CI/CD

WITH
KUBERNETES
The New Stack
CI/CD with Kubernetes
Alex Williams, Founder & Editor-in-Chief

Core Team:
Bailey Math, AV Engineer
Benjamin Ball, Marketing Director
Gabriel H. Dinh, Executive Producer
Judy Williams, Copy Editor
Kiran Oliver, Podcast Producer
Lawrence Hecht, Research Director
Libby Clark, Editorial Director
Norris Deajon, AV Engineer

© 2018 The New Stack. All rights reserved.

20181112
TABLE OF CONTENTS
Introduction .................................................................................................................................. 4
Sponsors ........................................................................................................................................ 7
Contributors .................................................................................................................................. 8

CI/CD WITH KUBERNETES

DevOps Patterns ........................................................................................................................... 9

KubeCon + CloudNativeCon: The Best CI/CD Tool For Kubernetes Doesn’t Exist ........39
Cloud-Native Application Patterns .........................................................................................40
Aqua Security: Improve Security with Automated Image Scanning Through CI/CD ....61
Continuous Delivery with Spinnaker ......................................................................................62
Google Cloud: A New Approach to DevOps With Spinnaker on Kubernetes .................88
Monitoring in the Cloud-Native Era ........................................................................................89
CloudBees: CI/CD with Jenkins X and Kubernetes ...........................................................115

Closing ........................................................................................................................................116
Disclosure ..................................................................................................................................118

CI/CD WITH KUBERNETES 3

INTRODUCTION
Kubernetes is the cloud orchestrator of choice. Its core is like a hive:
orchestrating containers, scheduling, serving as a declarative
infrastructure on self-healing clusters. With its capabilities growing at
such a pace, Kubernetes’ ability to scale forces questions about how an
organization manages its own teams and adopts DevOps practices.
Historically, continuous integration has offered a way for DevOps teams
to get applications into production, but continuous delivery is now a
matter of increasing importance. How to achieve continuous delivery will
largely depend on the use of distributed architectures that manage
services on sophisticated and fast infrastructure that use compute,
networking and storage for continuous, on-demand services. Developers
will consume services as voraciously as they can to achieve the most out
of them. They will try new approaches for development, deployment and,
increasingly, the management of microservices and their overall health
and behavior.

Kubernetes is similar to other large-scope, cloud software projects that are

so complex that their value is only determined when they are put into
practice. The container orchestration technology is increasingly being
used as a platform for application deployment defined by the combined
forces of DevOps, continuous delivery and observability. When employed
together, these three forces deliver applications faster, more efficiently
and closer to what customers want and demand. Teams start by building
applications as a set of microservices in a container-based, cloud-native
architecture. But DevOps practices are what truly transform the
application architectures of an organization; they are the basis for all of
the patterns and practices that make applications run on Kubernetes. And
DevOps transformation only comes with aligning an organization’s values
with the ways it develops application architectures.

CI/CD WITH KUBERNETES 4

INTRODUCTION

In this newly optimized means to cloud-native transformation, Kubernetes

is the enabler — it’s not a complete solution. Your organization must
implement the tools and practices best suited to your own business
needs and structure in order to realize the full promise of this open source
platform. The Kubernetes project documentation itself says so:

Kubernetes “does not deploy source code and does not build your
application. Continuous Integration, Delivery, and Deployment (CI/CD)
workflows are determined by organization cultures and preferences
as well as technical requirements.”

This ebook, the third and final in The New Stack’s Kubernetes ecosystem
series, lays the foundation for understanding and building your team’s
practices and pipelines for delivering — and continuously improving —
applications on Kubernetes. How is that done? It’s not a set of rules. It’s a
set of practices that flow into the organization and affect how application
architectures are developed. This is DevOps, and its currents are now
deep inside organizations with modern application architectures,
manifested through continuous delivery.

Section Summaries
• Section 1: DevOps Patterns by Rob Scott of ReactiveOps, explores
the history of DevOps, how it is affecting cloud-native architectures
and how Kubernetes is again transforming DevOps. This section traces
the history of Docker and container packaging to the emergence of
Kubernetes and how it is affecting application development and
deployment.

• Section 2: Cloud-Native Application Patterns is written by

Janakiram MSV, principal analyst at Janakiram & Associates. It reviews
how Kubernetes manages resource allocation automatically, according

CI/CD WITH KUBERNETES 5

INTRODUCTION

to policies set out by DevOps teams. It details key cloud-native

attributes, and maps workload types to Kubernetes primitives.

• Section 3: Continuous Delivery with Spinnaker by Craig Martin,

senior vice president of engineering at Kenzan, analyzes how
continuous delivery with cloud-native technologies requires deeper
understanding of DevOps practices and how that affects the way
organizations deploy and manage microservices. Spinnaker is given
special attention as an emerging CD tool that is itself a cloud-native,
microservices-based application.

• Section 4: Monitoring in the Cloud-Native Era by a team of

engineers from Container Solutions, explains how the increasing
complexity of microservices is putting greater emphasis on the need
for combining traditional monitoring practices to gain better
observability. They define observability for scaled-out applications
running on containers in an orchestrated environment, with a specific
focus on Prometheus as an emerging management tool.

While the book ends with a focus on observability, it’s increasingly clear
that cloud-native monitoring is not an endpoint in the development life
cycle of an application. It is, instead, the process of granular data
collection and analysis that defines patterns and informs developers and
operations teams from start to finish, in a continual cycle of improvement
and delivery. Similarly, this book is intended as a reference throughout
the planning, development, release, manage and improvement cycle.

CI/CD WITH KUBERNETES 6

 SPONSORS
We are grateful for the support of our ebook foundation sponsor:

And our sponsors for this ebook:

CI/CD WITH KUBERNETES 7

CONTRIBUTORS
Rob Scott works out of his home in Chattanooga as a Site
Reliability Engineer for ReactiveOps. He helps build and
maintain highly scalable, Kubernetes-based infrastructure
for multiple clients. He’s been working with Kubernetes since
2016, contributing to the official documentation along the way. When he’s
not building world-class infrastructure, Rob likes spending time with his
family, exploring the outdoors, and giving talks on all things Kubernetes.

Janakiram MSV is the Principal Analyst at Janakiram &

Associates and an adjunct faculty member at the
International Institute of Information Technology. He is also
a Google Qualified Cloud Developer; an Amazon Certified
Solution Architect, Developer, and SysOps Administrator; a Microsoft
Certified Azure Professional; and one of the first Certified Kubernetes
Administrators and Application Developers. His previous experience
includes Microsoft, AWS, Gigaom Research, and Alcatel-Lucent.

Craig Martin is Kenzan’s senior vice president of engineering,

where he helps to lead the technical direction of the company
ensuring that new and emerging technologies are explored
and adopted into the strategic vision. Recently, Craig has
been focusing on helping companies make a digital transformation by
building large-scale microservices applications. Prior to Kenzan, Craig was
director of engineering at Flatiron Solutions.

Ian Crosby, Maarten Hoogendoorn, Thijs Schnitger and

Etienne Tremel are engineers and experts in application
deployment on Kubernetes for Container Solutions, a
consulting organization that provides support for clients who are doing
cloud migrations.

CI/CD WITH KUBERNETES 8

DEVOPS PATTERNS
by ROB SCOTT

evOps practices run deep in modern application architectures.

D DevOps practices have helped create a space for developers and

engineers to build new ways to optimize resources and scale out
application architectures through continuous delivery practices. Cloud-
native technologies use the efficiency of containers to make
microservices architectures that are more useful and adaptive than
composed or monolithic environments. Organizations are turning to
DevOps principles as they build cloud-native, microservices-based
applications. The combination of DevOps and cloud-native architectures
is helping organizations meet their business objectives by fostering a
streamlined, lean product development process that can adapt quickly to
market changes.

Cloud-native applications are based on a set of loosely coupled

components, or microservices, that run for the most part on containers,
and are managed with orchestration engines such as Kubernetes.
However, they are also beginning to run as a set of discrete functions in
serverless architectures. Services or functions are defined by developer
and engineering teams, then continuously built, rebuilt and improved by
increasingly cross-functional teams. Operations are now less focused on

CI/CD WITH KUBERNETES 9

DEVOPS PATTERNS

the infrastructure and more on the applications that run light workloads.
The combined effect is a shaping of automated processes that yield
better efficiencies.

In fact, some would argue that an application isn’t truly cloud native
unless it has DevOps practices behind it, as cloud-native architectures are
built for web-scale computing. DevOps professionals are required to build,
deploy and manage declarative infrastructure that is secure, resilient and
high performing. Delivering these requirements just isn’t feasible with a
traditional siloed approach.

As the de facto platform for cloud-native applications, Kubernetes not

only lies at the center of this transformation, but also enables it by
abstracting away the details of the underlying compute, storage and
networking resources. The open source software provides a consistent
platform on which containerized applications can run, regardless of their
individual runtime requirements. With Kubernetes, your servers can be
dumb — they don’t care what they’re running. Instead of running a
specific application on a specific server, multiple applications can be
distributed across the same set of servers. Kubernetes simplifies
application updates, enabling teams to deliver applications and features
into users’ hands quickly.

In order to find success with DevOps, however, a business must be

intentional in its decision to build a cloud-native application. The
organizational transformation required to put DevOps into practice will
happen only if a business team is willing to invest in DevOps practices —
transformation comes with the alignment of the product team in the
development of the application. Together, these teams create the
environment needed to continually refine technical development into
lean, streamlined workflows that reflect continuous delivery processes
built on DevOps principles.

CI/CD WITH KUBERNETES 10

DEVOPS PATTERNS

For organizations using container orchestration technologies, product

direction is defined by developing a microservices architecture. This is
possible only when the organization understands how DevOps and
continuous development processes enable the creation of applications
that end users truly find useful.

Therein lies the challenge: You must make sure your organization is
prepared to transform the way all members of the product team work.
Ultimately, DevOps is a story about why you want to do streamlined, lean
product development in the first place — the same reason that you’re
moving to a microservices architecture on top of Kubernetes.

Our author for this chapter is Rob Scott, a site reliability engineer at
ReactiveOps. Scott is an expert in DevOps practices, applying techniques
from his learnings to help customers run services that can scale on
Kubernetes architectures. His expertise in building scaled-out architectures
stems from years of experience that has given him witness to:

• How containers brought developers and operators together into the

field of DevOps.

• The role a container orchestration tool like Kubernetes plays in the

container ecosystem.

• How Kubernetes resulted in a revolutionary transformation of the

entire DevOps ecosystem — which is ultimately transforming
businesses.

Traditional DevOps patterns before containers required different

processes and workflows. Container technologies are built with a DevOps
perspective. The abstraction containers offer is having an effect on how
we view DevOps, as traditional architecture development changes with
the advent of microservices. It means following best practices for running

CI/CD WITH KUBERNETES 11

DEVOPS PATTERNS

containers on Kubernetes, and the extension of DevOps into GitOps and

SecOps practices.

The Evolution of DevOps and CI/CD

Patterns
A Brief History of DevOps
DevOps was born roughly 10 years ago, though organizations have shown
considerably more interest in recent years. Half of organizations surveyed
implemented DevOps practices in 2017, according to Forrester Research,
which has declared 2018 “The Year of Enterprise DevOps.” Although
DevOps is a broad concept, the underlying idea involves development and
operations teams working more closely together.

Traditionally, the speed with which software was developed and deployed
didn’t allow a lot of time for collaboration between engineers and operations
staff, who worked on separate teams. Many organizations had embraced
lean product development practices and were under constant pressure to
release software quickly. Developers would build out their applications, and
the operations team would deploy them. Any conflict between the two
teams resulted from a core disconnect — the operations team was
unfamiliar with the applications being deployed, and the development team
was unfamiliar with how the applications were being deployed.

As a result, application developers sometimes found that their platform

wasn’t configured in a way that best met their needs. And because the
operations team didn’t always understand software and feature
requirements, at times they over-provisioned or under-provisioned
resources. What happened next is no mystery: Operations teams were
held responsible for engineering decisions that negatively impacted
application performance and reliability. Worse, poor outcomes impacted
the organization’s bottom line.

CI/CD WITH KUBERNETES 12

DEVOPS PATTERNS

A key concept of DevOps involved bringing these teams together. As

development and operations teams started to collaborate more
frequently, it became clear that automation would speed up deployments
and reduce operational risk. With these teams working closely together,
some powerful DevOps tooling was built. These tools automated what
had been repetitive, manual and error-prone processes with code.

Eventually these development and operations teams started to form their

own “DevOps” teams that combined engineers from development and
operations backgrounds. In these new teams, operations engineers
gained development experience, and developers gained exposure to the
behind-the-scenes ways that applications run. As this new specialization
continues to evolve, next-generation DevOps tooling is being designed
and built that will continue to transform the industry. Increased
collaboration is still necessary for improved efficiencies and business
outcomes, but further advantages of DevOps adoption are emerging.
Declarative environments made possible by cloud-native architectures
and managed through continuous delivery pipelines have lessened
reliance on collaboration and shifted the focus toward application
programming interface (API) calls and automation.

The Evolution of CI/CD Workflows

There are numerous models for developing iterative software, as well as
an infinite number of continuous integration/continuous delivery (CI/CD)
practices. While CI/CD processes aren’t new to the scene, they were more
complex at the start. Now, continuous delivery has come to the fore as the
next frontier for improved efficiencies as more organizations migrate to
microservices and container-based architectures. A whole new set of tools
and best practices are emerging that allow for increasingly automated
and precise deployments, using strategies such as red/black deployments
and automated canary analysis (ACA). Chapter 3 has more detail.

CI/CD WITH KUBERNETES 13

DEVOPS PATTERNS

Before the idea of immutable infrastructure gained popularity, servers

were generally highly specialized and difficult to replace. Each server
would have a specific purpose and would have been manually tuned to
achieve that purpose. Tools like Chef and Puppet popularized the notion
of writing reproducible code that could be used to build and tune these
servers. Servers were still changing frequently, but now code was
committed into version control. Changes to servers became simpler to
track and recreate. These tools also started to simplify integration with
CI/CD workflows. They enabled a standard way to pull in new code and
restart an application across all servers. Of course, there was always a
chance that the latest application could break, resulting in a situation
that could be difficult to recover from quickly.

With that in mind, the industry started to move toward a pattern that
avoided making changes to existing servers: immutable infrastructure.
Virtual machines combined with cloud infrastructure to dramatically
simplify creating new servers for each application update. In this
workflow, a CI/CD pipeline would create machine images that included
the application, dependencies and base operating system (OS). These
machine images could then be used to create identical, immutable
servers to run the application. They could also be tested in a quality
assurance (QA) environment before being deployed to production.

The ability to test every bit of the image before it reached production
resulted in an incredible improvement in reliability for QA teams.
Unfortunately, the process of creating new machine images and then
running a whole new set of servers with them was also rather slow.

It was around this time that Docker started to gain popularity. Based on
Linux kernel features, cgroups and namespaces, Docker is an open source
project that automates the development, deployment and running of
applications inside isolated containers. Docker offered a lot of the same

CI/CD WITH KUBERNETES 14

DEVOPS PATTERNS

advantages as machine images, but it did so with a much more lightweight

image format. Instead of including the whole base operating system,
Docker images simply included the application and its dependencies. This
process still provided the reliability advantages described earlier, but
came with some substantial improvements in speed. Docker images were
much faster to build, faster to pull in and faster to start up. Instead of
creating new servers for each new deployment, new Docker containers
were created that could run on the same servers.

With the lightweight approach Docker provided, CI/CD workflows really

started to take off. For example, each new commit to your Git repository
could have a corresponding Docker image built. Each Git commit could
trigger a multi-step, customizable build process that includes vulnerability
scanning for container images. Cached images could then be used for
subsequent builds, speeding up the build process in future iterations. One
of the most recent improvements in these workflows has come with
container orchestration tools like Kubernetes. These tools have
dramatically simplified deployment of application updates with
containers. In addition, they have had transformative effects on resource
utilization. Whereas before you might have run a single application on a
server, with container orchestration multiple containers with vastly
different workloads can run on the same server. With Kubernetes, CI/CD is
undergoing yet another evolution that has tremendous implications for
the business efficiencies gained through DevOps.

Modern DevOps Practices

Docker was the first container technology to gain broad popularity,
though alternatives exist and are standardized by the Open Container
Initiative (OCI). Containers allow developers to bundle up an application
with all of the dependencies it needs to run and package and ship it in a

CI/CD WITH KUBERNETES 15

DEVOPS PATTERNS

single package. Before, each server would need to have all the OS-level
dependencies to run a Ruby or Java application. The container changes
that. It’s a thin wrapper — single package — containing everything you
need to run an application. Let’s explore how modern DevOps practices
reflect the core value of containers.

Containers Bring Portability

Docker is both a daemon — a process running in the background — and
a client command. It’s like a virtual machine, but it’s different in
important ways. First, there’s less duplication. With each extra virtual
machine (VM) you run, you duplicate the virtualization of central
processing units (CPUs) and memory and quickly run out of local
resources. Docker is great at setting up a local development environment
because it easily adds the running process without duplicating the
virtualized resource. Second, it’s more modular. Docker makes it easy to
run multiple versions or instances of the same program without
configuration headaches and port collisions.

Thus, instead of a single VM or multiple VMs, you can link each

individual application and supporting service into a single unit and
horizontally scale individual services without the overhead of a VM. And
it does it all with a single descriptive Dockerfile syntax, improving the
development experience, speeding software delivery and boosting
performance. And because Docker is based on open source technology,
anyone can contribute to its development to build out features that
aren’t yet available.

With Docker, developers can focus on writing code without worrying

about the system on which their code will run. Applications become
truly portable. You can repeatedly run your application on any other
machine running Docker with confidence. For operations staff, Docker is
lightweight, easily allowing the running and management of applications

CI/CD WITH KUBERNETES 16

DEVOPS PATTERNS

with different requirements side by side in isolated containers. This

flexibility can increase resource utilization per server and may reduce
the number of systems needed due to lower overhead, which in turn
reduces cost.

Containers Further Blur the Lines Between

Operations and Development
Containers represent a significant shift in the traditional relationship
between development and operations teams. Specifications for building a
container have become remarkably straightforward to write, and this has
increasingly led to development teams writing these specifications. As a
result, development and operations teams work even more closely
together to deploy these containers.

The popularity of containers has led to significant improvements for CI/CD

pipelines. In many cases, these pipelines can be configured with some
simple YAML files. This pipeline configuration generally also lives in the
same repository as the application code and container specification. This
is a big change from the traditional approach in which code to build and
deploy applications is stored in a separate repository and entirely
managed by operations teams.

With this move to a simplified build and deployment configuration living

alongside application code, developers are becoming increasingly
involved in processes that were previously managed entirely by
operations teams.

Initial Challenges with Containers

Though containers are now widely adopted by most organizations, there
have historically been three basic challenges that prevented organizations
from making the switch. First, it takes a mindshift to translate a current
development solution into a containerized development solution. For

CI/CD WITH KUBERNETES 17

DEVOPS PATTERNS

example, if you think of a container as a virtual machine, you might want

to cram a lot of things in it, such as services, monitoring software and your
application. Doing so could lead to a situation commonly called “the
matrix of hell.” Don’t put many things into a single container image;
instead, use many containers to achieve the full stack. In other words, you
can keep your supporting service containers separate from your
application container, and they can all be running on different operating
systems and versions while being linked together.

Next, the way containers worked and behaved was largely undefined
when Docker first popularized the technology. Many organizations
wondered if containerization would really pay off, and some remain
skeptical.

And while an engineering team might have extensive experience in

implementing VM-based approaches, it might not have a conceptual
understanding of how containers themselves work and behave. A key
principle of container technology is that an image never changes, giving
you an immutable starting point each time you run the image and the
confidence that it will do the same thing each time you run it, no matter
where you run it. To make changes, you create a new image and replace
the current image with the newer version. This can be a challenging
concept to embrace, until you see it in action.

