wmf-auto-reimage fails 1 out of 2 times for me (not necessarily because the script, but faulty harware, installer, ipmi problems, not using --new after if failed the first time, etc.), and then it doesn't fail gracefully. Normally the neodymium process gets stuck (which is relatively easy to kill, just screen + control-c), but also at least 3 processes on puppetmaster1001.
Ideally, control-c would be captured and try to clean up better, alternatively, some parameter to clean up or, on start, make sure there are not other zombie installs ongoing on puppetmaster. Maybe some watchdog to detect stuck installations? Maybe this will not be needed if wmf-reimage gets deprecated.