This is the multi-page printable view of this section. Click here to print.

Return to the regular view of this page.

Kubernetes Guides

Management of a Kubernetes Cluster hosted by Talos Linux

1 - Configuration

How to configure components of the Kubernetes cluster itself.

1.1 - Ceph Storage cluster with Rook

Guide on how to create a simple Ceph storage cluster with Rook for Kubernetes

Preparation

Talos Linux reserves an entire disk for the OS installation, so machines with multiple available disks are needed for a reliable Ceph cluster with Rook and Talos Linux. Rook requires that the block devices or partitions used by Ceph have no partitions or formatted filesystems before use. Rook also requires a minimum Kubernetes version of v1.16 and Helm v3.0 for installation of charts. It is highly recommended that the Rook Ceph overview is read and understood before deploying a Ceph cluster with Rook.

Installation

Creating a Ceph cluster with Rook requires two steps; first the Rook Operator needs to be installed which can be done with a Helm Chart. The example below installs the Rook Operator into the rook-ceph namespace, which is the default for a Ceph cluster with Rook.

$ helm repo add rook-release https://charts.rook.io/release
"rook-release" has been added to your repositories

$ helm install --create-namespace --namespace rook-ceph rook-ceph rook-release/rook-ceph
W0327 17:52:44.277830   54987 warnings.go:70] policy/v1beta1 PodSecurityPolicy is deprecated in v1.21+, unavailable in v1.25+
W0327 17:52:44.612243   54987 warnings.go:70] policy/v1beta1 PodSecurityPolicy is deprecated in v1.21+, unavailable in v1.25+
NAME: rook-ceph
LAST DEPLOYED: Sun Mar 27 17:52:42 2022
NAMESPACE: rook-ceph
STATUS: deployed
REVISION: 1
TEST SUITE: None
NOTES:
The Rook Operator has been installed. Check its status by running:
  kubectl --namespace rook-ceph get pods -l "app=rook-ceph-operator"

Visit https://rook.io/docs/rook/latest for instructions on how to create and configure Rook clusters

Important Notes:
- You must customize the 'CephCluster' resource in the sample manifests for your cluster.
- Each CephCluster must be deployed to its own namespace, the samples use `rook-ceph` for the namespace.
- The sample manifests assume you also installed the rook-ceph operator in the `rook-ceph` namespace.
- The helm chart includes all the RBAC required to create a CephCluster CRD in the same namespace.
- Any disk devices you add to the cluster in the 'CephCluster' must be empty (no filesystem and no partitions).

Once that is complete, the Ceph cluster can be installed with the official Helm Chart. The Chart can be installed with default values, which will attempt to use all nodes in the Kubernetes cluster, and all unused disks on each node for Ceph storage, and make available block storage, object storage, as well as a shared filesystem. Generally more specific node/device/cluster configuration is used, and the Rook documentation explains all the available options in detail. For this example the defaults will be adequate.

$ helm install --create-namespace --namespace rook-ceph rook-ceph-cluster --set operatorNamespace=rook-ceph rook-release/rook-ceph-cluster
NAME: rook-ceph-cluster
LAST DEPLOYED: Sun Mar 27 18:12:46 2022
NAMESPACE: rook-ceph
STATUS: deployed
REVISION: 1
TEST SUITE: None
NOTES:
The Ceph Cluster has been installed. Check its status by running:
  kubectl --namespace rook-ceph get cephcluster

Visit https://rook.github.io/docs/rook/latest/ceph-cluster-crd.html for more information about the Ceph CRD.

Important Notes:
- You can only deploy a single cluster per namespace
- If you wish to delete this cluster and start fresh, you will also have to wipe the OSD disks using `sfdisk`

Now the Ceph cluster configuration has been created, the Rook operator needs time to install the Ceph cluster and bring all the components online. The progression of the Ceph cluster state can be followed with the following command.

$ watch kubectl --namespace rook-ceph get cephcluster rook-ceph
Every 2.0s: kubectl --namespace rook-ceph get cephcluster rook-ceph

NAME        DATADIRHOSTPATH   MONCOUNT   AGE   PHASE         MESSAGE                 HEALTH   EXTERNAL
rook-ceph   /var/lib/rook     3          57s   Progressing   Configuring Ceph Mons

Depending on the size of the Ceph cluster and the availability of resources the Ceph cluster should become available, and with it the storage classes that can be used with Kubernetes Physical Volumes.

$ kubectl --namespace rook-ceph get cephcluster rook-ceph
NAME        DATADIRHOSTPATH   MONCOUNT   AGE   PHASE   MESSAGE                        HEALTH      EXTERNAL
rook-ceph   /var/lib/rook     3          40m   Ready   Cluster created successfully   HEALTH_OK

$ kubectl  get storageclass
NAME                   PROVISIONER                     RECLAIMPOLICY   VOLUMEBINDINGMODE   ALLOWVOLUMEEXPANSION   AGE
ceph-block (default)   rook-ceph.rbd.csi.ceph.com      Delete          Immediate           true                   77m
ceph-bucket            rook-ceph.ceph.rook.io/bucket   Delete          Immediate           false                  77m
ceph-filesystem        rook-ceph.cephfs.csi.ceph.com   Delete          Immediate           true                   77m

Talos Linux Considerations

It is important to note that a Rook Ceph cluster saves cluster information directly onto the node (by default dataDirHostPath is set to /var/lib/rook). If running only a single mon instance, cluster management is little bit more involved, as any time a Talos Linux node is reconfigured or upgraded, the partition that stores the /var file system is wiped, but the --preserve option of talosctl upgrade will ensure that doesn’t happen.

By default, Rook configues Ceph to have 3 mon instances, in which case the data stored in dataDirHostPath can be regenerated from the other mon instances. So when performing maintenance on a Talos Linux node with a Rook Ceph cluster (e.g. upgrading the Talos Linux version), it is imperative that care be taken to maintain the health of the Ceph cluster. Before upgrading, you should always check the health status of the Ceph cluster to ensure that it is healthy.

$ kubectl --namespace rook-ceph get cephclusters.ceph.rook.io rook-ceph
NAME        DATADIRHOSTPATH   MONCOUNT   AGE   PHASE   MESSAGE                        HEALTH      EXTERNAL
rook-ceph   /var/lib/rook     3          98m   Ready   Cluster created successfully   HEALTH_OK

If it is, you can begin the upgrade process for the Talos Linux node, during which time the Ceph cluster will become unhealthy as the node is reconfigured. Before performing any other action on the Talos Linux nodes, the Ceph cluster must return to a healthy status.

$ talosctl upgrade --nodes 172.20.15.5 --image ghcr.io/talos-systems/installer:v0.14.3
NODE          ACK                        STARTED
172.20.15.5   Upgrade request received   2022-03-27 20:29:55.292432887 +0200 CEST m=+10.050399758

$ kubectl --namespace rook-ceph get cephclusters.ceph.rook.io
NAME        DATADIRHOSTPATH   MONCOUNT   AGE   PHASE         MESSAGE                   HEALTH        EXTERNAL
rook-ceph   /var/lib/rook     3          99m   Progressing   Configuring Ceph Mgr(s)   HEALTH_WARN

$ kubectl --namespace rook-ceph wait --timeout=1800s --for=jsonpath='{.status.ceph.health}=HEALTH_OK' rook-ceph
cephcluster.ceph.rook.io/rook-ceph condition met

The above steps need to be performed for each Talos Linux node undergoing maintenance, one at a time.

Cleaning Up

Rook Ceph Cluster Removal

Removing a Rook Ceph cluster requires a few steps, starting with signalling to Rook that the Ceph cluster is really being destroyed. Then all Persistent Volumes (and Claims) backed by the Ceph cluster must be deleted, followed by the Storage Classes and the Ceph storage types.

$ kubectl --namespace rook-ceph patch cephcluster rook-ceph --type merge -p '{"spec":{"cleanupPolicy":{"confirmation":"yes-really-destroy-data"}}}'
cephcluster.ceph.rook.io/rook-ceph patched

$ kubectl delete storageclasses ceph-block ceph-bucket ceph-filesystem
storageclass.storage.k8s.io "ceph-block" deleted
storageclass.storage.k8s.io "ceph-bucket" deleted
storageclass.storage.k8s.io "ceph-filesystem" deleted

$ kubectl --namespace rook-ceph delete cephblockpools ceph-blockpool
cephblockpool.ceph.rook.io "ceph-blockpool" deleted

$ kubectl --namespace rook-ceph delete cephobjectstore ceph-objectstore
cephobjectstore.ceph.rook.io "ceph-objectstore" deleted

$ kubectl --namespace rook-ceph delete cephfilesystem ceph-filesystem
cephfilesystem.ceph.rook.io "ceph-filesystem" deleted

Once that is complete, the Ceph cluster itself can be removed, along with the Rook Ceph cluster Helm chart installation.

$ kubectl --namespace rook-ceph delete cephcluster rook-ceph
cephcluster.ceph.rook.io "rook-ceph" deleted

$ helm --namespace rook-ceph uninstall rook-ceph-cluster
release "rook-ceph-cluster" uninstalled

If needed, the Rook Operator can also be removed along with all the Custom Resource Definitions that it created.

