Why Would I Want To Containerise My Life? Scott Coulton, Health Direct

Post on 13-Jan-2017

600 views 1 download

Transcript of Why Would I Want To Containerise My Life? Scott Coulton, Health Direct

Why containerise your life ?

The agenda

● So you decided to use containers● The pitfalls of starting to use containers● We have the solution to that problem ● Is this ready for production ?

Have you ever danced with the devil in the pale moonlight ?

The devil does not change, he changes you

Quote “The Joker, Batman 1989”

This quote is so relevant when starting out with containers.

If you want to be successful you have to change the way you think about solutions, your operational processes and CD/CI pipelines. Trying to fit old solution methodology will cause you technical debt.

So you decided to use containers

Where do you start ?

So there is a lot of buzz around containers at the moment. So why should you spend the time and money to move your applications to containers ?

We will answer this question from three different perspectives

● The Dev● Infrastructure● The business

The dev

Why would Dev’s want to use containers ?

● Quick build times● Run code locally that is the same code as production● More efficient unit test cycles ● The container becomes the artifact ● Makes the CD/CI process simple

Infrastructure

So what about the infra guys and girls ?

● The application is abstract from the host operating system

● In true ‘devops’, it’s infra as code● It creates a collaborative environment● Brings down operational overhead● Makes the CD/CI process simple

The business

It’s all about the bottom dollar, so how do we justify the change

● Collaboration breeds efficiency● Increased quality of releases ● Containers are platform agnostic● Language agnostic

Where do we start

Now we see the benefits, where do we start ?

● Pick a small application or service ● Make sure it works ● Gain business trust

The pitfalls of starting to use containers

It can’t be all fairy floss and rainbows

Now we have a few ideas of services or apps to containerise. What real world issues will come up.

● Do we run 1 container per compute resource● Where does our data or state reside● How do we handle failures ● What about the way my data is structured

Do we run 1 container per compute resource ?

If we run containers 1:1 with compute resources what’s the point ?

● So we need a solution to scale

That word ‘scale’ needs to be defined

● Are we talking on the same compute resource● Does this mean across AZ’s● What about completely different datacentres

What about application routing within a compute resource ?

Where does our data or state reside ?

Containers and stateless services are like a mirage in heaven. Unfortunately not all applications can be stateless.

So how do we handle this ?

● Should the data live inside the container● Maybe map a drive to the compute resource● Can we split the compute from the state

How do we handle failures

What happens to application failure within a container ?

● How can we design for failure ● What measures can we put in place to prevent failure● Can we auto heal containers

How do I structure my data

There are so many moving parts to building an image, creating the container and the release process.

● What configuration lives where● How can I test my container● With so many parts what does my source control structure

look like

We have the solution to that problem

This is how we scale

So let’s define scale !!!

For the same for the same compute node we have to following options

● Docker Compose https://docs.docker.com/compose/

This is how we scale

For more complex scaling needs like across multiple compute nodes or AZ’s. We would use a container scheduler.

● Docker Swarm https://docs.docker.com/swarm/● Google’s Kubernetes http://kubernetes.io/● Apache Mesos https://mesos.apache.org/● Hashicorp Nomad https://www.nomadproject.io/

Our data is safe

If we need to give our state to our container applications.

We have options depending on our design choices

● Kubernetes http://kubernetes.io/docs/user-guide/volumes/● Flocker https://clusterhq.com/flocker/introduction/

This is how we protect ourselves from failure

Service discovery is the backbone to our container schedulers. Service discovery dynamically updates the locations of our containers as they move between compute nodes.

● Consul https://www.consul.io/● Etcd https://coreos.com/etcd/

How is my data structured

Dockerfile

● Base configurations for the container● Arguments for runtime● Live in it’s own git repo

Run time

● Volume, CPU, Mem constraints ● Runtime args● Name of container

How is my data structured

CD/CI

● Unit test should live with the Dockerfile

Unit test can be written in various languages.

● Golang ● Ruby● Python

Is this production ready

Let’s tie it all together

Kubernetes

Let’s tie it all together

Docker Swarm with Flocker

Let’s tie it all together

Mesos

Questions

About mehttps://forge.puppet.com/scottyc

https://github.com/scotty-c

https://hub.docker.com/u/scottyc/

@scottcoulton scottyc