Re-reading the OP I see where you tried to communicate this fact but it is so contrary to the documentation and general use I hadn't considered that this might be what you had done.
I expected the client will ask the server using the service name.
How does the client even know where the server is in order to ask it? The whole point of having pg_service.conf is to avoid passing arguments on the command line and instead store them attached to a name in a local file.
If the service will be found on the server, the server should accept the connection from the client.
Having the server accept a connection just because the client knows the name of a service is insecure.
In pg_service.conf file can be stored user name and password.
Yes, the user name and password your client is going to present to the user to identify itself. The host field indicates where the server is located.
I want to use login of service type, because I won't to store any login information (ie password) on the client side.
What you describe is worse than having a password securely stored on the local machine.
What you want to do is possible but using the pg_service.conf file plays no role. And really the password should not be placed into pg_service.conf anyway - if you are going to use that file you probably should also use the .pgpass file.
Copping the pg_service.conf to the client and setting the system variable to this file, rather has no sense.
Which machine are you thinking the "system variable" is being set on?
