How to diagnose memory leaks in a legacy Java batch job from heap dumps?

[Thomas_R]

I'm maintaining a legacy Java application that's been running in production for years, but we've recently started experiencing gradual performance degradation and eventual crashes in our weekly batch processing jobs, which strongly points to a memory leak. I've taken heap dumps and see old object references piling up, but tracing the root cause back through layers of outdated frameworks and custom caching code is proving incredibly time-consuming. For developers who have successfully hunted down elusive memory leaks in complex systems, what is your step-by-step diagnostic approach once you have a heap dump? What tools or analysis techniques are most effective for identifying the specific chain of references holding objects alive, and how do you isolate the leak's origin in code you didn't write, especially when garbage collection behavior seems normal until a specific threshold is hit?