We have hundreds of queries (mostly SELECTs and some are calls to a Store procedure) to several MS SQL servers. All VMs are hosted in Azure. We have a variety of Windows Server versions and MS SQL server versions (2008, 2012, 2016.. etc).
I got this issue since Zabbix 5.0 as far I remember (now running Zabbix 6.0 still happening), and I'm mostly certain that the ODBC poller get stuck when is doing the query and simultaneously the vm or the database is down/unreachable.

I have this evidence, were you can see that at the exact time of the error, the utilization jumps to more than 75%.

If you check each poller, you'll notice that for example says "got # values in ## sec, getting values", and if you come back, even a day later, to check again, the same pollers still have the same value. It just stay in that state, like is waiting forever to get a value or something.
In this image, I used 100 pollers (normally I only use 10 and is more than enough).

This often occurs when there is an issue or update in Azure and many servers get restarted. All other pollers work normally.

I have no way to debug what the poller is doing (yes, I tried set logs to debug and nothing is out of normal), or reset an individual poller. And therefore I have to restart the zabbix server service.

I use FreeTDS, with default values, and I don't remember this happening in Zabbix 4.0.

I don't know if you can find and fix this issue, but maybe at least add a poller health check and see if it is responding to get new values.

Thanks in advance, and sorry if I can't provide more or exact evidence to reproduce, this is something I'm still trying find a solution for more than a year now.