Using Native OpenTelemetry Instrumentation to Make Client Libraries Better
summary
Shows how native OpenTelemetry traces and continuous profiling instrument Azure SDKs during long stress and load tests, driving concrete fixes: avoid redundant HTTP verification requests that return 416, eliminate repeated 401 token refreshes across chunks, shrink always-allocated 1MB buffers to precise sizes for big throughput gains, and fix thread-pool and prefetch bugs that create 4 million messages and OOMs; dashboards, traces and dumps prove fixes (throughput doubles, memory drops orders of magnitude) and validate retry and concurrency configs.
AI-written summary · about
chapters (10)
- 0:00OpenTelemetry for SDKs
- 2:30Observability for libraries
- 3:11Debugging with traces
- 5:21API behavior via traces
- 7:47Flaky tests with observability
- 9:17Performance testing beyond benchmarks
- 11:30Hidden performance issues
- 13:15Memory bugs via telemetry
- 15:03Observability for SDKs
- 16:55User zero
related talks
21mNative Instrumentation for Open Source Software with OpenTelemetry
01h 01mPractical OpenTelemetry in .NET 8
01h 00mPractical OpenTelemetry for .NET
47mHow I Solved Production Issues with OpenTelemetry (And How You Can Too)
31mDiscovering the Magic Behind OpenTelemetry Instrumentation
47mUsing metric and traces to diagnose our cloud-native applications
59mOpenTelemetry will save your day and night
23mModern Application Debugging: An Intro to OpenTelemetry