Requests to one endpoint hang forever while the rest of the service is fine. How do you diagnose it?

Take a thread dump with jstack or jcmd, ideally two a few seconds apart, and look at the stuck threads. The JVM detects lock cycles and prints a Java level deadlock section outright. If there is no cycle, threads are probably blocked on a resource such as a connection pool or a socket read with no timeout. Fix by ordering locks consistently, shrinking the critical section, or adding timeouts.

First thing is a thread dump, and I take two or three a few seconds apart so I can tell what is genuinely stuck from what is merely busy. The JVM does a lot of the work here: if two threads hold each other's monitors it prints a Java level deadlock section naming both threads and both locks, and then the stack traces tell me exactly which methods to look at. If there is no cycle, the pattern is usually different: dozens of threads all in a waiting state on the same connection pool, which means something is holding connections rather than a classic deadlock, or threads blocked in a socket read with no timeout configured, which hangs indefinitely when a peer stops responding. For prevention I have a few rules. Locks are acquired in a consistent order, and I document that order when there are two. I keep the critical section as small as possible and never make a network call or publish a message while holding a lock. And where the design allows it I use tryLock with a timeout, so contention degrades into a handled error rather than a hang.