Making Observability Work at the Edge
summary
Edge proof observability uses an ultra‑optimized OpenTelemetry Collector Contrib DaemonSet (tuned for K3s/MicroK8s) plus Fluent Bit on-device, applying tail-based sampling, lossless batching to disk via file_storage extension (PVC backed) and a 5s decision window with 10k trace retention and 200ms latency threshold. Results: ~82–86% fewer spans/logs sent, 100% of errors preserved, queue drains after reconnection, collector image trimmed from ~250MB to ~30MB and CPU ~70% / memory ~30% lower while keeping ~2,400 spans/s throughput.
AI-written summary · about
chapters (13)
- 0:00Edge observability challenges
- 2:48Cloud-native defaults at edge
- 4:52Edge observability architecture
- 6:20Tail-based sampling
- 7:27Synchronizing logs and traces
- 11:18Buffering network interruptions
- 13:03Data persistence reliability
- 17:02Telemetry signal correlation
- 18:26Resource usage optimization
- 20:11Collector performance tuning
- 21:37Key takeaways
- 22:19Resources and demo
- 23:44Q&A
related talks
11mObservability 2.0 - Much more than just logs,metrics and traces
41mResilient Observability at the Retail Edge: A Lightweight, Scalable, and...
33mReal-World Sampling – Lessons Learned...
24mTelemetry Showdown: Fluent Bit Vs. OpenTelemetry Collector - A Comprehensive Benchma...
30mPanel: Telemetry That Matters
13mManaging Observability Data at the Edge with the OpenTelemetry Collector and OTTL
19mTelemetry Showdown: Fluent Bit vs. OpenTelemetry Collector - A Comprehensive Benchma...
40mBuilding a Scalable Observability Platform for Edge Compute Nodes