Chapter 11. Nodes
11.1. Node maintenance
Nodes can be placed into maintenance mode by using the oc adm utility or NodeMaintenance custom resources (CRs).
The node-maintenance-operator (NMO) is no longer shipped with OpenShift Virtualization. It is deployed as a standalone Operator from the OperatorHub in the OpenShift Container Platform web console or by using the OpenShift CLI (oc).
For more information on remediation, fencing, and maintaining nodes, see the Workload Availability for Red Hat OpenShift documentation.
Virtual machines (VMs) must have a persistent volume claim (PVC) with a shared ReadWriteMany (RWX) access mode to be live migrated.
The Node Maintenance Operator watches for new or deleted NodeMaintenance CRs. When a new NodeMaintenance CR is detected, no new workloads are scheduled and the node is cordoned off from the rest of the cluster. All pods that can be evicted are evicted from the node. When a NodeMaintenance CR is deleted, the node that is referenced in the CR is made available for new workloads.
Using a NodeMaintenance CR for node maintenance tasks achieves the same results as the oc adm cordon and oc adm drain commands using standard OpenShift Container Platform custom resource processing.
11.1.1. Eviction strategies
Placing a node into maintenance marks the node as unschedulable and drains all the VMs and pods from it.
You can configure eviction strategies for virtual machines (VMs) or for the cluster.
- VM eviction strategy
The VM
LiveMigrateeviction strategy ensures that a virtual machine instance (VMI) is not interrupted if the node is placed into maintenance or drained. VMIs with this eviction strategy will be live migrated to another node.You can configure eviction strategies for virtual machines (VMs) by using the web console or the command line.
ImportantThe default eviction strategy is
LiveMigrate. A non-migratable VM with aLiveMigrateeviction strategy might prevent nodes from draining or block an infrastructure upgrade because the VM is not evicted from the node. This situation causes a migration to remain in aPendingorSchedulingstate unless you shut down the VM manually.You must set the eviction strategy of non-migratable VMs to
LiveMigrateIfPossible, which does not block an upgrade, or toNone, for VMs that should not be migrated.
- Cluster eviction strategy
- You can configure an eviction strategy for the cluster to prioritize workload continuity or infrastructure upgrade.
Configuring a cluster eviction strategy is a Technology Preview feature only. Technology Preview features are not supported with Red Hat production service level agreements (SLAs) and might not be functionally complete. Red Hat does not recommend using them in production. These features provide early access to upcoming product features, enabling customers to test functionality and provide feedback during the development process.
For more information about the support scope of Red Hat Technology Preview features, see Technology Preview Features Support Scope.
| Eviction strategy | Description | Interrupts workflow | Blocks upgrades |
|---|---|---|---|
|
| Prioritizes workload continuity over upgrades. | No | Yes 2 |
|
| Prioritizes upgrades over workload continuity to ensure that the environment is updated. | Yes | No |
|
| Shuts down VMs with no eviction strategy. | Yes | No |
- Default eviction strategy for multi-node clusters.
- If a VM blocks an upgrade, you must shut down the VM manually.
- Default eviction strategy for single-node OpenShift.
11.1.1.1. Configuring a VM eviction strategy using the command line
You can configure an eviction strategy for a virtual machine (VM) by using the command line.
The default eviction strategy is LiveMigrate. A non-migratable VM with a LiveMigrate eviction strategy might prevent nodes from draining or block an infrastructure upgrade because the VM is not evicted from the node. This situation causes a migration to remain in a Pending or Scheduling state unless you shut down the VM manually.
You must set the eviction strategy of non-migratable VMs to LiveMigrateIfPossible, which does not block an upgrade, or to None, for VMs that should not be migrated.
Procedure
Edit the
VirtualMachineresource by running the following command:$ oc edit vm <vm_name> -n <namespace>
Example eviction strategy
apiVersion: kubevirt.io/v1 kind: VirtualMachine metadata: name: <vm_name> spec: template: spec: evictionStrategy: LiveMigrateIfPossible 1 # ...- 1
- Specify the eviction strategy. The default value is
LiveMigrate.
Restart the VM to apply the changes:
$ virtctl restart <vm_name> -n <namespace>
11.1.1.2. Configuring a cluster eviction strategy by using the command line
You can configure an eviction strategy for a cluster by using the command line.
Configuring a cluster eviction strategy is a Technology Preview feature only. Technology Preview features are not supported with Red Hat production service level agreements (SLAs) and might not be functionally complete. Red Hat does not recommend using them in production. These features provide early access to upcoming product features, enabling customers to test functionality and provide feedback during the development process.
For more information about the support scope of Red Hat Technology Preview features, see Technology Preview Features Support Scope.