These challenges have been largely overcome, however, as adoption

spread and organizations began to realize the benefits of containers — or
see their competitors realize them.

DevOps with Containers

Docker runs processes in isolated containers — processes that run on a
local or remote host. When you execute the command docker run, the
container process that runs is isolated: It has its own file system, its own

CI/CD WITH KUBERNETES 18

DEVOPS PATTERNS

networking and its own isolated process tree separate from the host.

Essentially it works like this: A container image is a collection of file system

layers and amounts to a fixed starting point. When you run an image, it
creates a container. This container-based deployment capability is
consistent from development machine to staging to QA to production —
all the way through. When you have your application in a container, you
can be sure that the code you’re testing locally is exactly the same build
artifact that goes into production. There are no changes in application
runtime environments.

You once had specialized servers and were worried about them falling
apart and having to replace them. Now servers are easily replaceable and
can be scaled up or down — all your server needs to be able to do is run
the container. It no longer matters which server is running your container,
or whether that server is on premises, in the public cloud or a hybrid of
both. You don’t need an application server, web server or different
specialized server for every application that’s running. And if you lose a
server, another server can run that same container. You can deploy any
number of applications using the same tools and the same servers.
Compartmentalization, consistency and standardized workflows have
transformed deployments.

Containerization provided significant improvements to application

deployment on each server. Instead of worrying about installing
application dependencies on servers, they were included directly in the
container image. This technology provided the foundation for
transformative orchestration tooling such as Mesos and Kubernetes that
would simplify deploying containers at scale.

Containers Evolved DevOps and the Profession

Developers were always connected to operations, whether they wanted to

CI/CD WITH KUBERNETES 19

DEVOPS PATTERNS

be or not. If their application wasn’t up and running, they were brought in

to resolve problems. Google was one of the first organizations to
introduce the concept of site reliability engineering, in which talented
developers also have skill in the operations world. The book, Site
Reliability Engineering: How Google Runs Production Systems (2016),
describes best practices for building, deploying, monitoring and
maintaining some of the largest software systems in the world, using a
division of 50 percent development work and 50 percent operational
work. This concept has taken off over the past two to three years as more
organizations adopt DevOps practices in order to migrate to microservices
and container-based architectures.

What began as two disparate job functions with crossover has now
become its own job function. Operations teams are working with code
bases; developers are working to deploy applications and are getting
farther into the operational system. From an operational perspective,
developers can look backward and read the CI file and understand the
deployment processes. You can even look at Dockerfiles and see all the
dependencies your application needs. It’s simpler from an operational
perspective to understand the code base.

So who exactly is this DevOps engineer? It’s interesting to see how a

specialization in DevOps has evolved. Some DevOps team members have
an operational background, while others have a strong software
development background. The thing that connects these diverse
backgrounds is a desire for and an appreciation of system automation.
Operations engineers gain development experience, and developers gain
exposure to the behind-the-scenes ways the applications run. As this new
specialization continues to evolve, next-generation DevOps tooling is
continually being designed and built to accommodate changing roles and
architectures in containerized infrastructure.

CI/CD WITH KUBERNETES 20

DEVOPS PATTERNS

Running Containers with Kubernetes

In 2015, being able to programmatically “schedule” workloads into an
application-agnostic infrastructure was the way forward. Today, the best
practice is to migrate to some form of container orchestration.

Many organizations still use Docker to package up their applications,

citing its consistency. Docker was a great step in the right direction, but it
was a means to an end. In fact, the way containers were deployed wasn’t
transformative until orchestration tooling came about. Just as many
container technologies existed before Docker, many container
orchestration technologies preceded Kubernetes. One of the better
known tools was Apache Mesos, a tool built by Twitter. Mesos does
powerful things with regards to container orchestration, but it was —
and still can be — difficult to set up and use. Mesos is still used by
enterprises with scale and size, and it’s an excellent tool for the right use
case and scale.

Today, organizations are increasingly choosing to use Kubernetes instead

of other orchestration tools. More and more, organizations are recognizing
that containers offer a better solution than the more traditional tooling
they had been using, and that Kubernetes is the best container
deployment and management solution available. Let’s examine these
ideas further.

Introduction to Kubernetes
Kubernetes is a powerful, next generation, open source platform for
automating the deployment, scaling and management of application
containers across clusters of hosts. It can run any workload. Kubernetes
provides exceptional developer user experience (UX), and the rate of
innovation is phenomenal. From the start, Kubernetes’ infrastructure
promised to enable organizations to deploy applications rapidly at scale

CI/CD WITH KUBERNETES 21

DEVOPS PATTERNS

and roll out new features easily while using only the resources needed.
With Kubernetes, organizations can have their own Heroku running in
their own public cloud or on-premises environment.

First released by Google in 2014, Kubernetes looked promising from the

outset. Everyone wanted zero-downtime deployments, a fully
automated deployment pipeline, auto scaling, monitoring, alerting and
logging. Back then, however, setting up a Kubernetes cluster was hard.
At the time, Kubernetes was essentially a do-it-yourself project with lots
of manual steps. Many complicated decisions were — and are —
involved: You have to generate certificates, spin up VMs with the correct
roles and permissions, get packages onto those VMs and then build
configuration files with cloud provider settings, IP addresses, DNS
entries, etc. Add to that the fact that at first not everything worked as
expected, and it’s no surprise that many in the industry were hesitant to
use Kubernetes.

Kubernetes 1.2, released in April 2016, included features geared more

toward general-purpose usage. It was accurately touted as the next big
thing. From the start, this groundbreaking open source project was an
elegant, structured, real-world solution to containerization at scale that
solves key challenges that other technologies didn’t address. Kubernetes
includes smart architectural decisions that facilitate the structuring of
applications within containerization. Many things remain in your control.
For example, you can decide how to set up, maintain and monitor
different Kubernetes clusters, as well as how to integrate those clusters
into the rest of your cloud-based infrastructure.

Kubernetes is backed by major industry players, including Amazon,

Google, Microsoft and Red Hat. With over 14,000 individual contributors
and ever increasing momentum, this project is here to stay.

CI/CD WITH KUBERNETES 22

DEVOPS PATTERNS

In years past, think about how often development teams wanted visibility
into operations deployments. Developers and operations teams have
always been nervous about deployments because maintenance windows
had a tendency to expand, causing downtime. Operations teams, in turn,
have traditionally guarded their territory so no one would interfere with
their ability to get their job done.

Then containerization and Kubernetes came along, and software

engineers wanted to learn about it and use it. It’s revolutionary. It’s not a
traditional operational paradigm. It’s software driven, and it lends itself
well to tooling and automation. Kubernetes enables engineers to focus on
mission-driven coding, not on providing desktop support. At the same
time, it takes engineers into the world of operations, giving development
and operations teams a clear window into each other’s worlds.

Kubernetes Is a Game Changer

Kubernetes is changing the game, not only in the way the work is done,
but in who is being drawn to the field. Kubernetes has evolved into the
standard for container orchestration. And the impacts on the industry
have been massive.

In the past, servers were custom-built to run a specific application; if a

server went down, you had to figure out how to rebuild it. Kubernetes
simplifies the deployment process and improves resource utilization. As
we stated previously, with Kubernetes your servers can be dumb — they
don’t care what they’re running. Instead of running a specific application
on a specific server, you can stack resources. A web server and a
backend processing server might both run in Docker containers, for
example. Let’s say you have three servers, and five applications can run
on each one. If one server goes down, you have redundancy because
everything filters across.

CI/CD WITH KUBERNETES 23

DEVOPS PATTERNS

Some benefits of Kubernetes:

• Independently Deployable Services: You can develop applications

as a suite of independently deployable, modular services.
Infrastructure code can be built with Kubernetes for almost any
software stack, so organizations can create repeatable processes that
are scalable across many different applications.

• Deployment Frequency: In the DevOps world, the entire team

shares the same business goals and remains accountable for building
and running applications that meet expectations. Deploying shorter
units of work more frequently minimizes the amount of code you have
to sift through to diagnose problems. The speed and simplicity of
Kubernetes deployments enables teams to deploy frequent
application updates.

• Resiliency: A core goal of DevOps teams is to achieve greater system

availability through automation. With that in mind, Kubernetes is
designed to recover from failure automatically. For example, if an
application dies, Kubernetes will automatically restart it.

• Usability: Kubernetes has a well-documented API with simple,

straightforward configuration that offers phenomenal developer UX.
Together, DevOps practices and Kubernetes also allow businesses to
deliver applications and features into users’ hands quickly, which
translates into more competitive products and more revenue
opportunities.

Kubernetes Simplifies the Orchestration of

Your Application
In addition to improving traditional DevOps processes, along with the
speed, efficiency and resiliency commonly recognized as benefits of
DevOps, Kubernetes solves new problems that arise with container and

CI/CD WITH KUBERNETES 24

 DEVOPS PATTERNS

microservices-based application architectures. Said another way,

Kubernetes reinforces DevOps goals while also enabling new workflows
that arise with microservices architectures.

Powerful Building Blocks

Kubernetes uses pods as the fundamental unit of deployment. Pods
represent a group of one or more containers that use the same storage
and network. Although pods are often used to run only a single container,
they have been used in some creative ways, including as a means to build
a service mesh.

A common use of multiple containers in a single pod follows a sidecar

pattern. With this pattern, a container would run beside your core
application to provide some additional value. This is commonly used for

FIG 1.1: With Kubernetes, pods are distributed across servers with load balancing
and routing built in. Distributing application workloads in this way can dramatically
increase resource utilization.

The Evolution of Application Infrastructure

Before Containers With Containers With Kubernetes

API URL Web URL API URL Web URL API URL Web URL

Load Balancer Load Balancer Load Balancer Load Balancer Load Balancer

API API Web Web API API Web Web Kubernetes Ingress Controller
Code Code Code Code Container Container Container Container

Libraries Libraries Libraries Libraries

Server Server Server Server Server Server Server Server

API Pod API Pod

Web Pod Web Pod

Database Database Server Server

CI/CD WITH KUBERNETES Database 25

Source: ReactiveOps © 2018

DEVOPS PATTERNS

proxying requests, or even handling authentication.

With these powerful building blocks, it becomes quite straightforward to

map services that may have been running in a virtual machine before
containerization, into multiple containers running in the same pod.

Simplified Service Discovery

In one monolithic application, different services each have their own
purpose, but self-containment facilitates communication. In a
microservices architecture, microservices need to talk to each other —
your user service needs to talk to your post service and address service
and so on. Figuring out how these services can communicate simply and
consistently is no easy feat.

With Kubernetes, a DevOps engineer defines a service — for example, a

user service. Anything running in that same Kubernetes namespace can
send a request to that service, and Kubernetes figures out how to route
the request for you, making microservices easier to manage.

Centralized, Easily Readable Configuration

Kubernetes operates on a declarative model: You describe a desired state,
and Kubernetes will try to achieve that state. Kubernetes has easily
readable YAML files used to describe the state you want to achieve. With
Kubernetes YAML configuration, you can define anything from an
application load balancer to a group of pods to run your application. A
deployment configuration might have three replicas of one of your
applications’ Docker containers and two different environment variables.
This easy-to-read configuration is likely stored in a Git repository, so you
can see any time that the configuration changes. Before Kubernetes, it
was hard to know what was actually happening with interconnected
systems across servers.

In addition to configuring the application containers running in your

CI/CD WITH KUBERNETES 26

DEVOPS PATTERNS

cluster, or the endpoints that can be used to access them, Kubernetes can
help with configuration management. Kubernetes has a concept called
ConfigMap where you can define environment variables and configuration
files for your application. Similarly, objects called secrets contain sensitive
information and help define how your application will run. Secrets work
much like ConfigMaps, but are more obscure and less visible to end users.
Chapter 2 explores all of this in detail.

Real-Time Source of Truth

Manual and scripted releases used to be extremely stressful. You had one
chance to get it right. With the built-in deployment power of Kubernetes,
anybody can deploy and check on delivery status using Kubernetes’
unlimited deployment history: kubectl rollout history.

The Kubernetes API provides a real-time source of truth about

deployment status. Any developer with access to the cluster can quickly
find out what’s happening with the delivery or see all commands issued.
This permanent system audit log is kept in one place for security and
historical purposes. You can easily learn about previous deployments, see
the delta between deployments or roll back to any of the listed versions.

Simple Health Check Capability

This is a huge deal in your application’s life cycle, especially during the
deployment phase. In the past, applications often had no automatic
restart if they crashed; instead, someone got paged in the middle of the
night and had to restart them. Kubernetes, on the other hand, has
automatic health checks, and if an application fails to respond for any
reason, including running out of memory or just locking up, Kubernetes
automatically restarts it.

To clarify, Kubernetes checks that your application is running, but it

doesn’t know how to check that it’s running correctly. However,

CI/CD WITH KUBERNETES 27

DEVOPS PATTERNS

Kubernetes makes it simple to set up health checks for your application.

You can check the application’s health in two ways:

1. Using a liveness probe that checks if an application goes from a

healthy state to an unhealthy state. If it makes that transition, it will try
to restart your application for you.

2. Using a readiness probe that checks if an application is ready to

accept traffic. It won’t get rid of previously working containers until the
new containers are healthy. Basically, a readiness probe is a last line of
defense that prevents a broken container from seeing the light of day.

Both probes are useful tools, and Kubernetes makes them easy to
configure.

In addition, rollbacks are rare if you have a properly configured readiness

probe. If all the health checks fail, a single one-line command will roll back
that deployment for you and get you back to a stable state. It’s not
commonly used, but it’s there if you need it.

Rolling Updates and Native Rollback

To build further off the idea of a real-time source of truth and health check
capabilities, another key feature of Kubernetes is rolling updates with the
aforementioned native rollback. Deployments can and should be frequent
without fear of hitting a point of no return. Before Kubernetes, if you
wanted to deploy something, a common deployment pattern involved the
server pulling in the newest application code and restarting your
application. The process was risky because some features weren’t
backwards compatible — if something went wrong during the deployment,
the software became unavailable. For example, if the server found new
code, it would pull in those updates and try to restart the application with
the new code. If something failed in that pipeline, the application was likely
dead. The rollback procedure was anything but straightforward.

CI/CD WITH KUBERNETES 28

DEVOPS PATTERNS

These workflows were problematic until Kubernetes. Kubernetes solves

this problem with a deployment rollback capability that eliminates large
maintenance windows and anxiety about downtime. Since Kubernetes
1.2, the deployment object is a declarative manifest containing everything
that’s being delivered, including the number of replicas being deployed
and the version of the software image. These items are abstracted and
contained within a deployment declaration. Such manifest-based
deployments have spurred new CD workflows and are an evolving best
practice with Kubernetes.

Before Kubernetes shuts down existing application containers, it will start

spinning up new ones. Only when the new ones are up and running
correctly does it get rid of the old, stable release. Let’s say Kubernetes
doesn’t catch a failed deployment — the app is running, but it’s in some
sort of error state that Kubernetes doesn’t detect. In this case, DevOps
engineers can use a simple Kubernetes command to undo that
deployment. Furthermore, you can configure it to store as few as two
changes or as many revisions as you want, and you can go back to the last
deployment or many deployments earlier, all with an automated, simple
Kubernetes command. This entire concept was a game-changer. Other
orchestration frameworks don’t come close to handling this process in as
seamless and logical a way as Kubernetes.

Simplified Monitoring
While on the surface it might seem that monitoring Kubernetes would be
quite complex, there has been a lot of development in this space.
Although Kubernetes and containers add some levels of complexity to
your infrastructure, they also ensure that all your applications are running
in consistent pods and deployments. This consistency enables monitoring
tools to be simpler in many ways.

Prometheus is an example of an open source monitoring tool that has

CI/CD WITH KUBERNETES 29

DEVOPS PATTERNS

become very popular in the cloud-native ecosystem. This tool provides

advanced monitoring and alerting capabilities, with excellent
Kubernetes integrations.

When monitoring Kubernetes, there are a few key components to watch:

Kubernetes nodes (servers); Kubernetes system deployments, such as
DNS or networking; and, of course, your application itself. There are
many monitoring tools that will simplify monitoring each of these
components.

Kubernetes and CI/CD Complement Each Other

Setting up a CI/CD pipeline on top of Kubernetes will speed up your
release life cycle — enabling you to release multiple times a day — and
enable nimble teams to iterate quickly. With Kubernetes, builds become a
lot faster. Instead of spinning up entirely new servers, your build process is
quick, lightweight and straightforward.

Development speeds up when you don’t have to worry about building and
deploying a monolith in order to update everything. By splitting a
monolith into microservices, you can instead update pieces — this service
or that. Part of a good CI/CD workflow should also include a strong test
suite. While not unique to Kubernetes, a containerized approach can make
tests more straightforward to run. If your application tests depend on
other services, you can run your tests against those containers, simplifying
the testing process. A one-line command is usually all you need to update
a Kubernetes deployment.

In a CI/CD workflow, ideally you run many tests. If those tests fail, your
image will never be built, and you’ll never deploy that container.

However, if testing fails to uncover issues, Kubernetes offers better

protection because Kubernetes simplifies zero-downtime deployment.
For a long time, deployments meant downtime. Operations teams used to

CI/CD WITH KUBERNETES 30

 CI/CD Pipeline Workflow with Kubernetes
DEVOPS PATTERNS

DEVELOPER CI SERVER KUBERNETES

Commit code, Build new Create Restart

push to Git Docker image Pull new new pod new pod
Docker image

Check Let old pod

Run tests pod health continue running
GIT REPO

CI Server notices Push new DOCKER New pod is New pod is

new code in Docker image REPOSITORY healthy not healthy
Git repo & starts
running through
its pipeline.
Update Delete
Kubernetes old pod
deployment Kubernetes receives request
to use new image

Source: ReactiveOps © 2018

FIG 1.2: Before Kubernetes shuts down existing pods, it will start spinning up new
ones. Only when the new ones are up and running correctly does it get rid of the
old, stable release. Such rolling updates and native rollback features are a game-
changer for DevOps.

handle deployment efforts manually or via scripting, a live process that

could take hours, if not all night. Accordingly, one of the fears of CI/CD is
that a deployment will break and a site will go down.

Kubernetes’ zero-downtime deployment capability relieves anxieties

about maintenance windows, making schedule delays and downtime a
thing of the past — and saving money in the process. It also keeps
everyone in the loop while meeting the needs of development, operations
and business teams. The revolutionary Kubernetes deployment object has
built-in features that automate this operations effort.

In particular, the aforementioned tests and health checks can prevent bad
code from reaching production. As part of a rolling update, Kubernetes
spins up separate new pods running your application while the old ones
are still running. When the new pods are healthy, Kubernetes gets rid of

CI/CD WITH KUBERNETES 31

DEVOPS PATTERNS

the old ones. It’s a smart, simple concept, and it’s one less thing you have
to worry about for each application and in your CI/CD workflow.

Complementary Tools
As part of the incredible momentum Kubernetes has seen, a number of
DevOps tools have emerged that are particularly helpful in developing CI/
CD workflows with Kubernetes. Bear in mind that CI/CD tools and
practices are still evolving with the advent of cloud-native deployments on
Kubernetes. No single tool yet offers the perfect solution for managing
cloud-native applications from build through deployment and continuous
delivery. Although there are far too many to mention here, it’s worth
highlighting a few DevOps tools that were purpose-built for cloud-native
applications:

• Draft: This tool from Microsoft targets developer workflows. With a

few simple commands, Draft can containerize and deploy an
application to Kubernetes. The automated containerization of
applications here can be quite powerful. Draft uses best practices for
popular frameworks and languages to build images and Kubernetes
configuration that will work in most cases.

