pve kernel: Out of memory: Killed process

[Thatoo]

It's been few days I face an issue, a VM crash suddenly and I could not find any reason from within the VM. Then I check systemctl in pve and I could find following logs but I don't understand how it can run out of memory. pve is on a 16 GB RAM and there is only one CT set with 2Gb and this VM set (at first) with 12 GB + ballooning. It was working well until recently. When I first found this issue, I then set the VM to 12 GB (with 8 GB minimum, not 12 GB like before).
Is 12 GB to high? Why it didn't happen earlier? Is it link to ZFS?

pve systemctl Logs : https://paste.yunohost.org/icaliriyor.yaml

 

[VictorSTS]

What's the output of arc_summary?

 

[Thatoo]

here is the result of arc_summary from pve shell: https://paste.yunohost.org/edihepokis.yaml

What is arc about?

 

[Ernst T.]

What is arc about?

This is the cache for your ZFS file system. It is 6GB in size.

So you have too little RAM memory.

 

[VictorSTS]

ARC is the caché memory used by ZFS (instead of the typical disk buffer of other filesystems). In your case it is configured to use as much as 6G:

ARC size (current):                                    77.2 %    6.0 GiB
        Target size (adaptive):                        77.7 %    6.0 GiB
        Min size (hard limit):                          6.2 %  495.3 MiB
        Max size (high water):                           16:1    7.7 GiB

You could use as much as 16-6=10GiB for the host and the VM/LXC's, which in practical terms means you have ~8GB for the VMs/LXC. You should either lower the ZFS ARC memory [1], get more RAM or lower the RAM of the VM.

[1] https://pve.proxmox.com/wiki/ZFS_on_Linux#sysadmin_zfs_limit_memory_usage

 

[Thatoo]

Thank you for your explanation.
I can't get more ram.
I have little less than 1Tb of disk so if I understand well, only 3GB for ARC would be enough, isn't it?

 

[VictorSTS]

Should be perfectly fine