$ helm --namespace rook-ceph uninstall rook-ceph
W0328 12:41:14.998307  147203 warnings.go:70] policy/v1beta1 PodSecurityPolicy is deprecated in v1.21+, unavailable in v1.25+
These resources were kept due to the resource policy:
[CustomResourceDefinition] cephblockpools.ceph.rook.io
[CustomResourceDefinition] cephbucketnotifications.ceph.rook.io
[CustomResourceDefinition] cephbuckettopics.ceph.rook.io
[CustomResourceDefinition] cephclients.ceph.rook.io
[CustomResourceDefinition] cephclusters.ceph.rook.io
[CustomResourceDefinition] cephfilesystemmirrors.ceph.rook.io
[CustomResourceDefinition] cephfilesystems.ceph.rook.io
[CustomResourceDefinition] cephfilesystemsubvolumegroups.ceph.rook.io
[CustomResourceDefinition] cephnfses.ceph.rook.io
[CustomResourceDefinition] cephobjectrealms.ceph.rook.io
[CustomResourceDefinition] cephobjectstores.ceph.rook.io
[CustomResourceDefinition] cephobjectstoreusers.ceph.rook.io
[CustomResourceDefinition] cephobjectzonegroups.ceph.rook.io
[CustomResourceDefinition] cephobjectzones.ceph.rook.io
[CustomResourceDefinition] cephrbdmirrors.ceph.rook.io
[CustomResourceDefinition] objectbucketclaims.objectbucket.io
[CustomResourceDefinition] objectbuckets.objectbucket.io

release "rook-ceph" uninstalled

$ kubectl delete crds cephblockpools.ceph.rook.io cephbucketnotifications.ceph.rook.io cephbuckettopics.ceph.rook.io \
                      cephclients.ceph.rook.io cephclusters.ceph.rook.io cephfilesystemmirrors.ceph.rook.io \
                      cephfilesystems.ceph.rook.io cephfilesystemsubvolumegroups.ceph.rook.io \
                      cephnfses.ceph.rook.io cephobjectrealms.ceph.rook.io cephobjectstores.ceph.rook.io \
                      cephobjectstoreusers.ceph.rook.io cephobjectzonegroups.ceph.rook.io cephobjectzones.ceph.rook.io \
                      cephrbdmirrors.ceph.rook.io objectbucketclaims.objectbucket.io objectbuckets.objectbucket.io
customresourcedefinition.apiextensions.k8s.io "cephblockpools.ceph.rook.io" deleted
customresourcedefinition.apiextensions.k8s.io "cephbucketnotifications.ceph.rook.io" deleted
customresourcedefinition.apiextensions.k8s.io "cephbuckettopics.ceph.rook.io" deleted
customresourcedefinition.apiextensions.k8s.io "cephclients.ceph.rook.io" deleted
customresourcedefinition.apiextensions.k8s.io "cephclusters.ceph.rook.io" deleted
customresourcedefinition.apiextensions.k8s.io "cephfilesystemmirrors.ceph.rook.io" deleted
customresourcedefinition.apiextensions.k8s.io "cephfilesystems.ceph.rook.io" deleted
customresourcedefinition.apiextensions.k8s.io "cephfilesystemsubvolumegroups.ceph.rook.io" deleted
customresourcedefinition.apiextensions.k8s.io "cephnfses.ceph.rook.io" deleted
customresourcedefinition.apiextensions.k8s.io "cephobjectrealms.ceph.rook.io" deleted
customresourcedefinition.apiextensions.k8s.io "cephobjectstores.ceph.rook.io" deleted
customresourcedefinition.apiextensions.k8s.io "cephobjectstoreusers.ceph.rook.io" deleted
customresourcedefinition.apiextensions.k8s.io "cephobjectzonegroups.ceph.rook.io" deleted
customresourcedefinition.apiextensions.k8s.io "cephobjectzones.ceph.rook.io" deleted
customresourcedefinition.apiextensions.k8s.io "cephrbdmirrors.ceph.rook.io" deleted
customresourcedefinition.apiextensions.k8s.io "objectbucketclaims.objectbucket.io" deleted
customresourcedefinition.apiextensions.k8s.io "objectbuckets.objectbucket.io" deleted

Talos Linux Rook Metadata Removal

If the Rook Operator is cleanly removed following the above process, the node metadata and disks should be clean and ready to be re-used. In the case of an unclean cluster removal, there may be still a few instances of metadata stored on the system disk, as well as the partition information on the storage disks. First the node metadata needs to be removed, make sure to update the nodeName with the actual name of a storage node that needs cleaning, and path with the Rook configuration dataDirHostPath set when installing the chart. The following will need to be repeated for each node used in the Rook Ceph cluster.

$ cat <<EOF | kubectl apply -f -
apiVersion: v1
kind: Pod
metadata:
  name: disk-clean
spec:
  restartPolicy: Never
  nodeName: <storage-node-name>
  volumes:
  - name: rook-data-dir
    hostPath:
      path: <dataDirHostPath>
  containers:
  - name: disk-clean
    image: busybox
    securityContext:
      privileged: true
    volumeMounts:
    - name: rook-data-dir
      mountPath: /node/rook-data
    command: ["/bin/sh", "-c", "rm -rf /node/rook-data/*"]
EOF
pod/disk-clean created

$ kubectl wait --timeout=900s --for=jsonpath='{.status.phase}=Succeeded' pod disk-clean
pod/disk-clean condition met

$ kubectl delete pod disk-clean
pod "disk-clean" deleted

Lastly, the disks themselves need the partition and filesystem data wiped before they can be reused. Again, the following as to be repeated for each node and disk used in the Rook Ceph cluster, updating nodeName and of= in the command as needed.

$ cat <<EOF | kubectl apply -f -
apiVersion: v1
kind: Pod
metadata:
  name: disk-wipe
spec:
  restartPolicy: Never
  nodeName: <storage-node-name>
  containers:
  - name: disk-wipe
    image: busybox
    securityContext:
      privileged: true
    command: ["/bin/sh", "-c", "dd if=/dev/zero bs=1M count=100 oflag=direct of=<device>"]
EOF
pod/disk-wipe created

$ kubectl wait --timeout=900s --for=jsonpath='{.status.phase}=Succeeded' pod disk-wipe
pod/disk-wipe condition met

$ kubectl delete pod disk-clean
pod "disk-wipe" deleted

1.2 - Cluster Endpoint

How to explicitly set up an endpoint for the cluster API

In this section, we will step through the configuration of a Talos based Kubernetes cluster. There are three major components we will configure:

  • apid and talosctl
  • the controlplane nodes
  • the worker nodes

Talos enforces a high level of security by using mutual TLS for authentication and authorization.

We recommend that the configuration of Talos be performed by a cluster owner. A cluster owner should be a person of authority within an organization, perhaps a director, manager, or senior member of a team. They are responsible for storing the root CA, and distributing the PKI for authorized cluster administrators.

Talos runs great out of the box, but if you tweak some minor settings it will make your life a lot easier in the future. This is not a requirement, but rather a document to explain some key settings.

Endpoint

To configure the talosctl endpoint, it is recommended you use a resolvable DNS name. This way, if you decide to upgrade to a multi-controlplane cluster you only have to add the ip address to the hostname configuration. The configuration can either be done on a Loadbalancer, or simply trough DNS.

For example:

This is in the config file for the cluster e.g. controlplane.yaml and worker.yaml. for more details, please see: v1alpha1 endpoint configuration

.....
cluster:
  controlPlane:
    endpoint: https://endpoint.example.local:6443
.....

If you have a DNS name as the endpoint, you can upgrade your talos cluster with multiple controlplanes in the future (if you don’t have a multi-controlplane setup from the start) Using a DNS name generates the corresponding Certificates (Kubernetes and Talos) for the correct hostname.

1.3 - Deploying Metrics Server

In this guide you will learn how to set up metrics-server.

Metrics Server enables use of the Horizontal Pod Autoscaler and Vertical Pod Autoscaler. It does this by gathering metrics data from the kubelets in a cluster. By default, the certificates in use by the kubelets will not be recognized by metrics-server. This can be solved by either configuring metrics-server to do no validation of the TLS certificates, or by modifying the kubelet configuration to rotate its certificates and use ones that will be recognized by metrics-server.

Node Configuration

To enable kubelet certificate rotation, all nodes should have the following Machine Config snippet:

machine:
  kubelet:
    extraArgs:
      rotate-server-certificates: true

Install During Bootstrap

We will want to ensure that new certificates for the kubelets are approved automatically. This can easily be done with the Kubelet Serving Certificate Approver, which will automatically approve the Certificate Signing Requests generated by the kubelets.

We can have Kubelet Serving Certificate Approver and metrics-server installed on the cluster automatically during bootstrap by adding the following snippet to the Cluster Config of the node that will be handling the bootstrap process:

cluster:
  extraManifests:
    - https://raw.githubusercontent.com/alex1989hu/kubelet-serving-cert-approver/main/deploy/standalone-install.yaml
    - https://github.com/kubernetes-sigs/metrics-server/releases/latest/download/components.yaml

Install After Bootstrap

If you choose not to use extraManifests to install Kubelet Serving Certificate Approver and metrics-server during bootstrap, you can install them once the cluster is online using kubectl:

kubectl apply -f https://raw.githubusercontent.com/alex1989hu/kubelet-serving-cert-approver/main/deploy/standalone-install.yaml
kubectl apply -f https://github.com/kubernetes-sigs/metrics-server/releases/latest/download/components.yaml

1.4 - iSCSI Storage with Synology CSI

Automatically provision iSCSI volumes on a Synology NAS with the synology-csi driver.

Background

Synology is a company that specializes in Network Attached Storage (NAS) devices. They provide a number of features within a simple web OS, including an LDAP server, Docker support, and (perhaps most relevant to this guide) function as an iSCSI host. The focus of this guide is to allow a Kubernetes cluster running on Talos to provision Kubernetes storage (both dynamic or static) on a Synology NAS using a direct integration, rather than relying on an intermediary layer like Rook/Ceph or Maystor.

This guide assumes a very basic familiarity with iSCSI terminology (LUN, iSCSI target, etc.).

