Dmitriy Rabotyagov bc21a289b5 Drop nova-key from scale documentation and scripts

Since we've switched to using ssh cert-based auth, we don't need
to distribute ssh keys across all computes with adding another one
anymore. This change should be reflected in docs and add-compute script.

Change-Id: I2d6de2e70fa59f8a0fc70eff56fab586fa648166

2022-05-28 07:30:44 +02:00

13 KiB

Raw Blame History

Scaling your environment

This is a draft environment scaling page for the proposed OpenStack-Ansible operations guide.

Add a new infrastructure host

While three infrastructure hosts are recommended, if further hosts are needed in an environment, it is possible to create additional nodes.

Warning

Make sure you back up your current OpenStack environment before adding any new nodes. See backup-restore for more information.

Add the node to the infra_hosts stanza of the /etc/openstack_deploy/openstack_user_config.yml
```
infra_hosts:
[...]
  infra<node-ID>:
    ip: 10.17.136.32
```
Change to playbook folder on the deployment host.
```
# cd /opt/openstack-ansible
```

To prepare new hosts and deploy containers on them run setup-hosts.yml playbook with the limit argument.

# openstack-ansible playbooks/setup-hosts.yml --limit localhost,infra<node-ID>,infra<node-ID>-host_containers

Next we need to expand galera/rabbitmq clusters, which is done during setup-infrastructure.yml. So we will run this playbook without limits.
Warning

Make sure that containers from new infra host does not appear in inventory as first one for groups galera_all, rabbitmq_all and repo_all. You can varify that with ad-hoc commands:
```
# ansible -m debug -a "var=groups['galera_all'][0]" localhost
# ansible -m debug -a "var=groups['rabbitmq_all'][0]" localhost
# ansible -m debug -a "var=groups['repo_all'][0]" localhost
```
```
# openstack-ansible playbooks/setup-infrastructure.yml -e galera_force_bootstrap=true
```
Once infrastructure playboks are done, it's turn of openstack services to be deployed. Most of the services are fine to be ran with limits, but some, like keystone, are not. So we run keystone playbook separately from all others:
```
# openstack-ansible playbooks/os-keystone-install.yml
# openstack-ansible playbooks/setup-openstack.yml --limit '!keystone_all',localhost,infra<node-ID>,infra<node-ID>-host_containers
```

Test new infra nodes

After creating a new infra node, test that the node runs correctly by launching a new instance. Ensure that the new node can respond to a networking connection test through the ping command. Log in to your monitoring system, and verify that the monitors return a green signal for the new node.

Add a compute host

Use the following procedure to add a compute host to an operational cluster.

Configure the host as a target host. See the :deploy_guide:target hosts configuration section <targethosts.html> of the deploy guide. for more information.
Edit the /etc/openstack_deploy/openstack_user_config.yml file and add the host to the compute_hosts stanza.

If necessary, also modify the used_ips stanza.
If the cluster is utilizing Telemetry/Metering (ceilometer), edit the /etc/openstack_deploy/conf.d/ceilometer.yml file and add the host to the metering-compute_hosts stanza.
Run the following commands to add the host. Replace NEW_HOST_NAME with the name of the new host.
```
# cd /opt/openstack-ansible/playbooks
# openstack-ansible setup-hosts.yml --limit localhost,NEW_HOST_NAME
# openstack-ansible setup-openstack.yml --limit localhost,NEW_HOST_NAME
```
Alternatively you can try using new compute nodes deployment script /opt/openstack-ansible/scripts/add-compute.sh.

You can provide this script with extra tasks that will be executed before or right after OSA roles. To do so you should set environment variables PRE_OSA_TASKS or POST_OSA_TASKS with plays to run devided with semicolon:
```
# export POST_OSA_TASKS="/opt/custom/setup.yml --limit HOST_NAME;/opt/custom/tasks.yml --tags deploy"
# /opt/openstack-ansible/scripts/add-compute.sh HOST_NAME,HOST_NAME_2
```

Test new compute nodes

After creating a new node, test that the node runs correctly by launching an instance on the new node.

$ openstack server create --image IMAGE --flavor m1.tiny \
--key-name KEY --availability-zone ZONE:HOST:NODE \
--nic net-id=UUID SERVER

Ensure that the new instance can respond to a networking connection test through the ping command. Log in to your monitoring system, and verify that the monitors return a green signal for the new node.

Remove a compute host

The openstack-ansible-ops repository contains a playbook for removing a compute host from an OpenStack-Ansible environment. To remove a compute host, follow the below procedure.

Note

This guide describes how to remove a compute node from an OpenStack-Ansible environment completely. Perform these steps with caution, as the compute node will no longer be in service after the steps have been completed. This guide assumes that all data and instances have been properly migrated.

Disable all OpenStack services running on the compute node. This can include, but is not limited to, the nova-compute service and the neutron agent service.

Note

Ensure this step is performed first
```
# Run these commands on the compute node to be removed
# stop nova-compute
# stop neutron-linuxbridge-agent
```

Clone the openstack-ansible-ops repository to your deployment host:

$ git clone https://opendev.org/openstack/openstack-ansible-ops \
  /opt/openstack-ansible-ops

Run the remove_compute_node.yml Ansible playbook with the host_to_be_removed user variable set:

$ cd /opt/openstack-ansible-ops/ansible_tools/playbooks
openstack-ansible remove_compute_node.yml \
-e host_to_be_removed="<name-of-compute-host>"

After the playbook completes, remove the compute node from the OpenStack-Ansible configuration file in /etc/openstack_deploy/openstack_user_config.yml.

