On Tue, Feb 21, 2017 at 12:07:41PM +0800, Yunchih Chen wrote:

On 02/20/2017 09:10 PM, Michal Privoznik wrote:

On 17.02.2017 17:18, Yunchih Chen wrote:

`virsh list` hangs on my server that hosts a bunch of VMs.
This might be due to the Debian upgrade I did on Feb 15, which upgrades
`libvirt` from 2.4.0-1 to 3.0.0-2.
I have tried restarting libvirtd for a few times, without luck.

Attached below are some relevant logs; let me know if you need some more
for debugging.
Thanks for your help!!

root@vm-host:~# uname -a
Linux vm-host 4.6.0-1-amd64 #1 SMP Debian 4.6.4-1 (2016-07-18) x86_64
GNU/Linux

root@vm-host:~# apt-cache policy libvirt-daemon
libvirt-daemon:
   Installed: 3.0.0-2
   Candidate: 3.0.0-2
   Version table:
  *** 3.0.0-2 500
         500 http://debian.csie.ntu.edu.tw/debian testing/main amd64
Packages
         100 /var/lib/dpkg/status

root@vm-host:~# strace -o /tmp/trace -e trace=network,file,poll virsh
list  # hangs forever .....
^C
root@vm-host:~# tail -10 /tmp/trace
access("/etc/libvirt/libvirt.conf", F_OK) = 0
open("/etc/libvirt/libvirt.conf", O_RDONLY) = 5
access("/proc/vz", F_OK)                = -1 ENOENT (No such file or
directory)
socket(AF_UNIX, SOCK_STREAM, 0)         = 5
connect(5, {sa_family=AF_UNIX,
sun_path="/var/run/libvirt/libvirt-sock"}, 110) = 0
getsockname(5, {sa_family=AF_UNIX}, [128->2]) = 0
poll([{fd=5, events=POLLOUT}, {fd=6, events=POLLIN}], 2, -1) = 1
([{fd=5, revents=POLLOUT}])
poll([{fd=5, events=POLLIN}, {fd=6, events=POLLIN}], 2, -1) = ?
ERESTART_RESTARTBLOCK (Interrupted by signal)
--- SIGINT {si_signo=SIGINT, si_code=SI_KERNEL} ---
+++ killed by SIGINT +++

root@vm-host:~# lsof /var/run/libvirt/libvirt-sock # hangs too ...

This is very suspicious. Looks like the daemon is in some weird state
and hence virsh is unable to get list of domains.

# ps axf | grep libvirtd
# gdb -p $(pgrep libvirtd)
(gdb) t a a bt

if you could run those commands and share the output that might shed
more light.

Michal

Unfortunately, gdb also hangs when attaching to libvirt ....

root@vm-host:~# gdb -q -p $(pgrep libvirtd)
Attaching to process 9556
[New LWP 9557]
[New LWP 9558]
[New LWP 9559]
[New LWP 9560]
[New LWP 9561]
[New LWP 9562]
[New LWP 9563]
[New LWP 9564]
[New LWP 9565]
[New LWP 9566]
[New LWP 9567]
[New LWP 9568]
[New LWP 9569]
[New LWP 9570]
[New LWP 9571]
[New LWP 9572]
^C^C^C^C^C^C^C^C^C^C^C

It must be killed with SIGKILL, and libvirtd will die with gdb.

Here[1] is the output of the following command:

strace -o /tmp/gdb-full-trace.txt -s 1024 -f gdb -q -p $(pgrep libvirtd)

Err, that is useless - you've just straced GDB, not libvirtd

You need to strace libvirtd as it starts

   strace -o libvirt.log -f -s 1000 /usr/sbin/libvirtd

Regards,
Daniel

-- 
--
Yun-Chih Chen 陳耘志
Network/Workstation Assistant
Dept. of Computer Science and Information Engineering
National Taiwan University
Tel: +886-2-33664888 ext. 217/204
Email: ta217@csie.ntu.edu.tw
Website: http://wslab.csie.ntu.edu.tw/