Skip to content

Latest commit

 

History

History
266 lines (179 loc) · 13 KB

README.md

File metadata and controls

266 lines (179 loc) · 13 KB
Raw

dev-install

dev-install installs TripleO standalone on a remote system for development use.

Host configuration

dev-install requires that:

  • An appropriate OS has already been installed
  • EPEL repositories aren't deployed, they aren't compatible with OpenStack repositories
  • The machine running dev-install can SSH to the standalone host as either root or a user with passwordless sudo access
  • This machine has Ansible installed, and some dependencies like python3-netaddr.

You need to deploy the right RHEL version depending on which OSP version you want:

OSP version RHEL Version
16.2 8.4
17.1* 9.2

Current default.

There is no need to do any other configuration prior to running dev-install. When deploying on TripleO from upstream, you need to deploy on CentOS Stream. If CentOS is not Stream, dev-install will migrate it.

Local pre-requisites

dev-install requires up to date versions of ansible and make, both of which must be installed manually before invoking dev-install.

If installing OSP 16.2 with official RHEL 8.4 cloud images, it is required that the cloud-init service be disabled before deployment as per THIS

At present the deployment depends on a valid DHCP source for the external interface (br-ex) as per THIS

All other requirements should be configured automatically by Ansible. Note that dev-install does require root access (or passwordless sudo) on the machine it is invoked from to install certificate management tools (simpleca) in addition to the remote host.

Running dev-install

dev-install is invoked using its Makefile. The simplest invocation is:

make config host=<standalone host>
make osp_full

make config initialises two local state files:

  • inventory - this is an Ansible inventory file, initialised such that standalone is an alias for your target host.
  • local-overrides.yaml - this is an Ansible vars file containing configuration which overrides the defaults in playbooks/vars/defaults.yaml.

Both of these files can be safely modified.

make osp_full performs the actual installation. On an example system with 12 cores and 192GB RAM and running in a Red Hat data centre this takes approximately 65 minutes to execute.

If you deal with multiple OpenStack clouds and want to maintain a single local-overrides per cloud, you can create local-overrides-<name>.yaml and then use it when deploying with make osp_full overrides=local-overrides-<name>.yaml

Accessing OpenStack from your workstation

By default, dev-install configures OpenStack to use the default public IP of the host. To access this you just need a correct clouds.yaml, which dev-install configures with:

make local_os_client

This will configure your local clouds.yaml with two entries:

  • standalone - The admin user
  • standalone_openshift - The appropriately configured non-admin openshift user

You can change the name of these entries by editing local-overrides.yaml and setting local_cloudname to something else.

Validating the installation

dev-install provides an additional playbook to validate the fresh deployment. This can be run with:

make prepare_stack_testconfig

This can be used to configure some helpful defaults for validating your cluster, namely:

  • Configure SSH access
  • Configure routers and security groups to allow external network connectivity for created guests
  • Upload a Cirros image

Network configuration

dev-install will create a new OVS bridge called br-ex and move the host's external interface on to that bridge. This bridge is used to provide the external provider network if external_fip_pool_start and external_fip_pool_end are defined in local-overrides.yaml.

In addition it will create OVS bridges called br-ctlplane and br-hostonly. The former is used internally by OSP. The latter is a second provider network which is only routable from the host.

Note that we don't enable DHCP on provider networks by default, and it is not recommended to enable DHCP on the external network at all. To enable DHCP on the hostonly network after installation, run:

OS_CLOUD=standalone openstack subnet set --dhcp hostonly-subnet

make local_os_client will write a sshuttle script to scripts/sshuttle-standalone.sh which will route to the hostonly provider network over ssh.

Configuration

dev-install is configured by overriding variables in local-overrides.yaml. See the default variable definitions for what can be overridden.

Sizing

When idle, a standalone deployment uses approximately:

  • 16GB RAM
  • 15G on /
  • 3.5G on /home
  • 3.6G on /var/lib/cinder
  • 3.6G on /var/lib/nova

There is no need to mount /var/lib/cinder and /var/lib/nova separately if / is large enough for your workload.

Advanced features

NFV enablement

This section contains configuration procedures for single root input/output virtualization (SR-IOV) for network functions virtualization infrastructure (NFVi) in your Standalone OpenStack deployment. Unfortunately, most of these parameters don't have default values nor can they be automatically calculated in a Standalone-type environment.

SR-IOV Variables

To understand how the SR-IOV configuration works, please have a look at this upstream guide.

Name Default Value Description
sriov_services ['OS::TripleO::Services::NeutronSriovAgent', 'OS::TripleO::Services::BootParams'] List of TripleO services to add to the default Standalone role
sriov_interface [undefined] Name of the SR-IOV capable interface. Must be enabled in BIOS (e.g. ens1f0)
sriov_nic_numvfs [undefined] Number of Virtual Functions that the NIC can handle.
sriov_nova_pci_passthrough [undefined] List of PCI Passthrough allowlist parameters. Guidelines to configure it.

Note: when SR-IOV is enabled, a dedicated provider network will be created and bound to the SR-IOV interface.

Kernel Variables

It is possible to configure the Kernel to boot with specific arguments:

Name Default Value Description
kernel_services ['TripleO::Services::BootParams'] List of TripleO services to add to the default Standalone role
kernel_args [undefined] Kernel arguments to configure when booting the machine.

