kuryr-conroller and cni should igore system k8s resource

Bug #1755757 reported by QingchuanHao
6
This bug affects 1 person
Affects Status Importance Assigned to Milestone
kuryr-kubernetes
New
Undecided
QingchuanHao

Bug Description

If a pod in kube-system is deleted, kuryr-cni and kuryr-controller will get the notification and allocated thread from pool to handler the event, But kuryr-controller and kuryr-cni care only about tenant-created resources, not system ones. So it's reasonable to ignore event of system namespaces.

Changed in kuryr-kubernetes:
assignee: nobody → QingchuanHao (haoqingchuan-28)
Revision history for this message
OpenStack Infra (hudson-openstack) wrote : Fix proposed to kuryr-kubernetes (master)

Fix proposed to branch: master
Review: https://review.openstack.org/552873

Revision history for this message
Eunsoo Park (esevan.park) wrote :

Hello, QingchuanHao.

Could you explain why the resources in kube-system should be ignored?
As a simple example, Prometheous pod which is used in a very common use case is usually created in kube-system.
How it could be created unless kuryr-kubernetes watches pods in kube-system?

Thanks.

Revision history for this message
QingchuanHao (haoqingchuan-28) wrote :

Thank your comment, Eunsoo Park.

I also share some questions here:

Kubernetes should be deployed by neutron network, or by administrator?

If Kubernetes use neutron to establish the system network, then some tenant should be allocated for system network, seperatedly to other tenant resources? But it seems there is no "controller" on kubernetes to manage Kuberentes system pods.

If kubernetes is deployed by administor, apart from neutron, then tenant-created pod, e.g. Prometheous should be able to reach kubernetes system pod, and it needs to allow physical network for kerbernetes system network reach neutron tenant network pass.

Maybe after answering these question, we can address your question :)

Revision history for this message
Irena Berezovsky (irenab) wrote :

I think that handling of the port Add/Remove events should be handled according to the Pod network owner. As long as we support single network owner, kuryr should be handling the events. Once there are multiple network providers as it is currently discussed by k8s-net-sig team, it should be only kuryr destined Pods

Revision history for this message
OpenStack Infra (hudson-openstack) wrote : Change abandoned on kuryr-kubernetes (master)

Change abandoned by Antoni Segura Puimedon (<email address hidden>) on branch: master
Review: https://review.openstack.org/552873
Reason: Lack of activity

To post a comment you must log in.
This report contains Public information  
Everyone can see this information.

Other bug subscribers

Remote bug watches

Bug watches keep track of this bug in other bug trackers.