Skip to content

Latest commit

 

History

History
1193 lines (799 loc) · 36.4 KB

README.md

File metadata and controls

1193 lines (799 loc) · 36.4 KB
Raw

Kubernetes training

Learn Kubernetes: from deploying your first application to deploying hundreds of microservices in no time.

GitLab SikaLabs

About course

About lector

Hi, my name is Vojtěch Mareš. I'm freelance DevOps engineer, consultant, and lector. For more, see my website or explore more courses.

Having questions? Contact me at email [email protected].

Before we start

Install tooling

  • kubectl
  • helm
  • kubectx & kubens
  • k9s
  • Docker: Official installation guide

Aliases

k

For simplicity and our sanity, let's create an alias for kubectl.

Shell (bash, zsh,...)

# .bashrc / .zshrc / etc.
alias k="kubectl"

Windows (cmd.exe)

doskey k=kubectl $*

Windows (PowerShell)

Set-Alias -Name k -Value kubectl

Course

Cluster components

Kubernetes components

Control plane

Formerly master.

A node running components necessary to run the Kubernetes cluster.

  • kube-apiserver
  • etcd
  • kube-scheduler
  • kube-controller-manager
  • cloud-controller-manager (optional)

Node

Formerly worker.

Machine running our workload (applications).

  • kubelet
  • kube-proxy
  • container runtime (by default containerd)

Explain Kubernetes resources

Manpage for Kubernetes resources.

When you need to know something about resource and it's field.

kubectl explain node
kubectl explain node.spec

kubectl explain pod
kubectl explain pod.spec
kubectl explain pod.spec.containers.image

Nodes

kubectl get nodes

# or short form
kubectl get no

kubectl

A command line tool to interact with the cluster.

kubectl get

List resources of type.

kubectl get namespace

kubectl describe

Describes resource including status, recent events and other information about it.

kubectl describe namespace default

kubectl create

Creates new resource either in terminal or from file.

kubectl create namespace example-ns

# or from file
kubectl create -f ./example-ns.yaml

kubectl delete

kubectl delete namespace example-ns

# or target resource from file
kubectl delete -f ./example-ns.yaml

kubectl apply

Creates a resource if it does not exist or applies the configuration from file to an existing resource.

kubectl apply -f ./example-ns.yaml

# supports URL
kubectl apply -f https://raw.githubusercontent.com/vojtechmares/kubernetes-training/.../pod.yaml

Pod

Smallest deployable unit in Kubernetes. Can be made from multiple containers, usually one is enough.

List pods

kubectl get pods

Describe pod

kubectl describe pod $POD_NAME

See pod logs

kubectl logs -f $POD_NAME

Connect to pod

kubectl port-forward pod/$POD_NAME $LOCAL_PORT:$POD_PORT

Open bash in pod

kubectl exec runs a binary (or shell script if shell is available) within a Pod. It is also useful to launch a shell session inside the Pod, when you need to debug some issue.

kubectl exec -it $POD_NAME -- bash

Copy files from / to pod

# From local to pod
kubectl cp ./path/to/file $POD_NAME:/remote/path

# From pod to local
kubectl cp $POD_NAME:/remote/path ./local/path

Service

Service is a cluster abstraction a single in-cluster endpoint (DNS name and IP address) that distributes traffic to it's pods.

Create service

kubectl create -f ./examples/02-service/service.yaml

List services

kubectl get service

Describe service

kubectl describe service example-svc

Connect to service

kubectl port-forward service/example-svc 8080:8080

Delete service

kubectl delete service example-svc

Exposing Service

You can expose Service outside of the cluster in two ways, with .spec.type:

  • type=NodePort
  • type=LoadBalancer

NodePort opens a port on every node and Kubernetes routes all incoming traffic (on every node) on given port to this Service. The default NodePort range is from 30 000 to 32 767.

LoadBalancer requires cloud integration to provision a managed load balancer (AWS, Azure, GCP, DigitalOcean, Hetzner Cloud, and others) or software like kube-vip to managed Virtual IP attachment/announcement, usually over ARP (L2) or BGP (L3).

