KubeClusterMetrics: node and app cpu/mem usage. #72
Conversation
andrewd-zededa
force-pushed
the
kubeclustermetrics
branch
2 times, most recently
from
eaef607 to
4cbe160
Compare
proto/metrics/metrics.proto
Outdated
| } | ||
|
|
||
| // Resources used as reported by kubernetes api for pods running eve user apps. | ||
| message KubeEVEAppPodMetrics { |
There was a problem hiding this comment.
Why only pod metrics ? What about the VMI ? Do we still use pod metrics for VMI too ?
There was a problem hiding this comment.
Each vmi app has an associated pod (virt-launcher-*), vmi and vmirs objects in the eve-kube-app namespace.
The vmi lists cpu cores, memory Gi resources, and storage bytes requested.
The vmirs list cpu cores and memory Gi resources.
The virt-launcher pod lists two containers virt-launcher-monitor (the vm) and virt-tail (logs). Neither of these specifically list cores but instead a Millicore value which doesn't seem to match the core count. My virt-launcher reports 200millicores which is 1/5th of a core instead of the vmi and vmirs which report 2 cores. Using memory requests from the virt-launcher pod does seem to be the most complete view of all memory used by all containers associated with a running domain.
There was a problem hiding this comment.
@zedi-pramodh I've changed KubeEVEAppPodMetrics to a generically named KubeEVEAppMetrics and KubeClusterMetrics now has two lists of it eve_pod_apps and eve_vmi_apps. How does this look?
There was a problem hiding this comment.
@zedi-pramodh I've changed KubeEVEAppPodMetrics to a generically named KubeEVEAppMetrics and KubeClusterMetrics now has two lists of it eve_pod_apps and eve_vmi_apps. How does this look?
That sounds right.
There was a problem hiding this comment.
Not sure if that causes a change in controller API. Just cross check with @xyuria-zededa
andrewd-zededa
force-pushed
the
kubeclustermetrics
branch
from
4cbe160 to
809c11d
Compare
For kubevirt-eve images this adds per node and eve-user-app cpu/memory usage metrics. Fix: add cluster_id to ZInfoKubeClusterUpdateStatus Signed-off-by: Andrew Durbin <andrewd@zededa.com>
Signed-off-by: Andrew Durbin <andrewd@zededa.com>
andrewd-zededa
force-pushed
the
kubeclustermetrics
branch
from
809c11d to
d6f209b
Compare
For kubevirt-eve images this adds per node and eve-user-app cpu/memory usage metrics.
Fix: add cluster_id to ZInfoKubeClusterUpdateStatus