Why it works
Every SSH session has its own process. systemctl restart ssh restarts only the main sshd, the one that
accepts new connections — sessions already open are left alone. A config error, a wrong AllowUsers, passwords
closed while the key doesn’t work — all of that breaks new logins, and from the old session you put everything
back.
The firewall is the same: ufw enable warns Command may disrupt existing ssh connections, but an established
connection keeps working: one of the first rules in /etc/ufw/before.rules lets packets of open connections
through (--ctstate RELATED,ESTABLISHED -j ACCEPT). A new connection without a rule for port 22 won’t get in.
What it looks like
window 1 (keep open) window 2
sudo nano /etc/ssh/sshd_config.d/…
sudo sshd -t && sudo systemctl restart ssh
ssh web1
sudo -v ← logged in, sudo works
exit …carry on
If the second session can’t log in, fix it in the first: sudo rm your file or git checkout
(Server configs in git), restart again, try again from the second window. If both sessions are lost,
the provider console is what’s left (Provider console).
For changes that can cut you off completely (firewall, network, port) there’s also auto-rollback: a script puts the old config back if nobody confirms within 2 minutes that login still works (An edit script with automatic rollback).