What Are Containers - Container Fundamentals | OneNoughtOne

0/273

System Design (HLD)Container Fundamentals

Container Fundamentals

LevelIntermediate

Duration75 mins

TopicContainer Fundamentals

1 / 5

What Are Containers

The Revolution in Software Deployment

Before 2013, deploying software at scale was a nightmare. Development teams spent countless hours resolving dependency conflicts, environment inconsistencies, and the dreaded "it works on my machine" syndrome. Operations teams struggled with resource utilization, application isolation, and deployment reliability. Then Docker emerged and fundamentally transformed how we package, distribute, and run software.

Containers represent one of the most significant paradigm shifts in infrastructure history—rivaling the impact of virtual machines in the early 2000s. Today, containers power everything from small startups to the largest tech companies, underpinning platforms that serve billions of users daily.

What You Will Learn

By the end of this page, you will deeply understand what containers are at a fundamental level, how they differ from traditional deployment approaches, the Linux kernel features that make them possible, and why they became the dominant deployment paradigm in modern software engineering. You'll gain the conceptual foundation necessary to reason about containerized architectures.

What Is a Container, Really?

At its core, a container is a standard unit of software that packages code and all its dependencies so the application runs quickly and reliably across different computing environments. But this definition, while accurate, masks the elegant engineering underneath.

A container is not a virtual machine. It doesn't run its own operating system kernel. Instead, a container is a specially isolated process (or group of processes) running on the host operating system, with carefully controlled visibility into system resources.

The key insight: Containers use the host's kernel but create the illusion of being a separate system. They can have their own:

Filesystem hierarchy (their own /bin, /etc, /var)
Network stack (their own IP address, ports, routing tables)
Process tree (PID 1 inside the container isn't PID 1 on the host)
Resource limits (CPU, memory, I/O constraints)
User mappings (root inside may not be root outside)

The Mental Model

Think of containers as "super processes" with special powers. A regular process shares everything with other processes on the system—same filesystem, same network, same view of other processes. A container wraps a process in isolating walls, giving it a custom view of the system while still sharing the underlying kernel.

Why this matters for system design:

Understanding that containers are processes, not VMs, has profound implications:

Startup time: Containers start in milliseconds, not minutes, because there's no OS to boot
Resource overhead: Containers add minimal overhead since they're just processes
Density: You can run hundreds of containers where you might run tens of VMs
Isolation guarantees: Container isolation is process-level, not hardware-level (security implications)
Host compatibility: Containers must be compatible with the host kernel (Linux containers on Linux, Windows containers on Windows)

The Problem Containers Solve

To truly appreciate containers, we must understand the pain they alleviate. Before containerization, software deployment suffered from several fundamental challenges:

Pre-Container Deployment Challenges

•Environment Inconsistency — Development machines differed from staging, which differed from production. Each environment had subtly different library versions, configurations, and system settings. Bugs would appear only in specific environments.
•Dependency Hell — Applications requiring conflicting library versions couldn't coexist on the same server. Installing Application A's dependencies might break Application B. Teams spent days debugging mysterious failures.
•Slow Provisioning — Setting up new servers meant manually installing dependencies, configuring services, and hoping nothing was missed. This could take hours or days and was error-prone.
•Resource Waste — Applications were often deployed on dedicated servers running single workloads, leading to massive underutilization. Average server utilization was often below 15%.
•Configuration Drift — Manual changes accumulated over time, making servers unique "snowflakes." No one knew exactly what was installed where, making debugging and replication nightmarish.
•Security Boundaries — Running multiple applications on shared hosts meant they could potentially access each other's files, network connections, and memory—creating security risks.

