Using Compute Profiles (vSphere)
Page last updated:
This topic describes how to use compute profiles for Linux- and Windows-based Kubernetes clusters on vSphere with NSX-T networking and for Linux-based Kubernetes clusters on vSphere without NSX-T networking.
Compute profiles let you customize cluster resources parameters.
How Compute Profiles are Created
TKGI cluster administrators can create and delete compute profiles, as described in the Creating and Managing Compute Profiles with the CLI (vSphere) topic.
After an administrator creates a compute profile, cluster managers (pks.cluster.manage
) can create clusters with it or assign it to existing clusters.
List Compute Profiles
To list available compute profiles, run the following command:
tkgi compute-profiles
For example:
$ tkgi compute-profiles Name Description custom-nodes-compute-profile custom-nodes-compute-profile custom-node-pools-compute-profile custom-node-pools-compute-profile Dc-east-single-node-pool A profile for the east datacenter with a single node pool dc-east-mixed A profile for the east datacenter with heterogeneous workers
Create a Cluster with a Compute Profile
You can assign a compute profile to a Kubernetes cluster at the time of cluster creation.
To create an TKGI-provisioned Kubernetes cluster with a compute profile, run the following command:
tkgi create-cluster CLUSTER-NAME --external-hostname HOSTNAME --plan PLAN-NAME --compute-profile COMPUTE-PROFILE-NAME --node-pool-instances "NODE-POOL-NAME:INSTANCES"
Where:
CLUSTER-NAME
is a unique name for your cluster.HOSTNAME
is your external hostname used for accessing the Kubernetes API.PLAN-NAME
is the name of the TKGI plan you want to use for your cluster.COMPUTE-PROFILE-NAME
is the name of the compute profile you want to use for your cluster.- (Optional)
--node-pool-instances
overrides the worker node instance counts specified in the compute profile:NODE-POOL-NAME
is the name of the node pool you want to specify the number of worker node instances for.INSTANCES
is the number of worker nodes for the node pool name specified asNODE-POOL-NAME
.
For example:
tkgi create-cluster custom-node-pools -e test.tkgi.shep.api.com --compute-profile custom-node-pools-compute-profile -p "small" --node-pool-instances "tiny-1:3" TKGI Version: 1.9.0-build.11 Name: test K8s Version: 1.18.5 Plan Name: small UUID:1 Last Action: CREATE Last Action State: in progress Last Action Description: Creating cluster Kubernetes Master Host: test.tkgi.shep.api.com Kubernetes Master Port: 8443 Worker Nodes: 4 Kubernetes Master IP(s): In Progress Network Profile Name: Kubernetes Profile Name: Compute Profile Name: custom-node-pools-compute-profile Tags:
Assign a Compute Profile to an Existing Cluster
TKGI supports changing the compute profile for an already created cluster. You can use this procedure to:
- assign a compute profile to a cluster that does not have one, or
- change a cluster’s existing profile to a new one
Note: Compute profiles do not support moving a cluster’s nodes to different Availability Zones.
Warning: Do not scale out or scale in existing master nodes by reconfiguring the TKGI tile or by using a compute profile. Reducing a cluster’s number of control plane nodes may remove a master node and cause the cluster to become inactive.
This is the procedure to change a cluster’s compute profile:
Do one of the following
- Choose a new compute profile for the cluster. See List Compute Profiles.
- Have a TKGI cluster administrator define and create a new compute profile as described in Create a Compute Profile in Creating and Managing Compute Profiles with the CLI (vSphere).
- The name of the new compute profile must be unique and different from the previously assigned compute profile.
Run the following command to update the cluster with the new compute profile:
tkgi update-cluster CLUSTER-NAME --compute-profile NEW-COMPUTE-PROFILE-NAME ---node-pool-instances "NODE-POOL-NAME:INSTANCES"
Where:
CLUSTER-NAME
is the name of the existing Kubernetes clusterNEW-COMPUTE-PROFILE-NAME
is the name of the new compute profile you want to apply to the cluster.- (Optional)
--node-pool-instances "NODE-POOL-NAME:INSTANCES"
updates the number of worker nodes for node poolNODE-POOL-NAME
to beINSTANCES
.NODE-POOL-NAME
should be the name of the node pool in the new compute profile.
For example:
tkgi update-cluster test --compute-profile new-compute-profile Update summary for cluster test: Compute Profile Name: new-compute-profile Are you sure you want to continue? (y/n): y Use 'tkgi cluster test' to monitor the state of your cluster
Compute Profile Update and Resize Validation
There are strict validation rules for the tkgi update-cluster --compute-profile
command:
- You cannot use the compute profile to change Availability Zones of a cluster.
- If any optional field is unspecified at the time of cluster creation or cluster update, then the value of that field is set to the value of the corresponding field from the plan.
- Values passed to
--num-nodes
are ignored. - Values passed to
--node-pool-instances
must match the labels of the node pools specified in the compute profile.
Clusters created with compute profile in TKGI 1.8 and earlier:
- You cannot update a compute profile for a cluster that already has an associated profile created in TKGI v1.8 or earlier.
Resize a Cluster that Has an Existing Compute Profile
TKGI supports using the CLI to resize a cluster already created or assigned with a compute profile, without having to create a new compute profile.
To do this, use tkg resize
with the --node-pool-instances
option.
Pass in a comma-separated list that associates node pools defined in the compute profile with new instance counts.
This changes the number of instances from each node pool.
The node pools must already be defined in the cluster’s compute profile.
For example, to resize a cluster custom-node-pools-cluster
to use three nodes from the node pool named tiny-1
and seven from the pool named medium-2
:
tkgi resize custom-node-pools-cluster --node-pool-instances "tiny-1:3,medium-2:7"
Please send any feedback you have to pks-feedback@pivotal.io.