Another option to expose network interface outside the cluster of your workload is either to use an Ingress for HTTP, Gateway and given routes for other protocols like gRPC. Or in some cases you can expose Pods directly with hostPort.

Deployment

Deploying our application to Pod might be easy, but not a good idea. To deploy our app to Kubernetes and run it, we use Deployment. It is a layer of abstraction on top of Pods (and ReplicaSet).

ReplicaSet

ReplicasSet is a child resource to Deployment, which is used to keep track of revisions of pools of pods and allows to rollback to it if new revision of Deployment is failing.

Today, ReplicaSet is usually not interacted with by users.

Updates

Kubernetes native:

  • Recreate (deletes all pods and creates new ones)
  • RollingUpdate (zero downtime)

The Recreate strategy works pretty much as you expect: deletes all running Pods and then creates new ones.

On the other hand, Rolling Update has a few configuration options, see example:

apiVersion: apps/v1
kind: Deployment
metadata:
  name: nginx
spec:
  replicas: 3
  selector:
    matchLabels:
      app: nginx
  template:
    metadata:
      labels:
        app: nginx
    spec:
      containers:
      - name: nginx
        image: nginx:latest
        ports:
        - containerPort: 80
      strategy:
        type: RollingUpdate
        rollingUpdate:
          maxSurge: 1
          maxUnavailable: 0

Extended:

  • Blue/Green
  • Canary
  • A/B testing

See: Argo Rollouts or Flagger

Ingress

Ingress resource exposes our application network interface (HTTP, TCP,...) to public internet.

Ingress Controller

Kubernetes does not bring in an Ingress Controller by default and it is up to cluster administrator to choose and deploy one (or multiple).

Kubernetes project offers Ingress NGINX.

Ingress resource

Ingress is a Kubernetes resource that exposes Service outside the cluster. The resource is managed by Ingress Controller

Ingress Class

IngressClass is Kubernetes abstraction to map Ingress resources to given Ingress Controller, like Ingress NGINX.

IngressClass is often managed for you by the installer (like Helm) of Ingress Controller or the controller itself.

Gateway API

Gateway API - a new standard for ingress traffic handling. Kubernetes extension made by SIG-Network. Only specification, implementation is up to users.

Generally Available implementations:

StatefulSet

A special abstraction for running Pods running stateful applications like databases (for example MySQL or Redis) or message brokers like RabbitMQ and Apache Kafka.

StatefulSet also needs a "headless service", which is defined in it's spec.

Headless service

A service with type=ClusterIP and clusterIP=None configuration.

apiVersion: v1
kind: Service
metadata:
  name: my-headless-service
  labels:
    app: my-app
spec:
  # type: ClusterIP # default
  clusterIP: None
  selector:
    app: my-app
  ports:
    - protocol: TCP
      port: 80
      targetPort: 8080

That creates a service without a Virtual IP for load balancing and instead DNS will return all IPs of Pods.

That is important for many reasons. This allows you to distinguish between running Pods (in another words preserving network identity of a process). Or to do client-side load balancing.

Why would we want to know to which Pod are we talking to? For example with databases, we want to connect to primary instance for writes, but reads are fine from replicas.

And as for the case of client side load balancing. This eliminates the need of a dedicated load balancer such as HA Proxy. Making it cheaper to operate and allows clients more granular control over to which backends it connects to. For example in microservices, you want your clients to connect to multiple backends, but each client should connect to a subset of all backends available to increase resiliency.

Job

Deployment and StatefulSet manage Pods that run indefinitely (not counting restarts or crashes).

For one-time workload or just anything that does not need to run 24/7, Job is the right resource.

apiVersion: batch/v1
kind: Job
metadata:
  name: my-job
spec:
  template:
    spec:
      containers:
      - name: my-job-container
        image: my-job-image
        command: ["bash", "-c", "echo Hello from the Kubernetes cluster"]
      restartPolicy: Never
  backoffLimit: 4

CronJob

CronJobs allows for periodic scheduling of Jobs at given schedule (cron expression).