Prerequisites

  • Synology NAS running DSM 7.0 or above
  • Provisioned Talos cluster running Kubernetes v1.20 or above
  • (Optional) Both Volume Snapshot CRDs and the common snapshot controller must be installed in your Kubernetes cluster if you want to use the Snapshot feature

Setting up the Synology user account

The synology-csi controller interacts with your NAS in two different ways: via the API and via the iSCSI protocol. Actions such as creating a new iSCSI target or deleting an old one are accomplished via the Synology API, and require administrator access. On the other hand, mounting the disk to a pod and reading from / writing to it will utilize iSCSI. Because you can only authenticate with one account per DSM configured, that account needs to have admin privileges. In order to minimize access in the case of these credentials being compromised, you should configure the account with the lease possible amount of access – explicitly specify “No Access” on all volumes when configuring the user permissions.

Setting up the Synology CSI

Note: this guide is paraphrased from the Synology CSI readme. Please consult the readme for more in-depth instructions and explanations.

Clone the git repository.

git clone https://github.com/zebernst/synology-csi-talos.git

While Synology provides some automated scripts to deploy the CSI driver, they can be finicky especially when making changes to the source code. We will be configuring and deploying things manually in this guide.

The relevant files we will be touching are in the following locations:

.
├── Dockerfile
├── Makefile
├── config
│   └── client-info-template.yml
└── deploy
    └── kubernetes
        └── v1.20
            ├── controller.yml
            ├── csi-driver.yml
            ├── namespace.yml
            ├── node.yml
            ├── snapshotter
            │   ├── snapshotter.yaml
            │   └── volume-snapshot-class.yml
            └── storage-class.yml

Configure connection info

Use config/client-info-template.yml as an example to configure the connection information for DSM. You can specify one or more storage systems on which the CSI volumes will be created. See below for an example:

---
clients:
- host: 192.168.1.1   # ipv4 address or domain of the DSM
  port: 5000          # port for connecting to the DSM
  https: false        # set this true to use https. you need to specify the port to DSM HTTPS port as well
  username: username  # username
  password: password  # password

Create a Kubernetes secret using the client information config file.

kubectl create secret -n synology-csi generic client-info-secret --from-file=config/client-info.yml

Note that if you rename the secret to something other than client-info-secret, make sure you update the corresponding references in the deployment manifests as well.

Build the Talos-compatible image

Modify the Makefile so that the image is built and tagged under your GitHub Container Registry username:

REGISTRY_NAME=ghcr.io/<username>

When you run make docker-build or make docker-build-multiarch, it will push the resulting image to ghcr.io/<username>/synology-csi:v1.1.0. Ensure that you find and change any reference to synology/synology-csi:v1.1.0 to point to your newly-pushed image within the deployment manifests.

Configure the CSI driver

By default, the deployment manifests include one storage class and one volume snapshot class. See below for examples:

---
apiVersion: storage.k8s.io/v1
kind: StorageClass
metadata:
  annotations:
    storageclass.kubernetes.io/is-default-class: "false"
  name: syno-storage
provisioner: csi.san.synology.com
parameters:
  fsType: 'ext4'
  dsm: '192.168.1.1'
  location: '/volume1'
reclaimPolicy: Retain
allowVolumeExpansion: true
---
apiVersion: snapshot.storage.k8s.io/v1
kind: VolumeSnapshotClass
metadata:
  name: syno-snapshot
  annotations:
    storageclass.kubernetes.io/is-default-class: "false"
driver: csi.san.synology.com
deletionPolicy: Delete
parameters:
  description: 'Kubernetes CSI'

It can be useful to configure multiple different StorageClasses. For example, a popular strategy is to create two nearly identical StorageClasses, with one configured with reclaimPolicy: Retain and the other with reclaimPolicy: Delete. Alternately, a workload may require a specific filesystem, such as ext4. If a Synology NAS is going to be the most common way to configure storage on your cluster, it can be convenient to add the storageclass.kubernetes.io/is-default-class: "true" annotation to one of your StorageClasses.

The following table details the configurable parameters for the Synology StorageClass.

NameTypeDescriptionDefaultSupported protocols
dsmstringThe IPv4 address of your DSM, which must be included in the client-info.yml for the CSI driver to log in to DSM-iSCSI, SMB
locationstringThe location (/volume1, /volume2, …) on DSM where the LUN for PersistentVolume will be created-iSCSI, SMB
fsTypestringThe formatting file system of the PersistentVolumes when you mount them on the pods. This parameter only works with iSCSI. For SMB, the fsType is always ‘cifs‘.ext4iSCSI
protocolstringThe backing storage protocol. Enter ‘iscsi’ to create LUNs or ‘smb‘ to create shared folders on DSM.iscsiiSCSI, SMB
csi.storage.k8s.io/node-stage-secret-namestringThe name of node-stage-secret. Required if DSM shared folder is accessed via SMB.-SMB
csi.storage.k8s.io/node-stage-secret-namespacestringThe namespace of node-stage-secret. Required if DSM shared folder is accessed via SMB.-SMB

The VolumeSnapshotClass can be similarly configured with the following parameters:

NameTypeDescriptionDefaultSupported protocols
descriptionstringThe description of the snapshot on DSM-iSCSI
is_lockedstringWhether you want to lock the snapshot on DSMfalseiSCSI, SMB

Apply YAML manifests

Once you have created the desired StorageClass(es) and VolumeSnapshotClass(es), the final step is to apply the Kubernetes manifests against the cluster. The easiest way to apply them all at once is to create a kustomization.yaml file in the same directory as the manifests and use Kustomize to apply:

kubectl apply -k path/to/manifest/directory

Alternately, you can apply each manifest one-by-one:

kubectl apply -f <file>

Run performance tests

In order to test the provisioning, mounting, and performance of using a Synology NAS as Kubernetes persistent storage, use the following command:

kubectl apply -f speedtest.yaml

Content of speedtest.yaml (source)

kind: PersistentVolumeClaim
apiVersion: v1
metadata:
  name: test-claim
spec:
#  storageClassName: syno-storage
  accessModes:
  - ReadWriteMany
  resources:
    requests:
      storage: 5G
---
apiVersion: batch/v1
kind: Job
metadata:
  name: read
spec:
  template:
    metadata:
      name: read
      labels:
        app: speedtest
        job: read
    spec:
      containers:
      - name: read
        image: ubuntu:xenial
        command: ["dd","if=/mnt/pv/test.img","of=/dev/null","bs=8k"]
        volumeMounts:
        - mountPath: "/mnt/pv"
          name: test-volume
      volumes:
      - name: test-volume
        persistentVolumeClaim:
          claimName: test-claim
      restartPolicy: Never
---
apiVersion: batch/v1
kind: Job
metadata:
  name: write
spec:
  template:
    metadata:
      name: write
      labels:
        app: speedtest
        job: write
    spec:
      containers:
      - name: write
        image: ubuntu:xenial
        command: ["dd","if=/dev/zero","of=/mnt/pv/test.img","bs=1G","count=1","oflag=dsync"]
        volumeMounts:
        - mountPath: "/mnt/pv"
          name: test-volume
      volumes:
      - name: test-volume
        persistentVolumeClaim:
          claimName: test-claim
      restartPolicy: Never

If these two jobs complete successfully, use the following commands to get the results of the speed tests:

# Pod logs for read test:
kubectl logs -l app=speedtest,job=read

# Pod logs for write test:
kubectl logs -l app=speedtest,job=write

When you’re satisfied with the results of the test, delete the artifacts created from the speedtest:

kubectl delete -f speedtest.yaml

1.5 - KubePrism

Enabling in-cluster highly-available controlplane endpoint.

Kubernetes pods running in CNI mode can use the kubernetes.default.svc service endpoint to access the Kubernetes API server, while pods running in host networking mode can only use the external cluster endpoint to access the Kubernetes API server.

Kubernetes controlplane components run in host networking mode, and it is critical for them to be able to access the Kubernetes API server, same as CNI components (when CNI requires access to Kubernetes API).

The external cluster endpoint might be unavailable due to misconfiguration or network issues, or it might have higher latency than the internal endpoint. A failure to access the Kubernetes API server might cause a series of issues in the cluster: pods are not scheduled, service IPs stop working, etc.

KubePrism feature solves this problem by enabling in-cluster highly-available controlplane endpoint on every node in the cluster.

Video Walkthrough

To see a live demo of this writeup, see the video below:

Enabling KubePrism

As of Talos 1.5, KubePrism is not enabled by default.

To enable KubePrism, apply the following machine config patch either during the machine config generation, or to a running cluster (the patch should be applied to all nodes):

machine:
  features:
    kubePrism:
      enabled: true
      port: 7445

Note: the port specified should be available on every node in the cluster.

How it works

Talos spins up a TCP loadbalancer on every machine on the localhost on the specified port which automatically picks up one of the endpoints:

  • the external cluster endpoint as specified in the machine configuration
  • for controlplane machines: https://localhost:<api-server-local-port> (http://localhost:6443 in the default configuration)
  • https://<controlplane-address>:<api-server-port> for every controlplane machine (based on the information from Cluster Discovery)

KubePrism automatically filters out unhealthy (or unreachable) endpoints, and prefers lower-latency endpoints over higher-latency endpoints.

Talos automatically reconfigures kubelet, kube-scheduler and kube-controller-manager to use the KubePrism endpoint. The kube-proxy manifest is also reconfigured to use the KubePrism endpoint by default, but when enabling KubePrism for a running cluster the manifest should be updated with talosctl upgrade-k8s command.

When using CNI components that require access to the Kubernetes API server, the KubePrism endpoint should be passed to the CNI configuration (e.g. Cilium, Calico CNIs).

Notes

As the list of endpoints for KubePrism includes the external cluster endpoint, KubePrism in the worst case scenario will behave the same as the external cluster endpoint. For controlplane nodes, the KubePrism should pick up the localhost endpoint of the kube-apiserver, minimizing the latency. Worker nodes might use direct address of the controlplane endpoint if the latency is lower than the latency of the external cluster endpoint.

KubePrism listen endpoint is bound to localhost address, so it can’t be used outside the cluster.

