Cadvisor metrics differs a lot with CloudWatch metrics

[Alvaro Gil]

Hi,

I've installed Prometheus for first time a few days ago, I've followed the advice to use Cadvisor to monitor containers and also the host machine as it claims to do.

When I compare CloudWatch metrics with the Cadvisor/Prometheus collected ones I am seen a huge difference on for example NetworkIn/Out, I've read some articles talking about this but still not very clear to me what is happening behind scenes and which solution I should trust. Can anyone explain to me why exactly this is happening and if I at least can trust in Cadvisor for container metrics and perhaps use another solution for hosts machines?

[Brian Brazil]

The node exporter and CloudWatch would be a better comparison, as they're both concerned with machine stats. Keep in mind there's a 10 minute lag on data from the cloudwatch_exporter.

--

Brian Brazil

www.robustperception.io