v8.0.16 - using variables prevents correct use of available indexes

[Craig Watkinson]

I have a complex query that executed in <0.1s in v8.0.15, but takes >300s in v8.0.16! This is due to indexes not being used when I substitute values into a query using variables. Has this behaviour changed, or is it possibly a bug?

The following queries demonstrate the issue:

# Query 1 - performs a range scan on s3Filename index => 750 rows
SELECT count(*) FROM logfiles
WHERE s3Filename > "20190308T104730.760958700Z.log"
AND s3Filename <= "20190308T112911.460695287Z.log";

# Query 2 - performs a full index scan on s3Filename index => 7M rows
set @start_s3Filename = "20190308T104730.760958700Z.log";
set @end_s3Filename = "20190308T112911.460695287Z.log";
SELECT count(*) FROM logfiles
WHERE s3Filename > @start_s3Filename
AND s3Filename <= @end_s3Filename;

I would expect these to perform identically, and for the EXPLAIN to be the same, but on v8.0.16 it is not. On v8.0.15, both queries perform a range scan. Adding a "USE INDEX" clause did not help.

I'm also surprised that the client is able to affect the way that a query is executed. I presumed the execution plan would be determined server side. Is this done client side? If yes, then I presume this means that I could get different behaviour with different clients e.g. MySQL Workbench vs. SQLAlchemy, which is not good news.

Thanks Craig