tjmdigital, July 21, 2020
Kubernetes (often seen abbreviated as K8s) is an open-source platform built for the automated deployment, scaling and managing of containerized applications.
Google originally created Kubernetes as a small-scale project called ‘Borg’ around 2003, where they managed their own workloads. It was made Open Source in 2014 and presented to the world as ‘Kubernetes’. Kubernetes provides a way to scale your applications and infrastructure without having to scale your operations teams. You’re abstracted away from cloud providers, which gives the ability to easily manage applications across on-premises, hybrid, or public cloud infrastructure.
This guide is a comprehensive overview of important elements, considerations and functionalities of Kubernetes, plus real-world experience on how to start utilizing it.
You might not need Kubernetes to successfully run your applications, but it’s going to make your life a whole lot easier. There are reasons why it’s remained the ‘golden child’ of container management, namely due to…
There’s an enormous cost benefit to Kubernetes, which isn’t always realized. Because the same host is reused, you’re able to determine the number of free resources within a given node and automatically schedule something else on that node. In a non-orchestrated environment, you would undoubtedly have a lot of free (read: wasted) space on the host, whereas a Kubernetes environment optimizes that for you.
The elasticity of Kubernetes applications allows you to schedule workloads up and down, so that you can minimize your costs and scale only according to demand.
Scalability is a cornerstone of Kubernetes usability. Your infrastructure is able to automatically scale in line with your application without you needing to manually do something about it. Although it’s an add-on to Kubernetes, it’s considered to be a ‘straight out-of-the-box’ solution for managing your applications at scale.
Best practices with Kubernetes-based applications push you to architect applications for failure, which in turn improves the resilience and availability. This can be done at many levels … from the applications failing, the underlying infrastructure, to the availability zone within the cloud provider. If they’re designed well, your applications can recover from these without any loss of service to your users.
Kubernetes simplifies the ‘defence in depth’ approach by defining best practices at all levels of your application stack: Cloud, cluster, container and code. Together with the ‘immutable infrastructure’ concept made popular by containers, the security posture of your stack can be hard to navigate. But through automation and securing your clusters in line with best practices for each cloud provider, you can significantly improve your security footprint.
In Kubernetes, there is a segregation of workloads running on a single instance, and the attack surface is reduced because only a single node or container is able to be attacked at any given time. Securing your Kubernetes cluster also looks a little different on each cloud provider, which can be an involved process if you don’t automate it.
Because your application is built within a container, it can be moved around easily. There’s also a lot of freedom when it comes to choosing operating systems, container runtimes, processor architectures, cloud platforms and PaaS in Kubernetes. And with great freedom, comes great portability.
Is there ever a wrong time to utilize Kubernetes? Yes … and no. There are situations where you might not be able to realize the full value that Kubernetes can bring, at least not yet, and there are also potential roadblocks you could run into along the way.
Kubernetes is complicated
No doubt, there’s a steep learning curve to Kubernetes. If you don’t have an expert on your team who possesses an in-depth knowledge of K8s, you could end up spending a lot of time and resources getting up to speed.
Kubernetes is expensive
And having that expertise as a resource on your team comes at a hefty cost, despite Kubernetes itself being cost-efficient. Costs can also vary significantly depending on which cloud providers and managed Kubernetes services you use, for better or for worse.
But, is it worth it?
We’d say so. Kubernetes was designed to boost performance and reduce the operational effort of distributed systems. You can make decisions that will lower the cost of your Kubernetes ecosystem, and it’s complex usability doesn’t mean that your team has to be Kubernetes experts to realise its benefits – there are solutions for that.
Control plane (master)
The Control plane is made up of the kube-api server, kube scheduler, cloud-controller-manager and kube-controller-manager. Kube proxies and kubelets live on each node, talking to the API and managing the workload of each node.
The cloud-controller-manager runs in the control plane as a replicated set of processes (typically, these would be containers in Pods).
etcd is a distributed key-value store and the primary datastore of Kubernetes. It stores and replicates the Kubernetes cluster state.
The kubelet maintains a set of pods, composed of at least one container, in entirety. Its functionality is watching for pod specs via the Kubernetes API server.
Kube-proxy is a network proxy that runs on each node. It maintains network rules, which allow for network communication to Pods from network sessions inside or outside of a cluster.
The Kubernetes controller manager embeds the core control loops shipped with Kubernetes. In applications of robotics and automation, a control loop is a non-terminating loop that regulates the state of the system.
