Use index on int equal queries

[jedelbo]

No description provided.

[ironage]

Here's the benchmark results of this PR compared to core v5.18.0 using a column of 1 million unique integers.

Operation	v5.18.0 (avg)	This PR (avg)	Approximate speedup
1000 conditions	230 ms	22 ms	10x
1000 conditions, indexed	231 ms	20 ms	10x
1 condition	615 us	370 us	1.6x
1 condition, indexed	617 us	15 us	40x

This shows a nice improvement across the board. The largest improvement in practice will be for users who chain many "OR" conditions together or heavily query an int primary key column (indexed automatically).

[ironage]

💯

[astigsen]

Any idea why the index makes such little difference at 1000 conditions? Is there some other overhead that dominates in that case?

On Mon, 1 Apr 2019 at 23.02, James Stone ***@***.***> wrote: Merged #3272 <#3272> into master. — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#3272 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AAEPLvDFDswvXYZInjzh2j0U06Yfx-Ryks5vcnPXgaJpZM4cWFvr> .

-- *Alexander Stigsen, CTO* Realm Inc

[jedelbo]

If you have a 1000 conditions on a column with index, the query engine must visit 1000 nodes that each must lookup a value.
If you don't have an index, you will just have one node that will have to lookup 1 millon values in a hash table. This is most ly done in a tight loop.