Procedure
Edit the
hyperconvergedresource by running the following command:$ oc edit hyperconverged kubevirt-hyperconverged -n openshift-cnv
Set the cluster eviction strategy as shown in the following example:
Example cluster eviction strategy
apiVersion: hco.kubevirt.io/v1beta1 kind: HyperConverged metadata: name: kubevirt-hyperconverged spec: evictionStrategy: LiveMigrate # ...
11.1.2. Run strategies
A virtual machine (VM) configured with spec.running: true is immediately restarted. The spec.runStrategy key provides greater flexibility for determining how a VM behaves under certain conditions.
The spec.runStrategy and spec.running keys are mutually exclusive. Only one of them can be used.
A VM configuration with both keys is invalid.
11.1.2.1. Run strategies
The spec.runStrategy key has four possible values:
Always-
The virtual machine instance (VMI) is always present when a virtual machine (VM) is created on another node. A new VMI is created if the original stops for any reason. This is the same behavior as
running: true. RerunOnFailure- The VMI is re-created on another node if the previous instance fails. The instance is not re-created if the VM stops successfully, such as when it is shut down.
Manual-
You control the VMI state manually with the
start,stop, andrestartvirtctl client commands. The VM is not automatically restarted. Halted-
No VMI is present when a VM is created. This is the same behavior as
running: false.
Different combinations of the virtctl start, stop and restart commands affect the run strategy.
The following table describes a VM’s transition between states. The first column shows the VM’s initial run strategy. The remaining columns show a virtctl command and the new run strategy after that command is run.
| Initial run strategy | Start | Stop | Restart |
|---|---|---|---|
| Always | - | Halted | Always |
| RerunOnFailure | - | Halted | RerunOnFailure |
| Manual | Manual | Manual | Manual |
| Halted | Always | - | - |
If a node in a cluster installed by using installer-provisioned infrastructure fails the machine health check and is unavailable, VMs with runStrategy: Always or runStrategy: RerunOnFailure are rescheduled on a new node.
11.1.2.2. Configuring a VM run strategy by using the command line
You can configure a run strategy for a virtual machine (VM) by using the command line.
The spec.runStrategy and spec.running keys are mutually exclusive. A VM configuration that contains values for both keys is invalid.
Procedure
Edit the
VirtualMachineresource by running the following command:$ oc edit vm <vm_name> -n <namespace>
Example run strategy
apiVersion: kubevirt.io/v1 kind: VirtualMachine spec: runStrategy: Always # ...
11.1.3. Maintaining bare metal nodes
When you deploy OpenShift Container Platform on bare metal infrastructure, there are additional considerations that must be taken into account compared to deploying on cloud infrastructure. Unlike in cloud environments where the cluster nodes are considered ephemeral, re-provisioning a bare metal node requires significantly more time and effort for maintenance tasks.
When a bare metal node fails, for example, if a fatal kernel error happens or a NIC card hardware failure occurs, workloads on the failed node need to be restarted elsewhere else on the cluster while the problem node is repaired or replaced. Node maintenance mode allows cluster administrators to gracefully power down nodes, moving workloads to other parts of the cluster and ensuring workloads do not get interrupted. Detailed progress and node status details are provided during maintenance.
11.1.4. Additional resources
11.2. Managing node labeling for obsolete CPU models
You can schedule a virtual machine (VM) on a node as long as the VM CPU model and policy are supported by the node.
11.2.1. About node labeling for obsolete CPU models
The OpenShift Virtualization Operator uses a predefined list of obsolete CPU models to ensure that a node supports only valid CPU models for scheduled VMs.
By default, the following CPU models are eliminated from the list of labels generated for the node:
Example 11.1. Obsolete CPU models
"486" Conroe athlon core2duo coreduo kvm32 kvm64 n270 pentium pentium2 pentium3 pentiumpro phenom qemu32 qemu64
This predefined list is not visible in the HyperConverged CR. You cannot remove CPU models from this list, but you can add to the list by editing the spec.obsoleteCPUs.cpuModels field of the HyperConverged CR.
11.2.2. About node labeling for CPU features
Through the process of iteration, the base CPU features in the minimum CPU model are eliminated from the list of labels generated for the node.
