Kubernetes Full Course 2026 ☸️ | Zero to Job-Ready 🚀 | Projects + 31 LPA Career Path | Class 5

Summary

The class combines a live Kubernetes/EKS demonstration with explanations of workload management, node maintenance, storage, networking, and autoscaling. The instructor also discusses career motivation, personal safety, and free IT training.

Kubernetes Workloads and Replica Management

• Deployment → ReplicaSet → Pods: A Deployment declares the desired number of replicas and manages a ReplicaSet, which creates the Pods. If a Pod is deleted, the ReplicaSet detects the shortfall and creates a replacement.
• DaemonSets: A DaemonSet is intended to run a Pod on each eligible node. It is commonly used for node-level agents, such as monitoring or storage components. Unlike a Deployment’s fixed replica count, the number of DaemonSet Pods generally follows the number of eligible nodes.
• StatefulSets: These are suitable for workloads that need stable Pod identities and persistent storage, such as databases. The instructor demonstrates that StatefulSet Pods have ordered, predictable names, unlike the generated names used by Deployments.
• Jobs and CronJobs: Jobs run tasks intended to finish successfully. CronJobs schedule such tasks to run at particular times.

Rolling Updates and Disruption Protection

• A Deployment’s rolling update strategy controls how many Pods may be unavailable during an update (maxUnavailable) and how many extra Pods may be created temporarily (maxSurge).
• The instructor illustrates these settings with eight replicas: a 25% unavailable allowance means up to two Pods may be unavailable, while a 25% surge allows two extra Pods. A larger surge can make an update faster but requires capacity for more Pods.
• A PodDisruptionBudget (PDB) limits voluntary disruptions, such as evictions during node maintenance. Its selector identifies the Pods it applies to, usually by matching labels. Settings such as minAvailable specify how many Pods should remain available.
• Although the class does not always distinguish these concepts precisely, the practical takeaway is that a Deployment’s rolling-update settings govern that Deployment’s rollout, while a PDB helps protect availability during voluntary evictions. A PDB is not a general safeguard against every kind of deletion.

Safely Maintaining or Upgrading Nodes

The instructor demonstrates the following node-maintenance workflow:

1. Cordon the node to prevent new Pods from being scheduled there. Cordoning alone does not remove existing Pods.
2. Drain the node to evict eligible Pods so they can be rescheduled elsewhere. A drain may be blocked if it would violate a PDB.
3. Account for DaemonSet Pods, which are generally not evicted by a normal drain. The drain command may require an option to ignore them.
4. Once the node is empty, perform maintenance or remove it. For a cluster upgrade, the instructor describes bringing up newer nodes, moving workloads, and then retiring the old nodes.

There must be enough capacity on the remaining or replacement nodes for workloads to be rescheduled.

Scheduling: Labels, Selectors, Taints, and Affinity

• Labels and selectors connect resources. For example, a Service or PDB uses a selector to find Pods with matching labels.
• Node affinity specifies which nodes a Pod should or must run on. Pod affinity and pod anti-affinity influence whether Pods are placed near one another or spread apart.
• Taints mark nodes so that Pods are kept away unless they have a matching toleration. The instructor uses the master/control-plane node as an example of a node commonly protected from ordinary workload scheduling.
• The class contrasts strict scheduling requirements with preferences. A strict requirement can leave a Pod pending if no suitable node exists, whereas a preference can allow it to run elsewhere.

Networking and Services

The session reviews Kubernetes Service types:

• ClusterIP: Provides internal access to a group of Pods.
• NodePort: Exposes a service through a port on each node. The instructor gives the usual range as approximately 30000–32767.
• LoadBalancer: Provides an external load-balancing endpoint, commonly used for external access in a cloud environment.
• ExternalName and headless Services are also mentioned.

The instructor demonstrates a Service and external IP in an EKS environment and discusses allowing traffic through the relevant security groups. The class also compares Ingress, which provides HTTP/HTTPS routing into a cluster, with a service mesh such as Istio, which offers more extensive traffic-management capabilities.

Storage

• A PersistentVolume (PV) represents storage available to the cluster, while a PersistentVolumeClaim (PVC) is a workload’s request for storage.
• A Pod can use a PVC so its data persists beyond the lifetime of an individual Pod. The instructor explains that a PVC is namespace-scoped, while a PV is a cluster-level resource.
• StorageClasses define how storage can be provisioned. GP3 and GP2 are compared, with GP3 presented as the newer option.
• A PV’s reclaim policy affects what happens to the underlying storage when its claim is released. The examples discussed include retaining storage for manual recovery or deleting it.

Architecture and Autoscaling

• The class reviews the roles of the control plane and worker nodes. The API server receives requests, etcd stores cluster state, controllers reconcile desired state, and the scheduler assigns Pods to nodes. On worker nodes, kubelet, kube-proxy, and the container runtime participate in running workloads.
• For horizontal scaling, the instructor explains that the Horizontal Pod Autoscaler (HPA) changes the number of Pod replicas based on metrics. The Metrics Server supplies CPU and memory metrics used by the HPA.
• KEDA is introduced for event-driven scaling, while Karpenter is discussed as a way to provision or manage nodes in response to workload needs.
• Horizontal scaling adds or removes Pods; vertical scaling changes the resources available to a Pod. The instructor warns that changing capacity at the Pod level can involve disruption.

Troubleshooting and Hands-On Work

The instructor uses Kubernetes commands and resource descriptions to investigate Pods, nodes, and events, particularly when Pods are pending. The general troubleshooting lesson is to inspect resource status and events rather than guess.

The session also includes Terraform-based EKS cluster setup and cleanup. The instructor notes that resources created separately may need to be cleaned up separately.

Other Themes

Outside the technical lesson, the instructor encourages students to build their skills and employability, promotes free IT training, and asks viewers to register for an upcoming orientation and subscribe to the channel.

A lengthy discussion addresses women’s safety: do not harass or intimidate others, offer help where appropriate, speak up when someone is in danger, and contact the police rather than taking the law into one’s own hands.

Speakers and Sources Featured

• Instructor: Leads the class and demonstrations; the stream refers to the instructor as Neeraj.
• Student participants: Unnamed students ask questions and respond during the live session. Some names heard include Ayush, Keshav, Harsh, Omkar, and Pankaj, though the auto-generated subtitles may have misrecognized them.
• Technical sources shown or discussed: Kubernetes/EKS command-line output and YAML manifests, AWS/EKS resources, and a brief ChatGPT consultation related to a PDB example.