• Helm: Known as the Kubernetes package manager, this framework

simplifies deploying applications to Kubernetes. Deployment
configuration for many popular projects is available in well maintained
“Charts.” This means that helm install prometheus is all that’s
needed to get a project like Prometheus running in your cluster. Helm
can also provide the same kind of conveniences when deploying your
own custom applications.

• Skaffold: Similar to Draft, this is a new tool from Google that enables
exciting new development workflows. In addition to supporting more
standard CI/CD workflows, this has an option to build and deploy code

CI/CD WITH KUBERNETES 32

DEVOPS PATTERNS

to a Kubernetes development environment each time the code

changes locally. This tool is highly configurable, and even supports
using Helm for deployments.

• Spinnaker: This open source continuous delivery platform was

developed by Netflix to handle CD operations at high scale over its
cloud network. It is a cloud-native pipeline management tool that
supports integrations into all the major cloud providers: Amazon Web
Services (AWS), Azure, Google Cloud Platform and OpenStack. It
natively supports Kubernetes deployments, but its scope extends
much farther beyond Kubernetes.

Extensions of DevOps
Continuous deployment of cloud-native applications has transformed the
way teams collaborate. Transparency and observability at every stage of
development and deployment are increasingly the norm. It’s probably no
surprise then that GitOps and SecOps, both enabled by cloud-native
architecture, are building on current DevOps practices by providing a
single source of truth for changes to the infrastructure and changes to
security policies and rules. The sections below highlight these
evolutionary developments.

GitOps
Git is becoming the standard for distributed version control, wherein a Git
repository contains the entire system: code, config, monitoring rules,
dashboards and a full audit trail. GitOps is an iteration of DevOps, wherein
Git is a single source of truth for the whole system, enabling rapid
application development on cloud-native systems, and using Kubernetes
in particular. “GitOps” is a term developed by Weaveworks to describe
DevOps best practices in the age of Kubernetes, and it strongly
emphasizes a declarative infrastructure.

CI/CD WITH KUBERNETES 33

DEVOPS PATTERNS

The fundamental theorem of GitOps is that if you can describe it, you can
automate it. And if you can automate it, you can control and accelerate it.
The goal is to describe everything — policies, code, configuration and
monitoring — and then version control everything.

With GitOps, your code should represent the state of your infrastructure.
GitOps borrows DevOps logic:

• All code must be version-controlled.

• Configuration is code.

• Configuration must also be version-controlled.

The idea behind GitOps is transparency. A declarative environment

captures state, enabling you to compare the observed state and the
desired state easily. In fact, you can observe the system at all times. In
short, every service has two sources of truth: the desired state of the
system and the observed state of the system.

In the truest sense of GitOps, Git maintains a repository that describes

your infrastructure and all of your Kubernetes configuration, and your
local copy of code gives you a complete version control repository. When
you push new code or configuration to Git, something on the other side
listens for this new push and makes the changes for you. All changes
happen the same way. All of your infrastructure and configuration live in a
centralized repository, and every single change is driven by a commit and
push of that repository.

How it works: Code is committed and pushed to GitHub, and then you
have a CI/CD workflow listening on the other side, and that CI/CD workflow
makes those changes and commits those changes in the configuration.
The key difference: Instead of engineers interacting with Kubernetes or the
system configuration directly — say, using Kubernetes CLI — they’re doing

CI/CD WITH KUBERNETES 34

DEVOPS PATTERNS

everything through Git: writing configuration, pushing configuration and

then applying those changes through the CI/CD workflow.

The three key goals of GitOps include:

1. Pipelines: Build completely automated CI/CD pipelines, enabling Git

to become a source of truth for the desired system state.

2. Observability: Implement 24/7 monitoring, logging and security —

observe and measure every service pull request to gain a holistic view
of the current system state.

3. Control: Version control everything and create a repository that

contains a single source of truth for recovery purposes.

With GitOps, Weaveworks is seeing people go from a few deployments a

week using CI systems to 30 to 50 deployments per day. In addition,
DevOps teams are fixing bugs twice as fast. Managing the state of the code
in Git versus Kubernetes allows for better tracking and recovery. It also
allows for continuous experimentation, such as A/B testing and response
to customer ideas, on the Kubernetes architecture.

SecOps
Security teams traditionally hand off security test results and vulnerability
scans to operations teams for review and implementation, often as an
application is being deployed. So long as an application is running and
performing as expected, security’s involvement in the process gets a green
light. However, information exchange and approval cycles can lead to delays
and slow what would otherwise be an agile DevOps workflow. It’s not
surprising that this occurs as the two teams are dealing with two very different
sets of goals. Operations tries to get the system running in as straightforward
and resilient a manner as possible. Security, on the other hand, seeks to
control the environment — the fewer things running, the better.

CI/CD WITH KUBERNETES 35

DEVOPS PATTERNS

In reality, these late-stage handoffs between operations and security are

problematic for many reasons, not the least of which is that the insights
and expertise of both teams are siloed rather than shared. As a result,
potential threats can grow into showstoppers, and security-related
problems often fester, going undetected for longer periods than
warranted.

When organizations don’t have a mechanism for communicating and

transferring key security data on an ongoing basis, those organizations
inevitably struggle to mitigate security risks, prioritize and remediate
security threats and vulnerabilities, and ultimately protect their
application environment. However, it’s not necessary for organizations to
sacrifice security in order to maintain uptime and performance. That’s
where SecOps comes in.

SecOps bridges the efforts of security and operations teams in the same
way that DevOps bridges the efforts of software developers and
operations teams. Just as a DevOps approach allows product developers
to strategically deploy, manage, monitor and secure their own
applications, a SecOps approach gives engineers a window into both
operations and security issues. It’s a transition from individual tacticians
— system administrators and database administrators — to more
strategic roles within the organization. These teams share priorities,
processes, tools and, most importantly, accountability, giving
organizations a centralized view of vulnerabilities and remediation actions
while also automating and accelerating corrective actions.

In a GitOps approach, all your configuration and infrastructure are stored

in a central Git repository. DevOps engineers write development and
deployment policies, whereas security engineers write security firewall
rules and network policies. And all of these rules end up in the same
repository. Collaboration among these teams — DevOps and SecOps, or

CI/CD WITH KUBERNETES 36

DEVOPS PATTERNS

SecDevOps — ups the ante, increasing efficiency and transparency into

what has happened and what should happen in the future.

In replacing disconnected, reactive security efforts with a unified,

proactive CI/CD-based security solution for both cloud and on-premises
systems, SecOps gains a more cohesive team from a diversity of
backgrounds working toward a common goal: frequent, fast, zero-
downtime, secure deployments. This goal empowers both operations and
security to analyze security events and data with an eye toward reducing
response times, optimizing security controls and checking and correcting
vulnerabilities at every stage of development and deployment. Blurring
the lines between the operations and security teams brings greater
visibility into any development or deployment changes warranted, along
with the potential impacts of those changes.

Conclusion
The story of DevOps and Kubernetes is one of continued, fast-paced
evolution. For example, in the current state of DevOps, technologies that
may be only a few years old can start to feel ancient. The industry is
changing on a dime.

Kubernetes is still very new and exciting, and there’s incredible demand in
the market for it. Organizations are leveraging DevOps to migrate to the
cloud, automate infrastructure and take Software as a Service (SaaS) and
web applications to the next level. Consider what you might accomplish
with higher availability, autoscaling and a richer feature set. Kubernetes
has vastly improved CI/CD workflows, allowing developers to do amazing
things they couldn’t do before. GitOps, in turn, offers a centralized
configuration capability that makes Kubernetes easy. With a transparent
and centralized configuration, changes aren’t being individually applied
willy nilly, but are instead going through the same pipeline. Today, product

CI/CD WITH KUBERNETES 37

DEVOPS PATTERNS

teams are centralizing development, operations and security, working

towards the same business goals: easier and faster deployments, less
downtime, fewer outages and faster time to recovery.

Organizations are increasingly turning to Kubernetes because they have

determined it’s the right tool for the job. Kubernetes just works — right out
of the box. Kubernetes makes it hard to do the wrong thing and easy to do
the right thing. More than half of Fortune 100 companies are using
Kubernetes, according to RedMonk. But the story doesn’t end there.
Mid-sized companies use Kubernetes. Ten-person startups use
Kubernetes. Kubernetes isn’t just for enterprises. It’s the real deal for
forward-thinking companies of all sizes. And it’s completely transforming
the way software is innovated, designed, built and deployed.

DevOps and Kubernetes are the future. Together, they just make good
business sense.

CI/CD WITH KUBERNETES 38

THE BEST CI/CD
TOOL FOR KUBERNETES
DOESN’T EXIST
There is no single, best set of tools for continuous
integration / continuous delivery (CI/CD) with
Kubernetes — each organization will use the tools
that are best suited for its specific use case.

“The hardest thing [about running on Kubernetes] is gluing all the

pieces together. You need to think more holistically about your
systems and get a deep understanding of what you’re working
with,” says Chris Short, a DevOps consultant and Cloud Native
Computing Foundation (CNCF) ambassador.

We talk with Short and Ihor Dvoretskyi, developer advocate at the

CNCF, about the trends they’re seeing in DevOps and CI/CD with
Kubernetes, the role of the Kubernetes community in improving CI/
CD and some of the challenges organizations face as they consider
the plethora of tools available today. Listen on SoundCloud »

Chris Short has spent more than two decades in various IT

disciplines, and has been an active proponent of open source
solutions throughout his time in the private and public sectors. Read
more at chrisshort.net, or follow his DevOps, cloud native, and open source-
focused newsletter DevOps’ish.

Ihor Dvoretskyi is a developer advocate at the Cloud Native

Computing Foundation. He is a product manager for Kubernetes,
co-leading the Product Management Special Interest Group, focused
on enhancing Kubernetes as an open source product. In addition, he
participates in the Kubernetes release process as a features lead.

CI/CD WITH KUBERNETES 39

CLOUD-NATIVE
APPLICATION PATTERNS
by JANAKIRAM MSV

ontainer technologies tell a story about a new wave of devel-

C opers. Their ranks are filled with people who are creating micros-
ervices that run on sophisticated underlying infrastructure tech-
nologies. Developers are drawn to the utility and packaging capabilities in
containers which improve efficiency and fit with modern application
development practices.

Developers use application architectures with container technologies to

develop services that reflect an organizational objective. Kubernetes runs
under the application architecture and is used to run services across multiple
clusters, providing the abstraction of orchestration to manage microservices
following DevOps practices. The flexibility of today’s cloud-native, modern
services running on infrastructure managed by software gives developers
capabilities that had previously not been possible. Developers now have
resources for connecting more endpoints through integrations that feed into
declarative infrastructure. Dynamic, declarative infrastructure is now a
foundation for development, and will increasingly serve as a resource for
event-driven automation in modern application architectures.

CI/CD WITH KUBERNETES 40

CLOUD-NATIVE APPLICATION PATTERNS

In all of this, the container can be defined as a unit. Each unit holds code,
a payload that in more complex operations gets orchestrated across
distributed architectures and infrastructure managed by cloud services.
More developers are now testing how to optimize the resources available
on different types of infrastructure to take advantage of the benefits
offered by containers and virtualization.

Cloud native is a term used to describe container-based environments.

Cloud-native technologies are used to develop applications built with
services packaged in containers, deployed as microservices, and
managed on elastic infrastructure through agile DevOps processes and
continuous delivery workflows. It’s important to note that in this equation,
the teams and processes that allow for increased speed and agility are as
important as the technology itself.

“ Cloud Native is structuring teams,

culture and technology to utilize
automation and architectures to
manage complexity and unlock
velocity.”
- Joe Beda, CTO and co-founder of Heptio.
Where operations teams would manage the infrastructure resource
allocations to traditional applications manually, cloud-native applications
are deployed on infrastructure that abstracts the underlying compute,
storage and networking primitives. Developers and operators dealing with
this new breed of applications don’t directly interact with application
programming interfaces (APIs) exposed by infrastructure providers.
Instead, the orchestrator handles resource allocation automatically,

CI/CD WITH KUBERNETES 41

CLOUD-NATIVE APPLICATION PATTERNS

according to policies set out by DevOps teams. The controller and

scheduler, which are essential components of the orchestration engine,
handle resource allocation and the life cycle of applications. Cloud-native
platforms, like Kubernetes, expose a flat network that is overlaid on
existing networking topologies and primitives of cloud providers. Similarly,
the native storage layer is often abstracted to expose logical volumes that
are integrated with containers. Operators can allocate storage quotas and
network policies that are accessed by developers and resource
administrators. The infrastructure abstraction not only addresses the
need for portability across cloud environments, but also lets developers
take advantage of emerging patterns to build and deploy applications.
Orchestration managers become the deployment target, irrespective of
the the underlying infrastructure that may be based on physical servers or
virtual machines, private clouds or public clouds.

Kubernetes is an ideal platform for running contemporary workloads

designed as cloud-native applications. It’s become the de facto operating
system for the cloud, in much the same way Linux is the operating system
for the underlying machines. Our author for this chapter is Janakiram MSV,
the Principal Analyst at Janakiram & Associates and an adjunct faculty
member at the International Institute of Information Technology. He is an
Ambassador for the Cloud Native Computing Foundation — home of the
Kubernetes open source project — and was also one of the first Certified
Kubernetes Administrators and Certified Kubernetes App Developers. Here,
he brings his considerable expertise consulting with organizations on their
cloud strategies and Kubernetes implementations to map the building
blocks of cloud-native applications with the constructs and primitives of
Kubernetes. The chapter acts as a guide to choosing the right Kubernetes
object for each of the components of a mature cloud-native application.

As long as developers follow best practices of designing and developing

CI/CD WITH KUBERNETES 42

CLOUD-NATIVE APPLICATION PATTERNS

software as a set of microservices that comprise cloud-native applications,

DevOps teams will be able to package and deploy them in Kubernetes.
This chapter is intended to help guide DevOps teams deploying cloud-
native applications in Kubernetes.

10 Key Attributes of Cloud-Native

Applications
Before taking a look at the big picture of cloud-native application design,
let’s review the ten key attributes of cloud-native applications.

1. Packaged as lightweight containers: Cloud-native applications are

a collection of independent and autonomous services that are
packaged as lightweight containers. Unlike virtual machines,
containers can scale out and scale in rapidly. Since the unit of scaling
shifts to containers, infrastructure utilization is optimized.

2. Developed with best-of-breed languages and frameworks: Each

service of a cloud-native application is developed using the language
and framework best suited for the functionality. Cloud-native
applications are polyglot; services use a variety of languages, runtimes
and frameworks. For example, developers may build a real-time
streaming service based on WebSockets, developed in Node.js, while
choosing Python and Flask for exposing the API. The fine-grained
approach to developing microservices lets them choose the best
language and framework for a specific job.

3. Designed as loosely coupled microservices: Services that belong

to the same application discover each other through the application
runtime. They exist independent of other services. Elastic
infrastructure and application architectures, when integrated
correctly, can be scaled out with efficiency and high performance.

CI/CD WITH KUBERNETES 43

CLOUD-NATIVE APPLICATION PATTERNS

Loosely coupled services allow developers to treat each service

independent of the other. With this decoupling, a developer can focus
on the core functionality of each service to deliver fine-grained
functionality. This approach leads to efficient life cycle management of
the overall application, because each service is maintained
independently and with clear ownership.

4. Centered around APIs for interaction and collaboration: Cloud-

native services use lightweight APIs that are based on protocols such
as representational state transfer (REST), Google’s open source remote
procedure call (gRPC) or NATS. REST is used as the lowest common
denominator to expose APIs over hypertext transfer protocol (HTTP).
For performance, gRPC is typically used for internal communication
among services. NATS has publish-subscribe features which enable
asynchronous communication within the application.

5. Architected with a clean separation of stateless and stateful

services: Services that are persistent and durable follow a different
pattern that assures higher availability and resiliency. Stateless
services exist independent of stateful services. There is a connection
here to how storage plays into container usage. Persistence is a factor
that has to be increasingly viewed in context with state, statelessness
and — some would argue — micro-storage environments.

6. Isolated from server and operating system dependencies:

Cloud-native applications don’t have an affinity for any particular
operating system or individual machine. They operate at a higher
abstraction level. The only exception is when a microservice needs
certain capabilities, including solid-state drives (SSDs) and graphics
processing units (GPUs), that may be exclusively offered by a subset of
machines.

CI/CD WITH KUBERNETES 44

CLOUD-NATIVE APPLICATION PATTERNS

7. Deployed on self-service, elastic, cloud infrastructure: Cloud-

native applications are deployed on virtual, shared and elastic
infrastructure. They may align with the underlying infrastructure to
dynamically grow and shrink — adjusting themselves to the varying
load.

8. Managed through agile DevOps processes: Each service of a

cloud-native application goes through an independent life cycle, which
is managed through an agile DevOps process. Multiple continuous
integration/continuous delivery (CI/CD) pipelines may work in tandem
to deploy and manage a cloud-native application.

9. Automated capabilities: Cloud-native applications can be highly

automated. They play well with the concept of infrastructure as code.
Indeed, a certain level of automation is required simply to manage
these large and complex applications.

10. Defined, policy-driven resource allocation: Finally, cloud-native

applications align with the governance model defined through a set of
policies. They adhere to policies such as central processing unit (CPU)
and storage quotas, and network policies that allocate resources to
services. For example, in an enterprise scenario, central IT can define
policies to allocate resources for each department. Developers and
DevOps teams in each department have complete access and
ownership to their share of resources.

Overview of Cloud-Native Application

Design
Cloud-native applications are composed of various logical layers, grouped
according to functionality and deployment patterns. Each layer runs
specific microservices designed to perform a fine-grained task. Some of

CI/CD WITH KUBERNETES 45

 Mapping Layers to Cloud-Native Workloads
CLOUD-NATIVE APPLICATION PATTERNS

Scalable Durable Parallelizable Event Driven Legacy Layer

Traditional/ Workload
Stateless Stateful Batch Processing Serverless Monolithic Type

UI Scheduled
Jobs
Legacy
RDBMS LOB App

API Batch
Jobs
Legacy
SCM

Graph NoSQL Event-Driven Legacy

Database Database Code ERP

Object Parallel Legacy

Storage Jobs CRM

Source: Janakiram MSV © 2018

FIG 2.1: Today’s modern application architectures bridge with monolithic legacy
systems.

these microservices are stateless, while others are stateful and durable.
Certain parts of the application may run as batch processes. Code
snippets may be deployed as functions that respond to events and alerts.

The depiction here attempts to identify layers of a cloud-native

application. Though they are grouped together for representation, each
layer is independent. Unlike traditional three-tier applications which are
stacked in a hierarchy, cloud-native applications operate in a flat structure
with each service exposing an API.

The scalable layer runs stateless services that expose the API and user
experience. This layer can dynamically expand and shrink depending on
the usage at runtime. During the scale-out operation, where more
instances of the services are run, the underlying infrastructure may also
scale out to match the CPU and memory requirements. An autoscale

CI/CD WITH KUBERNETES 46

CLOUD-NATIVE APPLICATION PATTERNS

policy is implemented to evaluate the need to perform scale-in and

scale-out operations.

The durable layer has stateful services that are backed by polyglot
persistence. It is polyglot because of the variety of databases that may be
used for persistence. Stateful services rely on traditional relational
databases, NoSQL databases, graph databases and object storage. Each
service chooses an ideal datastore aligned with the structure of stored
data. These stateful services expose high-level APIs that are consumed by
both — the services from the scalable and durable layers.

