[systemd-devel] Alias for SMTP providers [ie. mutually exclusive service alternatives]

[Kenneth Porter]

--On Wednesday, June 16, 2021 11:46 AM +0300 Mantas Mikulėnas 
<grawity at gmail.com> wrote:

>> What does "needs to send mail" even mean? That /usr/sbin/sendmail can
>> be called to queue a message? That you can talk to localhost:25?
>>
>> A well behaving MTA actually make /usr/sbin/sendmail work without the
>> main mail daemon to be up. The mail is then only enqueued, but not
>> dispatched, but that'll be done once the service is fully up.
>>
>
> Hmm, I was going to post the same at first, but it doesn't really work in
> reverse -- if you want to send mail on shutdown and if the goal of
> After=postfix is "run my ExecStop before postfix gets stopped", then
> ability to queue doesn't help all that much.

This isn't that intractable. The objective is to allow the message to 
escape the box to the next hop before shutdown. (On startup there's no 
issue.) So I can issue a queue flush in the ExitStop and then poll the 
queue for empty until a timeout (say, 5 seconds) is reached to allow the 
queue to drain. Admittedly this won't work on a system that's getting 
flooded with mail from other sources. On a lightly-loaded system, just 
sleeping for 2-5 seconds is enough to let the message escape.

My current workflow for a reboot is to mail myself the result of uptime and 
then start a ping from another box to monitor when it comes back up. So I'm 
trying to script away from that to just looking at that server's mail 
folder.

My bigger concern is how to use one Unit file to handle different MTAs.