Art of Debugging
summary
Shows concrete debugging steps and new tooling: use stats API for aggregates, performance-analyzer plugin for shard/index-level I/O and network hotspots, profile API plus hot threads to find query CPU sinks, and slowlog/trace logs for offending queries. Introduces a Telemetry plugin (released experimental in 2.11) that emits distributed traces and a metric framework with exponential histograms, default aggregation 60s, and sampling tuned to export when queue hits 2048 or time hits 2s and export in batch size 512. Telemetry is usable now but requires sampling to keep overhead low.
AI-written summary · about
chapters (11)
- 0:00Debugging OpenSearch
- 2:24Cluster stats API
- 6:18Search timeouts with Performance Analyzer
- 8:27Slow queries profiling
- 10:55Logging indexing/search issues
- 12:02Debugging tools limitations
- 13:47Telemetry framework
- 18:45Distributed tracing architecture
- 21:29Tracing request sampling
- 23:33Granular metric framework
- 26:28Future enhancements
related talks
19mBest Practices for Rapid Insights from Metrics, Logs, Traces and Profiles
01h 05mThe art and methodology of troubleshooting
30mOpenTelemetry: The Road Ahead + Meet the Co...
01h 01mPractical OpenTelemetry in .NET 8
36mMetrics - What's your code actually doing?
33mDistributed Tracing - All The Warning Signs Were Out There!
45mFrom Noise To Signal: Building Smart Observability Pipelines With OpenTelemetry
33mThe Observant Developer - OpenTelemetry from code to prod