Found a strange combination of multiple factor but fixing the issue could have been longer then just recreating the VM (as we have full ansible automation) and as node was removed from pool already (nobody was using that node) I decided to just delete the VM and start from fresh qcow2 image/filesystem.
Current status : node is reinstalled and under ansible control and is getting mirror content back.
Once fully validated, it will be added back and I'll reach out to @adrian and @kevin to ensure that at the Fedora infra side, it's pointing to a CNAME to a record we (centos-infra) can quickly update to transparently redirect elsewhere, should such issue would appear again

status update : node was fully reinstalled and under ansible control but it just takes time to rsync data down. So just keeping this ticket open as reminder to add it back when it will be having full content (and temporary removed from ansible automation through /etc/no-ansible for local rsync job in parallel to speed things up)

status update : node was fully reinstalled and under ansible control but it just takes time to rsync data down. So just keeping this ticket open as reminder to add it back when it will be having full content (and temporary removed from ansible automation through /etc/no-ansible for local rsync job in parallel to speed things up)

Node is fully back online with all data and in sync with primary reference mirror. DNS record updated and so we have again redundancy/workload balance for the mirror cdn (both mirror.centos.org and mirror.stream.centos.org)
Closing