Upgrade issues

This page documents upgrade issues and notes. These may apply to either of the three upgrade types (charms, OpenStack, series).

The items on this page are distinct from those found on the following pages:

The issues are organised by upgrade type.

Charm upgrades

rabbitmq-server charm

A timing issue has been observed during the upgrade of the rabbitmq-server charm (see bug LP #1912638 for tracking). If it occurs the resulting hook error can be resolved with:

juju resolved rabbitmq-server/N

openstack-dashboard charm: upgrading to revision 294

When Horizon is configured with TLS (openstack-dashboard charm option ssl-cert) revisions 294 and 295 of the charm have been reported to break the dashboard (see bug LP #1853173). The solution is to upgrade to a working revision. A temporary workaround is to disable TLS without upgrading.

Note

Most users will not be impacted by this issue as the recommended approach is to always upgrade to the latest revision.

To upgrade to revision 293:

juju upgrade-charm openstack-dashboard --revision 293

To upgrade to revision 296:

juju upgrade-charm openstack-dashboard --revision 296

To disable TLS:

juju config enforce-ssl=false openstack-dashboard

Multiple charms: option worker-multiplier

Starting with OpenStack Charms 21.04 any charm that supports the worker-multiplier configuration option will, upon upgrade, modify the active number of service workers according to the following: if the option is not set explicitly the number of workers will be capped at four regardless of whether the unit is containerised or not. Previously, the cap applied only to containerised units.

OpenStack upgrades

Nova RPC version mismatches: upgrading Neutron and Nova

If it is not possible to upgrade Neutron and Nova within the same maintenance window, be mindful that the RPC communication between nova-cloud-controller, nova-compute, and nova-api-metadata is very likely to cause several errors while those services are not running the same version. This is due to the fact that currently those charms do not support RPC version pinning or auto-negotiation.

See bug LP #1825999.

neutron-gateway charm: upgrading from Mitaka to Newton

Between the Mitaka and Newton OpenStack releases, the neutron-gateway charm added two options, bridge-mappings and data-port, which replaced the (now) deprecated ext-port option. This was to provide for more control over how neutron-gateway can configure external networking. Unfortunately, the charm was only designed to work with either ext-port (no longer recommended) or bridge-mappings and data-port.

See bug LP #1809190.

cinder/ceph topology change: upgrading from Newton to Ocata

If cinder is directly related to ceph-mon rather than via cinder-ceph then upgrading from Newton to Ocata will result in the loss of some block storage functionality, specifically live migration and snapshotting. To remedy this situation the deployment should migrate to using the cinder-ceph charm. This can be done after the upgrade to Ocata.

Warning

Do not attempt to migrate a deployment with existing volumes to use the cinder-ceph charm prior to Ocata.

The intervention is detailed in the below three steps.

Step 0: Check existing configuration

Confirm existing volumes are in an RBD pool called ‘cinder’:

juju run --unit cinder/0 "rbd --name client.cinder -p cinder ls"

Sample output:

volume-b45066d3-931d-406e-a43e-ad4eca12cf34
volume-dd733b26-2c56-4355-a8fc-347a964d5d55

Step 1: Deploy new topology

Deploy the cinder-ceph charm and set the ‘rbd-pool-name’ to match the pool that any existing volumes are in (see above):

juju deploy --config rbd-pool-name=cinder cinder-ceph
juju add-relation cinder cinder-ceph
juju add-relation cinder-ceph ceph-mon
juju remove-relation cinder ceph-mon
juju add-relation cinder-ceph nova-compute

Step 2: Update volume configuration

The existing volumes now need to be updated to associate them with the newly defined cinder-ceph backend:

juju run-action cinder/0 rename-volume-host currenthost='cinder' \
    newhost='cinder@cinder-ceph#cinder.volume.drivers.rbd.RBDDriver'

Keystone and Fernet tokens: upgrading from Queens to Rocky

Starting with OpenStack Rocky only the Fernet format for authentication tokens is supported. Therefore, prior to upgrading Keystone to Rocky a transition must be made from the legacy format (of UUID) to Fernet.

Fernet support is available upstream (and in the keystone charm) starting with Ocata so the transition can be made on either Ocata, Pike, or Queens.

A keystone charm upgrade will not alter the token format. The charm’s token-provider option must be used to make the transition:

juju config keystone token-provider=fernet

This change may result in a minor control plane outage but any running instances will remain unaffected.

The token-provider option has no effect starting with Rocky, where the charm defaults to Fernet and where upstream removes support for UUID. See Keystone Fernet Token Implementation for more information.

Neutron LBaaS: upgrading from Stein to Train

As of Train, support for Neutron LBaaS has been retired. The load-balancing services are now provided by Octavia LBaaS. There is no automatic migration path, please review the Octavia LBaaS page for more information.

Designate: upgrading from Stein to Train

When upgrading Designate to Train, there is an encoding issue between the designate-producer and memcached that causes the designate-producer to crash. See bug LP #1828534. This can be resolved by restarting the memcached service.

juju run --application=memcached 'sudo systemctl restart memcached'

Ceph BlueStore mistakenly enabled during OpenStack upgrade

The Ceph BlueStore storage backend is enabled by default when Ceph Luminous is detected. Therefore it is possible for a non-BlueStore cloud to acquire BlueStore by default after an OpenStack upgrade (Luminous first appeared in Queens). Problems will occur if storage is scaled out without first disabling BlueStore (set ceph-osd charm option bluestore to ‘False’). See bug LP #1885516 for details.

Placement: endpoints not updated in Keystone service catalog

When the placement charm is deployed during the upgrade to OpenStack Train (as described in placement charm: OpenStack upgrade to Train) the Keystone service catalog is not updated accordingly. This issue is tracked in bug LP #1928992, which also includes an explicit workaround (comment #4).

Ceph: option require-osd-release

Before upgrading Ceph its require-osd-release option should be set to the current Ceph release (e.g. ‘nautilus’ if upgrading to Octopus). Failing to do so may cause the upgrade to fail, rendering the cluster inoperable.

On any ceph-mon unit, the current value of the option can be queried with:

sudo ceph osd dump | grep require_osd_release

If it needs changing, it can be done manually on any ceph-mon unit. Here the current release is Nautilus:

sudo ceph osd require-osd-release nautilus

In addition, upon completion of the upgrade, the option should be set to the new release. Here the new release is Octopus:

sudo ceph osd require-osd-release octopus

The charms should be able to respond intelligently to these two situations. Bug LP #1929254 is for tracking this effort.

Series upgrades

DNS HA: upgrade to focal

DNS HA has been reported to not work on the focal series. See LP #1882508 for more information.

Upgrading while Vault is sealed

If a series upgrade is attempted while Vault is sealed then manual intervention will be required (see bugs LP #1886083 and LP #1890106). The vault leader unit (which will be in error) will need to be unsealed and the hook error resolved. The vault charm README has unsealing instructions, and the hook error can be resolved with:

juju resolved vault/N