CronJob embeds a jobTemplate in it's spec. Which is just a Job manifest without metadata. Therefore you can use it with a template engine like Helm, if you need to.

apiVersion: batch/v1beta1
kind: CronJob
metadata:
  name: my-cron-job
spec:
  schedule: "*/1 * * * *"
  timeZone: Etc/UTC # Kubernetes 1.27+
  jobTemplate:
    spec:
      template:
        spec:
          containers:
          - name: my-cron-job-container
            image: my-cron-job-image
            command: ["bash", "-c", "echo Hello from the Kubernetes cluster"]
          restartPolicy: OnFailure

Configuration and secrets

Contents of either ConfigMap or Secret are key-value pairs.

Since it's YAML, you can use even multiline strings with |.

Typically, keys are names of environment variables or file names.

ConfigMap

apiVersion: v1
kind: ConfigMap
metadata:
  name: my-config-env
data:
  myKey: myValue
  anotherKey: anotherValue

Or with values as a config file:

apiVersion: v1
kind: ConfigMap
metadata:
  name: my-config-file
data:
  config.ini: |
    [DEFAULT]
    SomeKey = someValue

Secret

If you want YAML secret in readable form, use stringData.

apiVersion: v1
kind: Secret
metadata:
  name: my-apps-db
type: Opaque
stringData:
  DATABASE_URL: mysql://wordpress:[email protected]:3306/wordpress

Otherwise, if you use data (default), values are base64 encoded.

Note

Secrets are not actually encrypted, but only base64 encoded! Therefore secrets are not actual secrets in security-sense, but the resource-level distinction allows RBAC for fine-gained access for applications and cluster administrator, application developers, and others.

Load environment variables from ConfigMap or Secret

apiVersion: v1
kind: Pod
metadata:
  name: my-app-pod
spec:
  containers:
  - name: app
    image: my-app-image
    envFrom:
    - configMapRef:
        name: my-config-env
    - secretRef:
        name: my-apps-db

Mount ConfigMap or Secret as volume

apiVersion: v1
kind: Secret
metadata:
  name: database-url-secret
  namespace: my-app
stringData:
  database.ini: |
    url = mysql://wordpress:[email protected]:3306/wordpress
---
apiVersion: v1
kind: Pod
metadata:
  name: my-app-pod
  namespace: my-app
spec:
  containers:
  - name: app
    image: my-app-image
    volumeMounts:
    - name: database-url-vol
      mountPath: /etc/secrets
      readOnly: true
  volumes:
  - name: database-url-vol
    secret:
      secretName: database-url-secret

The application can then read the file /etc/secrets/database.ini.

Tip

It is recommended to mount secrets to containers as volume mounts rather than using environment variables, since environment variables are sometimes included in error dumps (for example Sentry is doing that). And that would lead to secret exposure.

Persistent data storage

PersistentVolume and PersistentVolumeClaim

PersistentVolume is a Kubernetes resource representing an actual volume.

PersistentVolumeClaim is a Kubernetes resource, marking a PersistentVolume claimed for given workload (Pod). Not allowing anyone else claim the volume.

Access modes

  • ReadWriteOnce (RWO)
  • ReadWriteMany (RWX)
  • ReadOnlyMany (ROX)
  • ReadWriteOncePod (RWOP), Kubernetes 1.29+

Storage classes

Storage class represents a storage backend, connected to Kubernetes with a CSI Driver.

At StorageClass you define if you allow for volume size expansion.

Volume binding mode describes when a volume should be created when using dynamic provisioning. Volume binding modes are:

  • Immediate is the default, creates volume as soon as PersistentVolumeClaim resource is created, which can lead to provisioning issues like volume being created in wrong Available Zone
  • WaitForFirstConsumer solves issues of Immediate and delays volume provisioning until a Pod is created
$ kubectl get storageclass
NAME                 PROVISIONER          RECLAIMPOLICY   VOLUMEBINDINGMODE   ALLOWVOLUMEEXPANSION   AGE
longhorn (default)   driver.longhorn.io   Delete          Immediate           true                   92d

Reclaim policy

  • Retain - volume is kept even after requesting resource (such as StatefulSet) is deleted, preventing data loss. It's up to cluster administrator to delete the volume.
  • Delete - volume is automatically deleted
  • Recycle (rm -rf /mnt/volume/*). On Kubernetes 1.30 only available for nfs and hostPath volume types.

