Design and Implementation of Bistoury: An Enhanced Java Application Diagnostic Platform

Background: Online application diagnosis is challenging; after Alibaba open‑sourced Arthas in 2018, the team identified several shortcomings (tool‑like usage, lack of web UI, single‑machine focus, and isolation) and decided to build a more powerful one‑stop Java diagnostic solution called Bistoury.

Design and Implementation: Bistoury consists of a user system, agent, proxy, UI, registration center, load balancer and application center. The agent runs on the same host as the target application, receives commands from the proxy via Netty/WebSocket, and interacts with the user system.

ClassLoader design: Bistoury introduces a dedicated BistouryClassLoader to load its attach JARs safely, avoiding version conflicts, and a MagicClassLoader to prioritize classes from Bistoury-magic-classes.jar , enabling lightweight source modifications.

Online Debug: By dynamically instrumenting bytecode at breakpoints, Bistoury captures snapshots of local variables, call stacks, and other context without pausing the application, offering an IDE‑like debugging experience. Example method to debug: protected ModelAndView detailView(String viewName, String code) { Application app = applicationManager.getAppByCode(code, From.master); applicationManager.checkOwner(app); return createView(viewName).addObject("app", app); }

CPU Monitoring: Bistoury collects per‑thread CPU usage from /proc at one‑minute intervals, correlates JVM thread IDs via jstack , and visualizes detailed CPU metrics for each thread.

Heap Object Statistics: An optional feature periodically runs a jmap -histo ‑like command to record object type counts, allowing trend analysis of memory growth.

Conclusion: The article summarizes Bistoury’s motivations, architecture, and key features, providing insights for developers seeking advanced Java application diagnostics.