More With Adaptive Joins

Erik Darling continues his adaptive joins exploration with two more posts.  First, how local variables can affect the query plan:

The easiest way to look at this is to compare Adaptive Joins with literal values to the same ones using local variables. The results are a little… complicated.

Here are three queries with three literal values. In my copy of the Super User database (the largest Stack Overflow sub-site), I’ve made copies of all the tables and added Clustered ColumnStore indexes to them. That’s the only way to get Adaptive Joins at this point — Column Store has to be involved somewhere along the line.

The last day of data in this dump is from December 11. When I query the data, I’m looking at the last 11 days of data, the last day of data, and then a day where there isn’t any data.

Then Erik takes on non-SARGable queries:

The queries with non-SARGable predicates on the Users table used Adaptive Joins.

The queries with non-SARGable predicates on the Posts table did not.

Now, there is an Extended Events… er… event to track this, called adaptive_join_skipped, however it didn’t seem to log any information for the queries that didn’t get Adaptive Joins.

Bummer! But, if I had to wager a guess, it would be that this happens because there’s no alternative Index Seek plan for the Posts table with those predicates. Their non-SARGableness takes that choice away from the optimizer, and so Adaptive Joins aren’t a choice. The Users table is going to get scanned either way — that’s the nature of ColumnStore indexes, so it can withstand the misery of non-SARGable predicates in this case and use the Adaptive Join.

Two more good posts in Erik’s series, and both definitely worth reading.

Related Posts

Table Variables And Parallelism

Erik Darling shows your brain on table variables: Inserts and other modifications to table variables can’t be parallelized. This is a product limitation, and the XML warns us about it. The select could go parallel if the cardinality estimate were more accurate. This could potentially be addressed with a recompile hint, or with Trace Flag […]

Read More

Non-Blocking Aggregations

Daniel Hutmacher tilts at windmills: It’s not entirely uncommon to want to group by a computed expression in an aggregation query. The trouble is, whenever you group by a computed expression, SQL Server considers the ordering of the data to be lost, and this will turn your buttery-smooth Stream Aggregate operation into a Hash Match […]

Read More