When Index Condition Pushdown is used, you cannot trust rows examined as reported by the slow query log (nor in some Performance Schema tables) and you cannot trust the global status Innodb_rows_read (nor the InnoDB Metrics dml_reads). These can lead to an incorrect query complexity analysis, to a mistaken estimation of the total load of a MySQL server, and to a lot of wasted time. As we go through the details in this post, we will understand why a SELECT * is sometimes faster than only selecting the columns from an index (we would normally expect the opposite) and we will visit inner workings of the MySQL Storage Engine Architecture. But first, let's start by exposing the problem.
