Why are processes run with nohup in rc.local killed?

Question

I have the following script under raspbian jessie:

file test.sh contains:

#!/bin/sh -e
nohup sleep 321 &
sleep 120
false
exit 0

When run as normal user in a terminal session, I have the following situation:

pi@raspberrypi:~ $ ps -ejf |grep sleep
pi        5073  5072  5072  4618  0 15:16 pts/1    00:00:00 sleep 321
pi        5074  5072  5072  4618  0 15:16 pts/1    00:00:00 sleep 120

and after the script exits, the sleep 321 process is reparented and still running as expected.

pi@raspberrypi:~ $ ps -ejf |grep sleep
pi        5073     1  5072  4618  0 15:16 pts/1    00:00:00 sleep 321

Why doesn't this work in rc.local? The command run with nohup gets just killed.

pi@raspberrypi:~ $ ps -ejf |grep sleep
root       450   449   449   449  0 15:45 ?        00:00:00 sleep 321
root       451   449   449   449  0 15:45 ?        00:00:00 sleep 120

and after 120 seconds the sleep 321 process just disappears.

It only works in rc.local when I replace false with false || true, so my guess is, it has something to do with running the script with -e, but why the difference between running test.sh from a terminal and rc.local at boot?

Answer 1

Because nohup would allow only for SIGHUP ignore. But there's also "session" concept in UNIX processes; typical way to daemonize a process is somewhat way more lengthy than just ignoring SIGHUPs.

As Gene Pavlovsky suggests in his comment: "Take a look at daemonize. Besides a nice tool, it has pretty good explanations on what a daemon is."

— You should indeed.