Recover a compute host failure

The following procedure addresses Compute node failure if shared storage is used.

Note

If shared storage is not used, data can be copied from the /var/lib/nova/instances directory on the failed Compute node ${FAILED_NODE} to another node ${RECEIVING_NODE}before performing the following procedure. Please note this method is not supported.

Re-launch all instances on the failed node.
Invoke the MySQL command line tool

Generate a list of instance UUIDs hosted on the failed node:

mysql> select uuid from instances where host = '${FAILED_NODE}' and deleted = 0;

Set instances on the failed node to be hosted on a different node:

mysql> update instances set host ='${RECEIVING_NODE}' where host = '${FAILED_NODE}' \
and deleted = 0;

Reboot each instance on the failed node listed in the previous query to regenerate the XML files:
```
# nova reboot —hard $INSTANCE_UUID
```

Find the volumes to check the instance has successfully booted and is at the login :

mysql> select nova.instances.uuid as instance_uuid, cinder.volumes.id \
as voume_uuid, cinder.volumes.status, cinder.volumes.attach_status, \
cinder.volumes.mountpoint, cinder.volumes,display_name from \
cinder.volumes inner join nova.instances on cinder.volumes.instance_uuid=nova.instances.uuid \
where nova.instances.host = '${FAILED_NODE}';

If rows are found, detach and re-attach the volumes using the values listed in the previous query:

# nova volume-detach $INSTANCE_UUID $VOLUME_UUID && \
# nova volume-attach $INSTANCE_UUID $VOLUME_UUID $VOLUME_MOUNTPOINT

Rebuild or replace the failed node as described in add-compute-host.

Replacing failed hardware

It is essential to plan and know how to replace failed hardware in your cluster without compromising your cloud environment.

Consider the following to help establish a hardware replacement plan:

What type of node am I replacing hardware on?
Can the hardware replacement be done without the host going down? For example, a single disk in a RAID-10.
If the host DOES have to be brought down for the hardware replacement, how should the resources on that host be handled?

If you have a Compute (nova) host that has a disk failure on a RAID-10, you can swap the failed disk without powering the host down. On the other hand, if the RAM has failed, you would have to power the host down. Having a plan in place for how you will manage these types of events is a vital part of maintaining your OpenStack environment.

For a Compute host, shut down the instance on the host before it goes down. For a Block Storage (cinder) host using non-redundant storage, shut down any instances with volumes attached that require that mount point. Unmount the drive within your operating system and re-mount the drive once the Block Storage host is back online.

Shutting down the Compute host

If a Compute host needs to be shut down:

Disable the nova-compute binary:

# nova service-disable --reason "Hardware replacement" HOSTNAME nova-compute

List all running instances on the Compute host:

# nova list --all-t --host <compute_name> | awk '/ACTIVE/ {print $2}' > \
/home/user/running_instances && for i in `cat /home/user/running_instances`; do nova stop $i ; done

Use SSH to connect to the Compute host.
Confirm all instances are down:
```
# virsh list --all
```
Shut down the Compute host:
```
# shutdown -h now
```
Once the Compute host comes back online, confirm everything is in working order and start the instances on the host. For example:
```
# cat /home/user/running_instances
# do nova start $instance
  done
```
Enable the nova-compute service in the environment:
```
# nova service-enable HOSTNAME nova-compute
```

Shutting down the Block Storage host

If a LVM backed Block Storage host needs to be shut down:

Disable the cinder-volume service:

# cinder service-list --host CINDER SERVICE NAME INCLUDING @BACKEND
# cinder service-disable CINDER SERVICE NAME INCLUDING @BACKEND \
cinder-volume --reason 'RAM maintenance'

List all instances with Block Storage volumes attached:

# mysql cinder -BNe 'select instance_uuid from volumes where deleted=0 '\
'and host like "%<cinder host>%"' | tee /home/user/running_instances

Shut down the instances:

# cat /home/user/running_instances | xargs -n1 nova stop

Verify the instances are shutdown:

# cat /home/user/running_instances | xargs -n1 nova show | fgrep vm_state

Shut down the Block Storage host:
```
# shutdown -h now
```
Replace the failed hardware and validate the new hardware is functioning.

Enable the cinder-volume service:

# cinder service-enable CINDER SERVICE NAME INCLUDING @BACKEND cinder-volume

Verify the services on the host are reconnected to the environment:

# cinder service-list --host CINDER SERVICE NAME INCLUDING @BACKEND

Start your instances and confirm all of the instances are started:

# cat /home/user/running_instances | xargs -n1 nova start
# cat /home/user/running_instances | xargs -n1 nova show | fgrep vm_state

Destroying Containers

To destroy a container, execute the following:

# cd /opt/openstack-ansible/playbooks
# openstack-ansible lxc-containers-destroy --limit localhost,<container name|container group>
Note

You will be asked two questions:

Are you sure you want to destroy the LXC containers? Are you sure you want to destroy the LXC container data?

The first will just remove the container but leave the data in the bind mounts and logs. The second will remove the data in the bind mounts and logs too.