You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
In short, the Topology Manager wants to allow k8s to deploy pods with hardware topology related requirement.
For example, some high speed VNF will require cpu pinning and cpu isolation, more over, the SRIOV vf that passthrough to pods by sriov-cni needs under the same numa node too.
But none of them has ability to see the system configuration for containers,
like, which cpu core that container process runs on,
and if there's any system cpu affinity set for this proccess.
And the numa node information about container's nic.
User Stories
Monitor tool like prometheus/grafana can get related information and tell the admin if the VNF is under proper NUMA configuration.
Topology Manager is able to make sure if the pod is deployed exactly like what it should be.
Goals
Provide unma/topology related information for containers.
able to tell which core that container process runs on.
able to tell the numa node of container's NIC.
able to tell if there is any cpu affinity set for the container.
Related work
Here's my simple tool to help me get the numa information for containers.
And I think if cadvisor will support this, it would be a big help for topology manger. https://github.com/mJace/numacc
Current work on cadvisor
I've done some work for cadvisor in my personal fork.
cadvisor is able to provide PSR information when doing 'ps' to containers now.
PSR is a ps output option, it shows which cpu that current process runs on. mJace@6e36207
I've create a PR for this.
Future work
to tell the numa node of container's NIC.
to tell if there is any cpu affinity set for the container processes.
The text was updated successfully, but these errors were encountered:
Motivation
The Topology Manager, CPU Manager, Device Manager is now in milestone for kubernetes future release.
https://github.com/kubernetes/enhancements/blob/master/keps/sig-node/0035-20190130-topology-manager.md#new-interfaces.
https://github.com/kubernetes/kubernetes/tree/master/pkg/kubelet/cm/devicemanager
https://github.com/kubernetes/kubernetes/tree/master/pkg/kubelet/cm/cpumanager.
In short, the Topology Manager wants to allow k8s to deploy pods with hardware topology related requirement.
For example, some high speed VNF will require cpu pinning and cpu isolation, more over, the SRIOV vf that passthrough to pods by sriov-cni needs under the same numa node too.
But none of them has ability to see the system configuration for containers,
like, which cpu core that container process runs on,
and if there's any system cpu affinity set for this proccess.
And the numa node information about container's nic.
User Stories
Goals
Related work
Here's my simple tool to help me get the numa information for containers.
And I think if cadvisor will support this, it would be a big help for topology manger.
https://github.com/mJace/numacc
Current work on cadvisor
I've done some work for cadvisor in my personal fork.
cadvisor is able to provide PSR information when doing 'ps' to containers now.
PSR is a
ps
output option, it shows which cpu that current process runs on.mJace@6e36207
I've create a PR for this.
Future work
The text was updated successfully, but these errors were encountered: