T-SQL - How to write a conditional join

Question

I have a stored procedure with a number of parameters. I would like to write my query so that it joins with certain tables but only if a particular parameter has a value. Take

Answer 1

Yes, it's very simple. Do left joins on the address and groups. Then in the where clause...

(@group_id is null or g.group_id = @group_id)
and (@address_id is null or a.address_id = @address_id)

Answer 2

Uh, probably all of you have resolved this so far.

As i understand You, you want to have 'dynamic' query, to join table if parameter exists, or to omit join if parameter is null. Secret is in using left outer join. Like:

SELECT p.*
FROM Parent AS p
LEFT OUTER JOIN Child AS c ON p.Id = c.ParentId
WHERE
        (@ConditionId IS NULL OR c.ConditionId = @ConditionId)

How this works?

• If filter parameter @ConditionId is null, then there's no child for outer join, and result will have all Parent's.
• If filter parameter @ConditionId is not null, then outer join will join Child's with this parent, and condition (@ConditionId IS NULL OR c.ConditionId = @ConditionId) will throw out Parent's which didn't have joined Child's with condition c.ConditionId = @ConditionId.

LEFT OUTER JOIN for sure have performance issue, but as much as this works fast, i don't want to concatenate string's to create query.

Answer 3

The simple ways are actualy not good solutions. As bad as it sounds, the best solution is to have explicit IF in the code and separate queries:

IF (condition) 
  SELECT ... FROM Person WHERE ...
ELSE IF (otherCondition)
  SELECT ... FROM Person JOIN ... ON ... WHERE ...
ELSE IF (moreCondition)
  SELECT ... FROM Persons JOIN ... JOIN ... WHERE ...

The reason for this is that if you're trying to build one single query that matches all three (or more) conditions then the engine has to produce one single query plan that works in all conditions. In T-SQL one statement equals one plan. Remember that plans are created for the generic case, for any variable value, so the result is always a very, very bad plan.

While is is counterintuitive and seems like a horrible solution to any programmer, this is how databases work. The reason why this is not a problem 99.99% of the times is that after trying what you ask and seeing what it has to be done, developers quickly come to their senses and revise their requirements so that they never have to run queries that optionaly join based on runtime variable values ;)