Argh.. I think this is done via a reverse DNS lookup in the koji code. We don't control reverse DNS for this space so it is not an easy fix. I will see if we can get this corrected in Red Hat's DNS.

If the address range for the s390x builders is unique, then the internal hostnames are some historical relics, I guess ...

Why is rpm using reverse dns here? Seems pretty odd, it seems like it would be going out of it's way to do so.

CCing the rpm team - @pmatilai @ffesti

Rpm has "canonicalized" the buildhost by doing gethostname() followed by gethostbyname() from day one. Literally - that code is present in the initial revision (cvs back then) from 1996.

So, not wanting to die on the hill of "lets remove this", I went and added a reverse dns map to our nameservers for these hosts.

However, we need to get RHIT to delegate it to us internally to make that work. So, keeping this to track that.

Ah, no, we don't need to wait. It's working now.

Thanks!

Taking a random build from today, the result seems OK:

$ wget https://kojipkgs.fedoraproject.org//packages/perl-Alien-Libxml2/0.15/1.fc33/s390x/perl-Alien-Libxml2-0.15-1.fc33.s390x.rpm
$ rpm -qpi perl-Alien-Libxml2-0.15-1.fc33.s390x.rpm|grep Host
Build Host  : buildvm-s390x-18.s390.fedoraproject.org

So, not wanting to die on the hill of "lets remove this", I went and added a reverse dns map to our nameservers for these hosts.

The association for preservation of strange software artifacts it's appreciation.

Argh, no edits. expresses its appreciation.