The "It Works on My Machine" Problem
Scenario	What Goes Wrong	Root Cause
Dev → Production	Application crashes on startup	Different glibc version in production
Developer A → Developer B	Tests pass for A, fail for B	Different Python packages installed locally
Today → Tomorrow	Deployment worked yesterday, fails today	System package auto-updated overnight
Server 1 → Server 2	Works on one production server, not another	Configuration drift over months of patches
Dev → CI Pipeline	Code works locally, CI fails	CI has different environment variables

The fundamental insight:

These problems all stem from the same root cause: incomplete specification of the runtime environment. When you deploy source code or even compiled binaries, you're implicitly depending on the target environment having exactly the right libraries, configurations, and system state. Containers solve this by packaging the application with its environment.

Beyond Code Portability

Java promised "write once, run anywhere" through the JVM. Docker delivers "build once, run anywhere" by including everything above the kernel. The container becomes the unit of deployment, carrying its entire filesystem and dependencies like a hermetically sealed package.

How Containers Work: The Linux Foundations

Containers aren't magic—they're built on Linux kernel primitives that have existed for years. Understanding these primitives demystifies containers and helps you reason about their behavior and limitations.

The three pillars of containerization:

Namespaces — Isolation of what a process can see
Control Groups (cgroups) — Limitation of what a process can use
Union Filesystems — Efficient, layered storage

Let's examine each in depth:

Linux Namespaces

•PID Namespace — Isolates process IDs. A container sees its own process tree, with its init process as PID 1. The same process has a completely different PID when viewed from the host.
•Network Namespace — Isolates network interfaces, IP addresses, routing tables, and port bindings. A container can bind to port 80 even if the host or other containers also use port 80.
•Mount Namespace — Isolates the filesystem mount points. A container sees its own root filesystem, /proc, /sys, and any mounted volumes, independent of the host's mounts.
•UTS Namespace — Isolates hostname and domain name. Each container can have its own hostname, useful for application identification and networking.
•IPC Namespace — Isolates inter-process communication resources. Containers can't access each other's shared memory segments, message queues, or semaphores.
•User Namespace — Isolates user and group IDs. A process can be root (UID 0) inside a container but an unprivileged user outside, adding a security layer.
•Cgroup Namespace — Isolates the view of control groups. Containers see a virtualized cgroup hierarchy, preventing them from discovering host resource limits.

Control Groups (cgroups):

While namespaces control visibility, cgroups control resources. They allow the system to limit and account for CPU time, memory, disk I/O, and network bandwidth used by groups of processes.

Key cgroup Controllers
Controller	What It Limits	Example Use Case
cpu	CPU cycles (time slices)	Allocate 2 CPU cores to a container
cpuacct	CPU usage accounting	Track how much CPU a container has used
memory	Memory usage (RAM + swap)	Limit container to 4GB RAM, kill if exceeded
blkio	Block I/O bandwidth	Limit disk reads to 100 MB/s
devices	Device access	Prevent container from accessing GPU
pids	Number of processes	Limit container to 1000 processes
net_cls	Network traffic classification	Prioritize container traffic with QoS

The Kernel Is Shared

All containers on a host share the same Linux kernel. This has security implications—kernel vulnerabilities affect all containers. It also means containers are only as isolated as the kernel enforces. Container escape vulnerabilities exploit gaps in this isolation.

Union Filesystems:

The third pillar enables the efficient layered image system that makes containers practical. A union filesystem overlays multiple directories (layers) to present a unified view:

Base layers are read-only (e.g., Ubuntu base image, Python runtime)
Upper layers capture changes (your application code, configuration)
Writes go to a thin writable layer on top
Reads check layers from top to bottom until the file is found

This layering means:

10 containers based on the same image share the base layers
Only differences between containers consume additional space
Images can be constructed incrementally from common bases

Container Architecture: A Deep Dive

Understanding the components that make containerization work helps you diagnose issues, optimize performance, and make informed architectural decisions.