Temporary storage

Not persisted between Pod deletions, but persisted between Pod restarts.

Volume with type emptyDir.

Local storage

  • local-storage storage class
  • hostPath

Warning

Do not use hostPath volumes in production! Since you are directly mounting a directory on host, it can lead to security issues or node errors.

CSI plugins

Kubernetes on it's own only implements APIs to support container storage, the implementation itself is left for vendors.

This brings the Container Storage Interface API. Allowing cluster administrators to install only what you need for your workload, if you need any.

The implementation is called a Driver, which is responsible for dynamically provisioning volumes, mounting them to nodes and setting up file system. Driver is typically

CSI drivers for on-premise:

CSI drivers for cloud:

Namespace

Namespace is a way to separate resources from each other. By default this induces no boundaries between resources. But application (Pod) in one namespace can still connect to a Service in another namespace.

The DNS name then looks like so: <service name>.<namespace>[.svc.<cluster domain>]. The .svc and .[cluster domain] are both optional. Default cluster domain is cluster.local. So the valid names are like so (for service my-service and namespace my-namespace):

  • my-service.my-namespace
  • my-service.my-namespace.svc
  • my-service.my-namespace.svc.cluster.local

You can manage network access from Pods to another (Pods, Namespaces, Services, Ingresses and egress routing rules) with a NetworkPolicy.

Only some resources are "namespaced" and some are cluster-wide.

Namespaced resources:

  • Pods
  • Services
  • Deployments
  • StatefulSets
  • Ingresses
  • ConfigMap and Secrets
  • ServiceAccount
  • Role and RoleBinding
  • ResourceQuota
  • and more

Cluster namespaces:

  • PersistentVolume
  • StorageClass
  • IngressClass
  • PriorityClass
  • PriorityClass
  • RuntimeClass
  • and more

In short, cluster namespaces are non-application specific and tied to cluster itself.

Create namespace:

kubectl create namespace <name>

# or use ns instead of namespace for short
kubectl create ns <name>

Delete namespace:

kubectl delete namespace <name>

Kubeconfig

Kubeconfig is a configuration file on your workstation, which describes to kubectl how to talk to the cluster. It contains the kubeapi endpoint, cluster CA (since by default kubeapi is using self-signed certificate) and user credentials, either token or certificate.

Context

Context represents a current cluster you are connected to, since Kubeconfig file can contain connection information to multiple clusters.

For example, imagine you are a DevOps engineer and you have a cluster per environment: dev, stage, qa, prod. That will leave you with four clusters in your kubeconfig (if you choose to have configuration in a single file).

Merge kubeconfig files

# Backup existing kubeconfig
cp ~/.kube/config ~/.kube/config.bak

# add new config and existing kubeconfig paths to $KUBECONFIG environment variable
# note the colon character between the paths
# RECOMMENDATION: use absolute paths
export KUBECONFIG="~/.kube/config.bak:/path/to/new/kubeconfig"

# merge and flatten kubeconfig, and save output to a file
kubectl config view --flatten > /tmp/new-kubeconfig

# replace your kubeconfig with new kubeconfig
mv /tmp/new-kubeconfig ~/.kube/config

# cleanup: unset $KUBECONFIG variable
unset KUBECONFIG

kubectx

Easily switch between contexts (clusters).

kubectx demo

Rename context:

kubectx <new name>=<old name>

Delete context:

kubectx -d <name>

kubens

Easily switch between namespaces.

kubens kube-system

RBAC

RBAC = Role Based Access Control

Impersonate ServiceAccount

kubectl auth can-i

Resources

Configure resources

You can configure requests and limits.

Resource requests are used for when scheduling Pods and limits are enforced by Kubelet in form of CPU throttling or Out of Memory Kills (exit code 137).

apiVersion: v1
kind: Pod
metadata:
  name: my-app-pod
spec:
  containers:
  - name: app
    image: my-app-image
    # Quality of Service class: Burstable
    resources:
      requests:
        cpu: "100m"
        memory: "256Mi"
      limits:
        cpu: "500m"
        memory: "1Gi"

You can also expand on resource types, for example with NVIDIA Device Plugin, you can use nvidia.com/gpu resource type. You can define values smaller than one. In that case your application must be ready for "GPU slicing".

