As described in the third item from OSPRH-6932, the BGP CI jobs will create an extra OCP worker with a special configuration:

• test-operator pods will always be scheduled on this worker
• no other pods will be scheduled on this worker

This will be needed not only for BGP, but also for Pidone. Otherwise, the HA tests running disruptive actions could affect the OCP workers running the test itself.

The node configuration can be applied from the ci-fmw and/or architecture's automation file. According to [1], it could be something like this:

spec:
  taints:
    - effect: NoSchedule
      key: testOperator
      value: "true"
    - effect: NoExecute
      key: testOperator
      value: "true"

The test-operator resources CRs will have to support the following options:
A) tolerations
See [1]. Something like this could work for our case:

spec:
  tolerations:
    - key: "testOperator"
      value: "true"
      effect: "NoSchedule"
    - key: "testOperator"
      value: "true"
      effect: "NoExecute"

B) nodeSelector
Something like this could work for our case:

spec:
  nodeSelector:
    kubernetes.io/hostname: worker-3

 
PRs:

• https://github.com/openstack-k8s-operators/test-operator/pull/95
• https://github.com/openstack-k8s-operators/ci-framework/pull/1716
   

[1] https://docs.openshift.com/container-platform/4.9/nodes/scheduling/nodes-scheduler-taints-tolerations.html