To round up this topic, we upgraded to SQL 2019 and the issue is gone or at least not a problem anymore. There is a bit of a decrease but...

Thanks for your response Jeff. The number of connections is stable. I should have mentioned that we have several of these database servers (1 instance per server) with comparable workload...

Use BCP. You will have to enable xp_cmdshell for it though which requires elevation. And be sure to enable it only for a minimum amount of time. It has...

There must be a way for you guys to reproduce this. I have now done this on 4 different (but all 2016) installations. What I get is what you see...

Thanks for the link but I know my isolation levels ;). The fact that the transactions run in snapshot is not relevant for the estimates. If you are argueing that...

The software refreshes data to subscribers. Various sources can generate a change. These changes are identified by a set of id's. These id's are first inserted into t1, queried against...

Updating the statistic manually indeed helps produce a better estimate. Not sure why it's 1 though as I would expect 5.

But this is good to know that we could have...

There is a void statistic on t1 with norecompute to prevent recompilations. It cannot be stale by design.

The issue here is that the estimates of the nested loop are much...

My company sells a software package of which the official statement is that we do support SQL Server Clustering but not always a failover. A failover can sometimes require a...