There is in-house application in our environment that has a crashing issue. It is housed in a floating pool (~90VMs) that is using a few Tesla M10 GPUs (app used a 3D PDF viewer). They have been tracking the crashes for a couple months now and have recently noticed a trend that anytime the hosts are restarted the application crashes less for about a week and then slowly gets worse. I know it's not normal to reboot hosts and have a hard time believing that this is "fixing" the crash issue.
A couple weeks ago I began restarting the VM's every night in an effort to help with the crash rate. This didn't help. There were a couple issues out of my control that required us to restart these hosts in the past 2 months and they noticed a correlation with crash rates being less after it restarted.
I guess my question could there be any validity to this? Even though I delete, recompose, restart VM's frequently there is something else that is reset when the host restarts?
Sorry if this is making little to no sense. I've been tasked to explain why the app crashes less after a host reboot and at a loss. I don't have a lot of experience with managing ESXi hosts and clusters. I mostly manage our View environment and generally don't have to interact much with the hosts because they just work!
