Support sql_calc_found_rows with limit in sharded keyspace

[harshit-gangal]

Support queries using SQL_CALC_FOUND_ROWS. This is done by rewriting the queries according to https://dev.mysql.com/worklog/task/?id=12615:

Frequently, it would be cheaper to run the query with LIMIT and then a separate SELECT COUNT() for the same query, since COUNT() can make use of optimizations that can't be done when searching for the entire result set (e.g. filesort can be skipped for COUNT(*), whereas with CALC_FOUND_ROWS, we must disable some filesort optimizations to guarantee the right result). If you search online, you'll frequently find the advice to run two separate queries instead of using SQL_CALC_FOUND_ROWS, precisely for performance reasons.

So, vtgate will plan these queries by first solving the original query, without the SQL_CALC_FOUND_ROWS option, and then following up with a COUNT(*) query.

SELECT SQL_CALC_FOUND_ROWS col1, col2 FROM user WHERE col3 > "foo" LIMIT 10

will get be rewritten into two queries:

SELECT col1, col2 FROM user WHERE col3 > "foo" LIMIT 10
SELECT count(*) FROM user WHERE col3 > "foo" 

The first query will be used to return the results to the user, and the second will be used to save the found_rows state on the session in vtgate.

[harshit-gangal]

reset this file change.

[systay]

Hmm... Weird. On my machine, with 723fef02e checked out, that file has an extra line at the end, and git does not see any changes. Not sure what's going on...