Dfyjkt

We have a SQL Server 2016 SP1 with max memory set to 24GB.

This server has a high numbers of compiles, only 10% of these compiles are from Ad-Hoc queries. So the newly compiled plans should be stored in the plan cache but the size of the plan cache is not increasing (approx 3.72GB).


I suspect that there is local memory pressure that lead to removal of plans from the cache. The plan cache pressure limit is 5GB. (75% of visible target memory from 0-4GB + 10% of visible target memory from 4GB-64GB + 5% of visible target memory>64GB). When a cachestore reaches 75% of the pressure limit, plans should be removed from the cache. In my case 75% of 5 GB is 3.75GB. So it is plausible this is the cause of the high compiles.


Is there a way to measure (perfmon, extended events, ...) the removal from plans out of the cache? So I can be certain local memory pressure is really the cause of the high compiles?

There's an XEvent for that:

query_cache_removal_statistics

Occurs when a query plan is removed from the plan cache and the
historical statistics for the object are about to be destroyed

So something like:

CREATE EVENT SESSION [PlanCacheEvictions] ON SERVER 
ADD EVENT sqlserver.query_cache_removal_statistics(
 ACTION(sqlserver.sql_text))

Also if your plan cache has a large number of single-use plans consider setting optimize for ad hoc workloads.