DPDK Variables

It is possible to configure the deployment to be ready for DPDK:

Name Default Value Description
dpdk_services ['OS::TripleO::Services::ComputeNeutronOvsDpdk'] List of TripleO services to add to the default Standalone role
dpdk_interface [undefined] Name of the DPDK capable interface. Must be enabled in BIOS. e.g. ens1f0
tuned_isolated_cores [undefined] List of logical CPU ids which need to be isolated from the host processes. This input is provided to the tuned profile CPU-partitioning to configure systemd and repin interrupts (IRQ repinning).

When deploying DPDK, it is suggested to configure these options:

extra_heat_params:
  # A list or range of host CPU cores to which processes for pinned instance
  # CPUs (PCPUs) can be scheduled:
  NovaComputeCpuDedicatedSet: ['6-47']
  # Reserved RAM for host processes:
  NovaReservedHostMemory: 4096
  # Determine the host CPUs that unpinned instances can be scheduled to:
  NovaComputeCpuSharedSet: [0,1,2,3]
  # Sets the amount of hugepage memory to assign per NUMA node:
  OvsDpdkSocketMemory: "2048,2048"
  # A list or range of CPU cores for PMD threads to be pinned to:
  OvsPmdCoreList: "4,5"

SSL for public endpoints

This sections contains configuration procedures for enabling SSL on OpenStack public endpoints.

Name Default Value Description
ssl_enabled false Whether or not we enable SSL for public endpoints
ssl_ca_cert [undefined] CA certificate. If undefined, a self-signed will be generated and deployed
ssl_ca_key [undefined] CA key. If undefined, it will be generated and used to sign the SSL certificate
ssl_key [undefined] SSL Key. If undefined, it will be generated and deployed
ssl_cert [undefined] SSL certificate. If undefined, a self-signed will be generated and deployed
ssl_ca_cert_path /etc/pki/ca-trust/source/anchors/simpleca.crt Path to the CA certificate
update_local_pki false Whether or not we want to update the local PKI with the CA certificate

Multi-node deployments

It is possible to deploy Edge-style environments, where multiple AZ are configured.

Deploy the Central site

Deploy a regular cloud with dev-install, and make sure you set these parameters:

  • dcn_az: has to be central.
  • tunnel_remote_ips: list of known public IPs of the AZ nodes.

Once this is done, you need to collect the content from /home/stack/exported-data into a local directory on the host where dev-install is executed.

Deploy the "AZ" sites

Before deploying OSP, you need to scp the content from exported-data into the remote hosts into /opt/exported-data. Once this is done, you can deploy the AZ sites with a regular config for dev-install, except that you'll need to set these parameters:

  • dcn_az: must contains az in the string (e.g. az0, az1)
  • local_ip: choose an available IP in the control plane subnet (e.g. 192.168.24.10)
  • control_plane_ip: same as for local_ip, pick one that is available (e.g. 192.168.24.11)
  • hostonly_gateway: if using provider networks, you'll need to select an available IP (e.g. 192.168.25.2)
  • tunnel_remote_ips: the list of known public IPs that will be used to establish the VXLAN tunnels
  • hostname: you must make sure both central and AZ doesn't use the default hostname (standalone), so set it at least on the compute (e.g. compute1)
  • octavia_enabled: set to false

Notes:

  • Control plane IPs (192.168.24.x) are arbitrary, if in doubt just use the example ones.
  • The control plane bridges will be connected thanks to VXLAN tunnels, which is why we need to select control plane IP for AZ nodes that were not taken on the Central site.
  • If you deploy the clouds in OpenStack, you need to make sure that the security groups allow VXLAN (udp/4789).
  • If the public IPs aren't predictable, you'll need to manually change the MTU on the br-ctlplane and br-hostonly on the central site and the AZ sites where needed. You can do it by editing the os-net-config configuration file and run os-net-config to apply it.

After the installation you can "join" AZs to just have a regular multi-node cloud. E.g.:

openstack aggregate remove host az0 compute1.shiftstack
openstack aggregate add host central compute1.shiftstack

Then if you're using OVN (you probably are) you got to execute this on compute nodes:

ovs-vsctl set Open_vSwitch . external-ids:ovn-cms-options="enable-chassis-as-gw,availability-zones=central"

Post Deployment Stack Updates

It is possible to perform stack updates on an ephemeral standalone stack.

Copying the generated tripleo_deploy.sh in your deployment users folder (e.g. /home/stack/tripleo_deploy.sh) to tripleo_update.sh and add the parameter --force-stack-update. This will allow you to modify the stack configuration without needing to redeploy the entire cloud which can save you considerable time.

Post install script

It is possible to run any script in post-install with post_install parameter:

post_install: |
  export OS_CLOUD=standalone
  openstack flavor set --property hw:mem_page_size=large m1.smal

And then run make post_install.

Troubleshooting

If installation fails, examine the Ansible logs for more information. Failures will usually occur early in the process or during the tripleo_deploy stage. If the latter, you can look at the logs in /home/stack/standalone_deploy.log. Once the issue has been addressed, you can resume the deployment by stopping heat and running the openstack tripleo deploy command again via the tripleo_deploy.sh script:

sudo pkill -9 heat-all
./triple_deploy.sh