Apart from stateless and stateful layers, there are scheduled jobs, batch
jobs and parallel jobs that are classified as the parallelizable layer. For
example, scheduled jobs may run extract, transform, load (ETL) tasks
once per day to extract the metadata from the data stored in object
storage and to populate a collection in the NoSQL database. For services
that need scientific computing to perform machine learning training, the
calculations are run in parallel. These jobs interface with the GPUs
exposed by the underlying infrastructure.

To trigger actions resulting from events and alerts raised by any service in
the platform, cloud-native applications may use a set of code snippets
deployed in the event-driven layer. Unlike other services, the code
running in this layer is not packaged as a container. Instead, functions
written in languages such as Node.js and Python are deployed directly.
This layer hosts stateless functions that are event driven.

Cloud-native applications also interoperate with existing applications at

the legacy layer. Legacy, monolith applications — such as enterprise
resource planning, customer relationship management, supply chain
management, human resources and internal line-of-business applications
— are accessed by services.

CI/CD WITH KUBERNETES 47

CLOUD-NATIVE APPLICATION PATTERNS

Enterprises will embrace microservices for building API layers and user
interface (UI) frontends that will interoperate with existing applications. In
this scenario, microservices augment and extend the functionality of
existing applications. For example, they may have to talk to the relational
database that is powering a line-of-business application, while delivering
an elastic frontend deployed as a microservice.

Each service of a cloud-native application exposes a well-defined API,

which is consumed by other services. For intra-service communication,
protocols like gRPC or NATS are preferred due to their efficient
compression and binary compatibility. The REST protocol is used for
exposing services that interact with the external world.

DevOps teams map the deployment and communication patterns with

the primitives exposed by cloud-native platforms such as Kubernetes.
They are expected to package, deploy and manage these services running
in a production environment. The next section helps with the alignment
and mapping of the workload patterns with Kubernetes primitives.

Mapping Cloud-Native Workloads to

Kubernetes Objects
Kubernetes is more than just a container manager. It’s a platform
designed to handle a variety of workloads packaged in any number of
containers and combinations. There are multiple controllers built into
Kubernetes that map to the layers of cloud-native architecture.

DevOps engineers can think of Kubernetes controllers as the means for

dictating the infrastructure needs of the various workloads your team is
running. They can define the desired configuration state through a
declarative approach. For example, a container/pod deployed as a part of
a ReplicationController is guaranteed to be available all the time. A

CI/CD WITH KUBERNETES 48

CLOUD-NATIVE APPLICATION PATTERNS

container packaged as a DaemonSet is guaranteed to run on every node

of the cluster. The declarative approach enables DevOps teams to take
advantage of paradigms such as infrastructure as code. Some of the
deployment patterns discussed below follow the principles of immutable
infrastructure, where each new rollout results in an atomic deployment.

The control plane of Kubernetes constantly tracks the deployments to

ensure that they are adhering to the desired configuration state defined
by DevOps.

The fundamental unit of deployment in Kubernetes is a pod. It is the basic

building block of Kubernetes, which is the smallest and simplest unit in
the Kubernetes object model. A pod represents a running process on the
cluster. Irrespective of a service being stateful or stateless, it is always
packaged and deployed as a pod.

A controller can create and manage multiple pods within the cluster,
handling replication that provides self-healing capabilities at cluster scope.
For example, if a node fails, the controller might automatically replace the
pod by scheduling an identical replacement on a different node.

Kubernetes comes with multiple controllers to handle the desired state of

pods. ReplicationController, Deployment, DaemonSet and StatefulSet are
a few examples of controllers. Kubernetes controllers use a pod template
that is provided to create the pods for which it is responsible to maintain
the desired state. Pods, like other Kubernetes objects, are defined in a
YAML file and submitted to the control plane.

When running cloud-native applications in Kubernetes, operators need to

understand the use cases addressed by controllers to get the most out of
the platform. This helps them in defining and maintaining the desired
state of configuration of the application.

CI/CD WITH KUBERNETES 49

CLOUD-NATIVE APPLICATION PATTERNS

Each of the patterns explained in the previous section maps to specific

Kubernetes controllers which allow more precise, fine-grained control of
workloads on Kubernetes, but in an automated fashion.

The declarative configuration of Kubernetes encourages an immutable

infrastructure. The deployments are tracked and managed by the control
plane to ensure that the desired configuration state is maintained
throughout the application life cycle. When compared to traditional
deployments based on virtual machines, DevOps engineers will spend
significantly less time maintaining workloads. An effective CI/CD strategy
that takes advantage of Kubernetes primitives and deployment patterns
frees operators from performing mundane tasks.

Scalable Layer: Stateless Workloads

Stateless workloads are packaged and deployed as a ReplicaSet in
Kubernetes. A ReplicationController forms the basis of a ReplicaSet, which
ensures that a specified number of pod replicas are always running at any
given time. In other words, a ReplicationController makes sure that a pod
or a homogeneous set of pods is always up and available.

If there are too many pods, the ReplicationController may terminate the
extra pods. If there are too few, the ReplicationController proceeds to
launch additional pods. Unlike manually created pods, the pods
maintained by a ReplicationController are automatically replaced if they
fail, are deleted or terminated. The pods are re-created on a node after
disruptive maintenance such as a kernel upgrade. For this reason, it is
recommended to use a ReplicationController even if the application
requires only a single pod.

A simple use case is to create one ReplicationController object to reliably

run one instance of a pod indefinitely. A more complex use case is to run
several identical replicas of a scale-out service, such as web servers.

CI/CD WITH KUBERNETES 50

CLOUD-NATIVE APPLICATION PATTERNS

DevOps teams and operators package stateless workloads as

ReplicationControllers when deploying in Kubernetes.

In the recent versions of Kubernetes, ReplicaSets replaced

ReplicationControllers. Both of them address the same scenario, but
ReplicaSets use a set-based label selector which makes it possible to use
complex queries based on annotations. Additionally, Deployments in
Kubernetes rely on ReplicaSets.

Deployments are an abstraction of ReplicaSets. When a desired state is

declared in the Deployment object, the Deployment controller changes
the actual state to the desired state at a controlled rate.

Deployments are highly recommended to manage stateless services of

cloud-native applications. Though services can be deployed as pods and
ReplicaSets, Deployments make upgrading and patching your application
easier. DevOps teams can upgrade a pod in place using a Deployment,
which cannot be done with a ReplicaSet. This makes it possible to roll out
a new version of an application with minimal downtime. Deployments
bring Platform as a Service (PaaS)-like capabilities to application
management.

Durable Layer: Stateful Workloads

Stateful workloads can be classified into two categories: services that
need persistent storage (single instance) and services that need to run in a
highly reliable and available mode (replicated multi-instance). A pod that
needs access to a durable storage backend is very different from a set of
pods that run a cluster for a relational database. While the former needs
long-term, durable persistence, the latter needs high availability of the
workload. Kubernetes addresses both scenarios.

Individual pods can be backed by volumes that expose underlying storage

to the services. The volume may be mapped to an arbitrary node on

CI/CD WITH KUBERNETES 51

CLOUD-NATIVE APPLICATION PATTERNS

which the pod is scheduled. If multiple pods are scheduled across

different nodes of the cluster and need to share the backend, a distributed
file system such as Network File System (NFS) or Gluster is configured
manually before deploying applications. Modern storage drivers available
within the cloud-native ecosystem offer container-native storage where
the file system itself is exposed through containers. Use this configuration
when pods just need persistence and durability.

For scenarios where high availability is expected, Kubernetes offers

StatefulSets — a specialized set of pods that guarantees the ordering and
uniqueness of pods. This is especially useful in running primary/secondary
— previously known as master/slave — configurations of database clusters.

Like a Deployment, a StatefulSet manages pods that are based on an

identical container specification. Unlike a Deployment, a StatefulSet
maintains a unique identity for each of its pods. These pods are created
from the same spec, but are not interchangeable: Each pod has a
persistent identifier that it maintains across any rescheduling.

StatefulSets are useful for workloads that require one or more of the
following:

• Stable, unique network identifiers.

• Stable, persistent storage.

• Ordered, graceful deployment and scaling.

• Ordered, graceful deletion and termination.

• Ordered, automated rolling updates.

Kubernetes treats StatefulSets differently than other controllers. When

pods of a StatefulSet are being scheduled with N replicas, they are created
sequentially, in order from 0 to N-1. When pods of a StatefulSet are being

CI/CD WITH KUBERNETES 52

CLOUD-NATIVE APPLICATION PATTERNS

deleted, they are terminated in reverse order, from N-1 to 0. Before a

scaling operation is applied to a pod, all of its predecessors must be
running and ready. Kubernetes ensures that before a pod is terminated, all
of its successors are completely shut down.

StatefulSets are recommended when services need to run clusters of

Cassandra, MongoDB, MySQL, PostgreSQL or any database workloads
with a high availability requirement.

Not every persistent workload needs to be a StatefulSet. Certain

containers rely on a durable storage backend to store data. For adding
persistence to these type of applications, pods may rely on volumes
backed by either host-based storage or container-native storage
backends.

Parallelizable Layer: Batch Processing

Kubernetes has built-in primitives for batch processing, which is useful for
executing run to completion jobs or scheduled jobs.

Run to completion jobs are typically used for running processes that need
to perform an operation and exit. A big data workload that runs until the
data is processed is an example of such a job. Another example is a job
that processes each message in a queue until the queue becomes empty.

A Job is a controller that creates one or more pods and ensures that a
specified number of them successfully terminate. As pods successfully
complete, the Job tracks the successful completions. When a specified
number of successful completions is reached, the Job itself is complete.
Deleting a Job will clean up the pods it created.

A Job can also be used to run multiple pods in parallel, which makes it
ideal for machine learning training jobs. Jobs also support parallel
processing of a set of independent but related work items.

CI/CD WITH KUBERNETES 53

CLOUD-NATIVE APPLICATION PATTERNS

When Kubernetes runs on hardware with GPUs, machine learning training

can take advantage of Jobs. Emerging projects such as Kubeflow — a
project dedicated to making deployment of machine learning on
Kubernetes simple, portable and scalable — will expose primitives to
package machine learning training as Jobs.

Apart from running parallelized jobs, there may be a need to run

scheduled jobs. Kubernetes exposes CronJobs that can run once at a
specified point in time or periodically at a specified point in time. A
CronJob object in Kubernetes is similar to one line of a crontab (cron
table) file in Unix. It runs a job periodically on a given schedule, written in
cron format.

CronJobs are especially useful for scheduling periodic jobs such as

database backups or sending emails.

Event-Driven Layer: Serverless

Serverless computing refers to the concept of building and running
applications that do not require server management. It describes a more
fine-grained deployment model where applications, bundled as one or
more functions, are uploaded to a platform and then executed, scaled
and billed in response to the exact demand needed at the moment.

Functions as a Service (FaaS) runs within the context of serverless

computing to provide event-driven computing. Developers run and
manage application code with functions that are triggered by events or
HTTP requests. Developers deploy small units of code to the FaaS, which
are executed as needed as discrete actions, scaling without the need to
manage servers or any other underlying infrastructure.

Though Kubernetes doesn’t have an integrated event-driven primitive that

responds to alerts and events raised by other services, there are efforts to
bring event-driven capabilities. The Cloud Native Computing Foundation,

CI/CD WITH KUBERNETES 54

CLOUD-NATIVE APPLICATION PATTERNS

the custodian of Kubernetes, has a serverless working group focused on

these efforts. Open source projects such as Apache OpenWhisk, Fission,
Kubeless, OpenFaaS and Oracle’s Fn can be run within a Kubernetes
cluster as the event-driven, serverless layer.

Code deployed in the serverless environment is fundamentally different

from the code packaged as pods. It consists of autonomous functions that
can be wired to one or more events that may trigger the code.

When event-driven computing — serverless computing — becomes an

integral part of Kubernetes, developers will be able to deploy functions
that respond to both internal events generated by the Kubernetes control
plane along with custom events raised by applications service.

Legacy Layer: Headless Services

Even after your organization is regularly building and deploying
applications using a microservices architecture into containers on the
cloud, there may be applications that continue to live outside of
Kubernetes. Cloud-native applications and services will have to interact
with those traditional, monolithic applications.

The legacy layer exists for interoperability, to expose a set of headless

services pointing to the monolithic applications. Headless services allow
developers to reduce coupling to the Kubernetes system by allowing them
freedom to do discovery their own way. Headless services in Kubernetes
are different from ClusterIP, NodePort and LoadBalancer types of services.
They don’t have an internet protocol (IP) address assigned to them, but
have a domain name system (DNS) entry that points to an external
endpoint such as API servers, web servers and databases. The legacy layer
is a logical interoperability layer that maintains DNS records to well-
known, external endpoints.

Each layer of a microservices application can be mapped to one of the

CI/CD WITH KUBERNETES 55

 CLOUD-NATIVE APPLICATION PATTERNS

controllers of Kubernetes. Depending on the pattern they wish to deploy,

DevOps teams can choose the appropriate option.

The depiction below (Fig. 2.2) maps various layers to the cloud-native
application stack with Kubernetes primitives.

Best Practices for Deploying Cloud-Native

Applications in Kubernetes
Understanding the mapping between cloud-native applications and
Kubernetes primitives is important for DevOps engineers. But it’s equally
important to follow some recognized best practices in deploying those
workloads.

Here are 10 best practices for deploying and running cloud-native

FIG 2.2: DevOps engineers can define the desired configuration state through a
declarative approach — each workload maps to a controller.

Mapping Workloads to Kubernetes Primitives

Traditional/ Workload
Stateless Stateful Batch Processing Serverless Monolithic Type

Pod Pod with Jobs (run to Events Headless Service Kubernetes

ReplicationController Volumes completion) HTTP (not a controller) Primitive

Deployment CronJobs Requests

StatefulSets

UI Scheduled
Jobs
Legacy
RDBMS LOB App

API Batch
Jobs
Legacy
SCM

Graph NoSQL Event-Driven Legacy

Database Database Code ERP

Object Parallel Legacy

Storage Jobs CRM

CI/CD WITH KUBERNETES 56

Source: Janakiram MSV © 2018

CLOUD-NATIVE APPLICATION PATTERNS

applications in Kubernetes. These techniques help DevOps teams in

getting the optimal performance for their application deployments on
Kubernetes.

1. Never Deploy “Naked” Pods. Naked pods are pods that are not a
part of a ReplicaSet, ReplicationController or Deployment. Since it is
easy, a common practice for both developers and operations teams is
to package a container as a simple pod and deploy it in Kubernetes.
Naked pods suffer from single points of failure as Kubernetes will not
be able to reschedule them when a node fails. Always package pods
as a ReplicationController or a Deployment.

2. Choose an Appropriate Controller for the Workload. Kubernetes

has a variety of controllers mapped to specific workloads. Choose
from ReplicationController, Deployment, StatefulSet, DaemonSet and
Job controllers depending on the kind of workload.

3. Use an Init Container to Ensure That the Application is

Initialized. A pod is a collection of one or more containers. An Init
container in a pod runs before the application containers are started.
Init containers are similar to regular containers but they always run to
completion, and they run in a sequence. If a pod has multiple Init
containers, each one must complete successfully before the next one
is started.

Init containers are used to populate database tables, download files

from remote locations to local volumes and check for other pods to
become available. Avoid using a sidecar container when targeting
Kubernetes. Sidecar containers packaged in the same pod are treated
as normal containers by Kubernetes. Unlike Init containers, they start
along with other containers of the pod, making it difficult to ensure
that the initialization is done before starting the rest of the pods.

CI/CD WITH KUBERNETES 57

CLOUD-NATIVE APPLICATION PATTERNS

4. Launch Services Before the Workloads. Create a service before

creating its corresponding backend workloads, such as Deployments
or ReplicaSets, and before any workloads that need to access it. When
Kubernetes starts a container, it provides environment variables pointing
to all the services which were running when the container was started.
When a pod is scheduled, it automatically gets populated with the
environment variables of services created within the same namespace.

5. Use Deployment History to Rollback and Rollforward Versions.

One of the advantages of using Deployments over naked pods and
ReplicaSets is the ability to rollback and rollforward versions. Always
use Deployment Record to perform easier rollbacks. This capability
mimics the ease of dealing with a Platform as a Service (PaaS), like
Heroku or Engine Yard for deployments.

6. Use ConfigMaps and Secrets. Secrets securely store sensitive

information, such as passwords, OAuth tokens and secure shell (SSH)
keys. Moving this information to a secret is much safer and more
flexible than embedding it in a pod definition or baking it into the
container image.

ConfigMaps allows developers to decouple configuration artifacts

from image content to keep containerized applications portable. They
are a good replacement to hard-wired configuration supplied through
external files.

7. Add Readiness Probe and Liveness Probe to Pods. The

Kubernetes control plane relies on a readiness probe to know when a
container is ready to start accepting traffic. A pod is considered ready
when all of its containers are ready. One use of this signal is to control
which pods are used as backends for services. When a pod is not
ready, it is removed from service load balancers.

CI/CD WITH KUBERNETES 58

CLOUD-NATIVE APPLICATION PATTERNS

Kubernetes depends on the liveness probe to decide when to restart a

container. When the liveness probe encounters a situation like a
deadlock, where an application is running but unable to make
progress, Kubernetes restarts the container to reset the state.

8. Define CPU and Memory Resource Limits to Containers and

Pods. The Pod specification allows resource limits to be specified for
each container. A container is guaranteed to have as much memory as
it requests, but it is not allowed to use more memory than the defined
limit. This is also the case for CPU limits. The CPU resource is
measured in CPU units, which typically translates to the virtual CPU
(vCPU) or a core of the associated infrastructure.

9. Define Multiple Namespaces to Limit the Default Visibility of

Scope of Services. Kubernetes supports multiple virtual clusters
backed by the same physical cluster called namespaces. For each
service created in a Kubernetes cluster, there is a corresponding DNS
entry. This entry is of the form <service-name>.<namespace-
name>.svc.cluster.local, which means that if a container just
uses <service-name>, it will resolve to the service which is local to a
namespace. This is helpful for using the same configuration across
multiple namespaces such as development, staging and production.
To reach services from other namespaces, a fully qualified domain
name (FQDN) may be used.

10. Configure Horizontal Pod Autoscaling for Dynamic Scaling of

Stateless Workloads. The Horizontal Pod Autoscaler (HPA)
automatically scales the number of pods in a ReplicationController,
Deployment or ReplicaSet based on observed CPU utilization or with
custom metrics support. HPA is managed by the Kubernetes control
plane. The resource consumption influences the behavior of the
controller. The controller periodically adjusts the number of replicas in

CI/CD WITH KUBERNETES 59

CLOUD-NATIVE APPLICATION PATTERNS

a ReplicationController or Deployment to match the observed average

CPU utilization to the target specified by a user. This configuration
delivers optimal performance of stateless services.

Conclusion
One of reasons that Kubernetes has become successful is the flexibility
and control it offers to developers and operators. Developers stay focused
on shipping microservices without worrying about the deployment
environment. DevOps engineers take the software, map the layers to
appropriate primitives and deploy it in Kubernetes. This workflow and
decoupling of software design and deployment is what makes Kubernetes
unique.

This chapter focused on attributes of cloud-native applications, mapping

cloud-native workload types to Kubernetes primitives and best practices
for deploying and running applications in Kubernetes. In the next chapter
we will explore to how to build CI/CD pipelines that automate
deployments.

CI/CD WITH KUBERNETES 60

IMPROVE SECURITY WITH
AUTOMATED IMAGE
SCANNING THROUGH CI/CD
Automation through a CI/CD pipeline is key to
securing an application deployed on Kubernetes.
Using cloud-native security tools that hook right
into Jenkins or your favorite CI/CD tool, enterprise
security teams can set policies for developers who are building
container images. The pipeline enforces those policies through
automated vulnerability scanning of each image during the build
process. Developers only deploy images that the security team is
confident in because they’ve been scanned.

