Dear Maintainer,


The systemd service file for PostgreSQL currently uses "network.target", since a fixed wishlist https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=910991.
However, this does not guarantee that the network is fully available, which can lead to startup failures.

Suggested fix:
The service should instead use "network-online.target" to ensure that the network is fully up before PostgreSQL starts.

Besides, that is what postgresql proposes in its documentation:
https://www.postgresql.org/docs/current/server-start.html
```
[Unit]
Description=PostgreSQL database server
Documentation=man:postgres(1)
After=network-online.target
Wants=network-online.target
```

Here is the Systemd documentation
https://www.freedesktop.org/wiki/Software/systemd/NetworkTarget/#conceptsinsystemd
about the targets:
Steps to reproduce:
1. Install PostgreSQL on a system with a slow DHCP or Wi-Fi connection.
2. Reboot and observe that PostgreSQL may fail to start due to missing network resources.


Here is a timeline of what happened recently on my server:
```
$ cat systemctl status network.target
[...]
Mar 06 01:41:00 prod3 systemd[1]: Reached target Network.

$ cat /var/log/postgresql/postgresql-14-main.log
[...]
2025-03-06 01:41:03.982 CET [1990] LOG:  could not bind IPv4 address "172.25.2.13": Cannot assign requested address
2025-03-06 01:41:03.982 CET [1990] WARNING:  could not create listen socket for "172.25.2.13"

$ sudo systemctl status network-online.target
[...]
Mar 06 01:41:11 prod3 systemd[1]: Reached target Network is Online.
```

As you can see, the interface was available only after
network-online.target, so postgresql failed to bind on it before.
It happened on an 18.04 ubuntu server, on a 14 cluster, but I think the
issue is still relevant today.



Thanks for considering this issue (my first one! please excuse any
mistakes).

Re: Aurélien Pardon

Hi,

thanks for bringing this up.

I'm not sure changing this would really fix the problem. If wifi is
down, this would either still break, or delay starting the database
indefinitely.

Do you really need to bind to that IP directly? The standard setup
would be `listen_address = '*'`, then startup doesn't need any
specific IP to be available.

There is also the `net.ipv4.ip_nonlocal_bind` sysctl know that you
could enable.

Christoph

Hey.

I've stumbled over this bug when I thought to suffer from the same
issue, but mine turned out to rather be what I describe in #1122511.

Still systemd.special(7) describes network.target as:

in particular, no guarantees that physical devices are already there
and configured.

Cheers,
Chris.