Introduction to Kubernetes

Introduction to Kubernetes
and why you should care
by Paris Apostolopoulos (aka javapapo)

About me
I come from Athens Greece, recently moved to
Luxembourg.
I love Java, Judo and Apple devices :P
Java EE enthusiast
Java Champion (2007) & JBoss Hero
co-Founder and admin of the first Java User Group in
Greece - www.jhug.gr

Goals
This is a soft introduction
Targeting mostly developers that now are entering
this new world of containers, microservices etc.
Save you some time from endless research and
experimentation.
Validate why kubernetes is the right framework
for the job - at least make you consider it!
Introduce basic concepts of kubernetes towards
system design and architecture.

A lot of buzzwords
a lot of trends
we are really good
at this
Microservices
Cloud
Containers

Some remarks
An new exciting world of
technologies.
Containers, microservices and the
‘cloud way’ are taking over.
There is a demand, sometimes not
clearly justified, to move existing
and new apps into this new world.
Devops is taking over but a gap is
created at the same time, ops vs
software developers.

Microservices
Containers
Cloud
Cloud : your private or a public one, a pool of
resources (machines), waiting to execute your
code.
Containers : another way of packaging -
containing applications (services) and OS
dependencies . They are expected to run on top
your ‘cloud’ machines.
Microservices : A new architectural trend on
designing and implementing apps. Separation of
concerns, decoupling, modular.

How?
How do we move from this
To this

How?
How do we design applications that can not
make any assumptions about their runtime
environment? (filesystem/network)
How do we design applications that are
consisted of several moving parts, but can still
be deployed and orchestrated easily?
How do we design applications can move from
one cloud platform to another ?
How can we reuse our existing skills and
techniques on application design and
architecture without deep diving to other areas?

Common
misconceptions
Docker-izing your application
does not mean that you do
microservices.
2 processes (e.g wars) that
exchange JSON are not a
microservices platform.
Technical split of your good old
monolith, does not mean that you
actually moved to a microservices
architecture.

Common
misconceptions
A lot of technologies / frameworks claim to be
‘the operating system’ of the cloud.
The reality is that certain tools cover specific
areas of the overall problem.
Platforms that abstract the real hardware layer
(your good old server cluster).
Platforms that can ‘schedule’ and orchestrate
specific work on a cluster (aka schedulers)
Platforms with mix concerns that promise to act
as a ‘platform as a service’, where you will
design, host and operate your own service.

The reality
You need to invest money and skills on separate
tools and technologies.
The terms platform and infrastructure as a
service are sometimes interpreted in a very
obscure sense.
You find yourself deep diving on ops and
experimenting with OS level services rather than
your application
Currently most of the technologies are on their
infancy, ever changing and competing.

The question
Is there any technology that will abstract most of
the complexity imposed by many infrastructure
technologies?
Is there any technology that will enable me to
think again about application (parts) services and
their layout- rathern the specifics of the
underlying infrastructure?
Is there any technology that can be easy and
powerfull enough so that can be ‘setup’ by the
average Joe Developer in order to be evaluated
and then maybe use it actually?

The answer
Yes! Entering Kubernetes

Kubernetes
κυβερνήτης
the person controlling a ship (captain)
the person that controls / governs a city

Abstract View
(very abstract)
The internet
You Cloud
infrastructure
Your Containers
(services)
Kubernetes is
here!

Kubernetes
History
Google project started in 2014
10 year old experience derived from Google
internal container platform code name ‘Borg’
Project reached it’s 1.0 major release on July
2015.
Written in Go lang.
Heavily relies on Etcd

A less abstract
view
So in plain words either you have a cluster of VMs or a cluster of physical
machines, that are all Kubernetes aware, then they form a Kubernetes cluster.
Kubernetes then is the 'tool' so that you make use of the underlying docker engine
installed in each host (node) . (simplified description).

What is it?
“is an open source system for managing containerized
applications across multiple hosts, providing basic
mechanisms for deployment, maintenance, and scaling
of applications”
Is not a Infrastructure as a Service framework, it
abstracts the data center resources on a high level.
Is not a Platform as a Service framework, it can be use
as the basis so you can come up with one.
It enables you to deploy, manage and run on top of
abstracted pool of resources, your dockerized
applications.
It enables you to ‘design’ your application topology and
spread and interconnect your services with minimal or
changes to the actual service

Anatomy of a
Kubernetes Node
Each node basically runs
a number of services
(kubernetes related) and
a docker engine
We have 2 basic types of nodes
● master
● worker

The Master Node
The controller - controls and manages
the cluster
Run services of kubernetes
kubectl : command line client
rest api: rest api for communicating with the
workers
Scheduling and Replication logic

The Worker Node
The node that eventually host your
services - containers
Run services of kubernetes
kubelet : kubernetes agent (accepts commands
from master)
kubeproxy: network proxy service on a node
level
docker host

pod
Within the worker node, services will be
‘contained’ in a pod.
A pod can contain more than one
services (aka docker containers)
each pod has it’s own IP.
it is a logical host .
master schedules, replicates & creates

All together pods
one instance of
MySQL and one
Tomcat, scheduled in
different pods.

