Don't use SIGALRM for the scheduler. This can be done entirely with pthread native constructs. (Merged by PR501 in trunk).
I suspect using this signal causes problems in a few cases where we saw worker threads never wake up. They might have missed the signal. Maybe this can happen when one of our libraries also catch sigalrm?
Currently I have 2 2.0.1 instances running monitoring any differences between unpatched and patched version.