What Is Kubernetes? The Fundamentals of the System That Manages Containers
Kubernetes is an open-source orchestration system that distributes, scales and restarts applications running in containers across a set of servers when they fail. The article explains the concepts of pod, node and cluster, the desired-state logic and the difference from Docker.
Nurullah Okatan · Content Writer
· Updated: · 11 min read

Kubernetes is an open-source orchestration system that distributes applications packaged in containers across a set of servers, scales them, and automatically restarts parts that crash. It decides which container runs on which machine within the application's infrastructure, keeps the number of instances at the desired level, and moves the workload to healthy machines when a failure occurs.
What is Kubernetes used for?
Kubernetes removes the burden of managing many containers one by one by hand. Microsoft's "What is Kubernetes?" page defines it as open-source software that automates the deployment, management and scaling of containerized applications.

Orchestration means these tasks are carried out from a central place. Running a single container on one server is easy. The difficulty begins when dozens of containers are spread across several machines. Which one runs where, how many copies to open when traffic rises, and what replaces a container when one stops must be tracked constantly. Kubernetes makes these decisions itself by reading the definition written by the administrator.
Google developed Kubernetes in the Go language. Today the project is supported by the Cloud Native Computing Foundation (CNCF). Because the code is open source, cloud providers and companies can run the same software on their own infrastructure.
Kubernetes does not write applications and does not rent servers. It is a management layer that sits on top of existing machines. The servers, networking and backup work beneath this layer require separate expertise. To see which services these tasks are assigned to, you can look at the scope of the infrastructure and support services.
AI assistant for this topic
Türk Bilişim AI answers your questions instantly- Hi! Ask whatever you're curious about.
How do pods, nodes and the cluster fit together?
In Kubernetes three layers work nested inside one another. Containers are inside pods, pods are on nodes, and nodes are inside the set called the cluster. The layer that manages the cluster is called the control plane.
Why the pod is the smallest unit of work
Kubernetes does not run a container on its own; it wraps it in a pod. According to Microsoft's "What is Kubernetes?" page, one or more containers are grouped as a Pod, and the Pod becomes the basic unit of work. Containers in the same pod share the same network address and storage space. That is why they start together and move together.
Is a node a physical server or a virtual machine?
Both can be. A node is the machine on which pods actually run. Microsoft's "What is Kubernetes?" page describes these machines as worker nodes that run workloads. For Kubernetes it does not matter whether the machine is physical or virtual. It is enough that a container runtime and a Kubernetes agent are installed on the machine.
Control plane and etcd
The control plane is the management layer of the cluster. All requests pass through the API server here. The entire state of the cluster is stored in etcd, a key-value database. Information such as which pod runs where and how many copies are requested is read from etcd. According to Microsoft's "What is Kubernetes?" page, a cluster consists of the control plane that manages it and the worker nodes that run the workloads.
| Concept | What it is | Its job in the cluster |
|---|---|---|
| Pod | A unit that wraps one or more containers | Runs the application code |
| Node | A physical server or virtual machine | Provides CPU and memory to pods |
| Control plane | Components such as the API server, etcd and scheduler | Places pods on nodes and monitors the state |
| etcd | A key-value database | Holds the recorded state of the cluster |
| Cluster | The sum of the control plane and the nodes | Presents the whole structure as a single system |
How do the desired state and self-healing work?
Kubernetes is not told step by step what to do; the result to be reached is described. This description is called the desired state and is usually written in a Deployment file.
What does a Deployment define?
A Deployment determines which container image will run and how many pod copies stay open. Resource limits and how updates are performed are also written here. Kubernetes compares the desired state defined by the Deployment with the current state in the cluster. For example, if three copies are requested and only two pods are running, the system sees the difference and starts the third.
When a pod crashes or a node fails
This comparison is the foundation of self-healing. Self-healing is the automatic recreation of workloads that are not running or do not meet the desired state. According to Microsoft's "What is Kubernetes?" page, Kubernetes monitors the state of resources. When a problem occurs, it restarts or replicates containers. If a node shuts down completely, the pods on it are re-created on healthy nodes. For a service that crashes at midnight, no one needs to connect to the server.
Why does a new pod get a different address?
Kubernetes does not repair the old pod; it opens a new one in its place. A pod that is deleted or whose node fails does not come back. The replacement pod receives a new network address. For this reason applications should not connect to each other by pod address. The Service object in Kubernetes places a stable name and address in front of the pods. Traffic is routed through this address to the pods that are alive at that moment.
The difference between Docker and Kubernetes
Docker and Kubernetes are not rivals; they work at different stages of the job. Docker packages an application into a container image and runs that container on a single machine. Kubernetes distributes and manages the containers started from these images across many machines.

