cAdvisor prometheus integration returns container_cpu_load_average_10s as 0

Question

I have configured Prometheus to scrape metrics from cAdvisor. However, the metric "container_cpu_load_average_10s" only returns 0. I am able to see the CPU metrics under the cAdvisor web UI correctly but Prometheus receives only 0. It is working fine for other metrics like "container_cpu_system_seconds_total". Could someone point if I am missing something here?

Prometheus version: 2.1.0

Prometheus config:

scrape_configs: - job_name: cadvisor scrape_interval: 5s metrics_path: /metrics scheme: http static_configs: - targets: - 172.17.0.2:8080

cAdvisor version: 0.29.0

What does the metric look like on the /metrics endpoint directly? — karora, Mar 04 '18 at 12:20
It looks like this "container_cpu_system_seconds_total{some list of parameters/labels} 0" — aries, Mar 04 '18 at 21:47
if the metric prometheus can read is always zero then the issue is with whatever is calculating that. — karora, Mar 04 '18 at 23:17

score 9 · Accepted Answer · edited Jan 16 '20 at 00:57

9

In order to get the metric container_cpu_load_average_10s, the cAdvisor must run with the option

--enable_load_reader=true

which is set fo false by default. This is described here.

edited Jan 16 '20 at 00:57

Pang

9,564
146
81
122

answered Mar 06 '18 at 08:18

aries

849
3
11
24

it still doesn't work for me( entrypoint: /usr/bin/cadvisor -logtostderr -enable_load_reader=true – yuklia Nov 26 '18 at 11:19
2

Same, I'm getting errors like this: `W1128 07:45:13.721484 1 container.go:380] Could not initialize cpu load reader for "/docker": failed to create a netlink based cpuload reader: failed to get netlink family id for task stats: netlink request failed with error errno 0` – Wout Neirynck Nov 28 '18 at 07:55
2

have the same issue, found this: > the Task Stats netlink family (which cAdvisor uses to collect load stats) doesn't work within a network namespace. You'll need to either run your cAdvisor container in the host network (--net=host), or run outside of a container entirely. https://github.com/google/cadvisor/issues/1287 – ignivs Apr 29 '19 at 10:28

CodingFrog · Answer 2 · 2023-01-18T13:54:21.627

0

If the value is zero, it means the container is idle. You don't need 'enable_load_reader'. I don't enable it as it may make cAdvisor unstable. Some useful links:

Linux Load Averages: Solving the Mystery

High CPU utilization but low load average question

cAdvisor enable_load_reader

edited Jan 18 '23 at 13:54

answered Jan 18 '23 at 12:36

CodingFrog

1,225
2
9
17

cAdvisor prometheus integration returns container_cpu_load_average_10s as 0

2 Answers2

Linked