Comment 4 for bug 1874032

OK, I have a better understanding on what we manage in Nova.

So, in order to prevent intensive I/O operations to compete, we recently added in https://review.opendev.org/#/c/609180/ a new compute semaphore that would hold competitive accesses to disk within Nova thru a configuration option.

In our case here, you can see the semaphore there : https://github.com/openstack/nova/blob/58aaffade9f78c5fdc6b0d26ec26b70908b7a6aa/nova/virt/libvirt/driver.py#L2420

As you can see in the change, there is now a config option that allows you to refrain concurrent I/O ops in Nova by setting max_concurrent_disk_ops to other value but 0.

Closing the bug as Invalid/Wishlist as I think there is a mitigation to your existing problem but feel free to discuss this problem either by filing a blueprint for describing the problem or by raising this at our next virtual PTG : https://etherpad.openstack.org/p/nova-victoria-ptg (which I hope you could attend)