The main difference in signal handling is that by default, the JVM itself traps INT, so you need to explicitly trap it in Ruby code. MRI automatically traps INT and raises Interrupt from startup.

There are also a few other signals the JVM traps that are not related to this issue.

The only thing I can think is that somehow the trap is never getting set.

It seems as though my approach on the rspec test was causing a timing issue. I'd send an asynchronous signal, then hope the thing I wanted to happen did before the test ended. I believe the different performance characteristics explains this discrepancy.

I switched my test methodology, and all looks good. That PR was merged. From what I can tell, this issue can be simply closed.

@myrridin please feel free to reopen it again if you think it is JRuby issue