The Container Runtime Stack:

Containerization involves multiple layers of software, each with specific responsibilities:

Container Runtime Stack Components
Layer	Examples	Responsibility
Container Engine	Docker, Podman, CRI-O	High-level container management (build, pull, run)
Container Runtime	containerd, runc, gVisor	Low-level container lifecycle (create namespaces, cgroups)
OCI Specifications	runtime-spec, image-spec	Industry standards ensuring compatibility
Linux Kernel	Namespaces, cgroups, seccomp	Actual isolation and resource management

The Container Lifecycle:

A container goes through distinct states during its lifetime:

Created — Container process is set up but not yet running
Running — Container process is executing
Paused — Container process is frozen (SIGSTOP)
Stopped — Container process has exited (still exists on disk)
Deleted — Container is removed entirely

Understanding process hierarchy:

When Docker starts a container, the process hierarchy looks like this:

Container Process Hierarchy

Process Tree

# On the host system:
dockerd (1234)                    # Docker daemon
  └── containerd (1235)           # Container runtime
        └── containerd-shim (1236) # Shim process
              └── nginx (1237)    # Actual container process (PID 1 inside container)
                    └── nginx worker (1238)
                    └── nginx worker (1239)
 
# Inside the container (different PID namespace):
nginx (PID 1)                     # Same process, but PID 1 inside
  └── nginx worker (PID 2)
  └── nginx worker (PID 3)

The containerd-shim pattern:

The shim process is a critical design pattern that enables:

Daemonless containers — The shim becomes the parent, so dockerd can restart without killing containers
Stream handling — The shim keeps stdout/stderr connections open for logging
Exit code reporting — The shim captures the container's exit code
Zombie reaping — The shim acts as the init process for orphaned processes

This architecture allows Docker to be upgraded while containers keep running—essential for production environments where container uptime is critical.

OCI Standards

The Open Container Initiative (OCI) standardized container formats and runtimes. This means container images built with Docker can run on Podman, CRI-O, or any OCI-compliant runtime. Standards prevent vendor lock-in and ensure ecosystem interoperability.

Container Networking Fundamentals

Networking is often the most complex aspect of containerization. Understanding how containers communicate is essential for designing distributed systems.

Network Namespace Mechanics:

Each container gets its own network namespace, which includes:

Its own network interfaces (typically eth0)
Its own IP address(es)
Its own routing table
Its own iptables rules
Its own port bindings

Container Network Modes

•Bridge Mode (default) — Containers connect to a virtual bridge (docker0). They get private IPs (172.17.x.x) and communicate through the bridge. Port mapping exposes services to the host.
•Host Mode — Container shares the host's network namespace. No network isolation, but maximum performance. The container directly binds to host ports.
•None Mode — Container has a loopback interface only. Complete network isolation for security-sensitive workloads.
•Overlay Mode — Spans multiple hosts for container clusters. Uses VXLAN to create virtual networks across physical network boundaries. Essential for orchestrators like Kubernetes.
•Macvlan Mode — Containers get MAC addresses and appear as physical devices on the network. Useful for legacy applications expecting to be on a physical LAN.

How bridge networking works:

┌─────────────────────────────────────────────────────┐
│                    HOST SYSTEM                       │
│  ┌─────────────────────────────────────────────────┐ │
│  │           docker0 bridge (172.17.0.1)           │ │
│  └──────────┬─────────────────────┬────────────────┘ │
│             │                     │                  │
│  ┌──────────▼──────┐   ┌──────────▼──────┐          │
│  │   Container A    │   │   Container B    │          │
│  │   172.17.0.2     │   │   172.17.0.3     │          │
│  │   eth0 (veth)    │   │   eth0 (veth)    │          │
│  └─────────────────┘   └─────────────────┘          │
│                                                      │
│  eth0 (physical) ──────────────────────▶ Internet   │
│  NAT via iptables                                    │
└─────────────────────────────────────────────────────┘

Traffic between containers flows through the bridge. Traffic to the outside world goes through NAT (Network Address Translation) via iptables rules that Docker automatically configures.

