poor-mans-drs

A Red Hat Advanced Cluster Management(ACM) Policy that migrates VM's from nodes that have high cpu and memory usage.

How does it do this:
- ACM will loop through every worker node and get their allocatable CPU/Memory
- ACM will loop through every worker node and get their current CPU/Memory Usage
- ACM policy sets a value for usage % to start migrating nodes, default is set to 50%. Value can be changed in policy.
- ACM will create a configmap called drs-worker-node-name-info on openshift-cnv with the values obtained and calculated from above.
- ACM will loop through virtual machines and compare the status of the nodes they are running on with the values calculated above.
- If node VM is on has high memory and cpu usage, ACM will highlight VM for migration. - ACM will check if their is an existing migration for that VM that is not Failed or Succeded status i.e A running Migration. If there is nor running migration ACM will create a new VM Migration object to move the VM.

Tested On

• OCP 4.15
• ACM 2.11

Deployment steps

• Install ACM

• Install OpenShift Virtualization

• Steps must be run with cluster-admin privileges

• Steps might require subscription-admin privilege

  oc adm policy add-cluster-role-to-user open-cluster-management:subscription-admin $(oc whoami)

• Create Sample Deployment artifacts

  oc apply -k ./deploy

• To exclude a VM from being managed by this policy label the VM with "acm-drs/exclude"

• In the ACM UI , under infrastructure, clustersets,drs clusterset - add the clusters that should be inside the drs clusterset.

• You can run one of the below test scenarios

  Test1:
  Running a stress-ng test pod that moves with the VM. Script below will get VM node and apply a node selector for the same node to the stress-ng pod.The stress-ng pod will stress the node CPU which should trigger the ACM policy to create a VM Migration to move the VM to another node. After a small wait the process should start all over again.

  • Start test and create Infra

  ./test/test1/test1.sh start

  • Exit from Test- CTRL-C from previous command and run below

  ./test/test1/test1.sh stop
  

  Test1 Verification:

  • Once test starts you should see increased load on the stressed node.

  • You can confirm from the configmap created by the policy for tracking metrics. Will be the same as the node on openshift-cnv namespace.

  • Script logs will also tell the source and target VM's for Node migrations.

  • You can also list the VMIM objects to see the migrations.