atd, batch // Setting the load limiting factor

Question

I am launching non interactive jobs using batch, and I would like to increase the load limiting factor in order to use all 8 of my cores. I am on Ubuntu 16.04 LTS.

From what I understand, batch uses atd to do the jobs. Jobs start when the load factor goes under a threshold, called the load limiting factor. It is said in the man of atd that we can change this factor using the -l option.

My question: how can I use this atd -l XX option? When I type, for instance, atd -l 7.2 before batch, it doesn't seem to be changing anything.

What I have found so far:

• In this question How to run bash script via multithreading, one contributor proposes to do this in the 'atd service starting script'. I guess that it refers to the /etc/init.d/atd, but I do not know what to change there, cf next bullet point.
• I have found pages, such as this one http://searchitchannel.techtarget.com/feature/Understanding-run-level-scripts-in-Fedora-11-and-RHEL, where they propose to: "modify the following line (in the start section) of the /etc/init.d/atd script: daemon /usr/sbin/atd. Replace it with this line, using the -l argument to specify the new minimum system load value: daemon /usr/sbin/atd -l 1.6". However, there is no such a line in /etc/init.d/atd.

It seems that it can be introduced in the /etc/init.d/atd, but I do not know where. I have never changed such files.

So, how can I change the load limiting factor used by the batch command?

Answer 1

Found a solution:

• Create a file: /etc/init/atd.override
• Add a line exec atd -l 7.2
• Then sudo service atd restart

It has to do with how the 'Upstart init daemon' works. Explanations there: http://linux.die.net/man/5/init If the file /etc/init/atd.override already exists with an line starting with exec, edit this line.

Answer 2

Edit /lib/systemd/system/atd.service on Ubuntu 16.04 / systemd.

After appending -l «load avg» to the ExecStart:

$ cat /lib/systemd/system/atd.service 
[Unit]
Description=Deferred execution scheduler
Documentation=man:atd(8)
[Service]
ExecStart=/usr/sbin/atd -f -l 7.2
IgnoreSIGPIPE=false
[Install]
WantedBy=multi-user.targe

you'll see a message when running systemctl status atd or (service atd status):

Warning: atd.service changed on disk. Run 'systemctl daemon-reload' to reload units.

Running that command gets rid of the warning, but to actually pick up the change the service needs to be restarted by systemctl restart atd (service atd restart).

Answer 3

This builds upon the answer by Wesley B.. It applies if using systemctl. Instead of updating the service configuration directly, an override can be defined for it. It is better to do it this way because the OS can otherwise replace the service configuration at any time. This answer was tested with Ubuntu 21.10.

1. Run sudo systemctl edit atd and insert:

[Service]
ExecStart=
ExecStart=/usr/sbin/atd -f -l 3.0 -b 5

Above, the line which is just ExecStart= with no value is necessary to nullify the old value. In the next line, set the values of -l and -b as required, referring to its man page.

This writes the above lines to /etc/systemd/system/atd.service.d/override.conf. If this file is added without systemctl edit, then sudo systemctl daemon-reload may have to be run.

2. Run sudo systemctl restart atd.

3. Verify the status:

systemctl status atd
ps aux | grep [a]td

These should show the service and process status, both with the customized args.

Answer 4

I don't have Ubuntu 16.04 LTS nearby, but on Debian-based system /etc/init.d/atd will usually say something like:

case "$1" in
  start)
        log_daemon_msg "Starting deferred execution scheduler" "atd"
        start_daemon -p $PIDFILE $DAEMON -l 7
        log_end_msg $?

note where I added -l 7 part. Then you need to restart with /etc/init.d/atd restart, and check with ps auxwww|grep atd if it was accepted.

note that sometimes you'll have /etc/default/atd which is the easier way (and better, as your changes won't be silently wiped on next upgrade)