[pve-devel] Kernel 4.2.8-1 problems

[Fabian Grünbichler]

> Paul Penev <ppquant at gmail.com> hat am 18. März 2016 um 09:05 geschrieben:
> 
> 
> For a day I thought it was hardware related, but tonight the problem
> reappeared again. This time, however, not inside NFS but somewhere
> around fuse. The only fuse filesystems I have mounted are
> lxcfs-related.
> 
> I'm attaching the logs below. I tried attaching gdb to the relevant
> processes, but it was impossible. GDB would hang at the "attaching"
> stage, so it was impossible to get more insight.
> 
> Look at the end, when the cgroup was killed because of low-memory
> situation. This could've been inside an LXC container, because the
> server still had 5GB of free memory left (checked in /proc/meminfo,
> not the proxmox gui).
> 

Did you try attaching gdb to the lxcfs processes? Preferably after installing
lxcfs-dbg ;)

There are updated packages in pvetest (lxcfs and lxcfs-dbg) that fix a PID
namespace related issue in lxcfs that can lead to hanging FUSE accesses. If you
have stuff running regularly that accesses files provided by lxcfs (meminfo,
cpuinfo, uptime in /proc for example), the hanging processes can accumulate
until the memory limit in the container is reached and the OOM killer for that
cgroup is triggered.

See
https://forum.proxmox.com/threads/pve-suddunly-stopped-working-all-cts-unrecheable.26458/page-2
for howto collect debugging output and for the updated package links.

Regards,
Fabian