For WAN fail-over, the current recommendation is to use mwan3. However, there is a significant cognitive overhead even with a pure IPv4 setup, because mwan3 also covers a more exotic use case of load balancing, and needs configuration of interfaces, members, policies, and rules, which looks much more complicated than needed.
There is also an interesting setup, documented here, which uses only metrics for fail-over, and does not involve any policy-based routing. Basically, it sets a backup connection's metric higher than the main one, and relies on the main connection to disappear when it is broken. It even works for PPPoE, but not for static IP or DHCP.
Therefore a feature request: make connectivity checking a core feature. Connectivity should be possible to periodically check, via ping or HTTP, through all links that have a route to 0.0.0.0, even if this route does not have the lowest metric. On failure, the interface's metric should be increased by something like 20000.
Or in other words: please drastically simplify the fail-over configuration for people who only need WAN fail-over and do not care about load balancing.