Shift Into an Observability Mindset with OpenTelemetry

Speaker(s): Daniel Gomez Blanco

Video URL: https://www.youtube.com/watch?v=_0AorR5JUS8

Summary: This talk is about explaining that in today’s complex, distributed systems you can no longer debug by intuition and isolated metrics alone - you need rich, standards-based context propagated end‐to‐end. By adopting OpenTelemetry’s unified model for metrics, traces and logs, teams gain data ("evidence") instead of guesswork and can correlate user experience across services.

Timestamps

• 00:00 - Introduction to the Talk and Speaker
• 02:10 - Debuggning without context makes it difficult to connect the dots
• 04:00 - With context and correlation we have evidence and everyone can be on the same page
• 06:30 - Trace Sampling (Head / Tail) allows to only keep the most useful data
• 09:30 - Use each signal of your telemetry data for its intended purpose (Metrics vs Traces vs Logs)
• 12:30 - Metric Views to control metric streams
• 14:00 - Communicate value to the Leadership
• 17:05 - Change the engineer mindset for cross-organisation alignment
• 20:30 - Takeaway points

Key Takeaways

• By propagating OpenTelemetry’s trace context end-to-end we get concrete evidence rather than relying on hunches when debugging.
• Leverage metrics, traces, and logs for their intended purposes but also correlate them via exemplars and semantic conventions.
• Use head sampling for low-overhead probabilistic trace capture, and tail sampling (via a collector) to focus on the most interesting traces (errors, high-latency) without losing end-to-end context.
• Keep metrics cardinality low and use “metric views” at runtime to reshape or filter out unbounded labels, ensuring performant long-term storage and querying.
• Use telemetry to prove ROI and guide leadership decisions by tying service-level indicators (SLIs/SLOs) back to real user-experience and business KPIs.
• We learn by doing so it's a good idea to guide engineers to adoption through live labs or “observability game days,” making it easy—and fun—for to experience the value of context-driven debugging.

Questions/Discussion Points

• Easy talk to follow along, kind of like an overview/intro to OpenTelemetry’s APIs and SDKs. I would say the recommended audience is Platform/SRE/DevOps engineers who want to move from ad-hoc metrics to a unified, standards-based approach - or engineering leads/architects looking to align multiple teams around a common telemetry model.