In T384944, mw-wikifunctions migrated to using k8s ingress. Among other things, that means it not longer has its own discovery services - i.e., mw-wikifunctions.discovery.wmnet and mw-wikifunctions-ro.discovery.wmnet are really just CNAMEs for the respective k8s-ingress-wikikube services [0].
At the very least, the MEDIAWIKI_SERVICES and MEDIAWIKI_RO_SERVICES lists in [0] need to be updated to remove these not-actually-a-discovery-service services. If we're really careful about it, that can happen once [2] is merged (ideally not before, given how some of the filtering works in the sre.discovery.datacenter cookbook).
However, there's kind of a broader question here:
We now have a mediawiki instance whose active-passive traffic moves during the Day 1 services switchover (i.e., with k8s-ingress-wikikube-rw), rather than coordinated with the database primary switchover on Day 2.
Having primary database ops going cross-DC for 24h "should be fine" now-a-days given the rewriting we do in the secondary DC in mediawiki config [3]. In fact, we had that happen with mw-videoscaler during the last switchover, since transcode processing stays local now: with eventgate being depooled in the primary DC on Day 1, the jobs would be enqueued and processed in the then-secondary DC (see T372849#10653971).
In any case, we should probably document the desired behavior somewhere to make this explicit.
[0] https://gerrit.wikimedia.org/r/c/operations/dns/+/1133878
[2] https://gerrit.wikimedia.org/r/c/operations/puppet/+/1163856