
----------------------------------------------
1           Tim           Rabbit of Caerbannog
2           Bridgekeeper  Bridge of Death

MESSAGEPART
----------------------------------------------
MESSAGEID   PARTNO        CONTENT
----------------------------------------------
1           0             (BLOB)
1           1             (BLOB)
3           0             (BLOB)

(MESSAGEPART has a composite PRIMARY KEY("MESSAGEID", "PARTNO"))

Desired output

Given the data above I should get something like this:

MESSAGEID   COUNT(*)
-----------------------------------------------
1           2
2           0

It seems obvious that I need to do a left join on the MESSAGE table, but how do I return a count of 0 for rows where the joined columns from MESSAGEPART are NULL? I've tried the following:

Logic

I've tried

SELECT m.MESSAGEID, COUNT(*) FROM MESSAGE m
LEFT JOIN MESSAGEPART mp ON mp.MESSAGEID = m.MESSAGEID
GROUP BY m.MESSAGEID;

However, this returns

MESSAGEID   COUNT(*)
-----------------------------------------------
1           2
2           1

I've also tried

SELECT mp.MESSAGEID, COUNT(*) FROM MESSAGE m
LEFT JOIN MESSAGEPART mp ON mp.MESSAGEID = m.MESSAGEID
GROUP BY mp.MESSAGEID;

but this returns

MESSAGEID   COUNT(*)
-----------------------------------------------
1           2
            1

What am I doing wrong here?

Solution 1

How about something like this:

SELECT m.MESSAGEID, sum((case when mp.messageid is not null then 1 else 0 end)) FROM MESSAGE m
LEFT JOIN MESSAGEPART mp ON mp.MESSAGEID = m.MESSAGEID
GROUP BY m.MESSAGEID;

The COUNT() function will count every row, even if it has null. Using SUM() and CASE, you can count only non-null values.

EDIT: A simpler version taken from the top comment:

SELECT m.MESSAGEID, COUNT(mp.MESSAGEID) FROM MESSAGE m
LEFT JOIN MESSAGEPART mp ON mp.MESSAGEID = m.MESSAGEID
GROUP BY m.MESSAGEID;

Hope that helps.

Solution 2

You first want to count in your messaepart table before joining, i think. Try this:

   SELECT m.MessageId
        , COALESCE(c, 0) as myCount
     FROM MESSAGE m
LEFT JOIN (SELECT MESSAGEID
                , count(*) c 
             FROM MESSAGEPART 
            GROUP BY MESSAGEID) mp
       ON mp.MESSAGEID = m.MESSAGEID

Solution 3

Don't forget to use DISTINCT in case you will LEFT JOIN more than one table:

SELECT m.MESSAGEID, COUNT(DISTINCT mp.MESSAGEID) FROM MESSAGE m
LEFT JOIN MESSAGEPART mp ON mp.MESSAGEID = m.MESSAGEID
GROUP BY m.MESSAGEID;

Comments

[Nick Krasnov]

It can be simplified a little. sum(case....end) can be replaced with simple count(mp.messageid). Count(*) counts, well, everything, including nulls and count(col_name) counts only non-null values.

[errantlinguist]

This is more complicated than the solution above, but I have to embed this query in another one anyway, so your solution was also ultimately helpful; cheers

[Mark J. Bobak]

Thanks @NicholasKrasnov. I totally forgot that distinction. The older I get, the more I forget.... sigh.... :-)

[DCShannon]

This was useful. I needed to see some other fields beside the one linking key. I found you can do that by adding the fields to both the select list and the group by clause.

[godspeedelbow]

After having tried tens of answers to other similar questions, this is one finally worked for me. It is the GROUP BY that I hadn't included earlier that makes this work, without having to use subqueries (you want to avoid that). Note: if you want to add a WHERE clause it'll need to look like this: WHERE (mp.PARTNO = 1 OR mp.PARTNO IS NULL) and is added before the GROUP BY

[Andrew]

but what if you dont want to group by :(

[petrosmm]

however inefficient this may or may not be but for purposes of knowledge, it is extremely useful especially right now!