1.6 - Local Storage

Using local storage with OpenEBS Jiva

If you want to use replicated storage leveraging disk space from a local disk with Talos Linux installed, OpenEBS Jiva is a great option. This requires installing the iscsi-tools system extension.

Since OpenEBS Jiva is a replicated storage, it’s recommended to have at least three nodes where sufficient local disk space is available. The documentation will follow installing OpenEBS Jiva via the offical Helm chart. Since Talos is different from standard Operating Systems, the OpenEBS components need a little tweaking after the Helm installation. Refer to the OpenEBS Jiva documentation if you need further customization.

NB: Also note that the Talos nodes need to be upgraded with --preserve set while running OpenEBS Jiva, otherwise you risk losing data. Even though it’s possible to recover data from other replicas if the node is wiped during an upgrade, this can require extra operational knowledge to recover, so it’s highly recommended to use --preserve to avoid data loss.

Preparing the nodes

Find the matching iscsi-tools image reference for your Talos version by running the following command:

crane export ghcr.io/siderolabs/extensions:v1.5.5 | tar x -O image-digests | grep iscsi-tools

Create a machine config patch with the contents below and save as patch.yaml

- op: add
  path: /machine/install/extensions
  value:
    - image: ghcr.io/siderolabs/iscsi-tools:<version>@sha256:<digest>
- op: add
  path: /machine/kubelet/extraMounts
  value:
    - destination: /var/openebs/local
      type: bind
      source: /var/openebs/local
      options:
        - bind
        - rshared
        - rw

Apply the machine config to all the nodes using talosctl:

talosctl -e <endpoint ip/hostname> -n <node ip/hostname> patch mc -p @patch.yaml

To install the system extension, the node needs to be upgraded. If there is no new release of Talos, the node can be upgraded to the same version as the existing Talos version.

Run the following command on each nodes subsequently:

talosctl -e <endpoint ip/hostname> -n <node ip/hostname> upgrade --image=ghcr.io/siderolabs/installer:v1.5.5

Once the node has upgraded and booted successfully the extension status can be verified by running the following command:

talosctl -e <endpoint ip/hostname> -n <node ip/hostname> get extensions

An output similar to below can be observed:

NODE            NAMESPACE   TYPE              ID                                          VERSION   NAME          VERSION
192.168.20.61   runtime     ExtensionStatus   000.ghcr.io-siderolabs-iscsi-tools-v0.1.1   1         iscsi-tools   v0.1.1

The service status can be checked by running the following command:

talosctl -e <endpoint ip/hostname> -n <node ip/hostname> services

You should see that the ext-tgtd and the ext-iscsid services are running.

NODE            SERVICE      STATE     HEALTH   LAST CHANGE     LAST EVENT
192.168.20.51   apid         Running   OK       64h57m15s ago   Health check successful
192.168.20.51   containerd   Running   OK       64h57m23s ago   Health check successful
192.168.20.51   cri          Running   OK       64h57m20s ago   Health check successful
192.168.20.51   etcd         Running   OK       64h55m29s ago   Health check successful
192.168.20.51   ext-iscsid   Running   ?        64h57m19s ago   Started task ext-iscsid (PID 4040) for container ext-iscsid
192.168.20.51   ext-tgtd     Running   ?        64h57m19s ago   Started task ext-tgtd (PID 3999) for container ext-tgtd
192.168.20.51   kubelet      Running   OK       38h14m10s ago   Health check successful
192.168.20.51   machined     Running   ?        64h57m29s ago   Service started as goroutine
192.168.20.51   trustd       Running   OK       64h57m19s ago   Health check successful
192.168.20.51   udevd        Running   OK       64h57m21s ago   Health check successful

Install OpenEBS Jiva

helm repo add openebs-jiva https://openebs.github.io/jiva-operator
helm repo update
helm upgrade --install --create-namespace --namespace openebs --version 3.2.0 openebs-jiva openebs-jiva/jiva

This will create a storage class named openebs-jiva-csi-default which can be used for workloads. The storage class named openebs-hostpath is used by jiva to create persistent volumes backed by local storage and then used for replicated storage by the jiva controller.

Patching the Namespace

when using the default Pod Security Admissions created by Talos you need the following labels on your namespace:

    pod-security.kubernetes.io/audit: privileged
    pod-security.kubernetes.io/enforce: privileged
    pod-security.kubernetes.io/warn: privileged

or via kubectl:

kubectl label ns openebs pod-security.kubernetes.io/audit=privileged pod-security.kubernetes.io/enforce=privileged pod-security.kubernetes.io/warn=privileged

Number of Replicas

By Default Jiva uses 3 replicas if your cluster consists of lesser nodes consider setting defaultPolicy.replicas to the number of nodes in your cluster e.g. 2.

Patching the jiva installation

Since Jiva assumes iscisd to be running natively on the host and not as a Talos extension service, we need to modify the CSI node daemonset to enable it to find the PID of the iscsid service. The default config map used by Jiva also needs to be modified so that it can execute iscsiadm commands inside the PID namespace of the iscsid service.

Start by creating a configmap definition named config.yaml as below:

apiVersion: v1
kind: ConfigMap
metadata:
  labels:
    app.kubernetes.io/managed-by: pulumi
  name: openebs-jiva-csi-iscsiadm
  namespace: openebs
data:
  iscsiadm: |
    #!/bin/sh
    iscsid_pid=$(pgrep iscsid)

    nsenter --mount="/proc/${iscsid_pid}/ns/mnt" --net="/proc/${iscsid_pid}/ns/net" -- /usr/local/sbin/iscsiadm "$@"

Replace the existing config map with the above config map by running the following command:

kubectl --namespace openebs apply --filename config.yaml

Now we need to update the jiva CSI daemonset to run with hostPID: true so it can find the PID of the iscsid service, by running the following command:

kubectl --namespace openebs patch daemonset openebs-jiva-csi-node --type=json --patch '[{"op": "add", "path": "/spec/template/spec/hostPID", "value": true}]'

Testing a simple workload

In order to test the Jiva installation, let’s first create a PVC referencing the openebs-jiva-csi-default storage class:

kind: PersistentVolumeClaim
apiVersion: v1
metadata:
  name: example-jiva-csi-pvc
spec:
  storageClassName: openebs-jiva-csi-default
  accessModes:
    - ReadWriteOnce
  resources:
    requests:
      storage: 4Gi

and then create a deployment using the above PVC:

apiVersion: apps/v1
kind: Deployment
metadata:
  name: fio
spec:
  selector:
    matchLabels:
      name: fio
  replicas: 1
  strategy:
    type: Recreate
    rollingUpdate: null
  template:
    metadata:
      labels:
        name: fio
    spec:
      containers:
      - name: perfrunner
        image: openebs/tests-fio
        command: ["/bin/bash"]
        args: ["-c", "while true ;do sleep 50; done"]
        volumeMounts:
        - mountPath: /datadir
          name: fio-vol
      volumes:
      - name: fio-vol
        persistentVolumeClaim:
          claimName: example-jiva-csi-pvc

You can clean up the test resources by running the following command:

kubectl delete deployment fio
kubectl delete pvc example-jiva-csi-pvc

1.7 - Pod Security

Enabling Pod Security Admission plugin to configure Pod Security Standards.

Kubernetes deprecated Pod Security Policy as of v1.21, and it is going to be removed in v1.25. Pod Security Policy was replaced with Pod Security Admission. Pod Security Admission is alpha in v1.22 (requires a feature gate) and beta in v1.23 (enabled by default).

In this guide we are going to enable and configure Pod Security Admission in Talos.

Configuration

Talos provides default Pod Security Admission in the machine configuration:

apiVersion: pod-security.admission.config.k8s.io/v1alpha1
kind: PodSecurityConfiguration
defaults:
    enforce: "baseline"
    enforce-version: "latest"
    audit: "restricted"
    audit-version: "latest"
    warn: "restricted"
    warn-version: "latest"
exemptions:
    usernames: []
    runtimeClasses: []
    namespaces: [kube-system]

This is a cluster-wide configuration for the Pod Security Admission plugin:

  • by default baseline Pod Security Standard profile is enforced
  • more strict restricted profile is not enforced, but API server warns about found issues

This default policy can be modified by updating the generated machine configuration before the cluster is created or on the fly by using the talosctl CLI utility.

Verify current admission plugin configuration with:

$ talosctl get admissioncontrolconfigs.kubernetes.talos.dev admission-control -o yaml
node: 172.20.0.2
metadata:
    namespace: controlplane
    type: AdmissionControlConfigs.kubernetes.talos.dev
    id: admission-control
    version: 1
    owner: config.K8sControlPlaneController
    phase: running
    created: 2022-02-22T20:28:21Z
    updated: 2022-02-22T20:28:21Z
spec:
    config:
        - name: PodSecurity
          configuration:
            apiVersion: pod-security.admission.config.k8s.io/v1alpha1
            defaults:
                audit: restricted
                audit-version: latest
                enforce: baseline
                enforce-version: latest
                warn: restricted
                warn-version: latest
            exemptions:
                namespaces:
                    - kube-system
                runtimeClasses: []
                usernames: []
            kind: PodSecurityConfiguration

Usage

Create a deployment that satisfies the baseline policy but gives warnings on restricted policy:

$ kubectl create deployment nginx --image=nginx
Warning: would violate PodSecurity "restricted:latest": allowPrivilegeEscalation != false (container "nginx" must set securityContext.allowPrivilegeEscalation=false), unrestricted capabilities (container "nginx" must set securityContext.capabilities.drop=["ALL"]), runAsNonRoot != true (pod or container "nginx" must set securityContext.runAsNonRoot=true), seccompProfile (pod or container "nginx" must set securityContext.seccompProfile.type to "RuntimeDefault" or "Localhost")
deployment.apps/nginx created
$ kubectl get pods
NAME                     READY   STATUS    RESTARTS   AGE
nginx-85b98978db-j68l8   1/1     Running   0          2m3s