“CI/CD automation is key because of the scale,” said Liz Rice,

technology evangelist at Aqua Security. “You couldn’t possibly
manually check all these different images when you’re shipping
potentially hundreds or thousands of deploys in a day.”

In this podcast, learn about how distributed architectures change

the enterprise approach to security, how automation can improve
security at scale and how Aqua’s tools — free as well as commercial
enterprise versions — can help improve security through the entire
application life cycle. Listen on SoundCloud »

Liz Rice is the Technology Evangelist with container security

specialists Aqua Security, where she also works on container-related
open source projects including kube-bench and manifesto. This year
she is co-chair of the CNCF’s KubeCon + CloudNativeCon events taking place in
Copenhagen, Shanghai and Seattle.

CI/CD WITH KUBERNETES 61

CONTINUOUS DELIVERY
WITH SPINNAKER
by CRAIG MARTIN

odern application architecture releases should be frequent, fast

M and, above all, boring. This makes everyone happy — the

developers, the business managers and the client. To this end,
the growing tech movement to organize software teams and technologies
around the notions of DevOps has created great interest in continuous
delivery (CD) platforms.

DevOps, itself, promises golden opportunities: faster deployments for a

marketplace that demands new features, greater accuracy of deployments
through automation, faster feedback from clients through iteration, and
even higher job satisfaction as teams cooperate to get features and fixes
out the door faster. DevOps is a journey and not a destination. It means
building cross-functional teams with common goals, aligning the
organization around the architecture — reversing Conway’s Law — and
creating a culture of continuous improvement. One of the higher-level
achievements in a DevOps journey is continuous delivery.

ThoughtWorks encapsulates the ideal CD mindset. Here’s how they

describe it:

CI/CD WITH KUBERNETES 62

CONTINUOUS DELIVERY WITH SPINNAKER

“ Continuous Delivery is the natural extension of Continuous

Integration, an approach in which teams ensure that every change to
the system is releasable, and release any version with the push of a
button. Continuous Delivery aims to make releases boring, so that we
can deliver frequently and get quick feedback on what users care
about.”

Implementing Kubernetes on its own doesn’t magically achieve CD for

your organization. The features Kubernetes provides, however, in
modularity, available tooling and immutable infrastructure certainly make
CD much easier to put in place. Although Kubernetes will help you define a
container deployment and manage instances, it leaves it up to you as to
how you will automate those deployments into environments.

In this chapter, The New Stack asked Craig Martin, senior vice president of
engineering at Kenzan, to discuss Spinnaker and how it illustrates the way
CI/CD practices are evolving for cloud-native architectures. Kenzan is a top
contributor to the open source Spinnaker project and has built its own
open source framework for Kubernetes deployments with it. Here, Martin
draws on his experience working with organizations to enact digital
transformations through building large-scale microservices applications
on top of Kubernetes and Spinnaker to explain:

• A new approach to CI/CD that’s emerging for cloud-native architectures.

• The emergence of Spinnaker as a CD tool based on “state” management.

• The benefits and drawbacks of this new tool, which has not yet gained
widespread adoption.

• Best practices for setting up CI/CD pipelines with Spinnaker on Kubernetes.

• New strategies for CI/CD on a cloud-native stack.

CI/CD WITH KUBERNETES 63

CONTINUOUS DELIVERY WITH SPINNAKER

Ultimately, how companies employ CD onto Kubernetes is very important

to achieving the ideals of push-button, boring deployments. Development
groups have commonly answered the question of CD with established
tools they are familiar with, namely Jenkins. Some 45 percent of
respondents in The New Stack’s Kubernetes survey said they use Jenkins
to deploy applications to Kubernetes. This makes perfect sense.
Companies have historically solved the problems of continuous
integration (CI) first, through versioning code and repeatable builds in
Jenkins. It’s only natural that they would extend the job-running CI
capability of Jenkins to then solve the problem of automating
deployments. CD-savvy developers are even becoming familiar with new
Jenkins 2.0 pipelines. In 2.0, an included pipeline plugin allows you to
create Groovy scripts to orchestrate a fine-grained control over build, test
and deployment stages. Other plugins, such as Blue Ocean, allow you to
visualize these CI/CD pipelines in Jenkins.

While these highly customizable Jenkins pipelines can be built out

through complex scripting, it still begs the question as to whether Jenkins
is the right platform for the job of CD, or what platform, in fact, is the best
tool for Kubernetes deployments on the cloud.

Enter Spinnaker
CD differs from CI. CI is a mechanism to merge and test code changes on
an ongoing basis, often achieved by a tool like Jenkins. CD is the attempt
to speed up and automate deployments, where an operator can push out
multiple deployments in a week across numerous services, and know the
exact condition of the applications and infrastructure in the course of the
deployments. What is truly required for continuous delivery which is not
provided by CI tools is a “state” machine. Such a state machine will have
the ability to take an environment from one state to the next until it makes

CI/CD WITH KUBERNETES 64

CONTINUOUS DELIVERY WITH SPINNAKER

it all the way to production. The machine will move the environment, such
as Docker containers, through to production in an automated fashion, and
will even have the ability to do things like rollbacks, canary deployments
and scaling instances. This allows for the agile, push-button, automated
deployments that an ideal CD mindset drives towards.

Spinnaker provides such a state machine. Spinnaker is an open source

continuous delivery platform developed by Netflix to handle CD
operations at high scale over its cloud network. It is a cloud-native
pipeline management tool that supports integrations into all the major
cloud providers, namely Amazon Web Services (AWS), Azure, Google Cloud
Platform and OpenStack. It natively supports Kubernetes deployments.

Spinnaker is different than most other CI tools. CI technologies build code

and run tests against it. CD technologies focus more on displaying the
current state of the environment. CD takes software delivery a step further
by automatically testing the software and pushing it into production using
techniques such as canary testing and blue-green testing.

Spinnaker does not replace CI tools. It works alongside the tried-and-true

workhorse of Jenkins: A Jenkins job can still handle building and storing
artifacts for the CI portion, and once finished, the job can trigger a
Spinnaker pipeline that deploys the application onto Kubernetes — or
anywhere — for the CD portion.

We’ve worked with several companies to implement Jenkins side-by-side

with Spinnaker, and have, at times, received furrowed-brow, questioning
looks at the initial recommendation. Why go to all the trouble of learning
Spinnaker as a secondary CD tool and integrating with it?

Spinnaker Features
Using Jenkins alone to enable the fine-grained control of pipelines that
can do things like automate testing, rollbacks, visualization and templated

CI/CD WITH KUBERNETES 65

CONTINUOUS DELIVERY WITH SPINNAKER

reuse would take quite a bit of custom Groovy code using Jenkins 2.0
pipelines. And after all this work, you would still not have a true “state”
management tool.

Spinnaker already holds mastery at this state management, and provides

the mentioned pipeline features out of the box, going well beyond tools
like Jenkins.

There are many features that give it this mastery:

• Cloud-Native Delivery Tool: Spinnaker is cloud native. It was built

for the cloud and in the cloud. This means that all the benefits of the
cloud — autoresiliency, autoprovisioning and autoscaling — are native
to Spinnaker.

• Microservices based: Spinnaker follows microservice architectural

patterns. Along with utilizing microservices, each component is built to
solve the needs of its domain, utilizing domain-driven design. This
allows Spinnaker a high level of modularity and extensibility to meet
your specific needs.

• Delivery and Infrastructure Visibility: Spinnaker provides a user

interface (UI) that allows you to view your infrastructure and to see
exactly where your code is residing. This gives you views of your load
balancers, regions, ingress IPs, etc. The importance of seeing code
deployments and the infrastructure cannot be overstated, as it will
simplify your view into deployments.

• Conditional Pipelining: This is the ability to have pipelines spin off

when certain criteria are met, for example: infrastructure, load, regions
and failure types. Although this can be achieved with other tooling, we
find Spinnaker pipelines the easiest and most intuitive to configure
with conditional logic.

CI/CD WITH KUBERNETES 66

CONTINUOUS DELIVERY WITH SPINNAKER

• Robust API: Spinnaker was built to be a consumed application

programming interface (API) and is very thorough, allowing a great
deal of customization as your deployments require it.

• Scaling Capabilities: As a result of the microservices architectural

patterns that Spinnaker uses (more details in the next section), it can
achieve high levels of scalability. Each microservice composing
Spinnaker is capable of being horizontally scaled and each component
has quite a bit of resiliency built into it. If you have a need for high
scalability, then Spinnaker is the gold standard.

• Built-in Deployment Strategies: Out of the box, Spinnaker provides

many commonly used deployment strategies, including Highlander,
red/black and rolling red/black. You can drop these into your pipelines
and use them wherever you see fit. It allows an operator to roll back a
failed deployment right within the Spinnaker UI.

• Flexibility of Deployment Modalities: Spinnaker isn’t just for

deploying applications. It is meant to encompass all aspects of your
stack, including deployment of infrastructure components and
configurations. In this way, you can deploy infrastructure components,
and then the application.

• Multi-Region: Spinnaker natively supports delivery pipelines that

can deploy across regions in parallel. It can even support a multiple-
cloud setup.

• Kubernetes and Container Aware: The pipelines within Spinnaker

integrate easily with Kubernetes container deployments. The UI can
even be used to manage instances. You can scale deployed pods from
the UI during a peak, or terminate them as needed.

• Flexibility of Deployment Targets: Spinnaker supports a variety of

CI/CD WITH KUBERNETES 67

 CONTINUOUS DELIVERY WITH SPINNAKER

deployment targets, including mainstream cloud and container

platforms such as Amazon Web Services, Azure, Cloud Foundry, DC/
OS, Google Compute Engine (GCE), Kubernetes and OpenStack.
Irrespective of the target environment, the deployment experience is
always consistent.

Spinnaker Architecture
Understanding the component architecture of Spinnaker is important to
seeing its strengths. A typical Spinnaker installation includes a number of
microservices that all work together to create and manage pipeline-based
deployments.

Each component takes on specific roles and responsibilities within its

domain.

FIG 3.1: Each component in Spinnaker’s modular architecture has its own
responsibilities.

Spinnaker Components and Their Roles

HTML API GATEWAY MICROSERVICES

WEBSERVER
Integration Event
Human Layer Router
User
Abstraction
Deck Igor Echo Layer

User Interface: Front50

monitor deployments,
manual judgments,
remedy issues.

Gate Orca
Abstraction
Layer
Custom API Gateway: Orchestration
basic AuthN, routing Engine
Script Cloud Fiat
into microservices.
driver
</> Authorization
Service

Rosco

Create
CI/CD WITH KUBERNETES machine images 68

Source: https://blog.spinnaker.io/exposing-spinnaker-to-end-users-4808bc936698 © 2018

CONTINUOUS DELIVERY WITH SPINNAKER

• Deck: A user interface to monitor the deployments, make manual

judgements and remedy issues.

• Gate: The API Gateway that is used to perform basic AuthN and routing
into the microservices.

• Orca: The orchestration engine that is used to actually perform the

pipeline task and perform the work. It uses Redis to persist data while
it is executing.

• Clouddriver: The abstraction layer that Spinnaker uses to

communicate with cloud providers. Also stores local caches of
deployed resources.

• Front50: The abstraction layer for persisting metadata, config,

notifications, etc. By default it uses Cassandra, but many different
datastores can be used.

• Rosco: This is the microservice that is used to create machine images,

for example: AWS Amazon Machine Image (AMI), Azure virtual machine
(VM) images, and Google Compute Engine (GCE) images.

• Igor: An integration layer in your continuous integration tooling, such

as Jenkins and Travis; and Git repositories, such as Bitbucket, GitHub
and Stash.

• Echo: This is an event router that is used to communicate all critical

events to the properly configured listeners, such as Slack and short
message service (SMS).

• Fiat: Spinnaker’s authorization service. It handles the authentication

and authorization for users, applications and service accounts.

• Halyard: Spinnaker’s configuration service that is used to install,

maintain and upgrade Spinnaker itself. Halyard is an important

CI/CD WITH KUBERNETES 69

 CONTINUOUS DELIVERY WITH SPINNAKER

“Terraform-like” tool, creating a golden deployment configuration for

Spinnaker that can be used for recovery or spawning additional
instances.

Spinnaker has a smart component architecture where each microservice

domain has its own API and can extend in a modular way. This allows for
greater flexibility, more features and future integrations with any number
of technologies.

The Beauty of Pipelines

While we’ve mentioned the idea of pipelines in Spinnaker, we haven’t yet
described what they do or seen them in action. Pipelines are at the core of
CD capabilities, by orchestrating a repeatable deployment over stages.

Each pipeline in Spinnaker starts with a specific trigger — this could be a

Git commit, a manual start, a Jenkins build, a push to a Docker repository
or another conditional trigger. The pipeline proceeds through several user-
defined stages. Stages could involve the actual deployment, running
automated tests for the specific environment, or even rolling back a
deployment if smoke tests fail. What you want to accomplish depends on

FIG 3.2: A new image in the development repository triggers the Spinnaker develop-
ment (dev) pipeline to deploy the application to a Kubernetes dev cluster.

Spinnaker Development Pipeline

AUTOMATIC TRIGGER

Push Development Test Data Deploy Run

New App Newman
Image Repository (optional) App
Tests

Database
Development
Cluster
CI/CD WITH KUBERNETES 70

Source: Kenzan © 2018

 CONTINUOUS DELIVERY WITH SPINNAKER

the environment, so individual pipelines are typically created per the

environment you deploy in.

Development Environment
For example, in a development (dev) environment, you might set up a very
simple Spinnaker deployment which runs often. Triggering off a new image
being pushed to the container image repository, it deploys the application
to the dev Kubernetes cluster, then on post-deploy runs a number of curls
against the application’s REST API using a Newman script.

Staging Environment
In the staging environment, you might set up another pipeline that is a bit
more complex. Triggering off a new image pushed to your staging repository,
you first create some new test data using a database insert. The next few
stages deploy the application, run integration tests utilizing the test data,
perform security penetration tests and finally load tests to simulate peaks.

Production Environment
In the production (prod) environment, instead of triggering off of Jenkins,
we allow operations to start the pipeline manually. Using Spinnaker’s

FIG 3.3: A new image in the staging repository triggers the Spinnaker staging pipe-
line to deploy the application to a staging cluster.

Spinnaker Staging Pipeline

AUTOMATIC TRIGGER

Push Staging Stage Deploy Run Run Run

New App Integration Security
Image Repository Test Data App Load Test
Tests Tests

Database
Staging Cluster

CI/CD WITH KUBERNETES 71

Source: Kenzan © 2018

 Spinnaker Production Pipeline
CONTINUOUS DELIVERY WITH SPINNAKER

MANUAL TRIGGER

Push Staging Production Canary Smoke

New App Data Updates
Image Repository (optional) Deployment Tests

Database
Production
Cluster

Source: Kenzan © 2018

FIG 3.4: The operations team manually triggers the prod pipeline to deploy an appli-
cation to production on Kubernetes.

built-in capabilities, we’ve set up a canary deployment which gradually

routes requests to the new version of the application, and also includes a
stage that performs smoke tests for good measure. For more on canary
deployments, see best practices.

Throughout these various pipelines, Spinnaker’s infrastructure view

provides visibility regarding what did and did not happen along the
pipelines, what tests passed or failed and what versions were deployed.
Having a CD orchestration tool that provides an applications-on-
infrastructure view makes supporting the running application that much
easier for operations, giving a level of insight and control that creates
push-button deployments across environments.

All of the above pipelines were constructed inside the same Spinnaker UI.
Although scripting pipelines is possible for more complex operations, the
templates in the Spinnaker UI make stringing together pipeline stages
fairly easy, with most tools and options needed for robust deployments. In
our experience, pipelines and its concept of Pipeline Templates are one of
the areas where Spinnaker really shines. Creating custom pipeline
templates is an easy way to create reusable pipeline modules that can be

CI/CD WITH KUBERNETES 72

CONTINUOUS DELIVERY WITH SPINNAKER

applied in varying situations. We have found this to be a very useful way to

give each team control over their pipelines, but ensure that individual
components are only built once and built correctly. The templates make
reuse natural, and not a chore requiring best practices or lengthy
documentation, which custom code templates might require.

And while the above pipelines demonstrate what deployments might look
like across typical development, staging and production environments,
the ultimate focus of CD on fast deployments should cause you to
question having so many environments. Could all of the above stages be
accomplished in fewer environments? Deployment strategies, like canary,
combined with the automation and accuracy Spinnaker provides in its
pipeline stages, could very well provide a path towards having code pass
through fewer environments, making it to production in fewer steps.

Example Implementations
It’s helpful to look at a few actual implementations to see the flexibility
and power of Spinnaker. At Kenzan, we’ve helped implement many of the
following Spinnaker use cases with clients.

Case 1: All in with the IaaS

For organizations whose journey to the cloud is not burdened by
supporting legacy infrastructure and applications, they may choose to
leverage Infrastructure as a Service (IaaS) capabilities for building
applications. The motivation is that IaaS now provides various classic
continuous integration capabilities, allowing organizations to use their
services rather than implementing specific build tools. For example,
Google Cloud Container Builder — which is also used to create the
Spinnaker container images for release — is a Google Cloud service that
can be employed to perform various build and low-level integration
actions.

CI/CD WITH KUBERNETES 73

CONTINUOUS DELIVERY WITH SPINNAKER

Spinnaker integrates with both Google Cloud and Kubernetes. A container

image build in Google Cloud Container Builder can act as a pipeline trigger
into Spinnaker, which can then deploy your containers on a Kubernetes
cluster. An organization that uses such a cloud-first, pro-IaaS approach
can quickly build and deploy containerized applications with all the
appropriate systems development life cycle (SDLC) capabilities. This
provides the organization with the advantage of not having to commit to
several separate and distinct SDLC tools. Additionally, flexibility is still
retained through the existing custom Spinnaker stages, which allows
integrations with various SDLC tools.

The outcome of the approach is that the organization can then spend
more time supporting applications and less time on app tooling.

Case 2: Spawning Developer Environments

Spinnaker is able to deploy both code and infrastructure. Starting from
zero, it can deploy and manage an entire stack: a functional environment
that includes Spinnaker itself.

To demonstrate this capability, let’s assume I am an infrastructure-as-

code, microservices and containerization-embracing company. My
e-commerce platform is composed of 100 different services with various
instances of those services. I would like to enable the following SDLC
conditions and practices:

1. I would like to have developers who are working on features be able to

obtain a duplicate of the last known good (LKG) of the entire platform
environment, except for the service they are enhancing.

2. I would like to have disposable integration environments.

3. I would like to spawn disposable performance and pen testing

environments.

CI/CD WITH KUBERNETES 74

CONTINUOUS DELIVERY WITH SPINNAKER

These are goals that many organizations dream of, but are often incapable
of executing due to the complexity of replicating environments and the
cost of doing so.

Spinnaker allows an elegant solution to a potentially complex problem by

being both infrastructure-aware and having the ability to replicate itself. A
solution we’ve implemented is having an “Environment Creator Spinnaker”
instance that creates new environments and infrastructure for the
developer. It starts by deploying Kubernetes. Using a predefined Halyard
definition, it then deploys a new Spinnaker instance onto the Kubernetes
cluster. Then the pipeline definitions are run in the appropriate sequence
to redeploy the desired version of all the applications and their
infrastructure. Spinnaker deploys Kubernetes and Spinnaker, running
pipelines to create a complete, disposable environment. In this way,
separate environments can be spun up for development, integration
testing and performance testing.

You may be thinking, that’s great, but what about the cost of such
environments? You could unknowingly spin up a number of cloud
resources and leave them in place, only to find out later your usage bill has
tripled. To help solve this, a manual decision stage can be put at the end
of the Environment Creator pipeline, creating a manual pause that allows
adequate time for testing to occur, then finishing the pipeline by
destroying the entire environment. While there is no way to escape the full
cost of creating such disposable environments, Spinnaker can help out by
mitigating their time to live.

