Of course you catch and handle errors. Or at the very least you must think about whether it makes sense to handle an error on the current abstraction level that you are on, which means you must always consider option 3. Only if you decide that it's not possible or doesn't make sense to handle it ("I am too lazy" is not a reason to not handle it!) you decide whether you need to throw it one level up the stack. And if doing that is the sensible thing, you think about whether you want to make the exception part of the API of whatever you currently write or not. If you want to, you throw the exception without doing anything (most of the time you don't even log it, because you understand that the receiver at the end, the one who finally handles it, is responsible of logging it, otherwise you get multiple logs of the same exception written out just because someone sprinkled logging into several abstraction layers). If you want to hide the exception type somehow, you wrap it in a more fitting exception that's explicitly or implicitly part of the API of whatever you're writing, and throw that one. That can, but doesn't have to be, a runtime exception.

I really don't get what's so hard about exception handling. Java's exceptions are a pretty good analog to how one must model proper error handling in multi-layered software applications anyway, regardless of the programming model used to implement it.