Deleting snapshot on-line causes I/O errors in guest

[Timothe Litt]

Host: Fedora 28 Guests: Fedora 28 x2 + (2 older Fedora machines)

Both guests in this event are F28 & had a single snapshot + current state, and were up (but largely idle).

Deleted snapshot of first machine. (Using VirtualBox GUI). It took quite a while, as the snapshot was almost 2 yrs old.

While the delete was running, I had an ssh session into the VM. I got an error "Read only filesystem", then "ps not found". ls worked - for a while.

This typically means that the kernel got an unrecoverable Disk I/O error - e.g. a timeout. But as the logs live on the disk in question, and it went read-only, I can't prove it.

shutdown was also not found. I waited a while, and commands started getting bus errors. Seemed clear that the disks were off-line.

So I waited for the disk merge (snapshot delete) to finish, did a vboxmanage reset, and rebooted the VM.

Log file will be attached. Note that it also shows a lot of display resize events - I see no reason for this. The VM was started detachable; the GUI is visible, but it seems to blink for no particular reason...

Here is an 'ls' of the VDI involved:

 ls -lih /home/root/VirtualBox_VMs/Overkiller/*.vdi /home/root/VirtualBox_VMs/Overkiller/*/*.vdi
12323679 -rw-------. 1 root root 1.6G Mar 20  2016 /home/root/VirtualBox_VMs/Overkiller/Fedora23_Overkiller.vdi
12320772 -rw-------. 1 root root  35G Aug 18 16:04 /home/root/VirtualBox_VMs/Overkiller/Fedora_17_Overkill.vdi
13893639 -rw-------. 1 root root  22G Aug 18 15:49 /home/root/VirtualBox_VMs/Overkiller/Snapshots/{e981f25e-0fc7-4ba7-9666-9d00bc48eae3}.vdi

The I/O errors don't appear to be a one-off event. I had a second VM which also had a single snapshot that I deleted the same way. It suffered the same fate. Log also attached. Its ls:

ls -lih /home/root/VirtualBox_VMs/WikiWorld/*.vdi /home/root/VirtualBox_VMs/WikiWorld/*/*.vdi
9306120 -rw-------. 1 root root 30G Aug 18 16:08 /home/root/VirtualBox_VMs/WikiWorld/Snapshots/{1bb6e5e7-1e5a-40c8-8d61-daf4ff6611be}.vdi
9306115 -rw-------. 1 root root 49G Aug 18 16:08 /home/root/VirtualBox_VMs/WikiWorld/atto_wiki.vdi

Snapshot delete times were on the order of 30 minutes (each).

If deleting a snapshot on-line can cause this much havoc, it should not be allowed...

[Socratis]

Replying to tlhackque:

If deleting a snapshot on-line can cause this much havoc, it should not be allowed...

Deleting a snapshot on-line is something that I've done many, many times. And it's working.

But... at the same time I try to keep the load at a minimum. There's lots of I/O going on, outside of the knowledge of the guest. If you put a lot of workload to your guest, the Linux kernel starts to thing that there's something wrong with the underlying hardware (due to the delays) and goes into read-only mode. Search the forums for this issue...

BTW, it's usually better and faster, if issues get first addressed in the ​VirtualBox forums, a lot more eyes. More than 95% of the issues are resolved over there, which keeps the developers focusing on the bug fixes and enhancements, and there is no need for another ticket to keep track of.