If only there was a way to have a look at the source code...
// Next we try to use an alternative method which uses the host name, process ID, and a time stamp
// We then hash it with md5 to get 32 bytes, and then add 4 extra random bytes
// Collision is possible of course, but unlikely and suitable for most industry needs (e.g.. aggregate tracking)