We have been running into a potential problem
in the VM regarding the monitor table they use in the green
threads library. The problem shows up in our cartridge
when a Java program tries to create a new child process.
After debugging the source code, I found out that the fd_mon
table has not been dynamically resized properly and as a result
when new file descriptors are being allocated for the parent
and child processes (e.g. pipes) an ASSERT statement fails
because the fd_mon entry returned NULL instead of an initialized
monitor. After increasing the no. of preallocated fds slightly,
the problem goes away. But we'd still like to get a patch
from Sun if possible for our upcoming release.