Case 3: Database Migrations with Spinnaker

Spinnaker is more than just a tool for deploying applications. It can be
used to automate infrastructure-related tasks. For example, using
Spinnaker pipelines you can accomplish the painfully difficult task of
schema migrations — often just called database migrations. Most

CI/CD WITH KUBERNETES 75

CONTINUOUS DELIVERY WITH SPINNAKER

organizations perform database migrations as a pseudo-manual process.

Some may attempt to include a schema change as part of application
start up. However, when you have a running application attempting a
canary release, this could lead to indeterminate data representation. As a
result, database migrations are considered curious special cases that
often lead to “one off” deployments.

Spinnaker solves this by enabling database migrations to run directly in a

release pipeline. Before the deploy stage for your application, you can
employ a stage to perform the data migration task, something a database
administrator (DBA) would typically have to perform. You can even include
a “check schema for changes” step in your deployment pipeline on every
release. This makes the obscure one-off task that is operationally
expensive just as cheap and easy as deploying a new container image.
Automating database migrations this way also provides a level of
repeatable reliability; the alternative — custom scripts or manual steps —
are typically much more brittle.

Spinnaker as the Standard

We often get the question as to whether Spinnaker is going to emerge as
the standard CD tool going forward. It is hard to predict, but that hasn’t
stopped us from having an opinion based on experience and the general
landscape we see in software development. The short answer is that we
believe it will emerge as the preeminent continuous delivery tool, though
it will take some time.

Spinnaker already has the backing of some large, forward-thinking

companies which are top contributors to the project, including Google,
Microsoft, Netflix, Oracle and Pivotal. And it is one of the first CD tools
built from the start to allow DevOps teams to truly take advantage of
cloud-native architectures to improve efficiency, speed and agility. But it

CI/CD WITH KUBERNETES 76

CONTINUOUS DELIVERY WITH SPINNAKER

is still early days. We have seen other tools begin to evolve to include
cloud-native CD capabilities and more will follow. In addition, any
contender in the space will need to address the need to integrate with, or
incorporate, build tooling to allow for complete application lifecycle
management. Spinnaker is the closest we’ve seen that delivers the scaling
capabilities and features needed in a cloud-native, continuous delivery
tool but much development work is still needed to get it to a turnkey,
enterprise-ready state.

A Cultural Shift Towards Delivery

Most companies are just getting started on their journey towards
transforming into a DevOps-based organization. For these companies,
delivery is still not seen as a true priority or a business differentiator. At
Kenzan, we push the notion that speed and accuracy of deployments
should be just as important for a product as any actual feature.
Companies are generally slow to adopt this mentality.

Focusing on deployments is a cultural shift for companies that involves

organizational change. It involves changing team structures, changing
processes and changing culture, all of which take a longer period to root
and grow. And while companies may be aware of the ability to automate
deployments through CI tools like Jenkins, they may initially overlook a
long-term strategic plan for achieving faster, automated deployments and
the benefits a fully-featured CD tool like Spinnaker can provide. Because
of this, adopting a truly deployment-focused platform like Spinnaker may
not be realized for some time.

A Complexity Shift in Tooling

On the technology side of the equation, the initial versions of Spinnaker
proved overly complex and involved an elaborate setup. The Spinnaker
team is now working very hard to simplify the setup of Spinnaker, but the
early versions did make it less approachable for the community to adopt.

CI/CD WITH KUBERNETES 77

CONTINUOUS DELIVERY WITH SPINNAKER

Fortunately, the current Kubernetes installation and setup via Helm is

probably the easiest method yet to configure and set up Spinnaker. This
has made it much more approachable for those integrating with
Kubernetes, yet the general complexity of Spinnaker remains a boundary
for many organizations.

In general, we’ve gradually seen complexity shifting away from building

and assembly of code towards orchestration of releases. Build tools are
starting to commoditize and become much simpler, such as Travis CI and
Jenkins. This commoditization has been driven by the simplification of
using immutable infrastructures, particularly Docker containers, where
building can happen directly in the container environment that it will run
in, and the entire container is then promoted through the pipeline. As
more and more organizations get comfortable with building custom code
using containers and other immutable constructs, they will spend less
cycles on building that code and shift into solving the problems of
orchestrated releases. We are helping many of our customers see an
“orchestration-first” mindset when thinking about software. When
approaching a new development, we start by considering how it will be
deployed and automated before we consider how it is built or even the
architecture within the features. From this viewpoint, building code
becomes just another implementation decision. We eventually foresee
most organizations taking this view, and once the shift takes place,
Spinnaker will be well-suited for the new wave of users who will require
the orchestration-native capabilities that Spinnaker offers. There are other
tools out there, such as Jenkins Pipelines, that can do basic orchestration,
but they don’t scale or have nearly the capability of Spinnaker.

Standing Out in a Crowd

Overall, Spinnaker has jumped into what might appear to be a crowded
CI/CD marketplace. It includes a number of alternatives, alongside tools

CI/CD WITH KUBERNETES 78

CONTINUOUS DELIVERY WITH SPINNAKER

like the aforementioned Jenkins 2.0. Most of the solutions are CI tools
that have been extended with CD capability, with a few others that focus
on CD only.

• AWS CodePipeline is a CD service built specifically for AWS. It allows

the creation of pipelines for building and testing code with AWS
CodeBuild and then deploying applications. It also integrates with AWS
CloudFormation, allowing you to deploy complete application stacks.
It provides a single UI for deployments, but you need to use other AWS
features to get full visibility.

• CircleCI is a CI-centered tool that has expanded to pipeline

capabilities. It is available hosted and self-hosted. Its greatest
capabilities reside in not trying to mimic the architecture of legacy CI
tools like Jenkins. Being able to support a hosted configuration but
still isolate build runtimes make CircleCI an enterprise-grade solution
where runtime isolation is important. In addition to this, CircleCI
makes getting started with CI/CD easy.

• GitLab CI is the leading GitHub clone that an organization can deploy

themselves. If you are looking for a self-hosted Git and are not opting
for Bitbucket, GitLab is a good choice, particularly with the CI/CD
capabilities that GitLab 8+ provides. As the name suggests, GitLab CI
is more about CI, but one can use GitLab’s pipeline capabilities for
CD tasks. With these pipelines one can sequence a code build, a
Docker build and deploy the image to Kubernetes. However, these
are all file-based configurations that require familiarity with
Kubernetes manifests.

• Harness is a Software as a Service (SaaS) continuous delivery tool

that can integrate into your custom environments. Recent additions
to Harness have allowed it to start leveraging artificial intelligence

CI/CD WITH KUBERNETES 79

CONTINUOUS DELIVERY WITH SPINNAKER

(AI) to predict and see anomalies in your environments. The benefit

of SaaS is it will abstract away the complexity of establishing your
own CD tooling, but you could be limited in customizations to the
specific tools.

• Jenkins X is an exciting and new open source CI/CD toolset that was
released in March of 2018. It takes many of the core features from
Jenkins and enhances them to make them cloud native. It is built for
Kubernetes and is designed and optimized for deploying into that
environment. Jenkins X attempts to simplify the process of CI/CD by
automatically producing a number of predefined things: Jenkinsfiles,
Dockerfiles, Helm Charts, Kubernetes Clusters, namespaces and even
environments. It also uses predefined automation to trigger builds and
deploys from Git commits. While Jenkins X is showing a lot of promise
and adding functionality very quickly, it is still maturing as a tool. Most
notably it only runs from the command line, does not yet have a UI for
managing deployments, and is not easily used for managing the code
and infrastructure underlying the Kubernetes clusters (e.g. load
balancers, images and DNS).

The above tools include the most popular and relevant alternatives on the
market; there are a number of other CI/CD solutions that we have not
mentioned. What is important to realize is that tools that began as CI
solutions are now being adapted to do what Spinnaker has been built to
do from the beginning: focus on cloud-based deployments. While some CI
tools will likely provide a base level of capability for deployments, they
may not provide the scalability, extensibility and community support that
Spinnaker has proven itself with. And most importantly, it may take a long
time before these tools match Spinnaker’s capability as a true state
machine: allowing an operator to automate changes, perform canary
deployments, roll them back and scale instances, all while having a high-

CI/CD WITH KUBERNETES 80

CONTINUOUS DELIVERY WITH SPINNAKER

level view of which applications and infrastructure are in place. When

examining the CD capability of any tool, it is important to remember the
ideal CD goal of having a state machine that enables fast deployments for
your organization.

Build Tooling as Part of Spinnaker

It is possible that the Spinnaker community will attempt to roll its own CI
build tool that will sit alongside or be a part of Spinnaker. Whether it
replaces Jenkins or other tools is not as important as the opportunities it
would create. Using two tools to do the build and orchestration is often a
barrier for adopting Spinnaker, and people end up reverting to what they
know in tools like Jenkins. If a CI solution is built into Spinnaker, it would
open the door to doing the entire build and orchestration process with
one set of tooling.

Spinnaker Best Practices

Over the years, we’ve picked up a number of lessons learned that we try to
leverage for engagements where we are moving clients to a CD pipeline
with Spinnaker and Kubernetes.

1. Ensure Resilience
It is important to put as much emphasis in ensuring resiliency for your
Spinnaker stack as you would for any application in your infrastructure.
Ensuring uptime of your deployment tooling is as important as any feature
or application. Fortunately, Spinnaker is already built with resiliency in
mind, and it typically only requires configuration to achieve your specific
needs. We suggest the following setup:

• Multiregion: If you have a multiregion setup, then ensure that each

cluster has a full set of Spinnaker components.

• Data Replication: Depending on the data stores that you are using

CI/CD WITH KUBERNETES 81

CONTINUOUS DELIVERY WITH SPINNAKER

for your components — Cassandra and Redis — it is important to

configure them to replicate across regions and datastores. We typically
find that the out-of-the-box configuration for data replication, such as
Active Active, Active Passive and Eventual Consistency, tends to meet
the needs of our installs.

• Chaos Engineering: Spinnaker integrates very nicely with chaos

monkeys, and we find this very useful in our application pipelines to
ensure that they are developed with the proper level of resilience.

We recommend performing chaos engineering and testing on your

Spinnaker setup as well. This will show that it can handle failures so
that you can be sure that your deployment pipelines will not be
impacted by outages.

2. Employ Namespacing
We typically put our Spinnaker installation in a separate namespace within
Kubernetes. This permits us the ability to size the Spinnaker resources at
the entire namespace level, and also prevents against Spinnaker resources
taking away from other namespaces or vice versa. Having partitioned
Spinnaker off, we can then closely monitor the needed resources for our
deployment microservices.

From within its own namespace, it should be noted that Spinnaker can
deploy into other specific namespaces. This is a very nice ability that
allows deployments to only target one namespace at a time.

3. Monitor Spinnaker
Monitoring Spinnaker is as important as monitoring any application in
your infrastructure. Spinnaker currently supports three major monitoring
systems — Datadog, Prometheus and Stackdriver — but others could be
added relatively easily. We recommend using the same monitoring tools
that you use for the rest of your infrastructure. The next chapter will go

CI/CD WITH KUBERNETES 82

CONTINUOUS DELIVERY WITH SPINNAKER

into more detail on monitoring with Prometheus.

• Understanding Monitoring: Spinnaker, by design, has very powerful

monitoring capabilities, but understanding how it works is critical.
Each microservice is instrumented to capture and expose events and
metrics, while separate daemons exist to gather this data and push it
to the monitoring tool of choice. For a deeper dive, see how monitoring
works within Spinnaker.

• Dashboards: We typically use the supplied dashboards that are

provided by our monitoring tool, such as Prometheus, but focus on
two main views into Spinnaker. The first is the health of Spinnaker
overall. This includes data from the clusters, nodes and namespaces
running Spinnaker and the health of any integration points such as
Jenkins. The second view is monitoring each microservice individually.
We typically look for error rates and latency within the microservices.

• Weigh Value Differential Over Rates: A monitoring practice that

we took from the Spinnaker team was to look for trends in data and
to not care as much about the current rate or count. This has proven
to be very good advice as the volatility in load is very high in a
delivery platform.

4. Version Your ConfigMaps and Secrets

If you are using ConfigMaps and Secrets that are outside of your Docker
containers, then it is very important to version them. Without versioning
them, you will not be able to roll back in a Spinnaker pipeline should the
configuration prove bad. It will likely result in a “roll forward” situation that
will require a new build and new configurations.

5. Use Canary Deployments

While Spinnaker supports many deployment strategies within pipelines,
Kenzan is keen on using the canary strategy. This allows us to slowly roll

CI/CD WITH KUBERNETES 83

CONTINUOUS DELIVERY WITH SPINNAKER

out the feature and test it directly in production. We typically use some
form of automated canary analysis (ACA) to monitor the health of the
deployment and compare logs files of the newly deployed to the previous
deployment. Fortunately, this is getting even easier to accomplish now
that Google and Netflix have open sourced Kayenta, which integrates with
Spinnaker to monitor the quality of canary deployments. This was
released in April of 2018 and will make achieving continuous delivery —
and then continuous deployment — easier than ever.

6. Use Feature Flags

We tend towards using feature flags as a design practice for applications.
This gives the ability to still deliver the feature when ready, but turn it on at
a later date and not create a backlog of items waiting to be deployed via
Spinnaker.

7. Use Pipeline Templates

Using Spinnaker Pipeline Templates allows you to standardize a set of
pipeline components, but still give each team some flexibility to decide
which parts they need and even the order of events. This is a very powerful
way to ensure that reuse is happening across all of your pipelines.

Broader CD Best Practices

The move to using Spinnaker isn’t simply about putting a piece of
technology in place. It will likely go hand-in-hand with organizational shifts
in thinking as well as enacting some key CD practices.

1. Mindset Shift
The overall organization needs to support CD. This means that the
business, development and operations need to align their practices to
support continuous delivery. Over the years, we have seen several
common themes emerge that are key to this journey:

CI/CD WITH KUBERNETES 84

CONTINUOUS DELIVERY WITH SPINNAKER

• Focus on Releasing Code and Not Building Code: This has been
mentioned a couple of times but it is probably worth emphasizing
again. It is important that everyone sees quality and speed of delivery
as the most important aspect of any application.

• Fund as a Product and Not a Project: Too many organizations are

still funding and building budgets around projects. Each project can
have many features, and sometimes these budgets won’t allow
features to be released independently. To ensure that releases are
given priority, funds need to flow into a product as a whole.

• Establish a Culture of Automation: Your organization should

support a culture of automation. This involves looking at new
automation technologies, but also avoiding any infrastructure or
application code that cannot be automated. This will kill the
automation of your pipelines.

• You Build it You Own it: Creating ownership with the team that built
the code is paramount in our experience. Every piece of code —
microservice, pipeline and infrastructure — should have a clear owner,
and we find it best not to create separations between the teams that
manage and the teams that build the code. These types of separations
typically end up creating handoffs that halt or impede the speed and
automation CD pipelines are inherently designed for.

2. Infrastructure as Code
Automating everything is made much easier if your infrastructure is
managed as code and configuration. This will make it much easier to create
consistencies and parity across all of your environments. Accomplishing
this is dependent on your specific implementation, but we typically use
some sort of scripting, such as with Terraform, and also employ dedicated
infrastructure pipelines to automate the deployment. This ensures that

CI/CD WITH KUBERNETES 85

CONTINUOUS DELIVERY WITH SPINNAKER

individuals are never manually hand tweaking environments without the

proper checks and balances in place.

3. Fail Early with the Testing Pyramid

Automated testing and failing early are important to speeding up delivery
pipelines. At Kenzan, we subscribe heavily to the testing pyramid, where
the bulk of tests occur at the base with unit and integration tests, and
further up we use fewer end-to-end and UI tests for basic smoke testing.
End-to-end and UI tests tend to be the slowest and most brittle. By running
complex logic in the faster running unit tests and integration tests, this
allows us to fail quickly and catch bugs earlier in the lower environments.

4. Consistent Branching Model

While Spinnaker is certainly capable of handling multiple Git branching
models, we have found it easier to manage if all microservices and
applications are using a consistent branching and versioning model across
all of your code bases. Without a consistent branching model, you wind
up adding different branch triggers in Spinnaker for different applications
and potentially even different versioning and tagging. This burdens
development with unnecessary complexity, in our experience.

5. Backwards Compatible Changes

Every change must be backwards compatible! Period. As soon as you
start releasing features that are not backwards compatible, the rollback
strategies and coupling becomes very difficult. This should never happen.

6. Start Small
You won’t achieve full CD in one week. The process will take time to shift
over all projects. When moving towards implementing CD in an
organization, we typically start very small with a single application or
group. This allows you to prove out the ideas and organizational changes
on a smaller scale and then roll them out to a larger group.

CI/CD WITH KUBERNETES 86

CONTINUOUS DELIVERY WITH SPINNAKER

Spinnaker Tutorial
We hope this chapter has piqued your interest in Spinnaker as a CD
platform. If you’d like to get started with the tool, Kenzan has an open
source repository that we use to set up a fully functioning CD environment
using Spinnaker (with Jenkins) running within a Google Kubernetes Engine
(GKE). The tool uses some simple Terraform scripts to set up and configure
your environment, and is a great way to begin examining Spinnaker
hands-on. Check it out at: https://github.com/kenzanlabs/capstan

CI/CD WITH KUBERNETES 87

A NEW APPROACH TO
DEVOPS WITH SPINNAKER
ON KUBERNETES
As organizations look to DevOps as a means to
achieve digital transformation, they realize they
must accelerate the end-to-end software delivery
process, and also make it safer.

“It can seem like a boil-the-ocean type of problem. That makes it

hard to figure out how you’re going to incrementally derive value
from it,” said Andrew Phillips, a product manager in Google Cloud
Platform’s DevOps division.

By separating the developer feedback cycle from the rollout

process, organizations can find a manageable starting point. Tools
like Spinnaker were built for this purpose, Phillips said, “to provide
an abstraction so that development teams can have a simplified
experience, while still providing the operations teams with the ability
to manage and tweak and define it in exactly the way that makes
the most sense for the organization.” Listen on SoundCloud »

Andrew Phillips is a project manager at Google Cloud Platform and

a frequent contributor to the continuous delivery and DevOps space.
He’s been a software engineer, team lead, infrastructure builder
(a.k.a. head of duct tape) and community evangelist, and contributes to a
number of open source projects. He is a regular speaker, author and
co-organizer of ContainerDays Boston and NYC.

CI/CD WITH KUBERNETES 88

MONITORING IN THE
CLOUD-NATIVE ERA
by IAN CROSBY, MAARTEN HOOGENDOORN, THIJS SCHNITGER AND
ETIENNE TREMEL

o gain insight into cloud-native systems means knowing what to

T observe, more than anything else. It’s not solely about

monitoring, alerting, tagging and metrics. Rather, these four
capabilities combined allow DevOps engineers to observe scaled-out
applications running on containers in an orchestrated environment such
as Kubernetes.

Some early adopters of cloud-native technologies refer to observability as

the new monitoring. The rising demand for observability is real and it
comes from the legitimate need to understand raw data produced by the
complex infrastructure and multitude of components that run and
comprise cloud-native applications. It will gain prominence as more
organizations deploy cloud-native applications. A granular understanding
of the underlying microservices architectures will lead organizations to
accept and embrace concepts pertaining to observability. The interest will
drive demand for deeper capabilities to collect data using time-series
databases, which in turn will lead to better observability.