Create a daemonset which fails to meet requirements of the baseline policy:

apiVersion: apps/v1
kind: DaemonSet
metadata:
  labels:
    app: debug-container
  name: debug-container
  namespace: default
spec:
  revisionHistoryLimit: 10
  selector:
    matchLabels:
      app: debug-container
  template:
    metadata:
      creationTimestamp: null
      labels:
        app: debug-container
    spec:
      containers:
      - args:
        - "360000"
        command:
        - /bin/sleep
        image: ubuntu:latest
        imagePullPolicy: IfNotPresent
        name: debug-container
        resources: {}
        securityContext:
          privileged: true
        terminationMessagePath: /dev/termination-log
        terminationMessagePolicy: File
      dnsPolicy: ClusterFirstWithHostNet
      hostIPC: true
      hostPID: true
      hostNetwork: true
      restartPolicy: Always
      schedulerName: default-scheduler
      securityContext: {}
      terminationGracePeriodSeconds: 30
  updateStrategy:
    rollingUpdate:
      maxSurge: 0
      maxUnavailable: 1
    type: RollingUpdate

$ kubectl apply -f debug.yaml
Warning: would violate PodSecurity "restricted:latest": host namespaces (hostNetwork=true, hostPID=true, hostIPC=true), privileged (container "debug-container" must not set securityContext.privileged=true), allowPrivilegeEscalation != false (container "debug-container" must set securityContext.allowPrivilegeEscalation=false), unrestricted capabilities (container "debug-container" must set securityContext.capabilities.drop=["ALL"]), runAsNonRoot != true (pod or container "debug-container" must set securityContext.runAsNonRoot=true), seccompProfile (pod or container "debug-container" must set securityContext.seccompProfile.type to "RuntimeDefault" or "Localhost")
daemonset.apps/debug-container created

Daemonset debug-container gets created, but no pods are scheduled:

$ kubectl get ds
NAME              DESIRED   CURRENT   READY   UP-TO-DATE   AVAILABLE   NODE SELECTOR   AGE
debug-container   0         0         0       0            0           <none>          34s

Pod Security Admission plugin errors are in the daemonset events:

$ kubectl describe ds debug-container
...
  Warning  FailedCreate  92s                daemonset-controller  Error creating: pods "debug-container-kwzdj" is forbidden: violates PodSecurity "baseline:latest": host namespaces (hostNetwork=true, hostPID=true, hostIPC=true), privileged (container "debug-container" must not set securityContext.privileged=true)

Pod Security Admission configuration can also be overridden on a namespace level:

$ kubectl label ns default pod-security.kubernetes.io/enforce=privileged
namespace/default labeled
$ kubectl get ds
NAME              DESIRED   CURRENT   READY   UP-TO-DATE   AVAILABLE   NODE SELECTOR   AGE
debug-container   2         2         0       2            0           <none>          4s

As enforce policy was updated to the privileged for the default namespace, debug-container is now successfully running.

1.8 - Seccomp Profiles

Using custom Seccomp Profiles with Kubernetes workloads.

Seccomp stands for secure computing mode and has been a feature of the Linux kernel since version 2.6.12. It can be used to sandbox the privileges of a process, restricting the calls it is able to make from userspace into the kernel.

Refer the Kubernetes Seccomp Guide for more details.

In this guide we are going to configure a custom Seccomp Profile that logs all syscalls made by the workload.

Preparing the nodes

Create a machine config path with the contents below and save as patch.yaml

machine:
  seccompProfiles:
    - name: audit.json
      value:
        defaultAction: SCMP_ACT_LOG

Apply the machine config to all the nodes using talosctl:

talosctl -e <endpoint ip/hostname> -n <node ip/hostname> patch mc -p @patch.yaml

This would create a seccomp profile name audit.json on the node at /var/lib/kubelet/seccomp/profiles.

The profiles can be used by Kubernetes pods by specfying the pod securityContext as below:

spec:
  securityContext:
    seccompProfile:
      type: Localhost
      localhostProfile: profiles/audit.json

Note that the localhostProfile uses the name of the profile created under profiles directory. So make sure to use path as profiles/<profile-name.json>

This can be verfied by running the below commands:

talosctl -e <endpoint ip/hostname> -n <node ip/hostname> get seccompprofiles

An output similar to below can be observed:

NODE       NAMESPACE   TYPE             ID           VERSION
10.5.0.3   cri         SeccompProfile   audit.json   1

The content of the seccomp profile can be viewed by running the below command:

talosctl -e <endpoint ip/hostname> -n <node ip/hostname> read /var/lib/kubelet/seccomp/profiles/audit.json

An output similar to below can be observed:

{"defaultAction":"SCMP_ACT_LOG"}

Create a Kubernetes workload that uses the custom Seccomp Profile

Here we’ll be using an example workload from the Kubernetes documentation.

First open up a second terminal and run the following talosctl command so that we can view the Syscalls being logged in realtime:

talosctl -e <endpoint ip/hostname> -n <node ip/hostname> dmesg --follow --tail

Now deploy the example workload from the Kubernetes documentation:

kubectl apply -f https://k8s.io/examples/pods/security/seccomp/ga/audit-pod.yaml

Once the pod starts running the terminal running talosctl dmesg command from above should log similar to below:

10.5.0.3: kern:    info: [2022-07-28T11:49:42.489473063Z]: cni0: port 1(veth32488a86) entered blocking state
10.5.0.3: kern:    info: [2022-07-28T11:49:42.490852063Z]: cni0: port 1(veth32488a86) entered disabled state
10.5.0.3: kern:    info: [2022-07-28T11:49:42.492470063Z]: device veth32488a86 entered promiscuous mode
10.5.0.3: kern:    info: [2022-07-28T11:49:42.503105063Z]: IPv6: ADDRCONF(NETDEV_CHANGE): eth0: link becomes ready
10.5.0.3: kern:    info: [2022-07-28T11:49:42.503944063Z]: IPv6: ADDRCONF(NETDEV_CHANGE): veth32488a86: link becomes ready
10.5.0.3: kern:    info: [2022-07-28T11:49:42.504764063Z]: cni0: port 1(veth32488a86) entered blocking state
10.5.0.3: kern:    info: [2022-07-28T11:49:42.505423063Z]: cni0: port 1(veth32488a86) entered forwarding state
10.5.0.3: kern: warning: [2022-07-28T11:49:44.873616063Z]: kauditd_printk_skb: 14 callbacks suppressed
10.5.0.3: kern:  notice: [2022-07-28T11:49:44.873619063Z]: audit: type=1326 audit(1659008985.445:25): auid=4294967295 uid=0 gid=0 ses=4294967295 pid=2784 comm="runc:[2:INIT]" exe="/" sig=0 arch=c000003e syscall=3 compat=0 ip=0x55ec0657bd3b code=0x7ffc0000
10.5.0.3: kern:  notice: [2022-07-28T11:49:44.876609063Z]: audit: type=1326 audit(1659008985.445:26): auid=4294967295 uid=0 gid=0 ses=4294967295 pid=2784 comm="runc:[2:INIT]" exe="/" sig=0 arch=c000003e syscall=3 compat=0 ip=0x55ec0657bd3b code=0x7ffc0000
10.5.0.3: kern:  notice: [2022-07-28T11:49:44.878789063Z]: audit: type=1326 audit(1659008985.449:27): auid=4294967295 uid=0 gid=0 ses=4294967295 pid=2784 comm="runc:[2:INIT]" exe="/" sig=0 arch=c000003e syscall=257 compat=0 ip=0x55ec0657bdaa code=0x7ffc0000
10.5.0.3: kern:  notice: [2022-07-28T11:49:44.886693063Z]: audit: type=1326 audit(1659008985.461:28): auid=4294967295 uid=0 gid=0 ses=4294967295 pid=2784 comm="runc:[2:INIT]" exe="/" sig=0 arch=c000003e syscall=202 compat=0 ip=0x55ec06532b43 code=0x7ffc0000
10.5.0.3: kern:  notice: [2022-07-28T11:49:44.888764063Z]: audit: type=1326 audit(1659008985.461:29): auid=4294967295 uid=0 gid=0 ses=4294967295 pid=2784 comm="runc:[2:INIT]" exe="/" sig=0 arch=c000003e syscall=202 compat=0 ip=0x55ec06532b43 code=0x7ffc0000
10.5.0.3: kern:  notice: [2022-07-28T11:49:44.891009063Z]: audit: type=1326 audit(1659008985.461:30): auid=4294967295 uid=0 gid=0 ses=4294967295 pid=2784 comm="runc:[2:INIT]" exe="/" sig=0 arch=c000003e syscall=1 compat=0 ip=0x55ec0657bd3b code=0x7ffc0000
10.5.0.3: kern:  notice: [2022-07-28T11:49:44.893162063Z]: audit: type=1326 audit(1659008985.461:31): auid=4294967295 uid=0 gid=0 ses=4294967295 pid=2784 comm="runc:[2:INIT]" exe="/" sig=0 arch=c000003e syscall=3 compat=0 ip=0x55ec0657bd3b code=0x7ffc0000
10.5.0.3: kern:  notice: [2022-07-28T11:49:44.895365063Z]: audit: type=1326 audit(1659008985.461:32): auid=4294967295 uid=0 gid=0 ses=4294967295 pid=2784 comm="runc:[2:INIT]" exe="/" sig=0 arch=c000003e syscall=39 compat=0 ip=0x55ec066eb68b code=0x7ffc0000
10.5.0.3: kern:  notice: [2022-07-28T11:49:44.898306063Z]: audit: type=1326 audit(1659008985.461:33): auid=4294967295 uid=0 gid=0 ses=4294967295 pid=2784 comm="runc:[2:INIT]" exe="/" sig=0 arch=c000003e syscall=59 compat=0 ip=0x55ec0657be16 code=0x7ffc0000
10.5.0.3: kern:  notice: [2022-07-28T11:49:44.901518063Z]: audit: type=1326 audit(1659008985.473:34): auid=4294967295 uid=0 gid=0 ses=4294967295 pid=2784 comm="http-echo" exe="/http-echo" sig=0 arch=c000003e syscall=158 compat=0 ip=0x455f35 code=0x7ffc0000

Cleanup

You can clean up the test resources by running the following command:

kubectl delete pod audit-pod

1.9 - Storage

Setting up storage for a Kubernetes cluster

In Kubernetes, using storage in the right way is well-facilitated by the API. However, unless you are running in a major public cloud, that API may not be hooked up to anything. This frequently sends users down a rabbit hole of researching all the various options for storage backends for their platform, for Kubernetes, and for their workloads. There are a lot of options out there, and it can be fairly bewildering.

