3

I am developing a systemd shutdown script (unit file) for a postgresql database and I followed the below two posts but could not get it to work.

How to run a script with systemd right before shutdown?

https://superuser.com/questions/1016827/how-do-i-run-a-script-before-everything-else-on-shutdown-with-systemd

Here is what I am doing with my postgresql installation: I have written a startup script that starts the postgresql database cluster at boot time but the shutdown script shown below does not seem to work at shutdown time. While it works fine if I execute it with systemctl start service_name.

When I looked at the database log, I could see a message "LOG: received smart shutdown request" which is the message when database cluster goes down in normal mode which is what the shutting down operating system does when I shutdown the host with a "shutdown -h now". What I want to see is "LOG: received fast shutdown request" which is a fast (immediate mode) shutdown which occurs when I run the script with a systemctl start service_name

Here is the unit file definition

[Unit]
Description=Database Shutdown
Before=shutdown.target reboot.target halt.target
DefaultDependencies=no
[Service]
ExecStart=/usr/bin/su - postgres -c "/opt/postgres-95/bin/pg_ctl -D /database/inst1/data -w stop &"
Type=oneshot
RemainAfterExit=yes
[Install]
WantedBy=shutdown.target

I also tried to ditch the [Install] section but the systemctl would not let me enable the service without the [Install] section. could someone please point me in the right direction.

Community
  • 1
Nawaz Ahmed
  • 31
  • 1
  • 2
  • You could just use one of the existing postgresql service files (for example the one from [Arch Linux](https://git.archlinux.org/svntogit/packages.git/tree/trunk/postgresql.service?h=packages/postgresql) or [Debian](https://alioth.debian.org/scm/loggerhead/pkg-postgresql/postgresql-common/trunk/view/head:/systemd/postgresql%40.service)) instead of reinventing your own one. Also, having a service file just to shut down a service is not correct, you should have one service file that handles starting, stopping and reloading the service – Wieland Jun 23 '16 at 11:19
  • Your stop command looks like you're trying to shut something down. You'd likely be much better off going with standard practices of using a unit file to manage the whole service, both start and stop, not just stopping it. `ExecStop` would then do what you want without any hackery. Also, you shouldn't use `su` in a unit script, this is what the `User=` parameter is for. – phemmer Jun 23 '16 at 12:18
  • sorry for the late reply folks, i got tied up and juggling around other things @Wieland, thanks for pointing out to the already existing service file for this purpose, it gives me a clarity about where should i start from. Patrick, i did play around with the User= parameter but found that the processes are coming up as owned by the 'root' user which is not what i wanted. I wanted the processes to be owned by the 'postgres' user or any other user which will own the PostgreSQL database directory. – Nawaz Ahmed Jul 11 '16 at 23:54

1 Answers1

1

systemd is designed to cleanly handle this case. You can do so by creating a small file specifies a command you want to run as part of stopping the PostgreSQL service. First, create a directory named:

/etc/systemd/system/postgresql.d/

In it, create a file with any name that ends in .conf, like custom-shutdown.conf. Add to that file a couple lines like this:

[Service]
ExecStop=/usr/bin/su - postgres -c "/opt/postgres-95/bin/pg_ctl -D /database/inst1/data -w stop &"

You may wish to use ExecStopPost= depending your case.

See official docs for Extending the Default Configuration and also official docs for the ExecStop= and ExecStopPost=.

Mark Stosberg
  • 7,420
  • 1
  • 32
  • 42
  • 1
    thanks for the response, I shall try these steps and get back you all. Thank you all for your help guys, I appreciate it. – Nawaz Ahmed Jul 11 '16 at 23:55