CI/CD WITH KUBERNETES 89

MONITORING IN THE CLOUD-NATIVE ERA

Monitoring has a different meaning today. In the past, developers built the
applications. Deployments were done by operations teams who managed
the applications in production. The health of services was mainly
determined based on customer feedback and hardware metrics such as
disk, memory or central processing unit (CPU) usage. In a cloud-native
environment, developers are increasingly involved in monitoring and
operational tasks. Monitoring tools have emerged for developers who use
them, to set up their markers and fine tune application-level metrics to
suit their interests. This, in turn, allows them to detect potential
performance bottlenecks sooner.

Developers are applying techniques like continuous integration/

continuous delivery (CI/CD) to optimize programmable and immutable
infrastructure. As new infrastructure increasingly becomes available, so
does the demand for DevOps professionals and people with site
reliability engineering (SRE) experience. Observability gives engineers
the information they need to adapt systems and application
architectures to be more stable and resilient. This, in turn, provides a
feedback loop to developers which allows for fast iteration and
adaptation to changing market conditions and customer needs. Without
this data and feedback, developers are flying blind and are more likely
to break things. With data in their hands, developers can move faster
and with more confidence.

The integration of a graph database in a cloud-native monitoring tool,

such as Prometheus, lends such tools some considerable staying power.
By capturing data that can be viewed as a graph, in relation to time, such
tools allow developers to observe applications with more granular detail.
Graph databases will increasingly serve as a way to gain deeper visibility,
and that enables any number of monitoring use cases. The outcome is
deeper efficiencies in the application architecture. Organizations can

CI/CD WITH KUBERNETES 90

MONITORING IN THE CLOUD-NATIVE ERA

begin to construct self-remedying application architectures on diverse

infrastructure environments.

In this new cloud-native era, continuous understanding about an

infrastructure’s state of health defines how applications are built,
deployed and managed. It determines how components are modified
automatically in an elastic manner — up or down — depending on the
load. It tells the operator how to make a decision about a failover or the
rollback of a service. Cloud-native systems, by their very nature, are
ephemeral and short-lived. They may fail at any time, triggering new
events. They scale fast, requiring new monitoring capabilities to cover a
range of situations. Cloud-native monitoring must not treat any specific
component independently, but rather focus on the aggregate functions
that these components together are supposed to perform.

The topic of observability is fairly new, but highly pertinent. Our authors
are software engineers who have studied the new monitoring approaches
that are emerging with cloud-native architectures. Ian Crosby, Maarten
Hoogendoorn, Thijs Schnitger and Etienne Tremel are experts in
application deployment on Kubernetes for Container Solutions, a
consulting organization that provides support for clients who are doing
cloud migrations. These engineers have deep experience with monitoring
using Prometheus, which has become the most popular monitoring tool
for Kubernetes, along with Grafana as a visualization dashboard.

Monitoring in a cloud-native environment needs to move beyond checks

on the state of a resource; it must consider other factors besides “my
HTTP service is responding,” or “my disk is at 60 percent capacity.” The
cloud-native monitoring environment must provide insight into how a
service’s state is related to the state of other resources. This, in turn, must
point to the overall state of the system, which is reflected in error
messages that encompass multiple considerations, with statements such

CI/CD WITH KUBERNETES 91

MONITORING IN THE CLOUD-NATIVE ERA

as “HTTP service response times are increasing beyond threshold, and we

can’t scale up because we have hit CPU resource limits.” It’s these types
of insights that enable the system to change based on informed
decisions. These insights define observability and move DevOps teams
further along the path toward a true CI/CD feedback loop. Without deeper
insights, problems can go unseen. In this cloud-native era with scalability
in mind, monitoring is one of many factors that come with the broader
practice of observability.

Observability is About Context

Cloud native means a lot more than just hosting services on the cloud
with unlimited computing capacity. A cloud-native system is composed of
applications assembled as microservices. These loosely coupled services
run across cloud providers in data centers around the globe. There may
be millions of these services running at the same time, producing a hive of
interactions. At such scale it becomes impossible to monitor each one
individually, let alone their interdependencies and communications.

At scale, context is important. It shows how separate events in a system

relate to each other. The understanding of this interrelationship serves as
the foundation for building a model that helps determine how and why a
system is behaving in a particular manner. It is not just a matter of
gathering as much data as possible, but collecting meaningful data that
adds to an understanding of the behavior. Visualizing data and metrics,
and tracing events as they flow from one component to the next, is now a
reality of monitoring microservices environments. If monitoring is about
watching the state of the system over time, then observability is more
broadly about gaining insight into why a system behaves in a certain way.

Observability stems from control theory, where it serves as a measure of

how well internal states of a system can be inferred from knowledge of its

CI/CD WITH KUBERNETES 92

MONITORING IN THE CLOUD-NATIVE ERA

external outputs. For a modern SRE, this means the ability to understand
how a system is behaving by looking at the parameters it exposes through
metrics and logs. It can be seen as a superset of monitoring.

According to Twitter, one of the pioneering companies in web-scale

computing and microservices, there are four pillars to observability:

1. Logging.

2. Monitoring and metrics.

3. Tracing.

4. Alerting and visualization.

Collecting, storing and analyzing these new types of application

performance data raises new challenges.

Application Performance Management

The dynamic nature of cloud-native systems poses new challenges for
application performance management (APM). First of all, cloud-native
systems are by definition more transient and complex than traditional
systems. The components making up the system are no longer static, but
ephemeral — appearing on demand and disappearing when they are no
longer needed. A sudden increase in demand might lead to certain
components being scaled in large numbers. Any APM solution needs to be
able to accommodate these rapid and numerous changes.

Components in a cloud-native system also tend to change more often

with the increased use of continuous deployment techniques. This creates
the necessity for logging and metrics to be linked not only to the state of
the system at a certain point in time, but also the changes in the software
leading up to that state. Also, the increased number of components vastly

CI/CD WITH KUBERNETES 93

 MONITORING IN THE CLOUD-NATIVE ERA

increases the amount of data and metrics being logged. This increases
demand for storage and processing capacity when analyzing these data
and metrics. Both of these challenges lead to the use of time-series
databases, which are especially equipped to store data that is indexed by
timestamps. The use of these databases decreases processing times and
this leads to quicker results.

These large amounts of data also allow for gaining insights by applying
principles of artificial intelligence and machine learning. These techniques
can lead to increased performance, because they allow the system to
adapt the way it changes in response to the data it’s collecting by learning
from the effect of previous changes. This in turn leads to the rise of
predictive analytics, which uses data of past events to make predictions
for the future, thereby preventing errors and downtime.

FIG 4.1: In cloud-native systems, observability is the new monitoring.

The Four Pillars of Observability

Recording of
discrete events.

Logging
+ +

Notification when event

behavior falls outside = Aggregation of similar
of acceptable threshold
and could potentially
Alerting Observability Metrics events to gain a higher
level of insight.
become problematic.

+ +
Tracing

Recording, ordering and

CI/CD WITH KUBERNETES binding of data from connected 94
events to provide context.
Source: https://www.weave.works/technologies/monitoring-kubernetes-with-prometheus/ © 2018
MONITORING IN THE CLOUD-NATIVE ERA

The Four Pillars

Observability, and its resulting insights, comes from logging, metrics,
tracing and alerting. The value that comes from these pillars of
observability derives from using well-defined terms and clearly identifying
the purpose of each pillar. Data is captured from each pillar and used for
later evaluation. Let’s take a simple example, a 500 error, and see how
DevOps engineers would gain insight through each lens.

Logging
Logging in the simplest sense is about recording discrete events. This is
the first form of monitoring which any new developer gets exposed to,
usually in the form of print statements. In a modern system, each
application or service will log events as they occur, be it to standard out,
syslog or a file. A log aggregation system will then centralize all logs to be
viewed or searched as needed. In our example of a 500 error occuring, this
would be visible by a service, or possibly multiple services, logging an
error which resulted in the 500 status code. This error can be deciphered
through an evaluation of the other three pillars.

Metrics
By contrast, metrics are a combination of data from measuring multiple
events. Cloud-native monitoring tools cater to different types of
measurements by having various metrics such as counters, gauges,
histograms and meters.

• Counter: A counter is a cumulative metric that can only ever increase;

for example, requests served, tasks completed and errors occurred.
It should not be used for metrics that can also go down, such as
number of threads.

• Gauge: A gauge is a metric that can arbitrarily go up and down; for

CI/CD WITH KUBERNETES 95

MONITORING IN THE CLOUD-NATIVE ERA

example, temperature, memory usage and live number of users.

• Histogram: Histograms measure the statistical distribution of a set of

events; for example, request duration and response size. Histograms
track the number of observations and the sum of the observed values,
allowing a user to view the average of the observed values.

• Meter: Measures the rate at which an event occurs. The rate can be
measured over different time intervals. The mean rate spans the
lifetime of your application, while one-, five- and fifteen-minute rates
are generally more useful.

The idea is to aggregate similar events to gain a higher level of insight.

Metrics are generally time based, therefore we usually collect metrics
periodically, such as once per second. In our 500 error example, we can
see the rate of 500 errors which a particular service is omitting. If we have
a consistent rate of 500 errors, this would point to a different problem
than a sudden spike of 500s would.

Tracing
Tracing is about recording and ordering connected events. All data
transactions, or events, are tied together by injecting a unique ID into an
initial request, and passing that ID to all further events through the
system. In a distributed system, a single call will end up passing through
multiple services. Tracing provides a complete picture at the application
level. Again, coming back to our example of a 500 error response, we can
see the entire flow of the specific request which resulted in a 500. By
seeing which services the request passed through we gain valuable
context, which will allow us to find the root cause.

Alerting
Alerting uses pattern detection mechanisms to discover anomalies that
may be potentially problematic. Alerts are made by creating events from

CI/CD WITH KUBERNETES 96

MONITORING IN THE CLOUD-NATIVE ERA

data collected through logging, metrics and tracing. Once engineers have
identified an event, or group of events, they can create and modify the
alerts according to how potentially problematic they may be. Returning to
our example: How do we start the process of debugging the 500 error?
Establish thresholds to define what constitutes an alert. In this case, the
threshold may be defined by the number of 500 errors over a certain
period of time. Ten errors in five minutes means an alert for operations
managed by Container Solutions. Alerts are sent to the appropriate team,
marking the start of the debugging and resolution process. Take into
consideration that what constitutes an alert also depends on what the
normal state of the system is intended to be.

By establishing the four data pillars, observability is gained into the

system and the cloud-native applications it runs. Complexity will only
increase as the system is developed, requiring more observability that
comes from collecting more data in a manner that can be stored and
analyzed, providing a feedback loop for deeper optimizations and the
proper insights into applications.

Monitoring Patterns
Of the four pillars, metrics provide the most insight into how an
application performs. Without metrics, it is impossible to tell if an
application behaves the way it should in order to meet service-level
objectives. There are different strategies used to collect and analyze
metrics in order to report the health of cloud-native systems, which is the
foremost concern.

Blackbox and whitebox monitoring are two different strategies used to

report the health of a system. Both rely on different techniques which,
when combined, strengthen the reliability of the report.

CI/CD WITH KUBERNETES 97

MONITORING IN THE CLOUD-NATIVE ERA

Blackbox monitoring is a method to determine the state of a system

without having access to the application internals. The type of metrics
collected provide information about the hardware such as disk, memory
and CPU usage or probes — Transmission Control Protocol (TCP), Internet
Control Message Protocol (ICMP), Hypertext Transfer Protocol (HTTP), etc.
A health check is a typical example of blackbox monitoring. It determines
the status of a system by probing different endpoints using a particular
protocol such as TCP or ICMP. If a probe is successful then the application
is alive, otherwise we can assume that the system is down without
knowing the exact cause.

In contrast to blackbox monitoring, whitebox monitoring is more

sophisticated and relies on telemetry to collect application behavior
metrics, such as the total number of HTTP requests and latencies, or the
number of errors or runtimes specific via interfaces, like Java Virtual
Machine Profiling Interface (JVMPRI). In order to monitor an application
properly, this information must be specified and it’s up to developers to
instrument it with the right metrics.

Blackbox and whitebox monitoring are two patterns that complement

each other to report the overall health of systems. They play an important
role in cloud-native systems where modern SREs interpret these metrics
to identify server performance degradation and spot performance
bottlenecks early on.

Performance Metrics and Methodology

In a cloud-native environment and with complex distributed systems, it
takes time and effort to discover what caused a failure. Only a handful of
methodologies exist, which are intended to be simple and fast in order to
help SREs come to a conclusion. Each method relies on one of the
following key metrics:

CI/CD WITH KUBERNETES 98

MONITORING IN THE CLOUD-NATIVE ERA

• Error: rate of error events produced.

• Latency: duration of a request.

• Utilization: how busy the system is.

• Saturation: the threshold at which a service cannot process extra

work.

• Throughput: rate or quantity at which the system is being requested.

From these metrics you can apply one of the following four methodologies
to determine how performant the system is:

• USE (utilization, saturation and errors): This technique, developed by

Brendan Gregg, is a resource-oriented method which is intended to
detect resource bottlenecks in a system under load. It relies on three
metrics: utilization, saturation and errors.

• TSA (thread state analysis): This method is complementary to the USE

method. Also developed by Brendan Gregg, it focuses on threads
instead of resources and tries to find which state takes the most time.
It relies on six key sources of performances issues: executing, runnable,
anonymous paging, sleeping, lock and idle.

• RED (rate, errors and duration): This method is aimed at request-

driven services. Like TSA, the RED method is complementary to the
USE method and relies on three key metrics: rate, errors and duration.

• Golden signals: This method was promoted by the Google SRE team
and relies on four key metrics to determine the state of a system:
latency, throughput, errors and saturation.

In any given system, if the right metrics are collected, engineers — even if
they’re not aware of the entire architecture of the system they use — can

CI/CD WITH KUBERNETES 99

MONITORING IN THE CLOUD-NATIVE ERA

apply one of these methodologies to quickly find out which part of a

system has the potential to become a performance bottleneck and
cause failure.

Anomaly Detection
In modern production systems, observability is a core feature which is
needed to detect and troubleshoot any kind of failure. It helps teams
make decisions on actionable items in order to return the system to its
normal state. All the steps taken to resolve a failure should be
meticulously recorded and shared through a post-mortem, which can be
used later on to speed up the resolution time of recurrent incidents.

Recovery procedures that used to be handled by an operations team are

now handled by container orchestrators. When the recovery procedure is
more complex, additional tooling can be developed to automate the
recovery steps, which bring the system back to its normal state. The
decision to recover can be triggered based on metrics and threshold, or
some other predictive mechanism such as machine learning.
Implementing a self-healing capability based on recurrent problems is a
first step toward making a system resilient. Observations and the
resolution described in a post-mortem can be translated into actionable
items for future decision-making.

Analytics can tell a lot about the behavior of a system. Based on historical
data it is possible to predict a potential trend before it becomes a
problem. That’s where machine learning comes into play. Machine
learning is a set of algorithms which progressively improves performance
on a specific task. It is useful to interpret the characteristics of a system
from observed behavior. With enough data, finding patterns that do not
conform to a model of “normal” behavior is an advantage, which can be
used to reduce false positive alerts and help decide on actions that will

CI/CD WITH KUBERNETES 100

 MONITORING IN THE CLOUD-NATIVE ERA

attempt to bring the system back to its normal state.

Supervised and unsupervised are two types of anomaly detection. A

model labelled either normal or abnormal is called supervised. It gets its
name when a dataset is used to train a model and then subsequently gets
labelled. It comes with limitations, since labelling can be difficult and
expensive. In contrast, unsupervised detection doesn’t identify faulty
datasets. Based on events that rarely occur or don’t repeat, it is possible
to classify them as abnormal by using standard inference in Bayesian
networks and compute a rank using probability.

Since this can be complex to integrate into a monitoring system, a simpler

approach is to make use of triple exponential smoothing, also known as
the Holt-Winters method. This has the potential to deliver accurate
predictions since it incorporates seasonal fluctuations into the model. It is

FIG 4.2: The Holt-Winters method has the potential to deliver accurate predictions
since it incorporates seasonal fluctuations to predict data points in a series over time.

The Holt-Winters Method of Anomaly Detection

Abnormal behaviors
Event Data

Time
CI/CD WITH KUBERNETES 101

Source: https://docs.google.com/presentation/d/16hsV0Pyxxk7ta7gIif7Tfxcr1Oy1LmfUbKXaeXisDvw/edit#slide=id.g1174bd54a0_0_137 © 2018

 MONITORING IN THE CLOUD-NATIVE ERA

one of the many methods that can be used to predict data points in a
series over time. Figure 4.2 provides an overview of what the Holt-Winters
method evaluates.

Push vs. Pull Data Collection

We can distinguish two models when it comes to gathering metrics from
an application: push and pull. Many monitoring solutions expect to be
handed data, which is known as the push model. Others reach out to
services and scrape data, which is known as the pull model. In both cases,
developers need to instrument a specific part of their application in order
to measure its performance — time to execute a task, time an external
request takes, etc. — and optimize it later on. Depending on the use case,
one method may be a better fit than the other.

The push model works best for event-driven, time-series datasets. It’s
more accurate, as each event is sent when it’s triggered at the source.
With the push model, it takes some time to tell if a service is unhealthy,
as the instance health is based on the event it receives. The push
FIG 4.3: Many monitoring solutions expect to be handed data, which is known as the
push model. Others reach out to services and scrape data, which is known as the pull
model.

Two Methods to Collect Data

Push (receive data) vs. Pull (scrape data)

Monitoring Monitoring
Service Service

Multiple requests. Can be triggered in parallel.

More accurate, but server needs resource capacity to handle load.

One request every few seconds.

Monitoring Monitoring Monitoring Monitoring Monitoring Monitoring

Agent Agent Agent Agent Agent Agent

Application Application Application Application Application Application

CI/CD WITH KUBERNETES 102

Source: https://www.slideshare.net/OliverMoser5/prometheus-introduction-infracoders-vienna © 2018

MONITORING IN THE CLOUD-NATIVE ERA

model assumes an instance is unhealthy when it cannot be reached to

pull the metrics.

Each serves a different purpose. A pull model is a good fit for most use
cases, as it enforces convention by using a standard language, but it does
have some limitations. Pulling metrics from internet of things (IoT) devices
or browser events requires a lot of effort. Instead, the push model is a
better fit for this use case, but requires a fixed configuration to tell the
application where to send the data.

In a cloud-native environment, companies tend to favor the pull model

over the push model for its simplicity and scalability.

Monitoring at Scale
Observability plays an important role in any large distributed system. With
the rise of containers and microservices, what happens when you start
scraping so many containers that you need to scale out? How can you
make it highly available?

There are two ways to solve this problem of monitoring at scale. The first
is a technical solution to use a federated monitoring infrastructure.
Federation allows a monitoring instance to gather selected metrics from
other monitoring instances. The other option is an organizational
approach to improve monitoring by adopting a DevOps culture and
empowering teams by providing them with their own monitoring tools.
This reorganization could be further split into domains — frontend,
backend, database, etc. — or product. Splitting can help with isolation
and coupling issues that can arise when teams are split by role. By
deciding on roles ahead of time, you can prevent scenarios like, “I’m
going to ignore that frontend alert because I’m working on the backend
at the moment.” A third option, and the best yet, is a hybrid of both

CI/CD WITH KUBERNETES 103

MONITORING IN THE CLOUD-NATIVE ERA

approaches: adopt DevOps and federate some metrics to pull some

top-level service level indicators out of the various monitoring instances.

Federation
A common approach when having a set of applications running on
multiple data centers or air-gapped clusters is to run a single monitoring
instance for each data center. Having multiple servers requires a “global”
monitoring instance to aggregate all the metrics. This is called
hierarchical federation.

Much later, you might grow to the point where your scrapes are too slow
because the load on the system is too high. When this happens you can
enable sharding. Sharding consists of distributing data across multiple
servers in order to spread the load. This is only required when a
monitoring instance is handling thousands of instances. In general, it is
recommended to avoid this as it adds complication to the monitoring
system.

