Opened 6 years ago
Last modified 6 years ago
#17644 new defect
VBoxHeadless crash
Reported by: | srats | Owned by: | |
---|---|---|---|
Component: | other | Version: | VirtualBox 5.2.8 |
Keywords: | Cc: | ||
Guest type: | Linux | Host type: | Linux |
Description
Hi everyone.
Since our VirtualBox upgrade (5.0.40 -> 5.2.8), we started seeing some VBoxHeadless process disappearing.
Crash file attached.
We'll be glad to help you in any way we can to fix investigate the issue.
Attachments (1)
Change History (8)
comment:1 by , 6 years ago
comment:2 by , 6 years ago
Try to upgrade the guest additions running inside the guest. They're very old, according to the VM log.
comment:3 by , 6 years ago
See thread on vbox-dev:
https://www.virtualbox.org/pipermail/vbox-dev/2018-April/014992.html
comment:4 by , 6 years ago
Old Guest Additions should not crash a VBoxHeadless process. (Though I do realise that sometimes theory and practice can differ.)
comment:5 by , 6 years ago
As the issue is very hard to reproduce, I do not have any stacktrace yet. I have deployed debug build of VBoxHeadless on some machines, hoping to reproduce it.
In the meantime, I continue investigating the the log I have, and I found the same message appearing pretty much at the same time the VBoxHeadless process crashes. Not sure if it's a cause or a consequence.
Apr 12 15:28:05 <host> kernel: [21282.959309] threaded-ml[2900]: segfault at 1b1 ip 00007fd900a4dd8e sp 00007fd933c82ac0 error 4 in libpulse.so.0.19.0[7fd900a1b000+4e000]
We know we have some issue with audio in our guests since 5.2.
Thanks everyone for the tips, I'll keep you posted.
follow-up: 7 comment:6 by , 6 years ago
Does it work if you launch the VM as "normal"? From the GUI? Normal start?
comment:7 by , 6 years ago
Replying to socratis:
Does it work if you launch the VM as "normal"? From the GUI? Normal start?
Unfortunately, as I cannot reproduce the issue really I can't tell you the behavior of the GUI.
As I suspected, switching the audio backend from PulseAudio to Alsa fixes the issue. I'll continue to use my debug builds on my staging env as much as possible, but I won't spend a lot more time investigate this issue, as my workaround is acceptable for my use case.
We have to rework the audio layer of our guest, so I will give you more data when we start working on this, if the issue is still there.
Thanks everyone.
Crash report is a bit too big for Trac, so here is a Google Drive link to it: https://drive.google.com/file/d/1vqDlOl6mBv-Bhjkko7Qn9k-ZsGF0lGAd/view?usp=sharing