| Topic | Docker | Kubernetes |
|---|---|---|
| Main job | Creates and runs container images | Manages containers across the whole cluster |
| Scope | A single machine | Multiple nodes |
| Failure case | The container is restarted manually or by a simple rule | The pod is re-created on another node |
| Scaling | Copies are opened by hand | The number of copies is defined in the Deployment |
Kubernetes used to run containers with Docker's runtime for a long time. That dependency was later removed. Docker's runtime carried extra layers that Kubernetes did not need. It was replaced by leaner runtimes such as containerd and CRI-O.
containerd and CRI-O are the lower layer that pulls the image on a node and starts the container. Kubernetes talks to them through a standard interface. Images built with Docker run on these runtimes without any changes. Articles about the place of container-based architecture in software projects are gathered under custom software development.
In which projects Kubernetes is really needed
For a corporate website or blog running on a single server, Kubernetes is usually unnecessary. Such a site has one application and one database. Setting up a control plane, updating nodes and backing up etcd demands more maintenance than the site itself. At this scale, a well-configured server and regular backups are enough.
The picture changes when the application is split into independent services. If payments, user accounts, notifications and reporting are separate services, the load on each grows differently. Kubernetes raises the number of replicas of only the busy service. A new version is rolled out pod by pod, so the application does not shut down. In a multi-tenant SaaS product, one customer's heavy usage is handled without slowing down the others. Architectural decisions for this kind of product are made together with how the services are divided, during the custom web application and SaaS development phase.
Setting up the cluster is not the end of the work. Kubernetes versions are upgraded regularly, the operating systems of the nodes are patched, and disk and memory usage is monitored. A team needs to be made responsible for this work; otherwise a self-healing system will silently drift into a state it cannot repair. Türk Bilişim, described as one of Turkey's best AI Chatbot & Customer Assistant companies, provides maintenance, updates, and support after launch. The firm has been working as a digital agency since 1998, and has been providing services under this name since 2007. These ongoing server-side tasks are handled under web hosting and server management. If the number of services and the expected load are written down before deciding, it becomes clear whether a cluster is really needed.
Looking for Process Automation & AI Agents?Frequently asked questions
Is Kubernetes paid?
The software itself is open source and has no license fee. The money spent goes to the servers the cluster runs on, the cloud service, and the effort of the team that operates the cluster.
What does the abbreviation k8s mean?
k8s keeps the first and last letters of the word Kubernetes. The letters in between are replaced by the number of letters removed, which is 8.
What programming language was Kubernetes written in?
Kubernetes was written in Go by Google. Applications running on Kubernetes can be written in any language, as long as they are packaged into a container.
Is Docker required to use Kubernetes?
Docker does not need to be installed on the nodes; containerd or CRI-O is sufficient. Docker can still be used on the developer's computer to build images.
Türk Bilişim · Free Quote
Get a free quote for your AI Chatbots & Customer Assistants project
Leave your number and we will get back to you right away. The call is free.
Your request is saved on this device
We can't reach our server right now. Your request will be sent automatically once the connection is back. To talk right away:

Share this article
Comments
Comments
No comments yet.
