The Kube-API server validates and configures data for API objects, including pods and services. The API Server provides the frontend to the cluster’s shared state, which is where all of the other components interact.
Kube-scheduler is the default scheduler for Kubernetes, in charge of scheduling pods onto nodes. It runs as part of the control plane.
A node is a worker machine that contains the services necessary to run Pods. You’ll typically have several nodes in a cluster, and each node is managed by the Master.
Desired state in Kubernetes (vs actual state)
Desired state is a core concept of Kubernetes. It means that, through a declarative or an imperative API, you describe the state of the objects that will run your containers.
Kubernetes is able to handle constant change and will continuously try hard to match your desired state with your actual state, which is how your containers are actually running. You might not potentially ever reach your desired state, but it doesn’t matter because the controllers for your cluster should be constantly running and working to fix and recover from errors, if possible.
There’s more than a few words, terms and phrases that you’ll need to know as you get started with Kubernetes. You have the components related to the architecture of Kubernetes above, and below are two of the most common terms that will be integral to your start in using Kubernetes, but you should get familiar with all of these Kubernetes terms and phrases.
A cluster is a group of nodes that run your containerized applications. You manage the cluster and everything it includes with Kubernetes.
A virtual ‘slice’ of that cluster where you can provision resources, organise objects and deploy applications inside the cluster.
Single-tenancy vs Multi-tenancy Kubernetes
Single-tenancy in terms of Kubernetes refers to an instance where a single workload, application, team or environment is associated with a single Kubernetes cluster.
Multi-tenancy, on the other hand, is when you have multiple applications, teams or environments all running side-by-side on one, large cluster.
The two options are distinctly opposite and, although both are valid approaches, there are a few reasons why you would choose one over the other. You should base your choice around these five key areas: Security, Reliability, Cost and Operational Overhead.
Security of your clusters
Kubernetes has measures in place to prevent security breaches — like PodSecurityPolicies and NetworkPolicies — but, as is the running trend, it takes experience to tweak these tools in the right way. And even then, they can’t prevent every security breach.
The way you configure your clusters has a hefty impact on your security. While we’ve established that no system can ever be completely secure, single-tenant deployments are optimized for increased security, because each customer’s data is completely separate from the others. Hence, there’s almost no chance that one customer will accidentally access another customer’s data.
In a multi-tenant environment, where multiple applications are running side-by-side, apps share the hardware, network and operating system on the nodes of the cluster so there is a heightened risk of a breach.
Reliability of the architecture
Much like the security benefit to single tenancy architecture, the reliability risks are lessened when there is a separation of customer information. If you have everything running on a single cluster and that cluster breaks, then all your workloads will be down.
And, there are many simple mistakes that could cause a catastrophic failure. We get deeper into some of the most common errors below, but these could include: insufficient resources, failing liveliness/readiness probes, etc.
At first glance, operating fewer clusters is the cheaper route – there’s more resource overhead cost when you have more clusters. More clusters, more money. But, managing the costs of storage and network bandwidth are far easier to manage with just one account per cluster.
In a single-tenant environment, you have more control over backups and recovery, because one system is backed up to a dedicated part of a SaaS server. If an app has specific requirements, these requirements can be installed in its cluster without affecting any of the other clusters.
Every cluster can be equipped with exactly the configuration that the corresponding app needs — no more and no less — which improves the efficiency of both the development and operation of your applications.
There are several steps you’ll want to take when first diving into Kubernetes. Your first step should be to have a Kubernetes cluster to play with, which could be in the form of Minicube or Kind. Once you’ve done that, you’ll deploy an application through two K8s resources: deployment and service.
The Appvia team has created an example open source GitHub project for you to play around with. If you follow the step-by-step guide, it will take you from having no Kubernetes ecosystem to having a locally deployed Kubernetes cluster with a running example application.
When you’re getting started with Kubernetes, there are many reasons your deployments might fail, given a large amount of unfamiliar jargon to come to terms with and large margin for error.
Here’s an at-a-glance look at some of the most common errors in Kubernetes and how to overcome them.
1. You have insufficient cluster resources
When creating a pod, Kubernetes will schedule the pod to a node that has enough free CPU, memory and ephemeral storage to satisfy the pod’s requests. When the cluster doesn’t have sufficient resources available, the pod’s state will be ‘pending’ and when describing the pod, its events will contain “Failed Scheduling” messages.
