A service is restarting every few days with an OutOfMemoryError. How do you find the cause?

First read the message, since heap space, metaspace, direct buffer and GC overhead limit exceeded point at very different causes. Run with heap dump on out of memory so you capture the state at failure, then open the dump in a memory analyzer and look at the dominator tree to see what retains the most. Compare dumps over time to separate a genuine leak from a working set that is simply larger than the heap.

The variant matters. Java heap space means live objects genuinely fill the heap, metaspace means classes are being loaded and never released, and a direct buffer error points at NIO or a native library rather than my objects. So I read that first, then make sure the JVM is running with heap dump on out of memory and a path on a volume that survives the restart, because guessing without a dump wastes days. With the dump I look at retained rather than shallow size, and the dominator tree usually names the culprit in a couple of minutes. What I actually find is repetitive: a map used as a cache with no eviction, a thread local set on a pooled thread and never cleared so it lives as long as the pool, or a repository method returning every row because someone dropped the pagination. I also check garbage collection logs to see whether the heap was climbing steadily or spiking on particular requests. The fix is bounding something, a cache with a maximum size and expiry, or streaming the query, and then I watch the live set across a week to confirm it is flat.