On Tue, Jun 10, 2025 at 07:28:15PM +0200, Dag-Erling Smørgrav wrote: > Rick Macklem <rick.mack...@gmail.com> writes: > > Gleb Smirnoff <gleb...@freebsd.org> writes: > > > Konstantin Belousov <kostik...@gmail.com> writes: > > > > Apparently we already have the feature called 'warmstart', where rpcbind > > > > can be restarted and existing registrations are reloaded. So it is in > > > > fact already solved, assuming admin is careful enough to use -w. > > > I really don't have a strong opinion on what would be right here. I have > > > no > > > idea on how wide is the use of rpcbind w/o NFS. Kostik, Rick and > > > Dag-Erling, > > > may you together come to agreement on what is right here? > > I don't have a strong opinion on it either, since most kernel configs > > include NFS options, so the krpc is there anyhow. > > It is not in fact “already solved”, and I'm frankly disgusted at the > amount of energy being spent on bikeshedding this after the fact > compared with the complete lack of interest in the problem prior to my > commit. > > Not everybody runs GENERIC. Right.
> Practically nobody uses NIS. Wrong. > Something > needs to ensure that krpc is loaded when rpcbind starts in nearly 100% > of rpcbind use cases. My original choice was a small change to the rc > script, but Rick insisted on doing it directly in rpcbind instead, > although he now conveniently washes his hands of it. > > Konstantin, I don't care how you solve this, but I need NFS to work, and > I don't ever want to see “netlink: could not create service” on my > console again. And I do not want to get a host that cannot be logged in because rpcbind exited with an error that krpc cannot be loaded, thus making NIS nss unoperational. This happens somewhere not on my local machines, I would indeed not use NIS at home. Changing all fleet there because the proper fix is considered bikeshed is not feasible.