DNS Resolution in Docker

Docker provides built-in DNS for containers. Containers on user-defined networks can resolve each other by container name. The embedded DNS server runs at 127.0.0.11 inside containers. This enables service discovery without external tooling—containers can connect to 'database' instead of hardcoding IPs.

Container Storage: Ephemeral by Default

Container storage follows a principle that surprises many newcomers: containers are ephemeral. When a container is deleted, all data written to its filesystem is lost. This is by design and has profound implications for how we architect stateful applications.

The layered filesystem in action:

Container Filesystem Layers

Layer Structure

┌─────────────────────────────────────────┐
│    Container Writable Layer (thin)      │  ← Writes go here (deleted with container)
├─────────────────────────────────────────┤
│    Application Layer (read-only)        │  ← COPY app.py, npm install, etc.
├─────────────────────────────────────────┤
│    Runtime Layer (read-only)            │  ← Python, Node, Java runtime
├─────────────────────────────────────────┤
│    Base Image Layer (read-only)         │  ← Ubuntu, Alpine, Debian
└─────────────────────────────────────────┘
 
Copy-on-Write (CoW):
- Reading a file: Search layers top-to-bottom, return first match
- Writing a file: Copy file to writable layer, modify there
- Deleting a file: Add "whiteout" file to writable layer

Persistent Storage Options

•Volumes — Docker-managed storage in /var/lib/docker/volumes/. Best for data that should persist beyond container lifecycle. Portable and backup-friendly.
•Bind Mounts — Maps a host directory directly into the container. Good for development (live code reloading) but creates host-container coupling.
•tmpfs Mounts — In-memory filesystem. Extremely fast, but data is lost on container stop. Good for sensitive data that shouldn't hit disk.
•Named Volumes — Volumes with human-readable names, making them easier to manage across container recreations.

Storage drivers:

Docker supports multiple storage drivers that implement the layered filesystem:

Driver	Performance	Notes
overlay2	Excellent	Default and recommended for Linux
aufs	Good	Legacy, not in mainline kernel
devicemapper	Moderate	Common in RHEL/CentOS environments
btrfs	Good	For btrfs filesystems
zfs	Excellent	High memory usage, advanced features

The choice of storage driver affects image build time, container startup speed, and I/O performance.

Databases and Containers

Running databases in containers is possible but requires careful planning. You MUST use volumes for data persistence. Consider that container restarts should not cause data loss, and think carefully about backup, replication, and upgrade strategies. Many teams run databases on VMs or managed services while containerizing stateless applications.

Container Security: Trust Boundaries

Container security is fundamentally different from VM security. Understanding the trust model helps you make informed decisions about what workloads to containerize and how to configure them.

The security paradox:

Containers provide isolation, but they share a kernel with the host. This creates a different security posture than VMs:

VMs: Each VM has its own kernel; compromising a VM rarely affects the hypervisor
Containers: All containers share the host kernel; kernel exploits affect everyone

Container Security Layers

•Linux Capabilities — Instead of all-or-nothing root, Linux supports fine-grained capabilities (CAP_NET_BIND_SERVICE, CAP_SYS_ADMIN, etc.). Docker drops most capabilities by default.
•Seccomp Profiles — Filter which system calls a container can make. Docker's default profile blocks ~44 dangerous syscalls while allowing normal operation.
•AppArmor/SELinux — Mandatory Access Control (MAC) policies that restrict container actions even if other isolation fails. Provides defense-in-depth.
•Read-Only Filesystems — Containers can run with read-only root filesystems, preventing any persistent modification and limiting attack surface.
•User Namespaces — Map the container's root user to an unprivileged user on the host. Even if attackers achieve root inside the container, they're unprivileged outside.
•Network Policies — Restrict which containers can communicate with each other. Implement network segmentation at the container level.

