Kubernetes Introduction

class: center, middle

## Introduction to

<br>

<h3 class="center">Fernando Ripoll <span style="color: gray;">(CoBrowser.net)</span> & Adrián Moreno <span style="color: gray;">(Dell EMC)</span></h3>

<h3 class="center">Tarragona Developers Meetup @ 29th March 2017</h3>

---

### **Kubernetes** is a system for managing containers across multiple hosts, providing basic mechanisms for deployment, maintenance, and scaling of containers

---

# Facts

* Means "helmsman" or "pilot" in Greek
* a.k.a. `k8s`
* Project started by Google in 2014
* Open source, written in Go
* Contributors
  * Google, CoreOS, RedHat, Microsoft, etc.
* Very active development
  * 45k+ commits, 21k+ stars, 7k+ forks (as of Mar 2017)

---

## Not an IaaS, nor a PaaS

#### e.g. PaaS on top of Kubernetes
 * Deis
 * OpenShift v3
 * Gondor

---

# Kubernetes does not...

* have a service marketplace
* dictate application frameworks
* restrict the language runtimes (e.g., Java, Python, Ruby)
* restrict to only 12-factor apps
* distinguish “apps” from “services”
* provide middleware (e.g. message buses, databases, caches)

---

# Kubernetes does...

* Container grouping
* Load-balacing
* Health checking
* Self-healing
* Auto-scaling
* Naming and discovery
* have a extensible design (plugin architecture)
* ...

---

# Kubernetes architecture

---

# master node

* Contains the required control plane components
  * i.e. etcd, API server, scheduler, controller manager
* Contains other optional addons
  * e.g. DNS, User interface, monitoring, logging, etc.
* Support for HA multi-master environments

---

# master node

### etcd

* Distributed key-value store
* Used to store the cluster state and configuration data

### kube-apiserver

* Serves up the Kubernetes API
* Processes operations, validates them, and updates the corresponding objects in etcd

---

# master node

### kube-controller-manager

* Runs controllers that handle routine tasks in the cluster
* Each controller is a separate thread, but run as a single process
* e.g. Node Controller, Service Account Controller, Endpoints Controller, etc.

---

# master node

### kube-scheduler

* Binds unscheduled pods to nodes
* Tracks resource utilization
* Pluggable by design
* Applies predicates (filters) and priorities (sorting)

---

# node (aka minion)

* Worker machine in Kubernetes
* Has the container runtime installed (Docker by default)
* Also the `kubelet` and `kube-proxy` services

---

# node (aka minion)

### kubelet

* Node agent that runs on each node
* Receives commands from the k8s control plane
* Manages pods and their containers, their images, their volumes, etc

### kube-proxy

* Simple network proxy and load balancer
* Its configuration is defined by user-defined services
* Routes traffic to pods across the cluster

---

# Key Concepts

---

# pod

* Basic deployment unit in Kubernetes
* Contains one or more containers
* Run inside nodes
* Has a single IP address (shared by all containers)
* All containers in the pod are deployed to the same host

---

# pod

#### Example:

```yaml
apiVersion: v1
kind: Pod
metadata:
  name: redis-django
  labels:
    name: webapp
spec:
  containers:
  - name: key-value-store
    image: redis
    ports:
    - containerPort: 6379
  - name: frontend
    image: django
    ports:
    - containerPort: 8080
```

---

# pod

---

# label

* Key/value pair
* Attached to objects (e.g. pods)
* Specify identifying attributes of objects that are meaningful and relevant to users
* Can be modified at any time

```json
"labels": {
  "app": "webapp",
  "environment": "dev",
  "release": "canary",
  "tier": "frontend",
}
```

---

# label

---

# selector

* Core grouping primitive in Kubernetes
* Allows users to identify a set of objects
* Allows multiple requirements (comma-separated)
* empty label selector -> all objects
* null label selector -> no objects

#### Equality-based selectors

```yaml
environment = production
tier != frontend
```

#### Set-based selectors

```yaml
environment in (production, qa)
tier notin (frontend, backend)
!partition
```

---

# Replication Controller / Deployment

* Ensures that a specified number of pod "replicas" are running
  * too many pods -> kill some
  * too few pods -> start some
* Supervises multiple pods across multiple nodes
* **Deployments** are the next-gen RC
  * Rolling upgrades
  * Rollbacks

---

# Replication Controller / Deployment

#### Example:

```yaml
apiVersion: extensions/v1beta1
kind: Deployment
metadata:
  name: webapp
spec:
  replicas: 3
  template:
    metadata:
      labels:
        app: webapp
        environment: production
    spec:
      restartPolicy: always
      containers:
      - name: key-value-store
        image: redis
        ports:
        - containerPort: 6379
      - name: frontend
        image: django
        ports:
        - containerPort: 8080
```

---

# service

* Set of pods that are defined by the same selector
* Traffic to the service is load-balanced to pods by Kubernetes

#### Example:

```yaml
apiVersion: v1
kind: Service
metadata:
  name: webapp-service
spec:
  selector:
    app: webapp
    environment: production
  type: NodePort
  ports:
    - protocol: TCP
      port: 80
      targetPort: 8000
```

---

# service

---

---

# Networking

### Requirements

* Pods need to communicate with each other, no matter which host they run on
* Each pod must have a unique IP in the cluster
* Containers in a pod must talk using localhost

### Implementations

* Contiv, **Flannel**, GCE, L2 networks and linux bridging, OpenVSwitch, Weave, etc...

---

## Flannel

* **Overlay network** created by CoreOS
* Uses etcd to maintain a mapping between allocated subnets and real host IP addresses
* Configured with an IP range and the size of the subnet for each node
  * e.g. overlay to use 10.100.0.0/16, each node to receive a /24 subnet
  * Node A 10.100.1.0/24; Host B 10.100.2.0/24; etc.

---

# Demo

---

# Thank you

## Questions?