KubeCon + CloudNativeCon India 2024

OpenTelemetry has transformed the observability landscape by moving beyond the traditional three pillars of traces, metrics, and logs. With these core components now stable, the project is expanding its reach to encompass even more facets of system monitoring and performance analysis. In this session, we will explore the latest advancements in OpenTelemetry, including the integration of profiling, which offers deeper insights into application performance by capturing detailed execution data. Additionally, we'll discuss the emerging concept of entities, which promises to provide a more holistic view of system components and their interactions. By expanding its capabilities, OpenTelemetry is not just enhancing the granularity and scope of observability but also setting the stage for more comprehensive and actionable insights. Join us to learn how these new developments can help you achieve unparalleled visibility into your systems and drive better operational outcomes.

Have you felt left out of Kubernetes development because you're not a GoLang expert? The good news is you don’t need to be. While GoLang has traditionally been the go-to language for creating Kubernetes-native tools like Helm, Argo CD, and Prometheus, Kubernetes is language-agnostic and can be interacted with via its declarative REST API in any language. This session will explore the growing ecosystem of Kubernetes application development beyond Go. We’ll review popular Kubernetes libraries and frameworks in languages like Python, JavaScript, and Java. To illustrate, we’ll demo the Java Operator SDK, showcasing how to build Kubernetes applications without relying on Go. Whether you code in Python, JavaScript, or Java, this talk will give you the tools and confidence to develop Kubernetes-native applications in your preferred language.

Imagine the k8s resource model only had “spec” and not the “status”. How operationally challenging would it have been to infer this critical information from out of band approaches? While Gitops frameworks work very well to translate a pre-defined intended state, it fails to provide visibility into the actual state of the system. This entails a need to build tools to improve infrastructure observability. Gitops, also fails to address immediate operational control and recovery in the event of outages and offers a broken experience. Gitops has been the primary developer experience for Walmart running over 200k workloads and 10k deployments per day on 1000s of k8s clusters across public and private cloud. This talk highlights our learnings on building various ops tooling which provides deployment/running state of the workloads to complement our Gitops based deployment pipelines.

Rolling out a new service, adding databases clusters, are always challenging. The recommended way is to start with diverting small % of traffic to the new service and increase rollout % if things look well. If things don’t go well, the rollout % should be reduced. This is generally done through feature flags and manually monitoring the service with an observability tool. Feature flags enable remote update of rollout % without doing new deployments. OpenFeature is a new CNCF project which is trying to standardise feature flagging. It also has built-in hooks for OpenTelemetry so that feature flag values are available by default in tracing & metrics data captured by OpenTelemetry. In this talk, I will share how you can leverage open standards like OpenFeature (for feature flagging) and OpenTelemetry ( for observability) to better manage rollouts. You can enable automatic increase or decrease of rollout % by tying observability with feature flagging.

Inference engines are used to generate prediction or deduce new information based on certain rules and data. With rise in the number of applications that benefit from inferencing, there is a definitive need to build Inferencing Services that are robust, performant (latency optimized) and can seamlessly scale on demand. In this session, we will discuss a proven set of procedures and guidelines for building and managing inference pipelines on Kubernetes. We will discuss details of the underlying hardware (GPU/CPU/memory) and K8s configuration requirements of some of the well-known Inference engines. We will demonstrate how robust and fault tolerant pipelines for LLM and RAG can be built using basic K8s constructs like operators, statefulsets and persistent volumes; along with enabling automated monitoring, triaging and remediation of the failed hardware and software components.

We, Priyanka and Jason, have been involved in Kubernetes release cycles for years, taking on roles from shadow to Release Lead and Emeritus Advisor. While we've worked with Kubernetes Enhancement Proposals (KEPs) throughout this time, it wasn’t until the 1.29 release, while prepping for press interviews, that we really dug into them—and wow, did we learn a lot! KEPs aren’t just feature proposals; they give you a peek into the entire lifecycle of a feature—designs, alternatives, testing, and final code. They’re a goldmine for understanding Kubernetes, whether you’re new or a seasoned contributor. We’ll walk you through some real KEPs and share our approach to reading and making sense of them. If you’re curious about Kubernetes and want to explore its features more deeply, join us. There’s so much to discover, and we’re excited to share what we’ve learned with you!

Flatcar is an operating system purpose-built for securely running container workloads. It ensures security by providing essential tools, reducing the OS size with minimal packages, and making the /usr directory immutable. This tutorial is designed for ops and DevOps professionals, offering step-by-step guidance on getting started with Flatcar. You’ll explore how to deploy it using Terraform, configure it with Ignition, and automate OS updates via Nebraska, enabling you to "set it and forget it.

The future of AI will be on cloud because of various reasons like scalability, computing power, features like Dynamic Resource Allocation and being able to run AI workloads across multiple clouds. However, the fields of cloud native and machine learning are still somewhat separated. We aim to bridge this gap with this tutorial. We plan on giving a brief introduction to machine learning with scikit-learn and cloud native with Kubernetes, then dive into a hands-on tutorial. We will show different ways to train and run ML models on and off the cloud (CPUs vs GPUs, public cloud vs on-prem, different flavors of Kubernetes) and evaluate performance in terms of training speed and cost. Additionally, training ML models is a complicated task and requires a lot of research. Optimizing your cloud setup on top of this makes it more difficult. There is no single solution, and we will explore a few of these options during this tutorial.

Picture this: Your startup just landed a big client, and as your business starts to grow, so does your Kubernetes cluster. The number of microservices and the traffic that they serve has started increasing. You configure auto-scaling of Nodes and Pods. Do you think that is enough, or do you need to look into other areas? What other components need scaling? What strategies should be devised? How do we verify whether there will not be any impact during scaling? In this session, we answer all these questions and come up with a blueprint for scaling a Kubernetes cluster from our experience with scaling our production cluster from merely 20-25 nodes to around 250 nodes with more than 5000 pods. We will touch base on the issues we faced and their resolution that will form a good prescription for those who want to scale their Kubernetes infrastructure.

We need to keep in mind many principles & techniques while writing a Kubernetes operator and especially so when used in production. Projects like kubebuilder go a long way to help in building operators. However, talks focus on how to write operators. But in this talk, you will learn how testing can help you create Kubernetes operators that are reliable and resilient to failure and give you the confidence to deploy them to production clusters. Often writing unit tests is not enough for an operator that manages infrastructure. Introducing e2e tests will help, but writing and running e2e tests are time-consuming. This session will share how a user can use envtest to simulate an environment very similar to a real environment and run the tests on behalf of that. In addition, you’ll learn what to keep in mind while writing the tests, such as it doesn’t support garbage collection, and a user must be very careful while creating objects and deleting them after a test suite is done.