High Availability
High availability (HA) is a distributed setup which allows for the failure of
one or more services while keeping the service up and running at all
times. Some monitoring systems, like Prometheus, can be made highly
available by running two monitoring instances simultaneously. It scrapes
targets and stores metrics in a database. If one goes down, the other is
still available to scrape.

Alerting can be difficult on a highly available system, however. DevOps

engineers must provide some logic to prevent an alert from being fired
twice. Displaying a dashboard can also be tricky since you need a load
balancer to send traffic to the appropriate instance if one goes down. Then
there is a risk of showing slightly different data due to the fact that each
instance might collect data at a different time. Enabling “sticky session” on

CI/CD WITH KUBERNETES 104

MONITORING IN THE CLOUD-NATIVE ERA

the load balancer can prevent such flickering of unsynchronised time

series to be displayed on a dashboard.

Prometheus for Cloud-Native Monitoring

Businesses are increasingly turning to microservices-based systems to
optimize application infrastructure. When done at scale, this means
having a granular understanding of the data to improve observability.
Applications running microservices are complex due to the interconnected
nature of Kubernetes architectures. Microservices require monitoring,
tracing and logging to better measure overall infrastructure performance,
and require a deeper understanding of the raw data. Traditional
monitoring tools are better suited to legacy applications that are
monitored through instrumentation of configured nodes. Applications
running on microservices are built with components that run on
containers in immutable infrastructure. It requires translating complicated
software into complex systems. The complexity in the service-level
domain means that traditional monitoring systems are no longer capable
of ensuring reliable operations.

Prometheus is a simple, but effective, open source solution to that

problem. At its heart, it is a time-series database, but the key feature lies in
its use of a pull model. It scrapes and pulls metrics from services. This
alone makes it robust, simple and scalable, which fits perfectly with a
microservices architecture. Originally developed by SoundCloud for
internal use, Prometheus is a distributed monitoring tool based on the
ideas around Google’s Borgmon, which uses time-series data and metrics
to give administrators insights into how their operations are performing. It
became the second project adopted by the Cloud Native Computing
Foundation (CNCF) after Kubernetes, which allows for some beneficial
coordination between the projects’ communities.

CI/CD WITH KUBERNETES 105

 Monitor as a Service, Not as a Machine
MONITORING IN THE CLOUD-NATIVE ERA
Global server:
Global • Aggregates metrics from Prometheus server instances.
Prometheus • Re-groups compressed time series (aka recording rule).
Server • Leverages load on other instances and hooks up dashboards.

Server aggregates metrics

from Prometheus clients.

Prometheus Prometheus Prometheus

Server Server Server
Client exposes
metrics via
instrumentation
and/or exporter.

Client Client Client Client Client Client

Application Application Application Application Application Application

Client Client Client Client Client Client

Application Application Application Application Application Application

CLUSTER 1 CLUSTER 2 CLUSTER 3

Source: https://www.slideshare.net/brianbrazil/prometheus-overview © 2018

FIG 4.4: Representation of Prometheus in a hierarchical, federated architecture.

Key features of Prometheus are:

• Simplicity.

• Pulls data from services, services don’t push to Prometheus.

• No reliance on distributed storage.

• No complex scalability problems.

• Discovers targets via service discovery or static configuration.

• Powerful query language called PromQL.

Prometheus works well in a microservices architecture. It handles

multidimensional data simply and efficiently. It is also a good fit for
mission-critical systems. When other parts of your system are down,
Prometheus will still be running.

CI/CD WITH KUBERNETES 106

MONITORING IN THE CLOUD-NATIVE ERA

Prometheus also has some drawbacks: Accuracy is one of them.

Prometheus scrapes data and such scrapes are not guaranteed to occur. If
you have services that require accuracy, such as per-usage billing, then
Prometheus is not a good fit. It also doesn’t work well for non-HTTP
systems. HTTP is the dominant encoding for Prometheus, so if you don’t
use HTTP, and instead use Google remote protocol procedure (gRPC), for
example, you will need to add some code to expose the metrics (see
go-grpc-prometheus).

Alternatives to Prometheus
Grafana and Prometheus are the preferred monitoring tools among
Kubernetes users, according to the CNCF’s fall 2017 community survey.
The open source data visualization tool Grafana is used by 64 percent of
organizations that manage containers with Kubernetes, and Prometheus
follows closely behind at 59 percent. The two tools are complementary
and the user data shows that they are most often employed together:
Some 67 percent of Grafana users also use Prometheus, and 75 percent of
Prometheus users also use Grafana.

Kubernetes users often use more than one monitoring tool

simultaneously, due to varying degrees of overlapping functionality,
according to the CNCF survey. Grafana and Graphite are primarily
visualization tools, for example. And Prometheus can be set up to provide
functionality similar to a time-series database, but it doesn’t necessarily
replace the need for one. Among Prometheus-using Kubernetes shops,
InfluxDB’s adoption rate increases slightly, at the same time OpenTSDB’s
use drops several percentage points. CNCF did not ask about many
monitoring vendors’ offerings, such as Nagios and New Relic. However, 20
percent of all the respondents providing an “other” answer mentioned
New Relic. (See the second ebook in this series, Kubernetes Deployment &
Security Patterns, for a more detailed analysis.)

CI/CD WITH KUBERNETES 107

 Grafana and Prometheus Are the Most
Widely Used
MONITORING IN THE Tools for Monitoring
CLOUD-NATIVE ERA Among Kubernetes Users

Grafana 64%
Prometheus 59%
InfluxDB 29%
Datadog 22%
Graphite 17%
Other 14%
Sysdig 12%
OpenTSDB 10%
Stackdriver 8%
Weaveworks 5%
Hawkular 5%

% of Respondents Using Each Monitoring Tool

(select all that apply)

Source: The New Stack Analysis of Cloud Native Computing Foundation survey conducted in Fall 2017.
Q. What monitoring tools are you currently using? Please select all that apply. English n=489; Mandarin, n=187.
Note, only respondents managing containers with Kubernetes were included in the chart. © 2018

FIG 4.5: Grafana and Prometheus are the most commonly used monitoring tools,
with InfluxDB coming in third.

Based on our experience at Container Solutions, here’s our take on some

of the Prometheus alternatives:

• Graphite is a time-series database, not an out-of-the-box monitoring

solution. It is common to only store aggregates, not raw time-series
data, and has expectations for time of arrival that don’t fit well in a
microservices environment.

• InfluxDB is quite similar to Prometheus, but it comes with a

commercial option for scaling and clustering. It is better at event
logging and more complex than Prometheus.

• Nagios is a host-based, out-of-the-box monitoring solution. Each host

can have one or more services and each service can perform one
check. It has no notion of labels or query language. Unfortunately, it’s

CI/CD WITH KUBERNETES 108

MONITORING IN THE CLOUD-NATIVE ERA

not really suited towards microservices since it uses a form of blackbox

monitoring which can be expensive when used at scale.

• New Relic is focused on the business side and has probably better
features than Nagios. Most features can be replicated with open
source equivalents, but New Relic is a paid product and has more
functionality than Prometheus alone can offer.

• OpenTSDB is based on Hadoop and HBase, which means it gains

complexity on distributed systems, but can be an option if the
infrastructure used for monitoring already runs on an Hadoop-based
system. Like Graphite, it is limited to a time-series database. It’s not an
out-of-the-box monitoring solution.

• Stackdriver is Google’s logging and monitoring solution, integrated

with Google Cloud. It provides a similar feature set to Prometheus, but
provided as a managed service. It is a paid product — although Google
does offer a basic, free tier.

Components and Architecture Overview

The Prometheus ecosystem consists of multiple components, some of
which are optional. At its core, the server reaches out to services and
scrapes data through a telemetry endpoint, using the aforementioned
pull model.

Basic features offered by Prometheus itself include:

• Scrapes metrics from instrumented applications, either directly or

via an intermediary push gateway.

• Stores data.

• Aggregates data and runs rules to generate a new time series or

generate an alert.

CI/CD WITH KUBERNETES 109

 Prometheus Ecosystem Components
MONITORING IN THE CLOUD-NATIVE ERA
1
Scrape metrics from Service Discovery Core component
3
instrumented applications, ... or generate an alert.
Optional component
either directly or via an • DNS
intermediary push gateway. • Kubernetes PagerDuty, email,
• Consul etc.
• Other 3
Short-lived
• Custom Integration Aggregate data NOTIFY
jobs
and run rules
to generate a new
FIND TARGETS Alertmanager
time series...

Push gateway PUSH ALERTS

Global Prometheus Server
PULL METRICS 4
API to visualize and
Retrieval PromQL act upon data.
Storage VISUALIZE
Jobs/Exporters
Web UI

2
Grafana
Node HDD/SSD Store data
locally or
Server externally. API clients

Source: https://prometheus.io/docs/introduction/overview/ © 2018

FIG 4.6: Components outside of the Prometheus core provide complementary fea-
tures to scrape, aggregate and visualize data, or generate an alert.

• Visualizes and acts upon the data via application programming

interface (API) consumers.

Other components provide complementary features. These include:

• Pushgateway: Supports short-lived jobs. This is used as a work-

around to have applications push metrics instead of being pulled for
metrics. Some examples are events from IoT devices, frontend
applications sending browser metrics, etc.

• Alertmanager: Handles alerts.

• Exporters: Translate non-compatible Prometheus metrics into

compatible format. Some examples are Nginx, RabbitMQ, system
metrics, etc.

CI/CD WITH KUBERNETES 110

MONITORING IN THE CLOUD-NATIVE ERA

• Grafana: Analytics dashboards to complement the Prometheus

expression browser, which is limited.

Prometheus Concepts
Prometheus is a service especially well designed for containers, and it
provides perspective about the data intensiveness of this new, cloud-
native age. Even internet-scale companies have had to adapt their
monitoring tools and practices to handle the vast amounts of data
generated and processed by these systems. Running at such scale
creates the need to understand the dimensions of the data, scale the
data, have a query language and make it all manageable to prevent
servers from becoming overloaded and allow for increased observability
and continuous improvement.

Data Model
Prometheus stores all of the data it collects as a time series which
represents a discrete measurement, or metric, with a timestamp. Each
time series is uniquely identified by a metric name and a set of key-value
pairs, aka labels.

By identifying streams of data as key-value pairs, Prometheus aggregates

and filters specified metrics, while allowing for finely-grained querying to
take place. Its functional expression language, called PromQL, allows
users to select and aggregate time-series data in real time using the
Prometheus user interface (UI). Other services, such as Grafana, use the
Prometheus HTTP API to fetch data to be displayed in dashboards.

Its mature, extensible data model allows users to attach arbitrary

key-value dimensions to each time series, and the associated query
language allows you to do aggregation and slicing and dicing. This
support for multi-dimensional data collection and querying is a strength,
though not the best choice for uses such as per-request billing.

CI/CD WITH KUBERNETES 111

MONITORING IN THE CLOUD-NATIVE ERA

One common use case of Prometheus is to broadcast an alert when

certain queries pass a threshold. SREs can achieve this by defining alerting
rules, which are then evaluated at regular intervals. By default, Prometheus
processes these alerts every minute, but this can be adjusted by changing
the Prometheus configuration key to: global.evaluation_interval.

Whenever the alert expression results in one or more vector elements at a

given point in time, Prometheus notifies a tool called Alertmanager.

Alertmanager is a small project that has three main responsibilities:

1. Storing, aggregating and de-duplicating alerts.

2. Inhibiting and silencing alerts.

3. Pushing and routing alerts out to external sources.

With Alertmanager, notifications can be grouped — by team, tier, etc. —

and dispatched amongst receivers: Slack, email, PagerDuty, WebHook, etc.

Prometheus Optimization
If used intensively, a Prometheus server can quickly be overloaded
depending on the amount of rules to evaluate or queries run against the
server. This happens when running it at scale, when many teams make use
of query-heavy dashboards. There are a few ways to leverage the load on
the server, however. The first step is to set up recording rules.

Recording rules precompute frequently needed or computationally

expensive expressions and save the result as a new set of time series,
which is useful for dashboards.

Instead of running a single big Prometheus server which requires a lot of

memory and CPU, a common setup adopted by companies running
e-commerce websites is to provide one Prometheus server with little
memory and CPU per product team — search, checkout, payment, etc.

CI/CD WITH KUBERNETES 112

MONITORING IN THE CLOUD-NATIVE ERA

— where each instance scrapes its own set of applications. Such a setup
can easily be transformed into a hierarchical federation architecture,
where a global Prometheus instance is used to scrape all the other
Prometheus instances and absorb the load of query-heavy dashboards
used by the business, without impacting the performance of the
primary scrapers.

Installing Prometheus
Installing Prometheus and its components is really simple. Each
component is a binary which can be installed on any popular operating
system, such as Unix and Windows. The most common way to install
Prometheus is to use Docker. The official image can be pulled from Docker
Hub prom/prometheus. A step-by-step guide to install Prometheus is
available on the Prometheus website.

In a cloud-native infrastructure there is a concept called Operators which

was introduced by CoreOS in 2016. An Operator is an application which
has the capability to set up, upgrade and recover applications in order to
reduce the heavy scripting or manual repetitive tasks — usually defined by
site reliability engineers — to make it work. In Kubernetes, Operators
extend the Kubernetes API through a CustomResourceDefinition, which
lets users easily create, configure and manage complex applications.

The Prometheus Operator — also developed by the CoreOS team —

makes the Prometheus configuration Kubernetes native. It manages and
operates Prometheus and the AlertManager cluster. A complementary
tool, called Kube Prometheus, is used on top of the Prometheus Operator
to help get started with monitoring Kubernetes. It contains a collection of
manifests — Node Exporter, Kube State Metrics, Grafana, etc. — and
scripts to deploy the entire stack with a single command. Instructions to
install the Prometheus Operator are available on the project repository.

CI/CD WITH KUBERNETES 113

MONITORING IN THE CLOUD-NATIVE ERA

Conclusion
Cloud-native systems are composed of small, independent services
intended to maximize resilience through predictable behaviors. Running
containers in a public cloud infrastructure and taking advantage of a
container orchestrator to automate some of the operational routine is just
the first step toward becoming cloud native.

Systems have evolved, and bring new challenges that are more complex
than decades ago. Observability — which implies monitoring, logging,
tracing and alerting — plays an important role in overcoming the
challenges that arise with new cloud-native architectures, and shouldn’t
be ignored. Regardless of the monitoring solution you ultimately invest in,
it needs to have the characteristics of a cloud-native monitoring system
which enables observability and scalability, as well as standard
monitoring practices.

Adopting the cloud-native attitude is a cultural change which involves a lot

of effort and engineering challenges. By using the right tools and
methodology to tackle these challenges, your organization will achieve its
business goals with improved efficiency, faster release cycles and
continuous improvement through feedback and monitoring.

CI/CD WITH KUBERNETES 114

CI/CD WITH JENKINS X
AND KUBERNETES
Kubernetes has made application lifecycle
management much easier. Before Kubernetes,
only the CI half of continuous integration and
continuous delivery (CI/CD) was automated. CD
was assembled by hand with scripts, pipelines, metadata and
configurations. Kubernetes enables CD automation, and tools like
Jenkins X make it simple to deploy on Kubernetes.

“CI/CD should become increasingly like an appliance that just does

CI and CD for you. It’s not this kind of ninja scripty thing that only
experts can figure out,” said James Strachan, chief architect at
CloudBees. “Jenkins X is not so much hiding Kubernetes, not so
much hiding CI and CD and pipelines, but just automating it.”

In this podcast, Michael Neale, CloudBees co-founder & engineering

manager, and Strachan discuss how Kubernetes has changed CI/CD,
evolving workflows such as GitOps, and how CloudBees is working
to improve the developer experience for application deployments
on top of Kubernetes. Listen on SoundCloud »

Michael Neale is co-founder and development manager at

CloudBees. Michael is an open source polyglot developer. For his sins,
he spent time at JBoss, working on the Drools project, and then Red
Hat. He lives and works in the Blue Mountains, just outside Sydney, Australia.

James Strachan is the chief architect of Jenkins X and of its

commercial version, Kube CD. He’s also the creator of the Groovy
programming language and Apache Camel. James is actively
involved with the Jenkins community. Prior to CloudBees, he was with Red Hat
and FuseSource.

CI/CD WITH KUBERNETES 115

CLOSING
The narrative about continuous integration and continuous delivery in
Kubernetes starts with DevOps. It encompasses the new drive for faster
and continuous deployment, and a deeper understanding for how to
manage components running on microservices. The transition to modern,
application-oriented architectures inevitably leads organizations to find
people with the DevOps experience needed to manage Kubernetes and
relevant cloud-native services.

As teams grow, we now see more of this need for declarative

infrastructure. Application architectures built on DevOps practices work
better and run with less friction, but in the end, they just help make the
infrastructure boring. If it is boring, then great — it’s working. Then the
developer has more control over their own resources, and the
performance of the application becomes the primary focus. The better
the performance, the happier the end user and the more uniform the
feedback loop between users and developers. In this way, cloud-native
technologies, such as Kubernetes, provide game-changing business value.

With great execution can come great results. But the scope has changed.
There are historical barriers to overcome that inhibit Kubernetes use,
namely the social issues that surface when people from different
backgrounds and company experiences enter an open source project and
work together. The Kubernetes community is maturing, and defining
values has become a priority as they work to strengthen the project’s
core. Still, the downside to Kubernetes does have to be taken into context
when thinking through longer term business and technical goals. It is
imperative to have a trust in the Kubernetes project as it matures. There
will be conflicts and stubbornness. And it will all be deep in the project,
affecting testing and the ultimate delivery of updates to the Kubernetes
engine. It’s up to the open source communities to work through how the

CI/CD WITH KUBERNETES 116

CLOSING

committees and the Special Interest Groups align to move the project
forward. It’s a problem that won’t go away. Here, too, the feedback loop
becomes critical between users and the Kubernetes community.

In this comes some wisdom to glean about the nature of continuous

delivery and how it may change with the evolution of CI/CD platforms, and
new uses for Git to manage Kubernetes operations. These efforts
encompass discussions about security, identity, service meshes and
serverless approaches to use the resources in further abstracted manners.

Only when the abstraction becomes dysfunctional does true change

come. There has to be a continual feedback loop throughout the build
and deploy cycle to better know how the comparison of time-series
information shows anomalies, for example. Emerging patterns become
the best way to find clues to problems. Feedback loops are also key for
the developer experience, which is critical in order for them to build their
own images in the best manner possible. What’s become obvious to us in
editing this ebook, is that this feedback loop must be present both
between users and the Kubernetes community itself, and within the
organizations that build and deploy their applications on top of it.

Coming up next for The New Stack is a new approach to the way we
develop ebooks. Look for books on microservices and serverless this year
with corresponding podcasts, in-depth posts, and activities around the
world wherever pancakes are being served.

Thanks and see you again soon.

Alex Williams
Founder, Editor-in-Chief
The New Stack

CI/CD WITH KUBERNETES 117

DISCLOSURE
In addition to our ebook sponsors, the following companies are sponsors
of The New Stack:

Alcide, AppDynamics, Aspen Mesh, Blue Medora, Buoyant, CA

Technologies, Chef, CircleCI, Cloud Foundry Foundation, {code}, InfluxData,
LaunchDarkly, MemSQL, Mesosphere, Microsoft, Navops, New Relic,
OpenStack Foundation, PagerDuty, Pivotal, Portworx, Pulumi, Puppet,
Raygun, Red Hat, Rollbar, SaltStack, Stackery, StackRox, The Linux
Foundation, Tigera, Twistlock, Univa, VMware, Wercker and WSO2.

CI/CD WITH KUBERNETES 118

thenewstack.io