We are seeing an issue on Rocky Linux 9 where VLAN interfaces on bridges
have their MTU set to MTU-1 after rebooting. This is likely related to
how MTU is set by NetworkManager for bridges: it first sets the MTU to
MTU-1 before setting it to the correct value [1].

We can work around this issue by not explicitly setting the MTU
configuration on the VLAN interface if it is the same as the parent
bridge. It will be automatically inherited from the parent.

[1] https://gitlab.freedesktop.org/NetworkManager/NetworkManager/-/commit/864872e9a8c694ee7472154bf45f869040310b71

Closes-Bug: #2039947
Change-Id: I23366f4de7842e7c2fe40e431fac76f26e9892de

This release includes the following fixes:

* Remove nmconnection files generated by cloud-init
* Remove ens3 also from NetworkManager config

Closes-Bug: #2039975
Change-Id: Id3e7913f4ba9d05cec332dab560fc42d44dd4130

Closes-Bug: #2038889
Change-Id: I4609494d009c6e1f97d833a9a11529d107b3216f

Stop configuring the seed host as gateway when seed SNAT is disabled
(which is the default since Xena), otherwise overcloud hosts may take a
long time to gather facts until their network configuration is applied
(probably because DNS servers are thought to be reachable until the
gateway is changed or removed).

Closes-Bug: #2039461
Change-Id: Ib847d3420dee374cec067cd8af519b510be04120

Back in the Ussuri release, we changed the Bifrost kernel location to
use ipa.kernel instead of ipa.vmlinuz. While this works fine for newly
discovered nodes, any node added to Bifrost in Train or earlier will
have kept ipa.vmlinuz as its deploy kernel.

This can cause issues since upgrading Bifrost or building new deployment
images will update ipa.kernel, but not ipa.vmlinuz, resulting in nodes
booting with an old kernel and a new ramdisk.

Fix by adding a new rule updating the legacy kernel location and
documenting how to update node information.

Change-Id: I6ae44dcae78424b4638762d015cf24336444707f

This change adds a variable that can be used to disable kolla docker
registry login attempts, even when the registry username and password
are set.

This is required for deployments using a non-standard containerised
registry deployed on the seed during the deploy-container step,
since it takes place after the registry login attempt.

Change-Id: Ie17ef9ce1147cbaec2e42db932c7d59293b49b1b

This requires disabling libvirt_vm_trust_guest_rx_filters, which when
enabled triggers the following errors when booting baremetal instances
with Tenks on Libvirt 9 (and most likely since 8.9.0):

    Cannot set interface flags on 'macvtap1': Value too large for defined data type

This is apparently triggered by a Libvirt commit refreshing rx-filters
more often [1].

As explained in I71a2051d8acd63379bd70bc1287a059d4a7f6387, this setting
was added to allow traffic destined for other MAC addresses to reach VMs
when using a macvtap interface.

This will prevent multicast from working, but we don't need it for
baremetal tests in CI.

This setting will be enabled again once the issue is resolved in either
Libvirt or Tenks.

This reverts commit 21c68bbf.

Also increase timeout of upgrade jobs which is too short now due to the
added delay added by bare metal testing.

[1] https://gitlab.com/libvirt/libvirt/-/commit/060d4c83ef436cf56abfad51a4d64c39448e199d

Change-Id: I2cfd2667abb1ae8988b7a7fd9761b75c20a0eaa4

The importlib.metadata library is part of the Python standard library
since Python 3.8.

Change-Id: Iecf1d469caaa777c5b253eb0e44dc00692197ef7

Backports of I9dd5b97dec93c0e5393a1e7d9640f85003651b56 to stable
branches are failing CI checks because several unit tests mock
os.path.exists and would need to be updated. This does not affect master
because test code changed in I4d9f96ec4cf3c6cd0d28dfe5ddb239d863498a72.

Work around this issue by replacing os.path.exists by os.path.isfile. It
is a better check anyway since we are reading the file.

Change-Id: Ic53efd03cecbd53ad3e3b64b664e084f4e25be0e
Related-Bug: #2020135

The return value was set to the following path which is not valid:

    kayobe/kayobe/utils.py/..

