Kubernetes Network Observability
The Retina platform, which is cloud-native and allows Kubernetes users, administrators, and developers to visualize, observe, debug, and analyze Kubernetes workload traffic regardless of the Container Network Interface (CNI), operating system (OS), or cloud, is being released by the Microsoft Azure Container Networking team with great excitement. They are thrilled to make Retina available as an open-source repository that supports networking scenarios relating to DevOps and SecOps for your Kubernetes clusters. They cordially urge the open-source community to collaborate with us on future innovations.
Microsoft AI technology is shown in photography, which foretells a student’s likelihood of dropping out of school in order to support teachers proactively.
A framework for cloud-native container networking observability
Cloud native technologies such as Kubernetes have simplified the process of developing mobile apps. Simultaneously, a lot of apps have grown in complexity, making cloud management them harder and harder. Network-related observability, troubleshooting, and debugging have become harder when businesses create cloud-native apps made up of linked services and then deploy them to several public clouds in addition to their own infrastructure.
Retina aims to provide actionable network insights, such as how containerized micro-services interact, in a non-intrusive manner without requiring changes to the applications themselves, thanks to the extended Berkley Packet Filter (eBPF). Retina will emphasize application developers’ experience in a new way, democratizing network observability and debugging. Retina relieves developers of the burden of having to manage the intricate details of the underlying network architecture and transformations by giving them easy tools to monitor and debug their apps for problems like packet loss and latency.
They are eager to expand on this partnership and interact with more communities, as well as to strengthen their good community experience with eBPF and Cilium. We think that by making Retina available to the community, Azure will be able to gain from well-informed criticism, creative suggestions, and teamwork that will help us improve and broaden Retina’s capabilities.
Retina capabilities and solutions
Azure discovered significant gaps in network monitoring, or the gathering of network metrics and traces from Kubernetes clusters, by utilizing our vast expertise administering several container networking services for the Azure Kubernetes Service (AKS). Retina is a state-of-the-art solution that bridges these gaps by addressing the intricate problems associated with maintaining and sustaining Kubernetes networks. It offers site- and infrastructure-reliability engineers thorough understanding of cluster networking. Additionally, Retina offers comprehensive traffic analysis inside the framework of Kubernetes, converting measurements into network flow logs or the industry-standard Prometheus.
The usability and adaptability of existing open-source solutions are typically limited by their close coupling to certain CNIs, OSs, or data planes. Retina is a useful complement to any current toolkit as it was created to be a highly flexible, adaptive, and extendable framework of plugins that can function flawlessly with any CNI, OS, or cloud provider. Supporting both Linux and Windows data planes, Retina maintains a small memory and CPU footprint on the cluster, this is true even at scale while meeting the various demands of infrastructure- and site-reliability engineers. Azure can quickly modify and customize Retina to suit new use cases without relying on any particular CNI, OS, or data plane because to its pluggability design ethos.
Deep network traffic insights, including Layer 4 (L4) measurements, DNS metrics, and distributed packet captures, are one of Retina’s primary features. The Kubernetes app model is completely integrated, providing pod-level data along with comprehensive context. It provides node-level metrics (like forward, drop, Transmission Control Protocol (TCP), User Datagram Protocol (UDP), and Linux utility) and pod-level metrics (like basic metrics, DNS, and API server latency) with actionable networking observability data that is emitted into industry-standard Prometheus metrics.
Label-driven distributed packet captures in Retina let users choose what, where, and from whom to collect packets. It also offers sophisticated debugging tools and historical context for network flow logs, which improve network troubleshooting and performance optimization.
Azure’s plans for the retina
Many businesses use several clouds and want solutions that function well on Microsoft Azure as well as other clouds and on-premises. From the start, Retina is multi-cloud and open-source. They want to educate the larger cloud-native community about their expertise and ambition for Kubernetes networking observability by making Retina open-sourced. Through cooperation with other developers and organizations who have similar experiences and aspirations in this industry, we hope that Retina will continue to grow and expand.
Extensibility has always been important in architecture and will continue to be so. With Retina, data gathering is extensible, making it simple for users to add additional metrics and insights. Additionally, it provides flexibility in exporters, allowing users to interact with other tools and monitoring systems. Because of its adaptability to various use cases and circumstances, Retina is a strong and flexible platform for Kubernetes networking observability.
Finally, Azure will see Retina as a platform that everyone can use to contribute, expand, and develop, eventually leading to the creation of a reliable, well-thought-out, all-inclusive solution for Kubernetes networking observability.
Overview of Retina
Retina: What Is It?
Retina is an open-source, cloud-independent Kubernetes Network Observability platform that supports compliance, DevOps, and SecOps use cases. It serves as a single portal for cluster network administrators, cluster security administrators, and devops engineers to monitor the security and health of applications and networks.
Retina gathers telemetry that is configurable and can be exported to various storage platforms (like Prometheus, Azure Monitor, and other vendors) and displayed in many ways (like Grafana, Azure Log Analytics, and other vendors).
Features
- Network Observability platform for Kubernetes workloads based on eBPF.
- Both configurable and on-demand.
- Industry-standard Prometheus metrics that are actionable.
- Simplified Packet Captures for In-depth Analysis.
- Multi-OS support (such as Windows, Linux, and Azure Linux) and independence from cloud providers.
For what reason is Retina?
With Retina, you can keep an eye on your clusters and look into network problems whenever you choose. Here are a few situations where Retina excels, reducing trouble spots and research time.
Examining: Troubleshooting Network Connectivity
Why are my Pods no longer able to interact with one another? A typical inquiry takes a lot of time and entails doing packet captures, which need access to each node, identification of the nodes involved, running of tcpdump commands, and exporting of the data from each node.
With only one CLI command or CRD/YAML file, you can use Retina to automate this procedure. It can:
- Perform captures on every Node that is home to the relevant Pods.
- Upload the output from every Node to a storage blob.
Keeping an eye on the network Actionable insights are supported by Health Retina via Grafana dashboards, Prometheus alerts, and other features. As an example, you can:
- Track lost connections inside a namespace.
- Notify me when production DNS error rates increase.
- Assess the scalability of your application while monitoring variations in API Server latency.
- Should a Pod begin to transmit excessive amounts of traffic, notify your security team.
- Metrics and captures are the two forms of telemetry that Retina employs.
Metrics Retina metrics provide ongoing observation of:
- Traffic moving in and out
- Packets dropped
- TCP/UDP
- DNS
- API Server latency
- Statistics for nodes and interfaces
Retina offers the two
- Advanced/Pod-Level metrics (if enabled) and
- Basic metrics (default, Node-Level metrics).
Takes Pictures
Network traffic and information for the designated Nodes/Pods are recorded in a Retina capture.
The ability to output captures to numerous destinations is available on demand. See Captures for more details.
