I have three tables:
SmallTable
(id int, flag1 bit, flag2 bit)
JoinTable
(SmallTableID int, BigTableID int)
BigTable
(id int, text1 nvarchar(100), otherstuff...)
SmallTable
has, at most, a few dozen records. BigTable
has a few million, and is actually a view that UNIONS a table in this database with a table in another database on the same server.
Here's the join logic:
SELECT * FROM
SmallTable s
INNER JOIN JoinTable j ON j.SmallTableID = s.ID
INNER JOIN BigTable b ON b.ID = j.BigTableID
WHERE
(s.flag1=1 OR b.text1 NOT LIKE 'pattern1%')
AND (s.flag2=1 OR b.text1 <> 'value1')
Average joined size is a few thousand results. Everything shown is indexed.
For most SmallTable
records, flag1
and flag2
are set to 1
, so there's really no need to even access the index on BigTable.text1, but SQL Server does anyway, leading to a costly Indexed Scan and Nested Loop.
Is there a better way to hint to SQL Server that, if flag1
and flag2
are both set to 1
, it shouldn't even bother looking at text1
?
Actually, if I can avoid the join to BigTable completely in these cases (JoinTable is managed, so this wouldn't create an issue), that would make this key query even faster.
SQL Boolean evaluation does NOT guarantee operator short-circuit. See On SQL Server boolean operator short-circuit for a clear example showing how assuming operator short circuit can lead to correctness issues and run-time errors.
On the other hand the very example in my link shows what does work for SQL Server: providing an access path that SQL can use. So, as with all SQL performance problems and questions, the real problem is not in the way the SQL text is expressed, but in the design of your storage. Ie. what indexes has the query optimizer at its disposal to satisfy your query?
If you love us? You can donate to us via Paypal or buy me a coffee so we can maintain and grow! Thank you!
Donate Us With