You can resolve this by reducing the pod’s requests or by increasing the size of the cluster, either manually or by installing autoscaler into the cluster.
2. You’ve specified the wrong container image
After deploying a pod, you may see that the state is ‘ErrImagePull’ or ‘ImagePullBackoff’. There are two potential causes of this. Firstly, if the image you have specified in your pod manifest does not exist. Alternatively, if the image is hosted in a private repository, you might need to configure image pull secrets.
When the pod first fails to start, you’ll see the ‘ErrImagePull’ status, indicating that the image did not pull. After a few failed attempts, the status will change to ImagePullBackOff and kubernetes will add a delay between attempts to pull the image.
4. Your liveliness/readiness probes are failing
Kubernetes provides two kinds of probes: readiness probes and liveness probes. These probes make sure that Kubernetes is able to detect when your application is unhealthy or unable to serve traffic, even when the application process has not exited.
A readiness probe is used when a container is first started and is used to wait for a container to become ready to handle requests. A liveness probe is used to ensure that a container continues to serve traffic once it has started. These probes can be configured to restart the container if it fails to start in time or an error causes it to stop handling requests.
If your probes are failing, check your container logs for any errors that could be causing the failure. Also, make sure it’s possible for your probe to run … for example, an http based probe will fail if the container is not listening on the port being checked or is only listening on localhost or an exec based probe will fail if the target script/binary does not have execute permissions.
This error is caused by the file “healthcheck.sh” which is specified as the liveness probe not existing.
5. Your Pods are getting OOMKilled and restarting
When creating a Pod, you can specify both a request and a limit for cpu and memory resources. Kubernetes will always schedule a pod to a node that has enough cpu and memory available to satisfy your request. A container that exceeds its memory request is in danger of being restarted or rescheduled, and a container that exceeds its memory limit will be restarted immediately. A pod that exceeds its cpu requests will not be killed, but may be throttled if the node does not have enough available cpu, and will always be throttled if it exceeds its limit, which may significantly reduce its performance.
To avoid this, refine your containers requests to ensure that they are not exceeded by your container, and make use of a HorizontalPodAutoscaler to ensure that increases in application load are handled by an increased number of pod replicas.
Often you’ll see a CrashLoopBackoff error rather than an OOMKilled error. After too many OOMKilled errors in too short a time, Kubernetes will add a delay between retries and the status will change to CrashLoopBackoff as above. In order to diagnose the problem in that case, you can check the output of ‘kubectl describe pod’ where you should be able to see the reason for the pods termination.
You have the ‘pick of the litter’ of cloud providers so to speak. There has been a long-standing battle between AWS, Azure and Google Cloud, and you want to compare them all to make sure your team is utilizing the one(s) that work for you.
Spoiler: They’re all used for the same purpose and provide nearly the same services. But you should still take a good look at each to determine which might be the best for you, or which aspects to use of which – if you’re utilizing a Kubernetes Managed Service. Keep in mind when you look at each cloud provider that they each have varied price models, and you might also consider using a provider closer to your geographical region, or the region of your users.
Amazon Web Services
AWS has been in the game the longest and, because of that, is often considered the leading platform. With a vast toolset that continues to grow exponentially, the capabilities of Amazon Web Services (AWS) are unmatched.
A close competitor to AWS with an extremely capable cloud infrastructure, Microsoft Azure has the benefit of integrating with other Microsoft tools. It should be a strong candidate for enterprises with an existing Microsoft footprint with cloud services and SaaS licenses being managed in a single place.
A well-funded potential underdog in the competition, Google entered the cloud market after Amazon Web Service and doesn’t have the enterprise focus that helps draw corporate customers. But Google is Google, with strong expertise in AI and analytics that present as significant advantages, and having created Kubernetes their managed Kubernetes service is unrivaled.
Nothing worth having comes easy, right? Complex functionalities in Kubernetes like networking, managing authentication, connectivity between services, role-based access controls etc. are hard to get your head around if you’re completely new to the game.
But utilizing Kubernetes doesn’t need to be an absolute nightmare. Appvia Wayfinder is a cloud-based platform that enables the automation of development environments and security standards for teams using Kubernetes. Designed to make Kubernetes a commodity for any organization by removing the need for specialists.
Being in control doesn’t have to be complicated
With Wayfinder, both Dev and DevOps teams can harness the scalability and efficiency of Kubernetes, delivering their applications with speed and security.