Open Source Summit Europe 2024

Are you struggling to monitor Windows on Kubernetes as a developer? With the limited support stack available, it can be a challenging task to keep track of everything. Kubernetes now supports HostProcess containers on Windows nodes. As a result, the Windows exporter, a Prometheus exporter for Windows, can be deployed as a host process pod. This feature is introduced as an add-on in the kube-prometheus project. This talk will implement a monitoring stack using Windows exporter on a Windows node in Kubernetes. We start by diving into various metric collectors, their functionalities, and utilization. Subsequently, demonstrate how to implement the Windows exporter, configure a ServiceMonitor, and deploy a Prometheus instance. Once Prometheus is up and running, we highlight the differences in the monitoring approach compared to Linux machines. Finally, we complete the monitoring experience with a Grafana dashboard analyzing the collected metrics.

Cloud computing has been primarily used by companies for production environments for quite some time now. However, a small group of users have found ways to move their development environment to the cloud. This allows them to access all their shared libraries, build tools, and software development kits (SDKs) on the cloud. As a result, they can build and ship their applications natively on the cloud. Gartner reports that 60% of cloud workloads will be developed and deployed using Container Development Environments (CDEs). In this talk, you will primarily focus on how you can simplify the structure of your projects by utilizing modern standards like devcontainer.json and open source CDEs such as DevPod ( devpod.sh ) to reduce the cognitive load for contributors and eliminate unnecessary obstacles. Moreover, DevPod handles the lifecycle of the infrastructure it runs on, and it can even suspend cloud resources automatically to save on costs. By the end of the talk, you will make local development much easier while saving costs.

More and more Operators automate deployment of cloud native applications in Kubernetes. Operators often automate the whole process of setting up and maintaining applications, including setting up the necessary workload resources, such as Deployments, StatefulSets and more. This poses a challenge for us Operator developers. Oftentimes, users of our Operator have specific requirements for the deployed workloads, restricting them to specific nodes, requiring additional configuration, added sidecar containers and many more. This talk shows how using kustomize in our Operator, we can simplify our own Operators while still allowing users the full range of customization. This even also allows for quick iteration during feature development.

In this presentation, I aim to address critical challenges in Kubernetes cluster management, particularly focusing on optimizing multicluster backup and restore processes using Velero. The current state of Kubernetes cluster management emphasizes the critical importance of implementing robust backup and restore processes. However, a challenge arises when dealing with multiple clusters, requiring the installation of Velero on each cluster, resulting in complex and resource-intensive multicluster Kubernetes environments. To address this issue, the proposed solution advocates for a centralized Velero installation, allowing for the backup and restore of individual clusters using their corresponding kubeconfig. The talk delves into the design architecture of multicluster backups, outlining how to integrate this approach with Velero's existing features. This innovative approach not only simplifies management but also optimizes resource utilization in a way that aligns with the evolving landscape of Kubernetes cluster operations. I aim for the audience to leave with actionable insights and a clear roadmap for implementing centralized backup strategies in their own multicluster setups.

In the realm of distributed systems and microservices architecture,dynamic API tracing systems play a pivotal role in diagnosing API failures and performance issues.However, current implementations often sample only a fraction of traces(1-5%) due to storage and scale constraints,leading to a skewed distribution favoring normal execution traces.This bias hampers effective diagnosis,impacting both developers and SRE teams.Our proposed solution employs a Machine Learning and Cognitive approach to rectify this imbalance in trace collection. By dynamically adapting based on actual data, the system learns to capture higher-interest traces, helping root causing issues. This innovative approach has significantly improved the Mean Time to Resolve (MTTR) for SRE teams, revolutionizing issue triaging within the organization. The Adaptive Sampling technique, driven by data analysis, standardizes trace distribution, reduces storage requirements, and dynamically adjusts the sampling rate.It lowers costs & resource utilization for tracing services, also streamlines trace metric analysis for SRE,offering a forward-looking solution to common challenges in the Observability and Infrastructure space.

Istio Ambient introduces a sidecar-less paradigm, aiming to simplify workload operations. Lin, a founding member of Istio, will guide you through the progressive application delivery process without sidecars, leveraging Argo Rollout while maintaining security, traffic control, and resilience with ambient service mesh. Lin will demonstrate the use of ArgoCD and Kubernetes Gateway resources (Gateway and HTTPRoute) to systematically increment weights, automatically directing traffic to a new application version. Throughout this process, she'll showcase monitoring using Prometheus metrics provided by Istio to ensure optimal performance. If defined success criteria aren't met, Argo Rollouts automatically triggers a rollback. Lin will illustrate the power of ArgoCD rollout with sidecar-less applications in Istio through a live demo. Additionally, she will delve into cost-saving strategies during rollout and highlight the distinctions between sidecar and sidecar-less application rollouts.

Join us for an enlightening session where we dive deep into the heart of containerization. In this talk, we'll unravel the intricacies of the Container Runtime Interface (CRI) by embarking on a journey to craft our own CRI entirely from scratch. Get ready to explore the fundamentals of container runtimes as we break down the CRI architecture, dissect its components, and understand how it interfaces with Kubernetes. From understanding the core functionalities to handling container lifecycle management, we'll guide you through each step of the process, demystifying the CRI along the way. Whether you're a seasoned container enthusiast looking to deepen your understanding or a newcomer eager to explore containerization at its core, this talk offers invaluable insights and hands-on experience. Join us as we empower you to take control of container runtimes by writing your very own CRI.