Since we abandoned bind mounts on the NFS clusters per https://gerrit.wikimedia.org/r/c/operations/puppet/+/571821, nfs-exportd throws non-zero statuses on stretch+ on DRBD secondary nodes (which makes perfect sense). Why it didn't on Jessie is the real mystery.
As a bandaid fix, we switched from subprocess.check_call() to subprocess.call(), but that seems like a good way to mask serious errors in the future. Adding some cluster awareness like that used by maintain-dbusers should fix things in a more sensible way. While we are at it, it wouldn't hurt to set this to only run exportfs when changes are made. Re-running exportfs every 5 minutes could not possibly be good for the system's performance despite the fact that we've done it for years.