Guest VM becomes unresponsive after 2 or 3 days continuous uptime

[linus100]

Using CentOS 7.5.1804 host and Windows 7 guest with dual monitors full screen, USB mouse.

After the VM has been running continuously for about 2 or 3 days, the VM starts going into an abnormal wait state. The cursor can be moved with the mouse, but it changes from a pointer to a blue circle after clicking on something. At first, the "wait state" circle is short-lived, changing back to the normal pointer after less than a second. The problem gets worse over time and the cursor will begin to flash between normal pointer and blue circle a few times and then for longer and longer intervals. The longer I let it go, the more time it spends in the wait state. When the cursor is displayed as a blue circle, any button press on the mouse is ignored as well as key presses on the keyboard, although there was one time that I must have been clicking all over the place trying to get a response and on killing the VM, those clicks seemed to appear in the Linux host.

If I let the problem persist and don't reboot the VM, eventually the VM becomes completely unusable. The cursor stayed as a blue circle for over 30 minutes one time before I gave up and killed it. I have to reboot the VM to clear the problem and when I first started doing that, I would halt the VM and also reboot the Linux machine. More recently, I have only been shutting down the VM and restarting it. Earlier today, I shut down the VM and restarted, but noticed I was still having the problem. I shut the VM down again, exited the VB GUI console and made sure all VB processes were stopped, then started it back up. That cleared the problem.

When the guest system enters a wait state hang, CPU utilization on both host and guest still seem normal; no CPUs are pegged. When the guest seems to have entered a permanent wait state, it is still getting work done. I have had this happen when the Windows task manager was running and showing the CPU usage history. I had it set to slow updates, meaning it should update every 4 seconds but I was getting updates more like around 16 seconds.

I had been using the paravirtualized network driver and wondered if that could be a problem, so changed to the Intel PRO/1000 MT Desktop adapter with no change.

I first started having this problem in later versions of 5.0 on CentOS 6 and I had other problems with 5.1 so didn't run that for any appreciable amount of time. I have captured trace files of a hung VM if that could be useful to anyone.

I see similar reports on the forum about lost mouse clicks, and dual monitors may be one of the conditions required to reproduce.

[Socratis]

Related discussion in the forums: ​https://forums.virtualbox.org/viewtopic.php?f=7&t=88158

[linus100]

The issue is actually specific to the display windows opened by virtualbox. I can use rdesktop to RDP in to the machine that appears to be hung and successfully interact with the guest. I shut it down and then looked for processes still running and see this:

[root@lh2 ~]# ps -ef | grep -i virtual root 19364 15297 0 13:58 pts/34 00:00:00 grep -i virtual linush 30645 6213 0 Aug31 ? 00:43:33 /usr/lib/virtualbox/VirtualBox linush 30669 1 0 Aug31 ? 00:30:00 /usr/lib/virtualbox/VBoxXPCOMIPCD linush 30674 1 1 Aug31 ? 00:57:23 /usr/lib/virtualbox/VBoxSVC --auto-shutdown linush 30839 30674 99 Aug31 ? 4-06:54:36 /usr/lib/virtualbox/VirtualBox --comment ESG1-64 --startvm edb632c3-0604-4671-893a-92b3d9f91ac8 --no-startvm-errormsgbox

The virtual machine is no longer running; it's status in the VM VirtualBox Manager shows "Powered Off."

I'm uploading strace's of the remaining threads.

Is there anything else I can provide that might help diagnose the problem?

[linus100]

Trace files captured during hang.