Fika Interview with Andreas Kollegger and Johannes Bechberger

The video is an interview with Johannes Bechberger, a JVM Developer at SAP, at the JFocus conference. Andreas Kollegger (ABK) of Neo4j is conducting the interview.

Johannes Bechberger discusses a prototype he is working on called “JStall”.

About JStall:
Function: JStall is a tool intended to answer the question, “Why is my application currently stalling?”. It’s an AI-aided observability tool.
Mechanism: It works by taking two thread dumps (descriptions of what every thread in the application is doing) separated by a distance of about five seconds. It then compares them to see which threads are doing the most work.
AI Integration: This analysis information is fed into a local large language model (LLM). The LLM summarizes the results and provides a short overview, such as “your application might just be stalling in this thread because this thread is doing a lot of work”. The use of a local LLM (like the 30-billion parameter Quen model) running with Ollama is important for data privacy.
Development: JStall is built in Java. The developer aimed to make the jar file as small as possible (currently under 400 kilobytes). To achieve this, Bechberger had to rewrite his own command-line interface application, Lama interface provider, and JSON library. The command-line library, called “mini CLI,” is also available separately and is under 40 kilobytes in size.
Availability: There is already a blog post about the basic functionality of JStall. The AI-based prototype is expected to be released in about a month.