Labels
● Labels are metadata we can assign to
kubernetes resources - such as pods,services
when we create them (spin them).
● They are simple key value pairs.
● Labels are crucial to kubernetes since a lot of
core kubernetes functionality relies on
‘querying’ the cluster for ‘resources’ that have
certain label assigned! (Selectors)
we can query
the
kubernetes
cluster for all
the pods that
belong to an
application
named
‘testApp’

What if I want 3
instances of
tomcat running?

The replication
Controller
● Is responsible to maintain as many pods as they
have been requested by the user.
● It will start or kill pods depending on the replication
limit provided.
○ E.g if we ask for 3 pods running a tomcat
docker instance with a specific description,
then it will kill any attempt to spin a 4th one.
● The replication controller uses a ‘template’ which is
just a simple descriptor that describes exactly what
each pod should contain.
● We can dynamically call the replication controller of
the kubernetes cluster in order to scale up or down a
specific pod.
○ E.g I want to scale up these 3 tomcats and to
make them 6.

How I can group
these 3 pods and
load balance
them?

The Service
● A service is not - a running thing - there is no real
single load balancer.
● It is information stored in the kubernetes cluster
state and networking information propagated to all
the nodes.
● It groups a set of pods, providing a single point of
access.
● For example if we want to access the ‘RestApi’
service we no longer need to know each pod’s ip
address.
● By grouping similar pods into service(s) we
eventually solve the discoverability and connectivity
between our containers.
● A Service with 3 pods of MySQL and another
Service with 3 pods of Wildfly can ‘talk’ to each by
this single ‘domain’ name - cluster internal IP and on
predefined (if we wish ports).

The Service- env
variables?
● Once a service is declared then related
Environment Variables are accessible on the node.
● TOMCAT_SERVICE_HOST=10.0.0.11
TOMCAT_SERVICE_HOST_PORT=80
TOMCAT_PORT=tcp://10.0.0.11:6379

The Service- How?
● Every time we create a service, all the nodes of the
cluster - get aware of it.
● The kube-proxy service running on all workers
allocates a specific port.
● IP-Tables are being populated so that traffic can be
redirected to the related services
● A service is eventually a cluster wide ‘information’
that each node is aware.

Volumes
● Another powerful abstraction
● File access for your services.
● It lives as the pod lives!
● A container can die in the pod, the volume still
lives.
● The volume can point to the Node (host), pod only
or to an external file /storage provider.

Volumes
● The most important thing is to foresee the need of
the volume in your topology.
● Once you have selected the type and you have
integrated it on your app design, you won't have to
worry about, no matter how your pods or
containers run or scale.

Secrets
● Sensitive information, that containers need to read
or consume.
● Secrets are another kubernetes abstraction
● Technically they are end up as special volumes
mounted automatically so that your containers can
read their contents.
● Each entry has it’s own path.

So how I Install
kubernetes - for
the first time?
The easiest way is a local cluster - with vagrant managed
vms. But there are several other ways, as documented
here.
1. Install Vagrant on your machine, make sure it works
2. Download the latest release of kubernetes from here
(releases) e.g 1.0.6
3. Unzip the folder to a path of your choice e.g
~kubernetes_home
1. Set the following env variables on your environment
or just export them beforehand.
export KUBERNETES_PROVIDER=vagrant
export VAGRANT_DEFAULT_PROVIDER=virtualbox
export NUM_MINIONS=3
export KUBERNETES_MASTER_MEMORY=1536
export KUBERNETES_MINION_MEMORY=1024
1. Execute the following script
kubernetes_homeclusterkube-up.sh
1. Wait a bit (depending on the number of nodes) to
complete - you are done!!!
Number of
Nodes

So how do I issue
commands to my
cluster?
Kubectl is the command line tool to ‘talk’ to your cluster
master.
● Kubectl is already included in the download archive
export KUBECTL=~kubernetes_homeplatformsxxxamd64
● add KUBECTL to your $PATH, or always navigate to
the above path.

