Why do we sometimes want to keep duplicate indexes?
I’ve done dutiful DBA work in the past to identify and remove what are commonly called duplicate indexes. That is, those indexes that look like (a) and (a,b). The thought is that a query will utilize an index as easily on (a) as on (a,b), and removing (a) will save storage cost and write performance. I’ve had the experience, though, of removing (a) and seeing performance tank.
(As an aside, these are really redundant indexes. A duplicate index would be (a,b) and (a,b) by two different names – this can commonly be done by object relational mapping (ORM) or other automated schema creation tools. I’ll call (a) and (a,b) redundant indexes below.)
This test is on Percona Server 5.7.14 with the sys schema installed and performance schema enabled.
Given two tables with the same number of rows and …[Read more]