Draft: Try libvirt-metal with vlanif primary interface
What does this MR do and why?
Using the evolution for libvirt-metal introducing bonded interface, in sylva-projects/sylva-elements/container-images/libvirt-metal!42.
It proposes capm3-virt envs on libvirt-metal infra use vlanif as primary interface and active-backup bond mode.
Own notes:
-
tried
ip link add/del
workaround to get bond0.100 up for Ubuntu images (!2153 (comment 1948470371)) but it did not help;
Issue on Ubuntu images: bond0.100 does not come up. What we know:
- noticed that changing
/etc/netplan/50-cloud-init.yaml
will not reflect in bond config, was trying to change
bonds: bond0: parameters: mii-monitor-interval: 1000 # from 100, to reduce the sensitivity to brief link status changes
after noticing some possible link flaps in
dmesg
:[52171.528227] bond0: (slave ens5): link status definitely down, disabling slave [52171.528249] bond0: now running without any active interface! [52171.528286] bond0: (slave ens4): link status definitely down, disabling slave [52171.561796] 8021q: adding VLAN 0 to HW filter on device ens5 [52171.561968] 8021q: adding VLAN 0 to HW filter on device ens4 [52171.632301] bond0: (slave ens5): link status definitely up, 0 Mbps full duplex [52171.632312] bond0: (slave ens5): making interface the new active one [52171.632613] bond0: active interface up! [52171.632653] bond0: (slave ens4): link status definitely up, 0 Mbps full duplex [52681.056268] bond0: (slave ens5): link status definitely down, disabling slave [52681.056279] bond0: (slave ens4): making interface the new active one [52681.140252] 8021q: adding VLAN 0 to HW filter on device ens5 [52681.140585] 8021q: adding VLAN 0 to HW filter on device ens4 [52681.160295] bond0: (slave ens5): link status definitely up, 0 Mbps full duplex [52681.264332] bond0: (slave ens5): link status definitely down, disabling slave [52681.282691] 8021q: adding VLAN 0 to HW filter on device ens5 [52681.368295] bond0: (slave ens5): link status definitely up, 0 Mbps full dupl
due to netplan bug https://bugs.launchpad.net/ubuntu/+source/nplan/+bug/1746419, found through https://askubuntu.com/a/1216635/1178214, for which
ip link del dev bond0 && netplan apply
as root helped (cat /proc/net/bonding/bond0
showed then the newMII Polling Interval (ms): 1000
)- Francois found that
$ ip link add link bond0 name bond0.100 type vlan id 100 $ ip link del bond0.100 $ netplan apply
or
$ sed -i 's/ens4/enp0s4/g' /etc/netplan/50-cloud-init.yaml $ sed -i 's/ens5/enp0s5/g' /etc/netplan/50-cloud-init.yaml $ reboot
ran manually on a libvirt-metal Ubuntu VM where the issue happens are successful workarounds.
- noticed that changing
-
was using remote values for openSUSE in sylva-projects/sylva-elements/ci-tooling/ci-deployment-values!5 (diffs) to set
capm3.bondXmitHashPolicy: layer2
for openSUSE-based deployments (still use defaultcapm3.bondXmitHashPolicy: layer3+4
for Ubuntu), but moved to havinglayer2
for both.