For Talos, we try to limit the options somewhat to make the decision-making easier.

Public Cloud

If you are running on a major public cloud, use their block storage. It is easy and automatic.

Storage Clusters

Sidero Labs recommends having separate disks (apart from the Talos install disk) to be used for storage.

Redundancy, scaling capabilities, reliability, speed, maintenance load, and ease of use are all factors you must consider when managing your own storage.

Running a storage cluster can be a very good choice when managing your own storage, and there are two projects we recommend, depending on your situation.

If you need vast amounts of storage composed of more than a dozen or so disks, we recommend you use Rook to manage Ceph. Also, if you need both mount-once and mount-many capabilities, Ceph is your answer. Ceph also bundles in an S3-compatible object store. The down side of Ceph is that there are a lot of moving parts.

Please note that most people should never use mount-many semantics. NFS is pervasive because it is old and easy, not because it is a good idea. While it may seem like a convenience at first, there are all manner of locking, performance, change control, and reliability concerns inherent in any mount-many situation, so we strongly recommend you avoid this method.

If your storage needs are small enough to not need Ceph, use Mayastor.

Rook/Ceph

Ceph is the grandfather of open source storage clusters. It is big, has a lot of pieces, and will do just about anything. It scales better than almost any other system out there, open source or proprietary, being able to easily add and remove storage over time with no downtime, safely and easily. It comes bundled with RadosGW, an S3-compatible object store; CephFS, a NFS-like clustered filesystem; and RBD, a block storage system.

With the help of Rook, the vast majority of the complexity of Ceph is hidden away by a very robust operator, allowing you to control almost everything about your Ceph cluster from fairly simple Kubernetes CRDs.

So if Ceph is so great, why not use it for everything?

Ceph can be rather slow for small clusters. It relies heavily on CPUs and massive parallelisation to provide good cluster performance, so if you don’t have much of those dedicated to Ceph, it is not going to be well-optimised for you. Also, if your cluster is small, just running Ceph may eat up a significant amount of the resources you have available.

Troubleshooting Ceph can be difficult if you do not understand its architecture. There are lots of acronyms and the documentation assumes a fair level of knowledge. There are very good tools for inspection and debugging, but this is still frequently seen as a concern.

Mayastor

Mayastor is an OpenEBS project built in Rust utilising the modern NVMEoF system. (Despite the name, Mayastor does not require you to have NVME drives.) It is fast and lean but still cluster-oriented and cloud native. Unlike most of the other OpenEBS project, it is not built on the ancient iSCSI system.

Unlike Ceph, Mayastor is just a block store. It focuses on block storage and does it well. It is much less complicated to set up than Ceph, but you probably wouldn’t want to use it for more than a few dozen disks.

Mayastor is new, maybe too new. If you’re looking for something well-tested and battle-hardened, this is not it. However, if you’re looking for something lean, future-oriented, and simpler than Ceph, it might be a great choice.

Video Walkthrough

To see a live demo of this section, see the video below:

Prep Nodes

Either during initial cluster creation or on running worker nodes, several machine config values should be edited. (This information is gathered from the Mayastor documentation.) We need to set the vm.nr_hugepages sysctl and add openebs.io/engine=mayastor labels to the nodes which are meant to be storage nodes. This can be done with talosctl patch machineconfig or via config patches during talosctl gen config.

Some examples are shown below: modify as needed.

First create a config patch file named mayastor-patch.yaml with the following contents:

- op: add
  path: /machine/sysctls
  value:
    vm.nr_hugepages: "1024"
- op: add
  path: /machine/nodeLabels
  value:
    openebs.io/engine: mayastor

Using gen config

talosctl gen config my-cluster https://mycluster.local:6443 --config-patch @mayastor-patch.yaml

Patching an existing node

talosctl patch --mode=no-reboot machineconfig -n <node ip> --patch @mayastor-patch.yaml

Note: If you are adding/updating the vm.nr_hugepages on a node which already had the openebs.io/engine=mayastor label set, you’d need to restart kubelet so that it picks up the new value, by issuing the following command

talosctl -n <node ip> service kubelet restart

Deploy Mayastor

Continue setting up Mayastor using the official documentation.

Piraeus / LINSTOR

Install Piraeus Operator V2

There is already a how-to for Talos: Link

Create first storage pool and PVC

Before proceeding, install linstor plugin for kubectl: https://github.com/piraeusdatastore/kubectl-linstor

Or use krew: kubectl krew install linstor

# Create device pool on a blank (no partitation table!) disk on node01
kubectl linstor physical-storage create-device-pool --pool-name nvme_lvm_pool LVM node01 /dev/nvme0n1 --storage-pool nvme_pool

piraeus-sc.yml

apiVersion: storage.k8s.io/v1
kind: StorageClass
metadata:
  name: simple-nvme
parameters:
  csi.storage.k8s.io/fstype: xfs
  linstor.csi.linbit.com/autoPlace: "3"
  linstor.csi.linbit.com/storagePool: nvme_pool
provisioner: linstor.csi.linbit.com
volumeBindingMode: WaitForFirstConsumer

# Create storage class
kubectl apply -f piraeus-sc.yml

NFS

NFS is an old pack animal long past its prime. NFS is slow, has all kinds of bottlenecks involving contention, distributed locking, single points of service, and more. However, it is supported by a wide variety of systems. You don’t want to use it unless you have to, but unfortunately, that “have to” is too frequent.

The NFS client is part of the kubelet image maintained by the Talos team. This means that the version installed in your running kubelet is the version of NFS supported by Talos. You can reduce some of the contention problems by parceling Persistent Volumes from separate underlying directories.

Object storage

Ceph comes with an S3-compatible object store, but there are other options, as well. These can often be built on top of other storage backends. For instance, you may have your block storage running with Mayastor but assign a Pod a large Persistent Volume to serve your object store.

One of the most popular open source add-on object stores is MinIO.

Others (iSCSI)

The most common remaining systems involve iSCSI in one form or another. These include the original OpenEBS, Rancher’s Longhorn, and many proprietary systems. iSCSI in Linux is facilitated by open-iscsi. This system was designed long before containers caught on, and it is not well suited to the task, especially when coupled with a read-only host operating system.

iSCSI support in Talos is now supported via the iscsi-tools system extension installed. The extension enables compatibility with OpenEBS Jiva - refer to the local storage installation guide for more information.

2 - Network

Managing the Kubernetes cluster networking

2.1 - Deploying Cilium CNI

In this guide you will learn how to set up Cilium CNI on Talos.

Cilium can be installed either via the cilium cli or using helm.

This documentation will outline installing Cilium CNI v1.14.0 on Talos in six different ways. Adhering to Talos principles we’ll deploy Cilium with IPAM mode set to Kubernetes, and using the cgroupv2 and bpffs mount that talos already provides. As Talos does not allow loading kernel modules by Kubernetes workloads, SYS_MODULE capability needs to be dropped from the Cilium default set of values, this override can be seen in the helm/cilium cli install commands. Each method can either install Cilium using kube proxy (default) or without: Kubernetes Without kube-proxy

From Talos 1.5 we can use Cilium in combination with KubePrism. You can read more about it in Talos 1.5 release notes.

Machine config preparation

When generating the machine config for a node set the CNI to none. For example using a config patch:

Create a patch.yaml file with the following contents:

cluster:
  network:
    cni:
      name: none
machine:
  features:
    kubePrism:
      enabled: true
      port: 7445

talosctl gen config \
    my-cluster https://mycluster.local:6443 \
    --config-patch @patch.yaml

Or if you want to deploy Cilium without kube-proxy, you also need to disable kube proxy:

Create a patch.yaml file with the following contents:

cluster:
  network:
    cni:
      name: none
  proxy:
    disabled: true
machine:
  features:
    kubePrism:
      enabled: true
      port: 7445

talosctl gen config \
    my-cluster https://mycluster.local:6443 \
    --config-patch @patch.yaml

Installation using Cilium CLI

Note: It is recommended to template the cilium manifest using helm and use it as part of Talos machine config, but if you want to install Cilium using the Cilium CLI, you can follow the steps below.

Install the Cilium CLI following the steps here.

