Kubernetes: No GKE CPU usage for own containers on Stackdriver
Hello,
I have noticed the GKE Container CPU Usage in Stackdriver dashboards only reports CPU usage for containers that are created by default (kube2ky, glbc, etcd, heapster and so on) when creating a container cluster on GCP but not for containers I have created for my own applications. This used to work back in Kubernetes 1.1.x. The GKE Container Memory, Disk usage and Page Faults graphs correctly displays data for my own containers though.
My 5 clusters (4 on one account, 1 on my personal account) are all on Kubernetes 1.2.4 and all have Cloud Monitoring/Logging enabled.
Was this feature changed, disabled or removed in 1.2.x?
Thank you!
alexferl
All 3 comments
cc @a-robinson @mwielgus
piosz
on 12 Jun 2016
To be precise, those default graphs show CPU _utilization_, not usage time. Utilization is only a meaningful metric for containers that have a CPU limit specified, since it's effectively CPU usage time / reserved cores. The system containers you list have a CPU limit attribute specified on them, while presumably your pods don't.
On 1.1.x, IIRC the system would use an arbitrary low number as the limit if there wasn't one. This mean that a utilization could always be calculated, but it was often misleading given that containers could go well over 100% utilization.
a-robinson
on 12 Jun 2016
@a-robinson Thanks!
alexferl
on 12 Jun 2016
Related issues
rhohubbuild
路
3Comments
tyranron
路
3Comments
arun-gupta
路
3Comments
Seb-Solon
路
3Comments
jason-riddle
路
3Comments
Most helpful comment
To be precise, those default graphs show CPU _utilization_, not usage time. Utilization is only a meaningful metric for containers that have a CPU limit specified, since it's effectively
CPU usage time / reserved cores. The system containers you list have a CPU limit attribute specified on them, while presumably your pods don't.On 1.1.x, IIRC the system would use an arbitrary low number as the limit if there wasn't one. This mean that a utilization could always be calculated, but it was often misleading given that containers could go well over 100% utilization.