I wrote a DNS server that did global software load balancing. Essentially it just has a health checking component and a sort and uses that to determine the closest healthy endpoint to return.

Mostly used for cluster failover or in cloud terms it can keep traffic within a zone if possible, otherwise within a region, otherwise closest region.

The reason it was my favorite project is because I was unqualified, but nobody else on my team was a DNS expert. So I got to drink from the firehose and learn. I had a really good testing feedback pipeline where basically visitors to our website did a couple extra background requests on their first page load and we used the web performance timing API to track DNS lookup times and TCP/HTTP times. So I every time I made a change I had millions of performance reports. I could see the impact of my changes in about 60 seconds in grafana.

Between learning something totally new and tying it to a short feedback loop with beautiful graphs I had a great time. Plus that product literally allowed my company to start using the cloud and build multi cloud systems.