With kube-proxy

cilium install \
    --helm-set=ipam.mode=kubernetes \
    --helm-set=kubeProxyReplacement=disabled \
    --helm-set=securityContext.capabilities.ciliumAgent="{CHOWN,KILL,NET_ADMIN,NET_RAW,IPC_LOCK,SYS_ADMIN,SYS_RESOURCE,DAC_OVERRIDE,FOWNER,SETGID,SETUID}" \
    --helm-set=securityContext.capabilities.cleanCiliumState="{NET_ADMIN,SYS_ADMIN,SYS_RESOURCE}" \
    --helm-set=cgroup.autoMount.enabled=false \
    --helm-set=cgroup.hostRoot=/sys/fs/cgroup

Without kube-proxy

cilium install \
    --helm-set=ipam.mode=kubernetes \
    --helm-set=kubeProxyReplacement=true \
    --helm-set=securityContext.capabilities.ciliumAgent="{CHOWN,KILL,NET_ADMIN,NET_RAW,IPC_LOCK,SYS_ADMIN,SYS_RESOURCE,DAC_OVERRIDE,FOWNER,SETGID,SETUID}" \
    --helm-set=securityContext.capabilities.cleanCiliumState="{NET_ADMIN,SYS_ADMIN,SYS_RESOURCE}" \
    --helm-set=cgroup.autoMount.enabled=false \
    --helm-set=cgroup.hostRoot=/sys/fs/cgroup \
    --helm-set=k8sServiceHost=localhost \
    --helm-set=k8sServicePort=7445

Installation using Helm

Refer to Installing with Helm for more information.

First we’ll need to add the helm repo for Cilium.

helm repo add cilium https://helm.cilium.io/
helm repo update

Method 1: Helm install

After applying the machine config and bootstrapping Talos will appear to hang on phase 18/19 with the message: retrying error: node not ready. This happens because nodes in Kubernetes are only marked as ready once the CNI is up. As there is no CNI defined, the boot process is pending and will reboot the node to retry after 10 minutes, this is expected behavior.

During this window you can install Cilium manually by running the following:

helm install \
    cilium \
    cilium/cilium \
    --version 1.14.0 \
    --namespace kube-system \
    --set ipam.mode=kubernetes \
    --set=kubeProxyReplacement=disabled \
    --set=securityContext.capabilities.ciliumAgent="{CHOWN,KILL,NET_ADMIN,NET_RAW,IPC_LOCK,SYS_ADMIN,SYS_RESOURCE,DAC_OVERRIDE,FOWNER,SETGID,SETUID}" \
    --set=securityContext.capabilities.cleanCiliumState="{NET_ADMIN,SYS_ADMIN,SYS_RESOURCE}" \
    --set=cgroup.autoMount.enabled=false \
    --set=cgroup.hostRoot=/sys/fs/cgroup

Or if you want to deploy Cilium without kube-proxy, also set some extra paramaters:

helm install \
    cilium \
    cilium/cilium \
    --version 1.14.0 \
    --namespace kube-system \
    --set ipam.mode=kubernetes \
    --set=kubeProxyReplacement=true \
    --set=securityContext.capabilities.ciliumAgent="{CHOWN,KILL,NET_ADMIN,NET_RAW,IPC_LOCK,SYS_ADMIN,SYS_RESOURCE,DAC_OVERRIDE,FOWNER,SETGID,SETUID}" \
    --set=securityContext.capabilities.cleanCiliumState="{NET_ADMIN,SYS_ADMIN,SYS_RESOURCE}" \
    --set=cgroup.autoMount.enabled=false \
    --set=cgroup.hostRoot=/sys/fs/cgroup \
    --set=k8sServiceHost=localhost \
    --set=k8sServicePort=7445

After Cilium is installed the boot process should continue and complete successfully.

Method 2: Helm manifests install

Instead of directly installing Cilium you can instead first generate the manifest and then apply it:

helm template \
    cilium \
    cilium/cilium \
    --version 1.14.0 \
    --namespace kube-system \
    --set ipam.mode=kubernetes \
    --set=kubeProxyReplacement=disabled \
    --set=securityContext.capabilities.ciliumAgent="{CHOWN,KILL,NET_ADMIN,NET_RAW,IPC_LOCK,SYS_ADMIN,SYS_RESOURCE,DAC_OVERRIDE,FOWNER,SETGID,SETUID}" \
    --set=securityContext.capabilities.cleanCiliumState="{NET_ADMIN,SYS_ADMIN,SYS_RESOURCE}" \
    --set=cgroup.autoMount.enabled=false \
    --set=cgroup.hostRoot=/sys/fs/cgroup > cilium.yaml

kubectl apply -f cilium.yaml

Without kube-proxy:

export KUBERNETES_API_SERVER_ADDRESS=<replace with api server endpoint here> # e.g. 10.96.0.1
export KUBERNETES_API_SERVER_PORT=6443

helm template \
    cilium \
    cilium/cilium \
    --version 1.14.0 \
    --namespace kube-system \
    --set ipam.mode=kubernetes \
    --set=kubeProxyReplacement=true \
    --set=securityContext.capabilities.ciliumAgent="{CHOWN,KILL,NET_ADMIN,NET_RAW,IPC_LOCK,SYS_ADMIN,SYS_RESOURCE,DAC_OVERRIDE,FOWNER,SETGID,SETUID}" \
    --set=securityContext.capabilities.cleanCiliumState="{NET_ADMIN,SYS_ADMIN,SYS_RESOURCE}" \
    --set=cgroup.autoMount.enabled=false \
    --set=cgroup.hostRoot=/sys/fs/cgroup \
    --set=k8sServiceHost=localhost \
    --set=k8sServicePort=7445 > cilium.yaml

kubectl apply -f cilium.yaml

Method 3: Helm manifests hosted install

After generating cilium.yaml using helm template, instead of applying this manifest directly during the Talos boot window (before the reboot timeout). You can also host this file somewhere and patch the machine config to apply this manifest automatically during bootstrap. To do this patch your machine configuration to include this config instead of the above:

Create a patch.yaml file with the following contents:

cluster:
  network:
    cni:
      name: custom
      urls:
        - https://server.yourdomain.tld/some/path/cilium.yaml
machine:
  features:
    kubePrism:
      enabled: true
      port: 7445

talosctl gen config \
    my-cluster https://mycluster.local:6443 \
    --config-patch @patch.yaml

However, beware of the fact that the helm generated Cilium manifest contains sensitive key material. As such you should definitely not host this somewhere publicly accessible.

Method 4: Helm manifests inline install

A more secure option would be to include the helm template output manifest inside the machine configuration. The machine config should be generated with CNI set to none

Create a patch.yaml file with the following contents:

cluster:
  network:
    cni:
      name: none
machine:
  features:
    kubePrism:
      enabled: true
      port: 7445

talosctl gen config \
    my-cluster https://mycluster.local:6443 \
    --config-patch @patch.yaml

if deploying Cilium with kube-proxy disabled, you can also include the following:

Create a patch.yaml file with the following contents:

cluster:
  network:
    cni:
      name: none
  proxy:
    disabled: true
machine:
  features:
    kubePrism:
      enabled: true
      port: 7445

talosctl gen config \
    my-cluster https://mycluster.local:6443 \
    --config-patch @patch.yaml

To do so patch this into your machine configuration:

cluster:
  inlineManifests:
    - name: cilium
      contents: |
        --
        # Source: cilium/templates/cilium-agent/serviceaccount.yaml
        apiVersion: v1
        kind: ServiceAccount
        metadata:
          name: "cilium"
          namespace: kube-system
        ---
        # Source: cilium/templates/cilium-operator/serviceaccount.yaml
        apiVersion: v1
        kind: ServiceAccount
        -> Your cilium.yaml file will be pretty long....

This will install the Cilium manifests at just the right time during bootstrap.

Beware though:

  • Changing the namespace when templating with Helm does not generate a manifest containing the yaml to create that namespace. As the inline manifest is processed from top to bottom make sure to manually put the namespace yaml at the start of the inline manifest.
  • Only add the Cilium inline manifest to the control plane nodes machine configuration.
  • Make sure all control plane nodes have an identical configuration.
  • If you delete any of the generated resources they will be restored whenever a control plane node reboots.
  • As a safety measure, Talos only creates missing resources from inline manifests, it never deletes or updates anything.
  • If you need to update a manifest make sure to first edit all control plane machine configurations and then run talosctl upgrade-k8s as it will take care of updating inline manifests.

Known issues

Other things to know

3 - Upgrading Kubernetes

Guide on how to upgrade the Kubernetes cluster from Talos Linux.

This guide covers upgrading Kubernetes on Talos Linux clusters.

For a list of Kubernetes versions compatible with each Talos release, see the Support Matrix.

For upgrading the Talos Linux operating system, see Upgrading Talos

Video Walkthrough

To see a demo of this process, watch this video:

Automated Kubernetes Upgrade

The recommended method to upgrade Kubernetes is to use the talosctl upgrade-k8s command. This will automatically update the components needed to upgrade Kubernetes safely. Upgrading Kubernetes is non-disruptive to the cluster workloads.

To trigger a Kubernetes upgrade, issue a command specifiying the version of Kubernetes to ugprade to, such as:

talosctl --nodes <controlplane node> upgrade-k8s --to 1.28.3

Note that the --nodes parameter specifies the control plane node to send the API call to, but all members of the cluster will be upgraded.

