Linux in practice: catch a real Java deadlock

Create, observe and fix a real deadlock on your own machine (Linux, macOS or Windows with a JDK).

1. Write Deadlock.java with two lock objects inventory and order and two threads: thread A locks inventory, sleeps 100 ms, then locks order; thread B locks order, sleeps 100 ms, then locks inventory. Each prints a message when it holds both locks. Run it and describe what you observe (output, CPU usage in top/Task Manager).
2. While it hangs, find its process ID (jps -l) and run jstack <pid>. Copy the relevant part of the output and explain every line of the "Found one Java-level deadlock" section. Map it onto the four necessary conditions.
3. Fix the program in two ways: (a) lock ordering; (b) ReentrantLock.tryLock(timeout) with release-and-retry. Run each fix 20 times and report whether it ever hangs.
4. Add automatic detection: a watchdog thread that calls ManagementFactory.getThreadMXBean().findDeadlockedThreads() every second and prints the names of deadlocked threads. What should a production service do when this detects a deadlock?
5. Reflection: why is the deadlock not visible as high CPU usage, and why did the 100 ms sleep make the deadlock appear every time? What happens if you remove it?