Chapter 13. Replacing Controller Nodes

In certain circumstances a Controller node in a high availability cluster might fail. In these situations, you must remove the node from the cluster and replace it with a new Controller node.

Complete the steps in this section to replace a Controller node. The Controller node replacement process involves running the openstack overcloud deploy command to update the overcloud with a request to replace a Controller node.


The following procedure applies only to high availability environments. Do not use this procedure if using only one Controller node.

13.1. Preparing for Controller replacement

Before attempting to replace an overcloud Controller node, it is important to check the current state of your Red Hat OpenStack Platform environment. Checking the current state can help avoid complications during the Controller replacement process. Use the following list of preliminary checks to determine if it is safe to perform a Controller node replacement. Run all commands for these checks on the undercloud.


  1. Check the current status of the overcloud stack on the undercloud:

    $ source stackrc
    (undercloud) $ openstack stack list --nested

    The overcloud stack and its subsequent child stacks should have either a CREATE_COMPLETE or UPDATE_COMPLETE.

  2. Perform a backup of the undercloud databases:

    (undercloud) $ mkdir /home/stack/backup
    (undercloud) $ sudo mysqldump --all-databases --quick --single-transaction | gzip > /home/stack/backup/dump_db_undercloud.sql.gz
  3. Check that your undercloud contains 10 GB free storage to accommodate for image caching and conversion when provisioning the new node.
  4. Check the status of Pacemaker on the running Controller nodes. For example, if is the IP address of a running Controller node, use the following command to get the Pacemaker status:

    (undercloud) $ ssh heat-admin@ 'sudo pcs status'

    The output should show all services running on the existing nodes and stopped on the failed node.

  5. Check the following parameters on each node of the overcloud MariaDB cluster:

    • wsrep_local_state_comment: Synced
    • wsrep_cluster_size: 2

      Use the following command to check these parameters on each running Controller node. In this example, the Controller node IP addresses are and

      (undercloud) $ for i in ; do echo "*** $i ***" ; ssh heat-admin@$i "sudo mysql -p\$(sudo hiera -c /etc/puppet/hiera.yaml mysql::server::root_password) --execute=\"SHOW STATUS LIKE 'wsrep_local_state_comment'; SHOW STATUS LIKE 'wsrep_cluster_size';\""; done
  6. Check the RabbitMQ status. For example, if is the IP address of a running Controller node, use the following command to get the status:

    (undercloud) $ ssh heat-admin@ "sudo docker exec \$(sudo docker ps -f name=rabbitmq-bundle -q) rabbitmqctl cluster_status"

    The running_nodes key should only show the two available nodes and not the failed node.

  7. If you are using Open Virtual Switch (OVS) and replaced Controller nodes in the past without restarting the OVS agents, then restart the agents on the compute nodes before replacing this Controller. Restarting the OVS agents ensures that they have a full complement of RabbitMQ connections.

    Run the following command to restart the OVS agent:

    [heat-admin@overcloud-compute-0 ~]$ sudo docker restart neutron_ovs_agent
  8. Disable fencing, if enabled. For example, if is the IP address of a running Controller node, use the following command to disable fencing:

    (undercloud) $ ssh heat-admin@ "sudo pcs property set stonith-enabled=false"

    Check the fencing status with the following command:

    (undercloud) $ ssh heat-admin@ "sudo pcs property show stonith-enabled"
  9. Check the nova-compute service on the director node:

    (undercloud) $ sudo systemctl status openstack-nova-compute
    (undercloud) $ openstack hypervisor list

    The output should show all non-maintenance mode nodes as up.

  10. Make sure all undercloud services are running:

    (undercloud) $ sudo systemctl -t service

13.2. Removing a Ceph Monitor daemon

Follow this procedure to remove a ceph-mon daemon from the storage cluster. If your Controller node is running a Ceph monitor service, complete the following steps to remove the ceph-mon daemon. This procedure assumes the Controller is reachable.


Adding a new Controller to the cluster also adds a new Ceph monitor daemon automatically.


  1. Connect to the Controller you want to replace and become root:

    # ssh heat-admin@
    # sudo su -

    If the controller is unreachable, skip steps 1 and 2 and continue the procedure at step 3 on any working controller node.

  2. As root, stop the monitor:

    # systemctl stop ceph-mon@<monitor_hostname>

    For example:

    # systemctl stop ceph-mon@overcloud-controller-1
  3. Remove the monitor from the cluster:

    # ceph mon remove <mon_id>
  4. On the Ceph monitor node, remove the monitor entry from /etc/ceph/ceph.conf. For example, if you remove controller-1, then remove the IP and hostname for controller-1.


    mon host =,,
    mon initial members = overcloud-controller-2,overcloud-controller-1,overcloud-controller-0


    mon host =,
    mon initial members = overcloud-controller-2,overcloud-controller-0
  5. Apply the same change to /etc/ceph/ceph.conf on the other overcloud nodes.


