Created attachment 1959547 [details] NetworkManager trace log Description of problem: Given a NIC with both ipv4 and ipv6 enabled with auto method, only DHCPv4 provided nameserver. When reapply with ipv6 disabled. Then noticed DNS nameserver been purged. Version-Release number of selected component (if applicable): NetworkManager-1.42.2-1.el9 How reproducible: 100% Steps to Reproduce: 1. Please reach out to QE in https://issues.redhat.com/browse/OCPBUGS-11827 for a cluster. 2. 3. Actual results: DNS nameserver been purged after `oc apply` which then lead to rollback. Expected results: DNS nameserver retrieved from DHCPv4 unchanged. Additional info:
Wen, I headerd, you were able to reproduce this issue? Could you explain in detail how to reproduce it. Or attach a reproducer script. Or write a NM-ci test (choose whatever approach you want). Also, to my understanding you identified https://gitlab.freedesktop.org/NetworkManager/NetworkManager/-/commit/0a02995175e06e62924705393121a1c5efc3822d as the fix? In that case, it is fixed in upstream 1.43.5+, 1.42.6+, 1.40.18+. It is also fixed in rhel-9.3 package NetworkManager-1.43.5-1.el9. As of now, it is not yet fixed in latest rhel-9.2, or latest rhel-8.
(In reply to Thomas Haller from comment #6) > Wen, I headerd, you were able to reproduce this issue? > > Could you explain in detail how to reproduce it. Or attach a reproducer > script. Or write a NM-ci test (choose whatever approach you want). > > > Also, to my understanding you identified > > > https://gitlab.freedesktop.org/NetworkManager/NetworkManager/-/commit/ > 0a02995175e06e62924705393121a1c5efc3822d > > as the fix? > > > In that case, it is fixed in upstream 1.43.5+, 1.42.6+, 1.40.18+. > It is also fixed in rhel-9.3 package NetworkManager-1.43.5-1.el9. > As of now, it is not yet fixed in latest rhel-9.2, or latest rhel-8. I discussed this bug with bgalvani, the upstream nm-1-40 and nm-1-42 branches already had that commit and there was no customer request for backporting into rhel-9.2/rhel-8.8. About the steps to reproduce the bug: 1. I used the nm-in-container.sh script and added the eth1 interface 2. nmcli connection add type ethernet ifname eth1 con-name eth1+ ipv6.method manual ipv6.addresses $someaddr ipv6.dns $somedns 3. nmcli connection up eth1+ 4. nmcli | grep -A10 "DNS configuration:" (In this case, you should be able to see some IPv4 and IPv6 DNS) 5. nmcli connection modify eth1+ -ipv6.addresses $someaddr -ipv6.dns $somedns ipv6.method disabled (Remove the IPv6 addresses and DNS, and set the Ipv6 method into disabled) 6. nmcli device reapply eth1+; nmcli | grep -A10 "DNS configuration:" (The expected behavior is that the IPv4 DNS is not removed, but if the bug persists, then the IPv4 DNS is removed)
Hi Wen, Please prepare a scratch build of 9.2 for Ben Nemec to try out.
(In reply to Gris Ge from comment #12) > Hi Wen, > > Please prepare a scratch build of 9.2 for Ben Nemec to try out. Sure, I will do it!