The three villains to agentic observability: retention, sampling and rollups

The text discusses the challenges and limitations of modern observability practices, particularly focusing on the constraints posed by retention, sampling, and roll-ups. These practices, often seen as necessary optimizations, are argued to be workarounds for the limitations of current storage systems, which struggle with scalability, cost, and high-cardinality data. The text suggests that these constraints not only limit the effectiveness of human operators but also hinder the potential of AI-driven workflows in observability by removing critical data context. It advocates for a shift towards using columnar storage models, which can handle full-fidelity, high-cardinality data more economically, thereby reducing the need for trade-offs like sampling and roll-ups. ClickHouse is highlighted as a standout solution due to its real-time focus, efficient compression, and ability to handle large datasets without requiring storage tiers. The text envisions an observability future where data is retained in full fidelity, enabling faster, more accurate analysis and supporting the next generation of agent-driven workflows.