OpenStack Platform Support

This document discusses the requirements, current expected behavior, and how to try out what exists so far. In addition, it covers the installation with the default CNI (OpenShiftSDN), as well as with the Kuryr SDN.

Reference Documents

Privileges
Known Issues and Workarounds
Using the OSP 4 installer with Kuryr
Troubleshooting your cluster
Customizing your install
Installing OpenShift on OpenStack User-Provisioned Infrastructure
Learn about the OpenShift on OpenStack networking infrastructure design
Deploying OpenShift bare-metal workers

OpenStack Requirements

In order to run the latest version of the installer in OpenStack, at a bare minimum you need the following quota to run a default cluster. While it is possible to run the cluster with fewer resources than this, it is not recommended. Certain cases, such as deploying without FIPs, or deploying with an external load balancer are documented below, and are not included in the scope of this recommendation. If you are planning on using Kuryr, or want to learn more about it, please read through the Kuryr documentation. NOTE: The installer has been tested and developed on Red Hat OSP 13.

For a successful installation it is required:

Floating IPs: 2 (plus one that will be created and destroyed by the Installer during the installation process)
Security Groups: 3
Security Group Rules: 60
Routers: 1
Subnets: 1
Server Groups: 1
RAM: 112 GB
vCPUs: 28
Volume Storage: 175 GB
Instances: 7
Depending on the type of image registry backend either 1 Swift container or an additional 100 GB volume.
OpenStack resource tagging

NOTE: The installer will check OpenStack quota limits to make sure that the requested resources can be created. Note that it won't check for resource availability in the cloud, but only on the quotas.

You may need to increase the security group related quotas from their default values. For example (as an OpenStack administrator):

openstack quota set --secgroups 8 --secgroup-rules 100 <project>`

Once you configure the quota for your project, please ensure that the user for the installer has the proper privileges.

Master Nodes

The default deployment stands up 3 master nodes, which is the minimum amount required for a cluster. For each master node you stand up, you will need 1 instance, and 1 port available in your quota. They should be assigned a flavor with at least 16 GB RAM, 4 vCPUs, and 25 GB Disk (or Root Volume). It is theoretically possible to run with a smaller flavor, but be aware that if it takes too long to stand up services, or certain essential services crash, the installer could time out, leading to a failed install.

The Master Nodes are placed in a single Server Group with "soft anti-affinity" policy; the machines will therefore be creted on separate hosts when possible.

Worker Nodes

The default deployment stands up 3 worker nodes. Worker nodes host the applications you run on OpenShift. The flavor assigned to the worker nodes should have at least 2 vCPUs, 8 GB RAM and 25 GB Disk (or Root Volume). However, if you are experiencing Out Of Memory issues, or your installs are timing out, try increasing the size of your flavor to match the master nodes: 4 vCPUs and 16 GB RAM.

See the OpenShift documentation for more information on the worker nodes.

Bootstrap Node

The bootstrap node is a temporary node that is responsible for standing up the control plane on the masters. Only one bootstrap node will be stood up and it will be deprovisioned once the production control plane is ready. To do so, you need 1 instance, and 1 port. We recommend a flavor with a minimum of 16 GB RAM, 4 vCPUs, and 25 GB Disk (or Root Volume).

Image Registry Requirements

If Swift is available in the cloud where the installation is being performed, it is used as the default backend for the OpenShift image registry. At the time of installation only an empty container is created without loading any data. Later on, for the system to work properly, you need to have enough free space to store the container images.

In this case the user must have swiftoperator permissions. As an OpenStack administrator:

openstack role add --user <user> --project <project> swiftoperator

If Swift is not available, the PVC storage is used as the backend. For this purpose, a persistent volume of 100 GB will be created in Cinder and mounted to the image registry pod during the installation.

Note: Since Cinder supports only ReadWriteOnce access mode, it's not possible to have more than one replica of the image registry pod.

Disk Requirements

Etcd, which runs on the control plane nodes, has disk requirements that need to be met to ensure the stability of the cluster.

Generally speaking, it is advised to choose for the control plane nodes a flavour that is backed by SSD in order to reduce latency.

If the ephemeral disk that gets attached to instances of the chosen flavor does not meet etcd requirements, check if the cloud has a more performant volume type and use a custom install-config.yaml to deploy the control plane with root volumes. However, please note that Ceph RBD (and any other network-attached storage) can result in unpredictable network latencies. Prefer PCI passthrough of an NVM device instead.

In order to measure the performance of your disk, you can use fio:

sudo podman run \
	--volume "/var/lib/etcd:/mount:z" \
	docker.io/ljishen/fio \
		--directory=/mount \
		--name=iotest \
		--size=22m \
		--bs=2300 \
		--fdatasync=1 \
		--ioengine=sync \
		--rw=write

The command must be run as superuser.

In the command output, look for the the 99th percentile of fdatasync durations (fsync/fdatasync/sync_file_range -> sync percentiles). The number must be less than 10ms (or 10000µs: fio fluidly adjusts the scale between ms/µs/ns depending on the numbers).

Look for spikes. Even if the baseline latency looks good, there may be spikes where it comes up, triggering issues that result in API being unavailable.

Prometheus collects etcd-specific metrics.