Quality of Service

Kubernetes has three Quality of Service classes:

  • Guaranteed: requests equals limits (both cpu and memory)
  • Burstable: requests are smaller than limits (at least cpu or memory or both)
  • Best Effort: no resources defined at all

Quality of Service is used also when scheduling other Pods and mey trigger Pod eviction, see Pod evictions under Pod Disruption Budget.

The Quality of Service priority goes as follows: Guaranteed > Burstable > Best Effort.

If you want more fine gained control over Pod scheduling and it's priority, see Priority Class resource.

Tip

Use the Guaranteed QoS class for stateful workloads or anything important, that you know needs to run. For example asynchronous operations usually do not need QoS Guaranteed.

Resource consumption

Note

Requires metrics server to be installed in the cluster.

kubectl top pods

# or

kubectl top nodes

Startup, liveness, and readiness probes

Probes help Kubernetes determine the state/health of a Pod.

Kubernetes supports multiple types of probes:

  • Execute a binary/shell script
  • TCP connection
  • HTTP probe (status code based: 2xx vs 5xx, usually 200 and 503)
  • gRPC probe, Kubernetes 1.27+

Startup probe

Wait for Pod to start, useful when application start takes time, for example Java applications or machine learning models.

For example, Keycloak takes quite a while to start up and is a great use of startup probe, in order to to overload liveness probe.

Liveness probe

Is the program running? If not, restart the Pod.

Readiness probe

Is the program ready to accept traffic? If not, do not send traffic to the Pod.

Best practices

  • liveness probe is not dependent on external dependencies (database, cache, downstream services,...)
  • different liveness and readiness probes
  • readiness probe should stop being ready as soon as possible after receiving SIGTERM signal, allowing service to gracefully shutdown

Pod autoscaling

One of great Kubernetes strengths is Kubernetes capability of scaling workload up and down.

Horizontal Pod Autoscaler

Changing the number of Pods running to handle the incoming requests efficiently.

Vertical Pod Autoscaler

Unlike HPA, VerticalPodAutoscaler does not come with Kubernetes by default, but it's a separate project on GitHub that you need to install yourself.

Changing the requested amount resources on Pods.

Note

This will invoke a Pod restart to apply the new configuration.

KEDA

KEDA (Kubernetes Event-Driven Autoscaler) is custom controller allowing for scaling Pods on different criteria/triggers than just plain Kubernetes.

KEDA offers many "scalers", that allows to trigger a scale up or down of Pods. You can scale both Deployments or Jobs.

Cluster autoscaling

Second available option, which goes hand-in-hand with automatically scaling Pods, to efficiently utilize the cluster. So we do not need to provision number of Nodes to cover maximum number of Pods during peak times.

Cluster Autoscaler

Dynamically add or remove nodes from the cluster based on resource consumption. Aka how many nodes do we need to efficiently schedule all Pods.

Karpenter

A cluster autoscaler made by AWS, to achieve higher efficiency and reduce the time from determining that the cluster needs more nodes to actually nodes becoming ready within the cluster.

Pod Disruption Budget

PodDisruptionBudget puts another constraint on Scheduler, since it specifies minimum number of available Pods (or maximum unavailable).

When we are running multiple Pods across multiple Nodes, we want to determine some minimum required amount of Pods, that make service still available without failing.

For example, when we are scaling down the cluster and Pods are being reshuffled across nodes, not all Pods may be available. Pod Disruption Budget says how many Pods can be not ready, but our service is still functioning, perhaps with increased latency.

apiVersion: policy/v1beta1
kind: PodDisruptionBudget
metadata:
  name: my-pdb
spec:
  minAvailable: 2 # use absolute number or percentage
  # or alternatively set maxUnavailable
  #maxUnavailable: 33%
  selector:
    matchLabels:
      app: my-app

The PodDisruptionBudget is using a selector (in the example above a label selector), to select to which Pods it applies to.

