Bug 2207690

Created attachment 1964907 [details]
NetworkManager trace log

The content of http://file.apac.redhat.com/fge/OCPBUGS-11300/bug.sh is:


#!/bin/bash -x

ip netns add tmp
ip link add eth2 type veth peer name eth2peer
ip link add eth1 type veth peer name eth1peer
ip link set eth1 up
ip link set eth2 up
ip link set eth1peer netns tmp
ip link set eth2peer netns tmp
ip netns exec tmp ip link set eth1peer up
ip netns exec tmp ip link set eth2peer up

nmcli device set eth1 managed yes
nmcli device set eth2 managed yes


echo 'interfaces:
- name: bond0
  type: bond
  state: up
  ipv4:
    enabled: false
  ipv6:
    enabled: false
  link-aggregation:
    mode: active-backup
    options:
      miimon: 100
    port:
    - eth1
    - eth2
- name: bond0.1656
  type: vlan
  state: up
  ipv4:
    enabled: true
    dhcp: false
    address:
    - ip: 192.168.122.10
      prefix-length: 24
  ipv6:
    enabled: false
  vlan:
    base-iface: bond0
    id: 1656
- name: eth2
  type: ethernet
  state: up
- name: eth1
  type: ethernet
  state: up' | sudo nmstatectl set -

echo 'interfaces:
- name: bond0
  type: bond
  state: up
  ipv4:
    enabled: false
  ipv6:
    enabled: false
  link-aggregation:
    mode: active-backup
    options:
      miimon: 100
    port:
    - eth1
    - eth2
- name: bond0.1656
  type: vlan
  state: up
  ipv4:
    enabled: true
    dhcp: false
    address:
    - ip: 192.168.122.10
      prefix-length: 24
  ipv6:
    enabled: false
  vlan:
    base-iface: bond0
    id: 1656
- name: eth2
  type: ethernet
  state: up
- name: eth1
  type: ethernet
  state: up' | sudo nmstatectl set -

ip addr show bond0.1656

ip link set eth1 down
ip link set eth2 down
sleep 10
ip addr show bond0.1656

ip link set eth1 up
ip link set eth2 up
sleep 10

ip addr show bond0.1656
nmcli c
nmcli d

From the log and the discussion we had, it seems that the problem is that nmstate blocks autoconnect when adding/updating the connection; this prevents the connection to go up when the bond gets carrier again.

The solution is that nmstate should unblock autoconnect when it's done. It seems to me there is no API currently to do that, so we should either:

  1. unblock the connection when the Reapply() D-Bus is called. This probably solves the nmstate use case, but I can't think of a reason why this should be expected from an API point of view;

  2. introduce a new Update() flag to unblock autoconnect; this seems the best approach to me;

  3. maybe, use the AUTOCONNECT property on the device instead to block autoconnect?

Patch send to upstream https://gitlab.freedesktop.org/NetworkManager/NetworkManager/-/merge_requests/1649

NetworkManager will unblock the autoconnect on the success of reapply action considering the reapply serves the same purpose of activation.

Forgot to mention, to reproduce the problem, you also need to crate /etc/NetworkManager/conf.d/ignore_carrier.conf with content


[main]
ignore-carrier=no

RHEL 8.4 scratch build uploaded to https://people.redhat.com/fge/bz_2207690/

*** Bug 2211456 has been marked as a duplicate of this bug. ***