[Talk::Overflow #22] PromCon EU 2025
The Pragmatic Turn in Observability
Welcome to the 22th edition of Talk::Overflow!!
This week, we bring you a curated talks from the PromCon EU 2025.
After years of the industry being swept up in the complexity of the “three pillars” and the massive shift toward OpenTelemetry, this year’s conference signalled a return to pragmatic, performance-first engineering.
For platform engineers, the message was clear: standardized doesn’t always mean better, and the “boring” parts of the stack—storage formats, downsampling, and hardware counters—are where the real reliability gains are being made.
As Prometheus 3.0 moves into the spotlight, the ecosystem is doubling down on native performance while acknowledging that OTel is here to stay, creating a hybrid reality that requires a senior engineer’s eye for trade-offs.
Most Talks
Why I Recommend Native Prometheus Instrumentation over OpenTelemetry — Julius Volz
Julius Volz (creator of Prometheus) provides a high-signal argument for why native Prometheus instrumentation remains the superior choice for most Kubernetes-native environments. He highlights the performance overhead and cognitive load introduced by the OTel SDKs, urging engineers to evaluate if the complexity of a vendor-neutral standard actually solves their local reliability problems. This is an essential watch for teams feeling “OTel guilt” while their existing Prometheus setup works perfectly.
Eight definitions of “observability” vs. “monitoring” – and why none of them work — Björn Rabenstein
Björn Rabenstein cuts through the marketing fluff that has clouded the monitoring space for the last five years. By dismantling eight common industry definitions, he refocuses the conversation on the technical reality of telemetry and incident response. For platform leads, this talk provides the vocabulary to stop debating semantics and start building better diagnostic systems.
Beyond TSDB: Unlocking Prometheus with Parquet for Modern Scale — Jesus Vazquez, Alan Protasio, Michael Hoffmann
A deep dive into the architecture of using Apache Parquet as a long-term storage format for metrics. The speakers demonstrate how shifting to columnar storage allows for massive analytical queries that would crash a standard TSDB. This is the blueprint for engineers managing petabyte-scale observability who need to balance real-time alerting with deep historical trend analysis without blowing the budget.
CPU Hardware Counters — Bryan Boreham
Bryan Boreham explores the “final frontier” of observability: hardware-level metrics. He demonstrates how to surface CPU counters like cache misses and branch mispredictions into Prometheus to find performance bottlenecks that application-level profiling misses. This is a masterclass in low-level visibility for anyone running high-throughput, latency-sensitive services.
Downsampling in Prometheus — Jan Fajerski
Jan Fajerski breaks down the implementation of downsampling in the Mimir/Prometheus ecosystem. He explains how to maintain query speed over multi-year datasets by reducing resolution as data ages. For platform teams struggling with slow dashboards and high storage costs, this provides a practical path to sustainable long-term metric retention.
TL;DR — All Other Talks
Everything you need to know about OpenMetrics 2.0! — György Krajcsovits, Bartłomiej Płotka
A technical update on the evolution of the OpenMetrics standard and its impact on ecosystem interoperability.
Building an Observability Agent for Rapid Root Cause Analysis using Prometheus metrics — Pavan Yekbote
A look at how to build custom agents that leverage Prometheus data to automate incident diagnosis.
From Manual to Managed: Prometheus Agent Deployment at Scale — Mihail Mihaylov
Strategies for managing large-scale Prometheus agent fleets in complex, distributed environments.
Auto-generated coherent Observability — Arthur Sens, Dominik Süß
How to use metadata and automation to ensure telemetry is consistent across an entire platform.
No Jsonnet? No Problem! Prometheus in Perses, Powered by Go — Saswata Mukherjee, Hélia Barroso
An introduction to Perses and how it simplifies dashboard-as-code without the complexity of Jsonnet.
Engage Prometheus in Rust Ecosystem — Ruihang Xia
A guide to the best practices for instrumenting Rust applications with Prometheus.
Alertmanager Has Amnesia – Should We Fix It? — Joel Verezhak
An investigation into state persistence in Alertmanager and the trade-offs of fixing historical state loss.
OpenTelemetry developments in Prometheus over the last year — Arve Knudsen, Owen Williams
A status report on the ongoing integration of OTel OTLP support directly into the Prometheus server.
SAAFE - A prioritized alerting model to troubleshoot your incidents — Jorge Creixell, Manoj Acharya
A framework for reducing alert fatigue by prioritizing incidents based on blast radius and impact.
Reverse-Engineering PromQL Usage: A Proxy’s Tale — Nicolas Takashi
How to use proxy layers to analyze query patterns and optimize backend performance.
From Logs to Metrics: OpenSearch’s Prometheus support with Agentic AI and Natural Language to PromQL — Anirudha Jadhav, Kyle Hounslow
Exploring the intersection of AI and PromQL to bridge the gap between logs and metrics.
Debugging the Pipeline: Observability for Your Observability with OTel + Prometheus — Israel Blancas
Practical tips for monitoring your own observability pipeline to prevent blind spots.
A History of Automatic Aggregations — Raphaël Bizos
A retrospective on the attempts to automate metric aggregation and why it’s a difficult engineering problem.
OTEL delta temporality support in Prometheus — Fiona Liao
Technical details on how Prometheus handles OTel’s delta temporality for better compatibility.
UX Research Report: Prometheus and OTel’s Resource Attributes. — Victoria Nduka, Andrej Kiripolský
A rare look at the usability challenges engineers face when mapping OTel resources to Prometheus labels.
Counting What You Care About in Your Security Data Pipeline — Attila Szakács
Using Prometheus to monitor the health and throughput of security-critical data streams.
Welcome — Richard Hartmann
Opening remarks setting the stage for the conference and the state of the Prometheus project.
💬 Like what you read?
You’re reading Talk::Overflow #22 — the weekly digest for developers who want to stay sharp and skip the noise.
→ Browse past issues
→ Suggest a talk
→ Share it with the friends
Stay curious. Stay kind.
— Talk::Overflow