[PVE-User] [OT?] OOM...

[Fabian Grünbichler]

On Tue, Jan 10, 2017 at 02:00:35PM +0100, Falko Trojahn wrote:
> Hello Marco,
> 
> did you ever find out more about your OOMs?
> 
> Hello all,
> 
> I'd like to get some idea what we can do here.
> 
> Since last pve updates last week (no idea if related or not) we get OOMs
> sometimes during the night. We have 5 proxmox nodes with ceph and kvms,
> 3 nodes are servers with Supermicro Boards with >=60 GB RAM, two are
> only for transition process from old Proxmox 3.x to new 4.x cluster,
> Asus P6T6 Boards with 12GB (no kvms) and 24GB which will be sorted out
> later if possible.
> 
> When we first noticed the oom, two kvm processes were killed one after
> another, now at least two times a ceph osd process was involved
> (see lists / syslog excerpts further down.
> 
> Our munin graphs never show memory shortages at the time of the ooms,
> seems plenty of RAM available.
> 
> So why does rados kill the process with the most memory, and how
> can this be prevented?
> 
> If more info about our config is needed, please ask.
> 
> Many thanks in advance
> and best regards
> Falko

there is an issue with the 4.4.35-1 kernel in pve-enterprise and OOM,
you can install the 4.4.35-2 one currently in pve-no-subscription (which
should move to pve-enterprise very soon as well).

see
https://forum.proxmox.com/threads/proxmox-4-4-5-kernel-out-of-memory-kill-process-8543-kvm-score-or-sacrifice-child.31569/