I used Ecessa PowerLink WAN aggregator for the past 10 years. Their appliances have a number of features one being DNS SOA. We set our TTL settings for DNS records to 30 seconds, but could go lower, for faster IP address updates upon DNS cache expiration. We also setup DNS load balancing (alternating IPs), local and/or site-to-site fail-over redundancy with multiple IP addresses for the same FQDN.
With their appliances they claim BGP is no longer required due to their WAN aggregation, load balancing, link-state awareness, and DNS SOA features; now offers SD-WAN capabilities as well. Even though, I still implemented BGP multi-homing via a single ISP (iNAP formally interNap), who algorithmic-ally leveraged multiple tier 1 carriers for additional redundancy, used separate AS numbers out of their own and partner collocation data centers. In our case we used a /23 subnet split between two sites providing both DNS load balancing and site-to-site DNS redundancy. Note with the Ecessa's DNS SOA we experienced better fail-over response time than with our BGP multi-homing fail-over convergence times.
In production I had multiple appliances over the years in a HA fail-over configuration. My Initial implementation began load balancing two 3Mb circuits between with multiple ISPs, then a few years later upgraded to multiple 100Mb Internet connections, and then more recently multiple 1GB circuits. Basically, every three years had hardware and circuit upgrades.
(I do not work for Ecessa nor do I get compensated by them).