Weird node behaviour

Seems it’s a SMR disk

It would be better to avoid SMR, if that not possible, then try to move databases to more fast disk:

If you do not have more fast disk, then there is another option - add a new node with a separate disk, new generated identity, new authorization token and different external ports (the right part of the port mapping should not be changed). It will spread the same load to two nodes instead of one.