    The director updates the ceph.conf file on the relevant overcloud nodes when you add the replacement controller node. Normally, director manages this configuration file exclusively and you should not edit the file manually. However, you can edit the file manually to ensure consistency in case the other nodes restart before you add the new node.

  6. Optionally, archive the monitor data and save the archive on another server:

    # mv /var/lib/ceph/mon/<cluster>-<daemon_id> /var/lib/ceph/mon/removed-<cluster>-<daemon_id>

13.3. Preparing the cluster for Controller replacement

Before replacing the old node, you must ensure that Pacemaker is no longer running on the node and then remove that node from the Pacemaker cluster.


  1. Get a list of IP addresses for the Controller nodes:

    (undercloud) $ openstack server list -c Name -c Networks
    | Name                   | Networks              |
    | overcloud-compute-0    | ctlplane= |
    | overcloud-controller-0 | ctlplane= |
    | overcloud-controller-1 | ctlplane= |
    | overcloud-controller-2 | ctlplane= |
  2. If the old node is still reachable, log in to one of the remaining nodes and stop pacemaker on the old node. For this example, stop pacemaker on overcloud-controller-1:

    (undercloud) $ ssh heat-admin@ "sudo pcs status | grep -w Online | grep -w overcloud-controller-1"
    (undercloud) $ ssh heat-admin@ "sudo pcs cluster stop overcloud-controller-1"

    In case the old node is physically unavailable or stopped, it is not necessary to perform the previous operation, as pacemaker is already stopped on that node.

  3. After stopping Pacemaker on the old node (i.e. it is shown as Stopped in pcs status), delete the old node from the corosync configuration on each node and restart Corosync. For this example, the following command logs into overcloud-controller-0 and overcloud-controller-2 removes the node:

    (undercloud) $ for NAME in overcloud-controller-0 overcloud-controller-2; do IP=$(openstack server list -c Networks -f value --name $NAME | cut -d "=" -f 2) ; ssh heat-admin@$IP "sudo pcs cluster localnode remove overcloud-controller-1; sudo pcs cluster reload corosync"; done
  4. Log in to one of the remaining nodes and delete the node from the cluster with the crm_node command:

    (undercloud) $ ssh heat-admin@
    [heat-admin@overcloud-controller-0 ~]$ sudo crm_node -R overcloud-controller-1 --force
  5. The overcloud database must continue to run during the replacement procedure. To ensure Pacemaker does not stop Galera during this procedure, select a running Controller node and run the following command on the undercloud using the Controller node’s IP address:

    (undercloud) $ ssh heat-admin@ "sudo pcs resource unmanage galera-bundle"

13.4. Replacing a Controller node

To replace a Controller node, identify the index of the node that you want to replace.

  • If the node is a virtual node, identify the node that contains the failed disk and restore the disk from a backup. Ensure that the MAC address of the NIC used for PXE boot on the failed server remains the same after disk replacement.
  • If the node is a bare metal node, replace the disk, prepare the new disk with your overcloud configuration, and perform a node introspection on the new hardware.
  • If the node is a part of a high availability cluster with fencing, you might need recover the Galera nodes separately. For more information, see the article How Galera works and how to rescue Galera clusters in the context of Red Hat OpenStack Platform.

Complete the following example steps to replace the the overcloud-controller-1 node with the overcloud-controller-3 node. The overcloud-controller-3 node has the ID 75b25e9a-948d-424a-9b3b-f0ef70a6eacf.


To replace the node with an existing ironic node, enable maintenance mode on the outgoing node so that the director does not automatically reprovision the node.


  1. Source the stackrc file:

    $ source ~/stackrc
  2. Identify the index of the overcloud-controller-1 node:

    $ INSTANCE=$(openstack server list --name overcloud-controller-1 -f value -c ID)
  3. Identify the bare metal node associated with the instance:

    $ NODE=$(openstack baremetal node list -f csv --quote minimal | grep $INSTANCE | cut -f1 -d,)
  4. Set the node to maintenance mode:

    $ openstack baremetal node maintenance set $NODE
  5. If the Controller node is a virtual node, run the following command on the Controller host to replace the virtual disk from a backup:

    $ cp <VIRTUAL_DISK_BACKUP> /var/lib/libvirt/images/<VIRTUAL_DISK>

    Replace <VIRTUAL_DISK_BACKUP> with the path to the backup of the failed virtual disk, and replace <VIRTUAL_DISK> with the name of the virtual disk that you want to replace.

