SQL Server 2025 - Higher Worker Thread Consumption After Upgrade

Question

Hi all,

We're currently testing SQL Server 2025 (17.x) and have noticed significantly higher worker thread consumption compared to our SQL Server 2019 environment.

We first observed this on SQL Server 2025 CU5 and have since observed the same behaviour on CU7.

In our environment, worker thread consumption appears to be substantially higher after upgrading to 2025, to the point where some of our busier servers experienced THREADPOOL waits, intermittent connection timeouts, and occasional login/access failures.

During investigation we came across the following known issue:

https://learn.microsoft.com/en-us/sql/sql-server/sql-server-2025-known-issues?view=sql-server-ver17#sql-server-might-become-slow-or-unresponsive-after-creating-or-bringing-online-a-large-number-of-databases

We implemented the documented workaround using startup trace flag 15608 and restarted SQL Server. We did not see an immediate reduction in worker counts after restart. However, after approximately 30 minutes, worker consumption began to decrease and eventually settled at noticeably lower levels.

An additional observation is that on subsequent SQL Server restarts, worker counts initially climb back to their previous elevated levels before gradually dropping again after roughly 30 minutes. This behaviour has been repeatable during our testing.

Has anyone else observed:

• Higher worker thread consumption after upgrading from SQL Server 2019 to SQL Server 2025?
• Improvement after enabling trace flag 15608?
• A delay before worker counts decrease following restart?
• Worker counts initially rising again after restart before eventually settling lower?

Interested to hear whether others have seen similar behaviour, and whether Microsoft has provided any additional guidance beyond the known issue article.

Thanks.

Answer 1

You say you see this on several servers in your environment. But may I guess that they have similar workload, for instance many databases?

I am not surprised that you see the same behaviour with the 30-second delay after a restart, because where would SQL Server persist the information to avoid this?

I understand that you post here to see if someone else has experienced this situation, and we'll see if someone chimes in. But given the traffic in the forums these days, I would not really hold my breath. Therefore, I am inclined to recommend you to open a support case, so that Microsoft can investigate this case more closely. Given that it seems to block your upgrade to SQL 2025, I would expect them to be very interested.