(Check out my Pluralsight online training course: SQL Server: Performance Troubleshooting Using Wait Statistics and my comprehensive library of all wait types and latch classes.)

One of the problems with the SOS_SCHEDULER_YIELD wait type is that it’s not really a wait type. When this wait type occurs, it’s because a thread exhausted its 4ms scheduling quantum and voluntarily yielded the CPU, going directly to the bottom of the Runnable Queue for the scheduler, bypassing the Waiter List. A wait has to be registered though when a thread goes off the processor, so SOS_SCHEDULER_YIELD is used.

You can read more about this wait type:

  • Here on my blog
  • Here in my post on the SQLPerformance.com blog

You want to investigate these waits if they’re a prevalent wait on your server, as they could be an indicator of large scans happening (of data that’s already in memory) where you’d really rather have small index seeks.

The problem is that they’re not a real wait type, so you can’t use my script to look at sys.dm_os_waiting_tasks and get the query plans of threads incurring that wait type, because these threads aren’t waiting for a resource, so don’t show up in the output of sys.dm_os_waiting_tasks!

The solution is to use the sys.dm_exec_requests DMV, as that will show the last_wait_type for all running requests. Below is a script you can use.

FROM sys.dm_exec_requests [er]
INNER JOIN sys.dm_exec_sessions [es] ON
	[es].[session_id] = [er].[session_id]
OUTER APPLY sys.dm_exec_sql_text ([er].[sql_handle]) [est]
OUTER APPLY sys.dm_exec_query_plan ([er].[plan_handle]) [eqp]
    [es].[is_user_process] = 1
	AND [er].[last_Wait_type] = N'SOS_SCHEDULER_YIELD'

That will give you the code and query plan of what’s happening, but even with that it might not be obvious which exact operator is causing that wait so you may need to resort to capturing SQL Server call stacks, as I explain in the first blog post link above.