KubeCon + CloudNativeCon Europe 2025

Many organizations drown in terabytes of telemetry data but often use only 20% of it.

In this lightning talk, we'll sprint through battle-tested strategies in OpenTelemetry to trim your observability costs without compromising visibility. From intelligent sampling to SDK optimization, learn how to implement a lean observability practice that focuses on signal over noise.

Key Takeaways:
- Sampling : Tail based sampling, head based sampling for traces, probabilistic sampler processor for logs
- Using processors in the Otel collector with granular filtering/dropping of irrelevant attributes - to control volume going out of collector
- Reduce the volumes at the SDKs by controlling what attributes are sent - e.g.control what http.headers are sent
- Visibility on cardinality in time series (and suggestions on what attributes to drop)
- Use `Views` in SDKs to customise aggregation or which attributed are reported in metrics
- Using granular retention settings to reduce observability costs