To check what will be upgraded you can run talosctl upgrade-k8s with the --dry-run flag:

$ talosctl --nodes <controlplane node> upgrade-k8s --to 1.28.3 --dry-run
WARNING: found resources which are going to be deprecated/migrated in the version 1.28.3
RESOURCE                                                               COUNT
validatingwebhookconfigurations.v1beta1.admissionregistration.k8s.io   4
mutatingwebhookconfigurations.v1beta1.admissionregistration.k8s.io     3
customresourcedefinitions.v1beta1.apiextensions.k8s.io                 25
apiservices.v1beta1.apiregistration.k8s.io                             54
leases.v1beta1.coordination.k8s.io                                     4
automatically detected the lowest Kubernetes version 1.27.4
checking for resource APIs to be deprecated in version 1.28.3
discovered controlplane nodes ["172.20.0.2" "172.20.0.3" "172.20.0.4"]
discovered worker nodes ["172.20.0.5" "172.20.0.6"]
updating "kube-apiserver" to version "1.28.3"
 > "172.20.0.2": starting update
 > update kube-apiserver: v1.27.4 -> 1.28.3
 > skipped in dry-run
 > "172.20.0.3": starting update
 > update kube-apiserver: v1.27.4 -> 1.28.3
 > skipped in dry-run
 > "172.20.0.4": starting update
 > update kube-apiserver: v1.27.4 -> 1.28.3
 > skipped in dry-run
updating "kube-controller-manager" to version "1.28.3"
 > "172.20.0.2": starting update
 > update kube-controller-manager: v1.27.4 -> 1.28.3
 > skipped in dry-run
 > "172.20.0.3": starting update

<snip>

updating manifests
 > apply manifest Secret bootstrap-token-3lb63t
 > apply skipped in dry run
 > apply manifest ClusterRoleBinding system-bootstrap-approve-node-client-csr
 > apply skipped in dry run
<snip>

To upgrade Kubernetes from v1.27.4 to v1.28.3 run:

$ talosctl --nodes <controlplane node> upgrade-k8s --to 1.28.3
automatically detected the lowest Kubernetes version 1.27.4
checking for resource APIs to be deprecated in version 1.28.3
discovered controlplane nodes ["172.20.0.2" "172.20.0.3" "172.20.0.4"]
discovered worker nodes ["172.20.0.5" "172.20.0.6"]
updating "kube-apiserver" to version "1.28.3"
 > "172.20.0.2": starting update
 > update kube-apiserver: v1.27.4 -> 1.28.3
 > "172.20.0.2": machine configuration patched
 > "172.20.0.2": waiting for API server state pod update
 < "172.20.0.2": successfully updated
 > "172.20.0.3": starting update
 > update kube-apiserver: v1.27.4 -> 1.28.3
<snip>

This command runs in several phases:

  1. Every control plane node machine configuration is patched with the new image version for each control plane component. Talos renders new static pod definitions on the configuration update which is picked up by the kubelet. The command waits for the change to propagate to the API server state.
  2. The command updates the kube-proxy daemonset with the new image version.
  3. On every node in the cluster, the kubelet version is updated. The command then waits for the kubelet service to be restarted and become healthy. The update is verified by checking the Node resource state.
  4. Kubernetes bootstrap manifests are re-applied to the cluster. Updated bootstrap manifests might come with a new Talos version (e.g. CoreDNS version update), or might be the result of machine configuration change. Note: The upgrade-k8s command never deletes any resources from the cluster: they should be deleted manually.

If the command fails for any reason, it can be safely restarted to continue the upgrade process from the moment of the failure.

Manual Kubernetes Upgrade

Kubernetes can be upgraded manually by following the steps outlined below. They are equivalent to the steps performed by the talosctl upgrade-k8s command.

Kubeconfig

In order to edit the control plane, you need a working kubectl config. If you don’t already have one, you can get one by running:

talosctl --nodes <controlplane node> kubeconfig

API Server

Patch machine configuration using talosctl patch command:

$ talosctl -n <CONTROL_PLANE_IP_1> patch mc --mode=no-reboot -p '[{"op": "replace", "path": "/cluster/apiServer/image", "value": "registry.k8s.io/kube-apiserver:v1.28.3"}]'
patched mc at the node 172.20.0.2

The JSON patch might need to be adjusted if current machine configuration is missing .cluster.apiServer.image key.

Also the machine configuration can be edited manually with talosctl -n <IP> edit mc --mode=no-reboot.

Capture the new version of kube-apiserver config with:

$ talosctl -n <CONTROL_PLANE_IP_1> get kcpc kube-apiserver -o yaml
node: 172.20.0.2
metadata:
    namespace: config
    type: KubernetesControlPlaneConfigs.config.talos.dev
    id: kube-apiserver
    version: 5
    phase: running
spec:
    image: registry.k8s.io/kube-apiserver:v1.28.3
    cloudProvider: ""
    controlPlaneEndpoint: https://172.20.0.1:6443
    etcdServers:
        - https://127.0.0.1:2379
    localPort: 6443
    serviceCIDR: 10.96.0.0/12
    extraArgs: {}
    extraVolumes: []

In this example, the new version is 5. Wait for the new pod definition to propagate to the API server state (replace talos-default-controlplane-1 with the node name):

$ kubectl get pod -n kube-system -l k8s-app=kube-apiserver --field-selector spec.nodeName=talos-default-controlplane-1 -o jsonpath='{.items[0].metadata.annotations.talos\.dev/config\-version}'
5

Check that the pod is running:

$ kubectl get pod -n kube-system -l k8s-app=kube-apiserver --field-selector spec.nodeName=talos-default-controlplane-1
NAME                                    READY   STATUS    RESTARTS   AGE
kube-apiserver-talos-default-controlplane-1   1/1     Running   0          16m

Repeat this process for every control plane node, verifying that state got propagated successfully between each node update.

Controller Manager

Patch machine configuration using talosctl patch command:

$ talosctl -n <CONTROL_PLANE_IP_1> patch mc --mode=no-reboot -p '[{"op": "replace", "path": "/cluster/controllerManager/image", "value": "registry.k8s.io/kube-controller-manager:v1.28.3"}]'
patched mc at the node 172.20.0.2

The JSON patch might need be adjusted if current machine configuration is missing .cluster.controllerManager.image key.

Capture new version of kube-controller-manager config with:

$ talosctl -n <CONTROL_PLANE_IP_1> get kcpc kube-controller-manager -o yaml
node: 172.20.0.2
metadata:
    namespace: config
    type: KubernetesControlPlaneConfigs.config.talos.dev
    id: kube-controller-manager
    version: 3
    phase: running
spec:
    image: registry.k8s.io/kube-controller-manager:v1.28.3
    cloudProvider: ""
    podCIDR: 10.244.0.0/16
    serviceCIDR: 10.96.0.0/12
    extraArgs: {}
    extraVolumes: []

In this example, new version is 3. Wait for the new pod definition to propagate to the API server state (replace talos-default-controlplane-1 with the node name):

$ kubectl get pod -n kube-system -l k8s-app=kube-controller-manager --field-selector spec.nodeName=talos-default-controlplane-1 -o jsonpath='{.items[0].metadata.annotations.talos\.dev/config\-version}'
3

Check that the pod is running:

$ kubectl get pod -n kube-system -l k8s-app=kube-controller-manager --field-selector spec.nodeName=talos-default-controlplane-1
NAME                                             READY   STATUS    RESTARTS   AGE
kube-controller-manager-talos-default-controlplane-1   1/1     Running   0          35m

Repeat this process for every control plane node, verifying that state propagated successfully between each node update.

Scheduler

Patch machine configuration using talosctl patch command:

$ talosctl -n <CONTROL_PLANE_IP_1> patch mc --mode=no-reboot -p '[{"op": "replace", "path": "/cluster/scheduler/image", "value": "registry.k8s.io/kube-scheduler:v1.28.3"}]'
patched mc at the node 172.20.0.2

JSON patch might need be adjusted if current machine configuration is missing .cluster.scheduler.image key.

Capture new version of kube-scheduler config with:

$ talosctl -n <CONTROL_PLANE_IP_1> get kcpc kube-scheduler -o yaml
node: 172.20.0.2
metadata:
    namespace: config
    type: KubernetesControlPlaneConfigs.config.talos.dev
    id: kube-scheduler
    version: 3
    phase: running
spec:
    image: registry.k8s.io/kube-scheduler:v1.28.3
    extraArgs: {}
    extraVolumes: []

In this example, new version is 3. Wait for the new pod definition to propagate to the API server state (replace talos-default-controlplane-1 with the node name):

$ kubectl get pod -n kube-system -l k8s-app=kube-scheduler --field-selector spec.nodeName=talos-default-controlplane-1 -o jsonpath='{.items[0].metadata.annotations.talos\.dev/config\-version}'
3

Check that the pod is running:

$ kubectl get pod -n kube-system -l k8s-app=kube-scheduler --field-selector spec.nodeName=talos-default-controlplane-1
NAME                                    READY   STATUS    RESTARTS   AGE
kube-scheduler-talos-default-controlplane-1   1/1     Running   0          39m

Repeat this process for every control plane node, verifying that state got propagated successfully between each node update.

Proxy

In the proxy’s DaemonSet, change:

kind: DaemonSet
...
spec:
  ...
  template:
    ...
    spec:
      containers:
        - name: kube-proxy
          image: registry.k8s.io/kube-proxy:v1.28.3
      tolerations:
        - ...

to:

kind: DaemonSet
...
spec:
  ...
  template:
    ...
    spec:
      containers:
        - name: kube-proxy
          image: registry.k8s.io/kube-proxy:v1.28.3
      tolerations:
        - ...
        - key: node-role.kubernetes.io/control-plane
          operator: Exists
          effect: NoSchedule

To edit the DaemonSet, run:

kubectl edit daemonsets -n kube-system kube-proxy

Bootstrap Manifests

Bootstrap manifests can be retrieved in a format which works for kubectl with the following command:

talosctl -n <controlplane IP> get manifests -o yaml | yq eval-all '.spec | .[] | splitDoc' - > manifests.yaml

Diff the manifests with the cluster:

kubectl diff -f manifests.yaml

Apply the manifests:

kubectl apply -f manifests.yaml

Note: if some bootstrap resources were removed, they have to be removed from the cluster manually.

kubelet

For every node, patch machine configuration with new kubelet version, wait for the kubelet to restart with new version:

$ talosctl -n <IP> patch mc --mode=no-reboot -p '[{"op": "replace", "path": "/machine/kubelet/image", "value": "ghcr.io/siderolabs/kubelet:v1.28.3"}]'
patched mc at the node 172.20.0.2

Once kubelet restarts with the new configuration, confirm upgrade with kubectl get nodes <name>:

$ kubectl get nodes talos-default-controlplane-1
NAME                           STATUS   ROLES                  AGE    VERSION
talos-default-controlplane-1   Ready    control-plane          123m   v1.28.3