Tracing Discord's Elixir Systems (Without Melting Everything)
Summary
This article explains how Discord built end-to-end tracing across its Elixir-based stack to diagnose latency and outages at scale. It covers the introduction of a custom Transport envelope for trace context propagation, dynamic sampling to manage fanout span volume, and runtime rollout techniques to achieve zero-downtime instrumentation, ultimately improving observability without overwhelming systems.