Containers & Orchestration

Question 1

Why do containers use less memory and start faster than virtual machines?

Accepted Answer

Containers share the host operating system kernel instead of bundling a full guest OS — Each virtual machine bundles its own complete guest operating system (kernel, system libraries, etc.), which consumes significant memory and takes time to boot. Containers skip that overhead because they share the host OS kernel — only the application code and its dependencies are packaged, so they start in milliseconds instead of seconds.

Question 2

What is a container image?

Accepted Answer

A read-only, layered blueprint containing the app and all its dependencies — A container image is a read-only, layered snapshot: each layer adds files on top of the previous one (OS base, runtime, libraries, your app code). When you run an image, the container engine starts a live process from that blueprint. The image itself never changes — you run many containers from the same image simultaneously.

Question 3

What does "desired state" mean in the context of container orchestration?

Accepted Answer

A spec the orchestrator continuously matches — e.g., run 5 replicas — Desired state is the specification you declare — for example, "I always want 5 replicas of my-app running." The orchestrator watches the actual state of the cluster and acts whenever it drifts from that spec (e.g., a node crash reduces replicas to 3, so the orchestrator schedules 2 more). This control loop is the core idea behind systems like Kubernetes.

Question 4

A node hosting 2 of your 5 replicas crashes. What does a healthy orchestrator do?

Accepted Answer

Schedules 2 new pods on healthy nodes to restore 5 running replicas — This is self-healing in action. The orchestrator detects that only 3 replicas are running (below the desired 5) and schedules 2 replacement pods on the surviving healthy nodes. From a user perspective the service keeps running, just with a brief reduction in capacity during the transition.

Question 5

In Kubernetes, what is a "pod"?

Accepted Answer

The smallest deployable unit — one or more containers sharing a network namespace and optional storage — A pod is Kubernetes's atomic unit of scheduling. It wraps one or more containers that need to share the same network namespace (same IP, same localhost) and optionally shared storage volumes. Typically a pod holds one main container plus optional sidecar containers (e.g., a log collector). The scheduler places entire pods onto nodes, not individual containers.

Question 6

What problem does orchestration primarily solve when running hundreds of containers across many machines?

Accepted Answer

It automates placement, health-checking, and restarts across many machines — At small scale you might SSH into a server and manually start a container. At hundreds or thousands of containers across dozens of machines, manual management is impossible. Orchestrators handle placement decisions, detect and replace unhealthy containers, scale replicas up or down based on load, and roll out updates — all without human intervention for each action.

Question 7

How does horizontal scaling work in container orchestration?

Accepted Answer

Increasing the replica count so more container instances share the incoming load — Horizontal scaling means adding more instances (replicas) of your service rather than making one instance bigger (vertical scaling). Because containers start in milliseconds and are identical copies from the same image, an orchestrator can spin up ten more replicas in seconds when traffic spikes, then remove them again when load drops — paying only for what you use.

Question 8

What is "bin packing" in the context of a Kubernetes scheduler?

Accepted Answer

Placing pods on nodes to use CPU and memory efficiently across the cluster — Just like fitting irregularly shaped items efficiently into a bin, Kubernetes's scheduler tries to place pods onto nodes so that resource capacity is used well — not leaving one node idle while another is overloaded. Each pod declares its CPU and memory requests; the scheduler only assigns it to a node that has sufficient free capacity to honor those requests.

Question 9

Which of the following best describes Docker's role versus Kubernetes's role?

Accepted Answer

Docker is a container runtime for one machine; Kubernetes orchestrates containers across an entire cluster — Docker (and its underlying runtime containerd) handles the mechanics of building images and starting/stopping individual containers on one machine. Kubernetes operates at a higher level: it decides which node each pod runs on, tracks health, manages networking between pods, and enforces desired state across an entire cluster of machines. Kubernetes uses a container runtime (like containerd) under the hood.

Question 10

What is a significant drawback or caveat of adopting container orchestration?

Accepted Answer

The cluster — control plane, etcd, network plugins, and access controls — must be configured and maintained by your team — Orchestration is powerful but not free. A Kubernetes cluster has many moving parts (control plane, etcd, CNI plugins, storage classes, RBAC, etc.) that your team must understand, upgrade, and troubleshoot. For a small hobby project with one service, this complexity can easily outweigh the benefits. Managed services (EKS, GKE, AKS) reduce but do not eliminate that burden.

Container or VM?

Test your knowledge

Why do containers use less memory and start faster than virtual machines?

The short version

Interview Questions

References & Further Reading

Related topics

Love this visualization?