For example:
-
An environment might have two supported CPU models:
PenrynandHaswell. If
Penrynis specified as the CPU model forminCPU, each base CPU feature forPenrynis compared to the list of CPU features supported byHaswell.Example 11.2. CPU features supported by
Penrynapic clflush cmov cx16 cx8 de fpu fxsr lahf_lm lm mca mce mmx msr mtrr nx pae pat pge pni pse pse36 sep sse sse2 sse4.1 ssse3 syscall tsc
Example 11.3. CPU features supported by
Haswellaes apic avx avx2 bmi1 bmi2 clflush cmov cx16 cx8 de erms fma fpu fsgsbase fxsr hle invpcid lahf_lm lm mca mce mmx movbe msr mtrr nx pae pat pcid pclmuldq pge pni popcnt pse pse36 rdtscp rtm sep smep sse sse2 sse4.1 sse4.2 ssse3 syscall tsc tsc-deadline x2apic xsave
If both
PenrynandHaswellsupport a specific CPU feature, a label is not created for that feature. Labels are generated for CPU features that are supported only byHaswelland not byPenryn.Example 11.4. Node labels created for CPU features after iteration
aes avx avx2 bmi1 bmi2 erms fma fsgsbase hle invpcid movbe pcid pclmuldq popcnt rdtscp rtm sse4.2 tsc-deadline x2apic xsave
11.2.3. Configuring obsolete CPU models
You can configure a list of obsolete CPU models by editing the HyperConverged custom resource (CR).
Procedure
Edit the
HyperConvergedcustom resource, specifying the obsolete CPU models in theobsoleteCPUsarray. For example:apiVersion: hco.kubevirt.io/v1beta1 kind: HyperConverged metadata: name: kubevirt-hyperconverged namespace: openshift-cnv spec: obsoleteCPUs: cpuModels: 1 - "<obsolete_cpu_1>" - "<obsolete_cpu_2>" minCPUModel: "<minimum_cpu_model>" 2- 1
- Replace the example values in the
cpuModelsarray with obsolete CPU models. Any value that you specify is added to a predefined list of obsolete CPU models. The predefined list is not visible in the CR. - 2
- Replace this value with the minimum CPU model that you want to use for basic CPU features. If you do not specify a value,
Penrynis used by default.
11.3. Preventing node reconciliation
Use skip-node annotation to prevent the node-labeller from reconciling a node.
11.3.1. Using skip-node annotation
If you want the node-labeller to skip a node, annotate that node by using the oc CLI.
Prerequisites
-
You have installed the OpenShift CLI (
oc).
Procedure
Annotate the node that you want to skip by running the following command:
$ oc annotate node <node_name> node-labeller.kubevirt.io/skip-node=true 1- 1
- Replace
<node_name>with the name of the relevant node to skip.
Reconciliation resumes on the next cycle after the node annotation is removed or set to false.
11.3.2. Additional resources
11.4. Deleting a failed node to trigger virtual machine failover
If a node fails and machine health checks are not deployed on your cluster, virtual machines (VMs) with runStrategy: Always configured are not automatically relocated to healthy nodes. To trigger VM failover, you must manually delete the Node object.
If you installed your cluster by using installer-provisioned infrastructure and you properly configured machine health checks, the following events occur:
- Failed nodes are automatically recycled.
-
Virtual machines with
runStrategyset toAlwaysorRerunOnFailureare automatically scheduled on healthy nodes.
11.4.1. Prerequisites
-
A node where a virtual machine was running has the
NotReadycondition. -
The virtual machine that was running on the failed node has
runStrategyset toAlways. -
You have installed the OpenShift CLI (
oc).
11.4.2. Deleting nodes from a bare metal cluster
When you delete a node using the CLI, the node object is deleted in Kubernetes, but the pods that exist on the node are not deleted. Any bare pods not backed by a replication controller become inaccessible to OpenShift Container Platform. Pods backed by replication controllers are rescheduled to other available nodes. You must delete local manifest pods.
Procedure
Delete a node from an OpenShift Container Platform cluster running on bare metal by completing the following steps:
Mark the node as unschedulable:
$ oc adm cordon <node_name>
Drain all pods on the node:
$ oc adm drain <node_name> --force=true
This step might fail if the node is offline or unresponsive. Even if the node does not respond, it might still be running a workload that writes to shared storage. To avoid data corruption, power down the physical hardware before you proceed.
Delete the node from the cluster:
$ oc delete node <node_name>
Although the node object is now deleted from the cluster, it can still rejoin the cluster after reboot or if the kubelet service is restarted. To permanently delete the node and all its data, you must decommission the node.
- If you powered down the physical hardware, turn it back on so that the node can rejoin the cluster.
11.4.3. Verifying virtual machine failover
After all resources are terminated on the unhealthy node, a new virtual machine instance (VMI) is automatically created on a healthy node for each relocated VM. To confirm that the VMI was created, view all VMIs by using the oc CLI.
11.4.3.1. Listing all virtual machine instances using the CLI
You can list all virtual machine instances (VMIs) in your cluster, including standalone VMIs and those owned by virtual machines, by using the oc command-line interface (CLI).
Procedure
List all VMIs by running the following command:
$ oc get vmis -A
