CNCF-hosted Co-located Events Europe 2025

Logs are a goldmine of information, which is why they have become the backbone of business-critical monitoring and observability systems. Yet like gold, mining this value requires significant effort – sifting through endless entries is time-consuming, tiring, and costly. To reduce operational overhead and minimize mean time to resolution (MTTR), this talk explores advanced techniques for log summarization, offering methods to reduce log volume without losing critical insights for modern cloud-native and K8s environments.

We'll discuss the use of key attributes and metrics to make logs more meaningful and enable more rapid root cause analysis (RCA). The presentation will demonstrate semantic log understanding using advanced AI and contextual log analysis powered by large language models (LLMs) to automatically extract actionable insights and understand the deeper context of system behaviors and application flows.

In today’s distributed microservices landscape, Kubernetes environments generate vast volumes of logs, making troubleshooting complex and time-consuming. Operators often sift through massive data to identify issues, leading to prolonged downtime—a challenge that intensifies with multiple clusters. Discover how GenAI optimizes troubleshooting by transforming traditional logs into conversational insights. This session covers building an AI-driven observability solution with Large Language Models (LLMs). We start by configuring Fluent Bit collectors to gather systemd logs, Kubernetes events, and application logs, which are then streamed to a scalable object storage. By constructing a vector database, we enable users to query and interact with logs in natural language. We will provide a step-by-step guide that equips attendees with actionable knowledge to implement GenAI observability in their Kubernetes clusters.

Salesforce handles billions of transactions daily, generating over 50 trillion spans. These transactions represent a complex ecosystem. Failure of even a single transaction, can leave users frustrated while everything appears “green” in the system. Why? Sampling-based tracing often misses such edge cases.
In this talk, we unveil how SF overcame this challenge by enabling 100% sampling for critical flows, all while keeping costs low. We’ll share our groundbreaking migration from Zipkin to OTel and the lessons learned along the way
Discover how we equipped SF developers with a 360-degree view of service and API performance. With OTel, they can now pinpoint RED metrics, diagnose issues faster, and achieve visibility beyond the limits of logs and metrics
We’ll also dive into the backend challenges of scaling OTel, from managing high data volumes to optimizing storage and query performance. We’ll share the pros and cons of various approaches, and our experiences with open-source tools

In this session I’m going to take you on a journey of Delivery Hero - an international food delivery company - towards the goal of becoming observability vendor-independent. This journey will be long and hard and will take our hero through the forest of the unstable OpenTelemetry contrib components and the swamp of high memory and CPU consumption. It will require them to find new allies to overcome the challenge of routing for stateful collector components in non-federated environments and fight the metric temporality conversion monsters. I am going to demonstrate what keeps our hero motivated after all these hurdles and why they are still convinced that OpenTelemetry is the right tool for them to accomplish this mission.

At Dropbox, managing observability for systems producing terabytes of logs daily posed a unique challenge. Initially, developers accessed logs by logging into individual servers, a process further complicated when we moved to containers. Containers are short-lived, which caused logs to disappear on termination. This highlighted the need for a scalable, persisted, centralized solution using open-source tools.

In this session, I’ll discuss our journey to build a robust observability framework centered on Loki as our logging solution. Scaling Loki to Dropbox’s data volume required extensive optimizations for reliable, efficient query performance. I’ll cover our deployment, challenges, and strategies for achieving high-performance logging.

We also integrated Grafana to unify logs and metrics in a single view, enhancing troubleshooting and security. Join us to learn how Dropbox scaled its observability with open-source solutions and key lessons from our experience.

Migrating observability infrastructure for a 5,000-person financial services company is daunting enough - doing it with just four engineers might seem impossible. This session details MSCI's journey from a traditional Splunk infrastructure to a modern cloud-native observability stack built on OpenTelemetry, Prometheus, Jaeger, Grafana, and Elasticsearch. We'll share our architectural decisions, implementation strategy, and critical lessons learned while maintaining observability during the transition. Through real-world examples, we'll demonstrate how we overcame scaling challenges, managed the cultural shift, and achieved better visibility while significantly reducing costs. Learn practical strategies for planning your own observability migration, including how to phase the transition, train teams effectively, and avoid the pitfalls we encountered.

