Home / ubuntu / Questions / 1213854
In Process
Askerman
Asked: 2020-03-01 03:42:00 +0800 CST

Systemd unit fails to run a bash script which otherwise works

  • 772

I have the following script:

#!/bin/bash

echo "Merging pool...";

if !(mergerfs -o minfreespace=4G,defaults,allow_other,category.create=ff /media/pool/hdd1:/media/pool/hdd2:/media/pool/hdd4:/media/pool/hdd5 /media/pool/merged/)
then
        echo "Merging failed."
        exit 1;
fi

echo "Done";

exit 0;

It is saved in /home/me/scripts/mergePool.sh. If I manually run the script like this: sudo ./mergePool.sh, it works fine

I made this unit in /etc/systemd/system/mergePool.service:

[Unit]
Description=Merge all pool drives into one big virtual drive

[Service]
Type=oneshot
ExecStart=/bin/bash /home/me/scripts/mergePool.sh

[Install]
WantedBy=multi-user.target

If I run the unit like this: 

sudo systemctl start mergePool.service

It produces no error and it doesn't work at all.

This is the log:

sudo systemctl status mergePool.service
● mergePool.service - Merge all pool drives into one big virtual drive
   Loaded: loaded (/etc/systemd/system/mergePool.service; enabled; vendor preset: enabled)
   Active: inactive (dead) since Sat 2020-02-29 12:26:07 CET; 6min ago
  Process: 2066 ExecStart=/bin/bash /home/me/scripts/mergePool.sh (code=exited, status=0/SUCCESS)
 Main PID: 2066 (code=exited, status=0/SUCCESS)

Feb 29 12:26:07 junkbox systemd[1]: Starting Merge all pool drives into one big virtual drive...
Feb 29 12:26:07 junkbox bash[2066]: Merging pool...
Feb 29 12:26:07 junkbox bash[2066]: Done
Feb 29 12:26:07 junkbox systemd[1]: Started Merge all pool drives into one big virtual drive.

This is quite confusing to me, why doesn't it work when the script is ran by systemd?

server command-line bash systemd 18.04

2 Answers

  1. While I don't know why your script is failing, I do see a couple of things that should be fixed, which might get you closer to a solution even if they aren't the solution in and of themselves:

    • When launching an executable script that has a sha-bang specifying the appropriate shell/interpreter, as in your example, don't call /bin/bash explicitly when you want to run the script. Just call the script.
    • If and when you actually need to pass commands directly in a bash invokation, the syntax is /bin/bash -c "<commands to execute>". What your example is doing is first running /bin/bash with no arguments or commands, then (as a completely separate second command) running your script. In a oneshot type service that shouldn't actually break anything, but do keep in mind that oneshot is the only service type that allows more than one command in this way.
    • By wrapping a single command in a shell script in this context, all you really accomplish is making things opaque to systemd. Try simply putting your mergerfs command directly in the unit file's ExecStart= directive (you may need to enclose it in double quotes). That way systemd can capture mergerfs's exit status, and the journalctl log will contain any output and/or error messages produced by mergerfs. Then if it fails you should have some information as to why.

    Note: For example, the .desktop files used for autostarting desktop goodies only accept one command. If you want to, say, autostart a program after a little delay, that one command could be /bin/bash -c "sleep 10; <program to launch>" It works because you're actually passing the double-quoted string of commands to bash as an argument, so the only command being run from the .desktop file is bash itself.

    • 1

  2. You should put the whole thing you want to run into an infinite loop without the exit code and a sleep 15 eg. to make the service ongoing.

    • -1