How to prevent an LXC container from increasing the sysload of the proxmox hypervisor

[BenDDD]

Hello,

An LXC container had an incident (still undetermined) which caused a very large increase in its sysload average (36,32,19 on 1 CPU).

This led to a very strong increase in the sysload average of the hypervisor that hosts it and consequently impacted all the other containers of this hypervisor.

Is there a mechanism to prevent this spread?

Thank you in advance.

 

[oguz]

hi,

how is the container configured? pct config CTID

caused a very large increase in its sysload average (36,32,19 on 1 CPU)

where did you see this? can you post the output from command:

cat /proc/loadavg

on both container and host?

 

[BenDDD]

how is the container configured? pct config CTID

pct config 193
arch: amd64
cores: 1
hostname: be-resasmm
memory: 512
net0: name=eth0,bridge=vmbr1,firewall=1,gw=147.215.150.1,hwaddr=12:41:40:B3:35:92,ip=147.215.150.36/24,tag=150,type=veth
ostype: debian
rootfs: ceph-UPEM-SSD:vm-193-disk-0,size=20G
swap: 512
unprivileged: 1
unused0: KVM:193/vm-193-disk-0.raw

where did you see this?

On my Nagios + Centreon monitoring server.

cat /proc/loadavg

on both container and host?

root@galaxie7:~# cat /proc/loadavg
2.36 2.49 2.31 1/1205 23527


root@be-resasmm:~# cat /proc/loadavg
1.92 2.36 2.27 1/1206 24362

The incident is over on the be-resasmm container.

 

[oguz]

On my Nagios + Centreon monitoring server.

where does it receive the data from? host or container? the values that come from the container can be unreliable in some cases.

This led to a very strong increase in the sysload average of the hypervisor that hosts it and consequently impacted all the other containers of this hypervisor.

this depends on if the containers are scheduled for the same CPU. the loadavg tells you how many tasks are currently waiting for the CPU in an average time (1m, 5m, 15m) so having a high load average means there could be a lot of processes waiting for the CPU scheduler.

if you want to limit performance impact you could try setting cpuunits (larger number means more cpu time allotted for that VM/container) and see if it makes any difference in your case.

 

[BenDDD]

where does it receive the data from? host or container? the values that come from the container can be unreliable in some cases.

I supervise the host and the container but as the container uses the host's CPU, I normally have the same values.

if you want to limit performance impact you could try setting cpuunits (larger number means more cpu time allotted for that VM/container) and see if it makes any difference in your case.

I will find out about this option and try to configure it correctly.

Thank you for the information.