The article explains a kernel‑space memory profiler that uses a compile‑time code‑tagging library and special linker sections to record allocation callers and byte counts with minimal CPU overhead, allowing the tool to be deployed in production.
The team successfully employed eBPF dynamic tracing to obtain full Java call‑stack traces from any point in an OpenJDK 8 JVM process with microsecond‑level overhead, without requiring any JVM agents, bytecode injection, or code modifications, making it suitable for production environments.
This article explains how Lianjia designed and implemented LTrace, a zero‑intrusion, high‑performance distributed tracing system that captures full request chains across heterogeneous services, supports multi‑language environments, offers flexible sampling, and enables rapid fault isolation and performance optimization.
