Systemd-resolved works fine for us on Ubuntu 26.04

For a very long time, we've configured our DNS servers through the traditional /etc/resolv.conf method. This continued after systemd-resolved was introduced and even after it became the default in Ubuntu server installs, for various reasons (eg, also). However, for reasons covered in my entry from this past summer, we wound up making our Ubuntu 26.04 install use systemd-resolved instead of our traditional /etc/resolv.conf.

(The Ubuntu 26.04 nsswitch.conf uses 'hosts: files dns', so we're not using the nss-resolve NSS module and programs are still doing DNS resolution, talking to resolved's local nameserver on 127.0.0.53. In Ubuntu 26.04, the 'libnss-resolve' package isn't installed by default and to my surprise it lives in Ubuntu's less-supported 'universe' repository, although it's part of systemd and systemd itself is in the 'main' repository as you'd expect.)

At this point we've been running Ubuntu 26.04 on a significant number of our servers for several months, including our external mail gateway (which obviously does plenty of DNS queries). Our experience with systemd-resolved is that we haven't noticed anything. It's been effectively invisible to us that we're using resolved instead of directly querying our local caching DNS servers. Certainly we haven't seen any problems come up, including on that external mail gateway. This is of course what we expected and hoped, but it's good to confirm and note this sort of positive result.

We have two identical local DNS resolvers, which we consider to be a main one and its backup; the main DNS resolver is the one we list first and that we expect most queries to go to. Looking at the state of resolved on our 26.04 servers, a modest number of the servers have switched from the primary DNS resolver to the backup. There's no particular pattern to which servers have switched, so I assume it's simply a matter of those servers making DNS queries at a moment when the primary DNS resolver didn't answer but the backup one did.

(We configure the systemd-resolved upstream DNS servers through Netplan, which means that they're attached to a specific interface. For each such interface, systemd-resolved has a 'current' DNS server that is the one it normally sends queries to. Systemd-resolved only switches to another configured DNS server if a query to the current DNS server fails. The corollary to this is that if a system switches its current DNS server, you know systemd-resolved saw at least one failure that made it flip over. Unfortunately I don't think systemd-resolved logs any message when this happens.)