Pod evictions

  • Preemption evictions
    • Scheduling a Pod with higher Priority Class
  • Node pressure evictions
    • Node drain
    • Scheduling a Pod with higher Quality of Service
    • API initiated (for example: deleting Pod via kubectl)
    • Taint-based

Helm

Helm is a package manager for Kubernetes.

Helm Chart

A package of all manifest for an application. Containing everything, that you need to run the application. Chart can also determine a minimal Kubernetes version it supports, that is especially important when supporting multiple Kubernetes versions and you make breaking changes in the Chart.

Helm Repository

Supports public and private repositories.

Can be hosted on GitHub, GitLab, AWS S3, Google Cloud Storage, Azure Blob Storage, and more.

helm install

Installs Helm chart to the cluster, creating Helm "release".

helm install my-release ./my-chart

# upgrade
helm upgrade my-release ./my-chart

# install and upgrade
helm upgrade --install my-release ./my-chart

# install from repository
helm repo add stable https://charts.helm.sh/stable
helm repo update
helm install my-release stable/mysql

# install from oci repository
helm install my-release oci://registry.example.com/some/chart

# uninstall
helm uninstall my-release

helm rollback

helm rollback my-release 1

Note

Rollback will actually create a new release (incrementing sequence number) instead of going back. Values/configuration will be copied from the old release to the new one.

helm uninstall

helm uninstall my-release

Helm Controller

See: Helm Controller GitHub repository.

Helm Controller is an external addon not installed by Helm, you need to install it yourself.

Or on Kubernetes distributions like k3s or RKE2, Helm controller is available by default.

Installs Helm release from Kubernetes Custom Resource.

  • HelmRelease
  • HelmReleaseConfig

Important

k3s-io/helm-controller and fluxcd/helm-controller are two different projects, even though they carry the same name!

Kustomize

Kustomize is using overlays and hierarchy-based merging of manifests unlike Helm, which is creating packages.

GitOps

Static manifests, Helm charts, and Kustomize are stored in Git repository and are applied to the cluster from there on pull-based model. Usually a pro-active solution is hosted in the cluster.

Networking

Kubernetes by default runs two networks backed by CNI plugin and kube-proxy on each node.

Locally on each node, networking is handled by iptables, altho there is an ongoing effort to migrate to more modern and performant nftables.

Subnets

It is useful to know your CIDRs, when debugging issues, so you can spot where the network traffic is heading and if it's a correct location.

Subnet CIDRs:

  • Service subnet (default CIDR: 10.43.0.0/16)
  • Pod subnet (default CIDR: 10.42.0.0/16)

Note

CIDRs may vary depending on your Kubernetes distribution or cluster configuration.

Network Policy

NetworkPolicy is a Kubernetes resource describing L4 (TCP/UDP) policies of what kind of workload can talk to what.

Including in-cluster resources (other workload, DNS,...), ingress, and egress policies.

For example highly sensitive workload may not be allowed to connect to anything outside of the cluster, to prevent leaking of sensitive information in case of an attack.

Cluster DNS addon

  • kube-dns
  • coredns (mostly default today)

Cilium Network Policy

If you are using Cilium as your CNI plugin, you can use the CiliumNetworkPolicy, which allows for more fine-grained control over the network traffic. Thanks to introducing L7 (HTTP) policies.

CNI plugins

Kubernetes offers an API interface, that allows vendors to develop custom networking solutions, that will handle networking between Pods and Services.

List of commonly used CNI plugins:

Note

AWS VPC CNI is CNI plugin that allows your nodes to use AWS Elastic Network Interface on your EC2 instances for Kubernetes networking. Using this is recommended to utilized existing systems and not creating another networking layer on top of it.

Security

Pod Security Standards

Kubernetes 1.26+

The Pod Security Standards define three different policies to broadly cover the security spectrum. These policies are cumulative and range from highly-permissive to highly-restrictive. This guide outlines the requirements of each policy.

Policies:

  • Privileged: Unrestricted policy, providing the widest possible level of permissions. This policy allows for known privilege escalations.
  • Baseline: Minimally restrictive policy which prevents known privilege escalations. Allows the default (minimally specified) Pod configuration.
  • Restricted: Heavily restricted policy, following current Pod hardening best practices.