Some kubectl
commands
#Get the state of your cluster
$ kubectl cluster-info
#Get all the nodes of your cluster
$ kubectl get nodes -o wide
#Get info about the pods of your cluster
$ kubectl get pods -o wide
#Get info about the replication controllers of your cluster
$ kubectl get rc -o wide
#Get info about the services of your cluster
$ kubectl get services
#Get full config info about a Service
$ kubectl get service <NAME_OF_SERVICE> -o json
#Get the IP of a Pod
$ kubectl get pod <NAME_OF_POD> -template={{.status.podIP}}
#Delete a Pod
$ kubectl delete pod NAME
#Delete a Service
$ kubectl delete service NAME_OF_THE_SERVICE

Schedule a simple
pod
# official documentation here
# Schedule to start a pod that will contain the latest
# Wildfly Image.
$ kubectl run myjboss --image=jboss/wildfly --port=8080
# Check our created pod(s).
$ kubectl get pods
NAME READY STATUS RESTARTS
AGE
myjboss-jvqob 1/1 Running 0
3m

Schedule a simple
pod - with template
{
"kind": "Pod",
"apiVersion": "v1",
"metadata": {
"name": "anotherJboss",
"labels": {
"app": "myapp"
}
},
"spec": {
"containers": [
{
"name": "backend-core-jboss",
"image": "jboss/wildfly",
"ports": [
{
"containerPort": 8080,
"protocol": "TCP"
}
]
}
]
}
}
$ kubectl create -f ./single-pod.json
Save it as
single-pod.json

Schedule a simple
service - with
template
{
"kind": "Service",
"apiVersion": "v1",
"metadata": {
"name": "test-service"
},
"spec": {
"selector": {
"app": "myapp"
},
"ports": [
{
"protocol": "TCP",
"port": 80,
"targetPort": 8080
}
]
}
}
$ kubectl create -f ./single-service.json
Save it as
single-
service.json

Schedule a
replication
controller - with
template
{
"kind": "ReplicationController",
"apiVersion": "v1",
"metadata": {
"name": "jboss-controller"
},
"spec": {
"replicas": 2,
"selector": {
"app": "backend"
},
"template": {
"metadata": {
"labels": {
"app": "backend"
}
},
"spec": {
"volumes": null,
"containers": [
{
"name": "jboss",
"image": "jboss/wildfly",
"ports": [
{
"containerPort": 8080,
"protocol": "TCP"
}
],
"imagePullPolicy": "IfNotPresent"
}
],
"restartPolicy": "Always",
"dnsPolicy": "ClusterFirst"
}
}
}
}
$ kubectl create -f ./single-rc.json
Save it as
single-rc.json

Who is embracing
Kubernetes?
● Google Container Engine
○ You can spina a cluster of VMs that run
kubernetes in a matter of minutes
● OpenShift version 3
○ The new P.a.a.S of RedHat is based on
Kubernetes!
○ Available as a service and as private
cloud installation.
● Tectonic - by CoreOS
○ a new platform
○ CoreOS+Kubernetes
● Fabric 8
○ O.S platform based on Kubernetes
○ + extra services
● OpenStack Support
○ Murano
● Apache Mesos
○ Kubernetes

Available
documentation
and resources?
Documentation
● Official Documentation here
● Code
● Google Container Engine documentation here
● Google Group - for questions.
● kubernetes.slack.com
● StackOverflow #kubernetes
Books
● Kubernetes Up and Running Oreilly (K.H)
● Scheduling the future at cloud scale - OpenShift
● Kubernetes-book (upcoming)
Articles
● Kubernetes Design Patterns (Arun Gupta)
● Kubernetes Key Concepts (Arun Gupta)
● Recipes for deploying JavaEE Apps (Arun. Gupta)
● Intro to Kubernetes
● Kubernetes for Developers
● The new PAAS
Video
● Tech overview of Kubernetes
● Container Orchestration using CoreOS and Kubernetes (1 ,2
, 3)

Twitter accounts
related with
kubernetes - check
them out!
● #kubernetes (twitter)
● @kubernetesio
Devs
● @jbeda
● @thockin
● @brendandburns
● @TallMartin
● @kelseyhightower
● @kleban
● @asynchio
● @preillyme
● @KitMerker
● @jml3on
Other
● @tectonicStack
● @thenewstack
● @kubeweekly
● @kismatic
● @googlecloud

Well it was just a
start..
This was just a soft introduction, but you
need to understand the basic ideas and then
project them to your current or future project.
Don’t rush into every technology there is a
chance you lose your focus and forget about
the real problem, which is delivering your
app.
The power of kubernetes is that does not
distract you 100% from your application
design and topology.
It does try not to become yet another
technical milestone in order for you to reach
your end goal - which is eventually to deploy
a scalable and easily maintained application.

Thanks for
your time!
Contact :
@javapapo
javapapo@mac.com
javapapo.blogspot.com
lu.linkedin.com/in/javaneze

Introduction to Kubernetes

In this document