Indirect Checkpoint And Non-Yielding Scheduler Problems

Parikshit Savjani has a post describing an issue you might experience with indirect checkpoint post SQL Server-2012:

One of the scenarios where skewed distribution of dirty pages in the DPList is common is tempdb. Starting SQL Server 2016, indirect checkpoint is turned ON by default with target_recovery_time set to 60 for model database. Since tempdb database is derived from model during startup, it inherits this property from model database and has indirect checkpoint enabled by default. As a result of the skewed DPList distribution in tempdb, depending on the workload, you may experience excessive spinlock contention and exponential backoffs on DPList on tempdb. In scenarios when the DPList has grown very long, the recovery writer may produce a non-yielding scheduler dump as it iterates through the long list (20k-30k) and tries to acquire spinlock and waits with exponential backoff if spinlock is taken by multiple IOC routines for removal of pages.

This is worth taking a close read.

Related Posts

Expanding LVM Drives

David Klee shows how to expand an LVM drive on Linux: Next in our SQL Server on Linux series is one important question. On Windows, if you’re about to run out of space, you get your VM admin / storage admin to expand one or more of your drives, and you go to Disk Management […]

Read More

Filtering Alert Noise With A Leaky Bucket Algorithm

I have a post implementing a leaky bucket algorithm in T-SQL: Now that we have a table, we want to do something with it.  The most naive solution would be to fire off an alert every time a row gets added to this table.  The problem with this solution is that it can flood our […]

Read More

Categories

June 2017
MTWTFSS
« May Jul »
 1234
567891011
12131415161718
19202122232425
2627282930