Press "Enter" to skip to content

Missing Index Hints and Index Rebuilds

Forrest McDaniel abuses that poor server:

The prod version of this issue manifested as 1-3 minutes of SCH-M blocking for index ops, yet only on a single server; I couldn’t replicate it anywhere else. Of course, bad blocking meant nasty timeouts, and that in turn meant I was reluctant to run live diagnostics. Better to just stop index rebuilds. One does not take down production in order to test weird SQL behavior.

An important clue came from a tiny table with a rebuild, and is visible in the whoisactive screenshot above. Why would such a small rebuild need so much CPU? That worker was definitely doing something – the question was what.

Read on to find out the answer, a repro script (that you should not run in your own production environment!), and what you can do about it.