The Privileged policy is purposely-open, and entirely unrestricted. This type of policy is typically aimed at system- and infrastructure-level workloads managed by privileged, trusted users.

The Baseline policy is aimed at ease of adoption for common containerized workloads while preventing known privilege escalations. This policy is targeted at application operators and developers of non-critical applications.

The Restricted policy is aimed at enforcing current Pod hardening best practices, at the expense of some compatibility. It is targeted at operators and developers of security-critical applications, as well as lower-trust users.

Policy Installation:

For correct setup, consult Kubernetes documentation.

Pod Security Admission

Kubernetes 1.25+

Pod Security levels

Pod Security admission places requirements on a Pod's Security Context and other related fields according to the three levels defined by the Pod Security Standards: privileged, baseline, and restricted. Refer to the Pod Security Standards page for an in-depth look at those requirements.

Pod Security Admission labels for namespaces

Once the feature is enabled or the webhook is installed, you can configure namespaces to define the admission control mode you want to use for pod security in each namespace. Kubernetes defines a set of labels that you can set to define which of the predefined Pod Security Standard levels you want to use for a namespace. The label you select defines what action the control plane takes if a potential violation is detected:

  • enforce Policy violations will cause the pod to be rejected.
  • audit Policy violations will trigger the addition of an audit annotation to the event recorded in the audit log, but are otherwise allowed.
  • warn Policy violations will trigger a user-facing warning, but are otherwise allowed.
# MODE must be one of `enforce`, `audit`, or `warn`.
# LEVEL must be one of `privileged`, `baseline`, or `restricted`.
pod-security.kubernetes.io/<MODE>: <LEVEL>

As mentioned in Pod Security Standards, you can use different engines/implementations to enforce/audit/warn upon. For example, Kyverno.

Pod Security Policy

Metrics

All Kubernetes components are exposing metrics in Prometheus format.

Prometheus

See: Prometheus website.

Prometheus is open-source and CNCF Graduated project (like Kubernetes), built originally at SoundCloud. It is designed to scrape monitor applications metrics. Also evaluate user-defined rules and alert when rules are broken.

Prometheus Operator

See: Prometheus Operator website.

Prometheus Operator helps you managed Prometheus and it's monitoring targets on Kubernetes. Both Pods behind Services or Pods directly (stateful apps).

Over time, Prometheus Operator (and it's CRDs) become widely supported by various projects and is de-factor open-source monitoring standard on Kubernetes today.

kube-prometheus-stack Helm Chart

If you need something quickly and out-of-the-box, kube-prometheus-stack is a great choice!

The Helm Chart installs by default the following components:

  • Prometheus Operator CRDs
  • Prometheus Operator
  • Prometheus
  • Alertmanager
  • Grafana
  • Node exporter

Logging

Elastic Cloud on Kubernetes (ECK)

Grafana Loki

Other

Cloud integrations like AWS CloudWatch, Azure Monitor, Google Cloud Operations Suite.

  • Fluentd
  • Splunk
  • DataDog

Extending Kubernetes

Operators

Operators are extensions to Kubernetes that make use of custom resources to manage applications and their components.

For example, RedHat OpenShift (Kubernetes distribution by RedHat) is heavily utilizing operators.

Kubebuilder

Kubernetes SDK for building Operators and Controllers.

Operator Framework

If you want to use Operator Lifecycle Manager or integrate more with Kubernetes distribution like RedHat OpenShift, use Operator Framework.

Operator Framework is built on top of Kubebuilder, so you do not need to learn new APIs, it just brings some more functionality and integration with OLM.

The Operator Framework projects offers Operator SDK.

Operator Lifecycle Manager (OLM)

Operator to manage operators.

Integrated with RedHat OpenShift, making it easy installing operators from OpenShift Admin Console.

Cluster API

Cluster API website

Kubernetes project's solution to provisioning, operating, and upgrading multiple clusters.

Links

Questions?

Thank you, that's all 👋

Vojtěch Mareš