Cloud Native Valencia #5 - The Summer Release

Introduction to the "Locos de la IA" community. Learn more about them: https://www.instagram.com/loslocosdelaia/.

Kubernetes platforms often struggle with configuration drift and inconsistent environments across clusters. Reproducing the exact same setup across nodes and environments can become complex when operating systems and system components evolve over time.

What if the operating system itself was immutable and fully controlled through declarative configuration?

In this session, maintainers of the SIGHUP Distribution will share how we designed an immutable Kubernetes platform built on top of Flatcar Container Linux, optimized for on-premises environments.

By removing traditional package managers and mutable system components, the platform eliminates configuration drift while reducing the attack surface. The operating system is preconfigured and tuned specifically for Kubernetes, with managed kernel parameters and system-level settings aligned with the architecture of the SIGHUP Distribution.

Custom system extensions are used to deliver tightly integrated components, improving cohesion between the operating system and the Kubernetes stack while preserving immutability, security, and operational consistency across all cluster nodes.

The session will include a live demo showing how we bootstrap a cluster from a PXE boot to a fully operational immutable Kubernetes cluster with a single command.

Along the way, we will explore how immutable infrastructure enables reproducible environments, declarative system management, and consistent platform operations for Kubernetes clusters.

What actually happens to a metric between the moment your application emits it and the moment a retention policy deletes it? In this talk, we deconstruct a metric’s life and learn about the magic tricks that make them possible. We cover how a metric is created with the OpenTelemetry SDK and how its type shapes aggregation; how its name and labels become a unique time series; how backends absorb a firehose of writes (buffer in memory, flush to immutable sorted files, merge in the background, the LSM-tree family) and compress it with columnar layouts and schemes like Gorilla; what every query engine must do (select, filter, aggregate) and why some operations are cheap and others bring a backend to its knees; and finally downsampling and retention, where the metric grows old and dies. Every backend varies in the details, but they share the same fundamental ideas. You'll leave with a mental model of what's happening under the hood of any metrics system.