Do you enjoy using the OpenTelemetry Collector, but can’t find a distribution with the right set of components included? Do you want to write your own components for that niche use case that only you have? Or maybe you just want a Collector that has your name on it? Give the OpenTelemetry Collector Builder (OCB) a try!

OCB is developed by the Collector maintainers and is purpose-built for easily building your own Collector. This session will cover the basics of how OCB works, then will cover a wide range of use cases, including creating release pipelines using OCB, publishing Docker images, hotfixing upstream components when a change is needed immediately, and using your own components. To tie it all together, we’ll also show how OCB is used in the wild to publish popular Collector distributions.

Struggling to monitor and debug Windows containers on Kubernetes? You’re not alone!
Unlike Linux, managing Windows workloads often feels like solving a puzzle with missing pieces. But it doesn’t have to be this way.
With Kubernetes now supporting HostProcess containers on Windows nodes, a lot more monitoring and troubleshooting is now possible.
We’ll show you how to implement a complete monitoring stack using Windows Exporter on a Windows node in Kubernetes. We’ll start by exploring the metric collectors in Windows exporter, their functionalities, and practical use. Next, we’ll demonstrate deploying the Windows exporter as a HostProcess pod, configuring a ServiceMonitor, and setting up Prometheus to collect and visualize metrics.
Finally, we’ll elevate your debugging game by exploring the newly added kubectl debug support for Windows nodes, enabling you to diagnose and resolve issues faster at the node level. You'll be ready to troubleshoot Windows nodes in no time.

Explore a cutting-edge approach to enhancing observability for serverless applications on solutions like Google Cloud Run and Cloud Functions. This session delves into creating a scalable metrics pipeline using Shopify's internal app platform for seamless container configuration and a ingestion system capable of handling millions of datapoints per minute.

We'll dive into the architecture featuring OpenTelemetry collectors as sidecars and a regional workloads to ingest and manage metrics with varying temporality models. Discover how we integrated OTLP with our ingestion layer, transforming exponential histograms into DD Sketches for optimal performance and accuracy.

Gain insights into the challenges and solutions in building this comprehensive observability pipeline. This talk provides valuable lessons for teams aiming to enhance serverless monitoring in Kubernetes environments, leveraging Shopify's philosophy of efficient, resilient, and cost-effective cloud utilization.

At Intuit, we manage over 1500 web/mobile plugins serving our customer needs. However, detecting and quantifying real-time user impact during failures remains a significant challenge. Traditional approaches only highlight large-scale issues and don’t offer insights into the specific business workflows affected.

This talk covers,
1. Leveraging OpenTelemetry to develop a capability called “Failed Customer Interactions” (FCIs)
2. Computing real-time customer impact on business workflows
3. Reducing our Mean Time to Detect (MTTD) to less than 3 mins, powered by Anomaly detection
4. Designing cost effective, highly scable system that handles 130,000 spans per second

The OpenTelemetry Transformation Language (OTTL) is a powerful way to customize telemetry data transformation with the OpenTelemetry collector, but it can be daunting for new and experienced users alike.
Enter the OTTL Playground (https://ottl.run), a powerful and user-friendly tool designed to allow users to experiment with the OTTL effortlessly.

The playground provides a rich interface for users to create, modify, and test statements in real-time, making it easier to understand how different configurations impact the OpenTelemetry data transformation. Users can instantly validate OTTL transformations, from input to output, along with diffs. This allows new users to explore the nuances of OTTL without the risk of disrupting production environments.

This session provides a quick introduction to OTTL, and a live demo on how the OTTL Playground can help users to create, test and troubleshoot OTTL statements. Offering ideas for enhancements and community contributions.