With containerization changing the face of IT architecture, Kubernetes has become the most popular tool in the DevOps domain. CNCF’s 2020 survey of 1,324 respondents showed 83% use Kubernetes in a production environment, which helps practitioners orchestrate containers by automating their deployment, scaling, and load balancing needs.
Although popular, Kubernetes is not necessarily easy to work with. As your Kubernetes cluster grows, so does the complexity in managing it. To get the most out of your growing Kubernetes cluster (and to minimize its complexity), we recommend following the best practices covered in this article.
You must always have the latest stable version of Kubernetes in your production cluster. The new releases have many updates, additional features, and most importantly, patches to previous version security issues. It helps you in keeping your cluster away from vulnerabilities. Older versions also do not get enough support from the provider or the community. So, it is better to update the cluster on the latest version of Kubernetes.
Configuration files related to your deployment, ingress, services, and other files should be stored in a version control system (e.g., Github) before being pushed to a cluster. Doing so allows you to keep track of who made the changes and implement a change approval process to improve your cluster’s stability and security.
By default, there are three different namespaces in Kubernetes in the beginning: default, kube-public and kube-system. Namespaces are very important in organizing your Kubernetes cluster and keeping it secured from other teams working on the same cluster. If your Kubernetes cluster is large (hundreds of nodes) and multiple teams are working on it, you need to have separate namespaces for each team. For example, you should create different namespaces for development, testing and production teams. This way, the developer having access to only the development namespace won’t be able to make any changes in the production namespace, even by mistake. If you do not do this separation, there is a high chance of accidental overwrites by well-meaning teammates.
A Kubernetes cluster includes multiple elements like services, pods, containers, networks, etc. Maintaining all of these resources and keeping track of how they interact with each other in a cluster is cumbersome. This is where labels come in. Kubernetes labels are key-value pairs that organize your cluster resources.
For example, let’s say you are running two instances of one type of application. Both are similarly named, but each application is used by different teams (e.g., development and testing). You can help your teams differentiate between the similar applications by defining a label which uses their team’s name to demonstrate ownership.
Readiness and liveness probes are strongly recommended; it is almost always better to use them than to forego them. These probes are essentially health checks.
In this example, the probe pings an application to check if it is still running. If it gets the HTTP response, it then marks the pod as healthy.
Today, everything is hackable, and so is your Kubernetes cluster. Hackers often try to find vulnerabilities in the system in order to exploit them and gain access. So, keeping your Kubernetes cluster secure should be a high priority. The first thing to do is make sure you are using RBAC in Kubernetes. RBAC is role-based access control. Assign roles to each user in your cluster and each service account running in your cluster. Roles in RBAC contain several permissions that a user or service account can perform. You can assign the same role to multiple people and each role can have multiple permissions.
RBAC settings can also be applied on namespaces, so if you assign roles to a user allowed in one namespace, they will not have access to other namespaces in the cluster. Kubernetes provides RBAC properties such as role and cluster role to define security policies.
You can create a firewall for your API server to prevent attackers from sending connection requests to your API server from the Internet. To do this, you can either use regular firewalling rules or port firewalling rules. If you are using something like GKE, you can use a master authorized network feature in order to limit the IP addresses that can access the API server.
Occasionally deploying an application to a production cluster can fail due limited resources available on that cluster. This is a common challenge when working with a Kubernetes cluster and it’s caused when resource requests and limits are not set. Without resource requests and limits, pods in a cluster can start utilizing more resources than required. If the pod starts consuming more CPU or memory on the node, then the scheduler may not be able to place new pods, and even the node itself may crash.
For both requests and limits, It’s typical to define CPU in millicores and memory is in megabytes or mebibytes. Containers in a pod do not run if the request of resources made is higher than the limit you set.
In this example, we have set the limit of CPU to 800 millicores and memory to 256 mebibytes. The maximum request which the container can make at a time is 400 millicores of CPU and 128 mebibyte of memory.
A beginner developer often makes the mistake of going for the base image, which consists of up to 80% packages and libraries they won’t need. We recommend choosing smaller docker images that take less storage space. This helps you pull and build the image faster. Also, the smaller the docker image, the fewer the chances of security issues.
You can use alpine images (which are almost 10 times smaller than the base image), and then add any needed packages and libraries required to run your application.
The logs of any system have a lot to tell, you just have to store and analyze them well. In Kubernetes, auditing the logs regularly is very important in order to identify any vulnerability or threat in the cluster. All the request data made on the Kubernetes API is stored in the audit.log file. This file is stored at /var/log/audit.log with the Kubernetes cluster's audit policy present at /etc/kubernetes/audit-policy.yaml
You need to pass the parameter mentioned below while stating the kube-apiserver if you want the cluster to have audit logging:
One common mistake many practitioners make is to forget to monitor their control plane components. Control plane components include: Kubernetes API, kubelet, etcd, controller-manager, kube-proxy and kube-dns. These are the core of your Kubernetes cluster and you must keep a close eye on their performance. Kubernetes control plane components can emit metrics in Prometheus format that you can use to send alerts if these components have any issues. Monitoring control plane components helps you in keeping the overall workload and resource consumption in check.
Kubernetes is a popular containerization solution that continues to see climbing adoption rates. That being said, using it successfully requires thorough consideration of your workflows and departmental best practices. Want to learn more about Kubernetes? See the next chapter of our guide to containers.