Error handling is too fragile #8
Description
If there's an issue with one docker service (it should not but 😬), our systemd unit is configured to try and restart forever (with a 30s delay) until docker compose is happy.
There are little chances that this would resolve such an issue but the problem is not there. It is that since we don't persist certs, ATM we request certs every time the reverse-proxy is started (which is fine in a working scenario).
This means that in case of error, we're requesting it every 30s and quickly hit https://letsencrypt.org/docs/too-many-registrations-for-this-ip/ banning the IP for 24h.