The intent must have been to use dirname to produce:

    kayobe/kayobe/..

Change-Id: I9d7d71d42026f3c12da6de9c5ca55dc647554fd0

Using an editable installation of Kayobe fails on Rocky Linux 9 or
Ubuntu with an error such as:

    ERROR! The requirements file '/home/rocky/kayobe/kayobe/requirements.yml' does not exist.
    Failed to install Ansible roles from /home/rocky/kayobe/kayobe/utils.py/../requirements.yml via Ansible Galaxy: returncode 1
    Control host bootstrap failed - likely Ansible Galaxy flakiness. Sleeping 5 seconds before retrying

This is caused by recent changes to how pip manages editable
installations. The egg-link file that Kayobe was using to find the
source path does not exist anymore. Instead, there is a direct_url.json
file under the kayobe dist-info directory that can be parsed.

Change-Id: I9dd5b97dec93c0e5393a1e7d9640f85003651b56
Closes-Bug: #2020135

Some mentions of skydive were forgotten in the original change [1].

[1] https://review.opendev.org/c/openstack/kayobe/+/884471.

Change-Id: If62fe0db3c107c52f80a481aa267335224456064

The default value for ipa_build_dib_elements_default was previously
changed in Iefd2d0b7a3a3e07f5c112d58e2ec0b3da0a747d3. This change
updates the comments to match.

Change-Id: I47e5acb19c806a1067879c9cf5fa5ccb7f07ccd4

Kolla Ansible enabled RabbitMQ HA queues by default, which require a
manual migration step [1]. Adds these to the Kayobe upgrade CI.

[1] https://review.opendev.org/c/openstack/kolla-ansible/+/882825

Change-Id: I82c286fd17e3a1d7f31952442fa281302cda7ee4

Kolla Ansible enabled RabbitMQ HA queues by default, which require a
manual migration step [1]. Work around the failing precheck by enabling
HA queues in the previous release until we implement migration code in
Kayobe CI jobs.

[1] https://review.opendev.org/c/openstack/kolla-ansible/+/882825

Change-Id: Idbbe0dd57acc9b7a9440a67c2b677e526a6be917

For overcloud commands involving Bifrost (e.g. overcloud provision), we
use the Bifrost dynamic inventory. This runs bifrost_inventory.py which
gathers node and port information for all bare metal hosts. Because this
gets executed for each node in our inventory, it produces a large number
of API requests to Ironic, which grows as the number of hosts increases.

Halve the number of calls using the BIFROST_NODE_NAMES environment
variable, which restricts fetching port information to just the node
being queried. bifrost_inventory.py still performs one call for each
node, which must be fixed separately in Bifrost [1].

Time to query overcloud nodes' hardware introspection data for 44 hosts:

* before patch: 823 seconds
* after patch: 415 seconds

With the patch to Bifrost [1], this is further reduced to 46 seconds.

[1] https://review.opendev.org/c/openstack/bifrost/+/882950

Change-Id: I341075115442a38c327e3ade74b81b1b9e7e85c6

There was a stray double quote in the firewalld config for monitoring
hosts, resulting in the following error when firewalld is enabled:

  ValueError: No closing quotation

Closes-Bug: 2019867
Change-Id: I201faf14da5d143670250052ab15fc285f24868c

Update old code to use truthy filter

Closes-bug: #2026276
Change-Id: I2085f9d1fd457e5b54b17f37a5e725990a60e8bc

For Rocky Linux 9, Kayobe will now disable STP on a bridge by default,
to preserve compatibility with network scripts, as Network Manager
enables STP on all bridges by default.
Enabling STP can lead to port down event if BPDU guard is enabled
on the switch.

Closes-Bug: #2028775

Change-Id: I35eaa92f4243af00697306aa801e5a733885ce4f

Updates interfaces role to a fixed version.

Depends-On: https://review.opendev.org/c/openstack/kolla-ansible/+/890198

Change-Id: I6a9c521d43b21ab6b6c01e4688cab697608a667c

Change-Id: I4ea22dbe43cd4780335df5cf0eee5c4339855fac

Change-Id: Ie29137fe1ab4d50366b8a000d891db3d256d09f2