SQL join without losing rows - sql

Question

I have 2 tables with the same schema of userID, category, count. I need a query to sum the count of each userID/category pair. Sometimes a pair will exist in one table and not the other. I'm having trouble doing a join without losing the rows where a userID/category pair only exists in 1 table. This is what I'm trying (without success):
select a.user, a.category, count=a.count+b.count
from #temp1 a join #temp2 b
on a.user = b.user and a.category = b.category
Example:
Input:
user category count
id1 catB 3
id2 catG 9
id3 catW 17
user category count
id1 catB 1
id2 catM 5
id3 catW 13
Desired Output:
user category count
id1 catB 4
id2 catG 9
id2 catM 5
id3 catW 30
Update: "count" is not the actual column name. I just used it for the sake of this example, and I forgot it's a reserved word.

Answer 1

You need to:
Use a full outer join so you don't drop rows present in one table and not the other
Coalesce counts prior to addition, because 0 + NULL = NULL
Also, because COUNT is a reserved word, I would recommend escaping it.
So, using all of these guidelines, your query becomes:
SELECT COALESCE(a.user, b.user) AS user,
COALESCE(a.category, b.category) AS category,
COALESCE(a.[count],0) + COALESCE(b.[count],0) AS [count]
FROM #temp1 AS a
FULL OUTER JOIN #temp2 AS b
ON a.user = b.user AND
a.category = b.category

Answer 2

One way to approach this is with a full outer join:
select coalesce(a.user, b.user) as user,
coalesce(a.category, b.category) as category,
coalesce(a.count, 0) + coalesce(b.count, 0) as "count"
from #temp1 a full outer join
#temp2 b
on a.user = b.user and
a.category = b.category;
When using full outer join, you have to be careful because the key fields can be NULL when there is a match in only one table. As a result, the select tends to have a lot of coalesce()s (or similar constructs).
Another way is using a union all query with aggregation:
select "user", category, SUM(count) as "count"
from ((select "user", category, "count"
from #temp1
) union all
(select "user", category, "count"
from #temp2
)
) t
group by "user", category