Container Security Best Practices
Practice	Why It Matters	Implementation
Don't run as root	Limits damage from container compromise	USER directive in Dockerfile
Use minimal base images	Fewer packages = smaller attack surface	Alpine, distroless, scratch images
Scan images for vulnerabilities	Catch known CVEs before deployment	Trivy, Clair, Snyk integration
Don't embed secrets in images	Images are often shared/cached	Environment variables, secrets management
Use read-only containers	Prevents malware persistence	--read-only flag or securityContext
Drop unnecessary capabilities	Principle of least privilege	cap_drop: ALL, then whitelist
Limit resources	Prevent DoS attacks	Memory and CPU limits

The Privileged Container Danger

Running containers with --privileged disables almost all security features. Privileged containers have full access to host devices, can load kernel modules, and have minimal isolation. They're essentially root on the host. Never run privileged containers in production unless absolutely necessary, and always understand the implications.

Why Containers Changed Everything

Containers succeeded not just because of technology, but because they solved organizational and workflow problems that plagued software teams for decades.

The immutability principle:

Containers introduced immutability to infrastructure. Once an image is built, it doesn't change. You don't patch a running container—you build a new image and replace the container. This has profound benefits:

Benefits of Immutable Infrastructure

•Reproducibility — Every deployment is identical. If it worked in staging, it works in production (same image, same behavior).
•Rollback — Bad deployment? Roll back to the previous image tag. No wondering what changed—you have the exact previous state.
•Auditability — Images are versioned and tracked. You know exactly what code and dependencies ran at any point in time.
•Scalability — Scaling means starting more identical containers. No per-server configuration, no drift, no snowflakes.
•Dev-Prod Parity — Developers run the same images locally that run in production. Environment differences are eliminated.

The microservices enabler:

Containers make microservices architectures practical. Before containers, running dozens of services meant managing dozens of different runtime environments. With containers:

Each service is self-contained with its dependencies
Services can use different languages and frameworks
Deployment and scaling are independent
Failure isolation is automatic (one container crashing doesn't affect others)

The DevOps accelerator:

Containers bridge the gap between development and operations:

Developers own the Dockerfile—they define the environment
Operations owns the container orchestration—they control how containers run
The container image is the contract between them

This clear responsibility boundary reduced friction and enabled faster delivery.

The Industry Transformation

By 2023, over 90% of organizations were running containers in production. Containerization became the default, not the exception. Kubernetes, born from Google's internal container orchestration experience, became the de facto platform for running containers at scale. The container ecosystem spawned entire industries around monitoring, security, service meshes, and developer tools.

Summary: Understanding Containers

Let's consolidate what we've learned about containers:

Key Takeaways

•Containers are isolated processes — They use the host kernel but create the illusion of separate systems through namespaces and cgroups. They are not VMs.
•Containers solve environment inconsistency — By packaging the application with its dependencies, containers eliminate 'works on my machine' problems.
•Linux namespaces provide isolation — PID, network, mount, UTS, IPC, user, and cgroup namespaces each isolate different aspects of the system.
•Cgroups limit resources — CPU, memory, I/O, and process limits prevent containers from monopolizing host resources.
•Layered filesystems enable efficiency — Union filesystems allow image sharing, fast builds, and minimal storage overhead.
•Container storage is ephemeral by default — Volumes and mounts are required for data persistence beyond container lifecycle.
•Container security differs from VM security — The shared kernel means defense-in-depth is essential. Never trust a single isolation layer.
•Containers enable immutable infrastructure — Build once, deploy anywhere, roll back easily. No more configuration drift.

What's next:

Now that you understand what containers are and how they work at a fundamental level, we'll move on to Docker basics—the practical skills for building, running, and managing containers. You'll learn the tools that put these concepts into practice.

Page Complete

You now have a deep understanding of what containers are, the Linux primitives that make them possible, and why they revolutionized software deployment. This conceptual foundation will inform every decision you make when designing and operating containerized systems.