    If you do not have a backup of the outgoing node, you must use a new virtualized node.

    If the Controller node is a bare metal node, complete the following steps to replace the disk with a new bare metal disk:

    1. Replace the physical hard drive or solid state drive.
    2. Prepare the node with the same configuration as the failed node.
  6. List unassociated nodes and identify the ID of the new node:

    $ openstack baremetal node list --unassociated
  7. Tag the new node with the control profile:

    (undercloud) $ openstack baremetal node set --property capabilities='profile:control,boot_option:local' 75b25e9a-948d-424a-9b3b-f0ef70a6eacf

13.5. Triggering the Controler node replacement

Complete the following steps to remove the old Controller node and replace it with a new Controller node.


  1. Determine the UUID of the node that you want to remove and store it in the NODEID variable. Ensure that you replace NODE_NAME with the name of the node that you want to remove:

    $ NODEID=$(openstack server list -f value -c ID --name NODE_NAME)
  2. To identify the Heat resource ID, enter the following command:

    $ openstack stack resource show overcloud ControllerServers -f json -c attributes | jq --arg NODEID "$NODEID" -c .attributes.value | keys[] as $k | if .[$k] == $NODEID then "Node index \($k) for \(.[$k])" else empty end
  3. Create the following environment file ~/templates/remove-controller.yaml and include the node index of the Controller node that you want to remove:

        [{resource_list: [NODE_INDEX]}]
  4. Run your overcloud deployment command, including the remove-controller.yaml environment file along with any other environment files relevant to your environment:

    (undercloud) $ openstack overcloud deploy --templates \
        -e /home/stack/templates/remove-controller.yaml \

    Include -e ~/templates/remove-controller.yaml only for this instance of the deployment command. Remove this environment file from subsequent deployment operations.

  5. The director removes the old node, creates a new one, and updates the overcloud stack. You can check the status of the overcloud stack with the following command:

    (undercloud) $ openstack stack list --nested
  6. Once the deployment command completes, the director shows the old node replaced with the new node:

    (undercloud) $ openstack server list -c Name -c Networks
    | Name                   | Networks              |
    | overcloud-compute-0    | ctlplane= |
    | overcloud-controller-0 | ctlplane= |
    | overcloud-controller-2 | ctlplane= |
    | overcloud-controller-3 | ctlplane= |

    The new node now hosts running control plane services.

13.6. Cleaning up after Controller node replacement

After completing the node replacement, complete the following steps to finalize the Controller cluster.


  1. Log into a Controller node.
  2. Enable Pacemaker management of the Galera cluster and start Galera on the new node:

    [heat-admin@overcloud-controller-0 ~]$ sudo pcs resource refresh galera-bundle
    [heat-admin@overcloud-controller-0 ~]$ sudo pcs resource manage galera-bundle
  3. Perform a final status check to make sure services are running correctly:

    [heat-admin@overcloud-controller-0 ~]$ sudo pcs status

    If any services have failed, use the pcs resource refresh command to resolve and restart the failed services.

  4. Exit to the director

    [heat-admin@overcloud-controller-0 ~]$ exit
  5. Source the overcloudrc file so that you can interact with the overcloud:

    $ source ~/overcloudrc
  6. Check the network agents in your overcloud environment:

    (overcloud) $ openstack network agent list
  7. If any agents appear for the old node, remove them:

    (overcloud) $ for AGENT in $(openstack network agent list --host overcloud-controller-1.localdomain -c ID -f value) ; do openstack network agent delete $AGENT ; done
  8. If necessary, add your hosting router to the L3 agent on the new node. Use the following example command to add a hosting router r1 to the L3 agent using the UUID 2d1c1dc1-d9d4-4fa9-b2c8-f29cd1a649d4:

    (overcloud) $ openstack network agent add router -l3 2d1c1dc1-d9d4-4fa9-b2c8-f29cd1a649d4 r1
  9. Compute services for the removed node still exist in the overcloud and require removal. Check the compute services for the removed node:

    [stack@director ~]$ source ~/overcloudrc
    (overcloud) $ openstack compute service list --host overcloud-controller-1.localdomain
  10. Remove the compute services for the removed node:

    (overcloud) $ for SERVICE in $(openstack compute service list --host overcloud-controller-1.localdomain -c ID -f value ) ; do openstack compute service delete $SERVICE ; done
  11. If you are using Open Virtual Switch (OVS), and the IP address for the Controller node has changed, then you must restart the OVS agent on all compute nodes:

    [heat-admin@overcloud-compute-0 ~]$ sudo docker restart neutron_ovs_agent