SE Radio 717: Eric Tschetter on Decoupling Observability

In this episode of Software Engineering Radio, host Amay Mbadeh interviews Eric Tschetter, a founding team member of the Apache Druid project and VP of Engineering at Metamarkets, about the evolving landscape of observability in modern software systems. Tschetter argues that the industry is moving from tightly coupled, vertically integrated observability stacks—like Splunk, ELK, and Grafana—toward a decoupled architecture inspired by the historical evolution of business intelligence. He explains that decoupling observability into distinct layers—ingest, storage, query compute, and visualization—enables greater flexibility, reduces vendor lock-in, and supports multiple query languages like SPL, LogQL, and PromQL. The key to this decoupling lies not in standardizing data schemas, but in unifying around query language APIs, allowing teams to use their preferred tools while accessing shared data. Tschetter emphasizes that success is measured by whether teams can continue using their existing workflows while gaining access to more data. He also discusses practical concerns like cold start latency, caching strategies, sampling trade-offs, and the importance of governance in multi-team environments. Finally, he advocates for decoupling as an inevitable and necessary evolution, especially as systems grow in complexity and scale. Key takeaways include: 1) Decouple observability by separating data storage from query and visualization layers to avoid vendor lock-in; 2) Use open standards like OpenTelemetry for data ingestion and support multiple query languages to maintain flexibility; 3) Prioritize data accessibility over schema standardization—focus on query language interoperability instead; 4) Leverage cloud object stores for cost-effective, long-term data retention; 5) Implement governance and resource isolation to prevent noisy neighbor issues; 6) Start migration with high-value, isolated data sets to prove value without disrupting workflows; 7) Recognize that decoupling is not just technical but organizational—it enables innovation and collaboration across teams. The episode concludes with Tschetter’s strong endorsement of decoupling as the future of observability, with a call to action for teams to envision what’s possible when data is truly accessible.