Once the cluster is up, Prometheus provides useful metrics here:

https://prometheus-k8s-openshift-monitoring.apps.<cluster name>.<domain name>/graph?g0.range_input=2h&g0.stacked=0&g0.expr=histogram_quantile(0.99%2C%20rate(etcd_disk_wal_fsync_duration_seconds_bucket%5B5m%5D))&g0.tab=0&g1.range_input=2h&g1.expr=histogram_quantile(0.99%2C%20rate(etcd_disk_backend_commit_duration_seconds_bucket%5B5m%5D))&g1.tab=0&g2.range_input=2h&g2.expr=etcd_server_health_failures&g2.tab=0

Click "Login with OpenShift", enter kubeadmin and the password printed out by the installer.

The units are in seconds and should stay under 10ms (0.01s) at all times. The etcd_health graph should remain at 0.

In order to collect relevant information interactively, run the conformance tests:

git clone https://github.com/openshift/origin/
make WHAT=cmd/openshift-tests
export KUBECONFIG=<path/to/kubeconfig>
_output/local/bin/linux/amd64/openshift-tests run openshift/conformance/parallel

The entire test suite takes over an hour to complete. Run it and check the Prometheus logs afterwards.

Neutron Public Network

The public network should be created by the OpenStack administrator. Verify the name/ID of the 'External' network:

openstack network list --long -c ID -c Name -c "Router Type"
+--------------------------------------+----------------+-------------+
| ID                                   | Name           | Router Type |
+--------------------------------------+----------------+-------------+
| 148a8023-62a7-4672-b018-003462f8d7dc | public_network | External    |
+--------------------------------------+----------------+-------------+

NOTE: If the neutron trunk service plug-in is enabled, trunk port will be created by default. For more information, please refer to neutron trunk port.

Nova Metadata Service

Nova metadata service must be enabled and available at http://169.254.169.254. Currently the service is used to deliver Ignition config files to Nova instances and provide information about the machine to kubelet.

Glance Service

The creation of images in Glance should be available to the user. Now Glance is used for two things:

Right after the installation starts, the installer automatically uploads the actual RHCOS binary image to Glance with the name <clusterID>-rhcos. The image exists throughout the life of the cluster and is removed along with it.
The installer stores bootstrap ignition configs in a temporary image called <clusterID>-ignition. This is not a canonical use of the service, but this solution allows us to unify the installation process, since Glance is available on all OpenStack clouds, unlike Swift. The image exists for a limited period of time while the bootstrap process is running (normally 10-30 minutes), and then is automatically deleted.

OpenStack Credentials

You must have a clouds.yaml file in order to run the installer. The installer will look for a clouds.yaml file in the following locations in order:

Value of OS_CLIENT_CONFIG_FILE environment variable
Current directory
unix-specific user config directory (~/.config/openstack/clouds.yaml)
unix-specific site config directory (/etc/openstack/clouds.yaml)

In many OpenStack distributions, you can generate a clouds.yaml file through Horizon. Otherwise, you can make a clouds.yaml file yourself. Information on this file can be found here and it looks like:

clouds:
  shiftstack:
    auth:
      auth_url: http://10.10.14.42:5000/v3
      project_name: shiftstack
      username: shiftstack_user
      password: XXX
      user_domain_name: Default
      project_domain_name: Default
  dev-evn:
    region_name: RegionOne
    auth:
      username: 'devuser'
      password: XXX
      project_name: 'devonly'
      auth_url: 'https://10.10.14.22:5001/v2.0'

The file can contain information about several clouds. For instance, the example above describes two clouds: shiftstack and dev-evn. In order to determine which cloud to use, the user can either specify it in the install-config.yaml file under platform.openstack.cloud or with OS_CLOUD environment variable. If both are omitted, then the cloud name defaults to openstack.

Self Signed OpenStack CA certificates

If your OpenStack cluster uses self signed CA certificates for endpoint authentication, you will need a few additional steps to run the installer. First, make sure that the host running the installer trusts your CA certificates. If you want more information on how to do this, refer to the Red Hat OpenStack Plaform documentation. In the future, we plan to modify the installer to be able to trust certificates independently of the host OS.

sudo cp ca.crt.pem /etc/pki/ca-trust/source/anchors/
sudo update-ca-trust extract

Next, you should add the cacert key to your clouds.yaml. Its value should be a valid path to your CA cert that does not require root privilege to read. The path can be either absolute, or relative to the current working directory while running the installer.

clouds:
  shiftstack:
    auth: ...
    cacert: "/etc/pki/ca-trust/source/anchors/ca.crt.pem"

Standalone Single-Node Development Environment

If you would like to set up an isolated development environment, you may use a bare metal host running CentOS 7. The following repository includes some instructions and scripts to help with creating a single-node OpenStack development environment for running the installer. Please refer to this documentation for further details.

Running The Installer

Known Issues

OpenStack support has known issues. We will be documenting workarounds until we are able to resolve these bugs in the upcoming releases. To see the latest status of any bug, read through bugzilla or github link provided in that bug's description. If you know of a possible workaround that hasn't been documented yet, please comment in that bug's tracking link so we can address it as soon as possible. Also note that any bug listed in these documents is already a top priority issue for the dev team, and will be resolved as soon as possible. If you find more bugs during your runs, please read the section on issue reporting.

Initial Setup

Please head to openshift.com/try to get the latest versions of the installer, and instructions to run it.

Before running the installer, we recommend you create a directory for each cluster you plan to deploy. See the documents on the recommended workflow for more information about why you should do it this way.

mkdir ostest
cp install-config.yaml ostest/install-config.yaml

API Access

All the OpenShift nodes get created in an OpenStack tenant network and as such, can't be accessed directly in most OpenStack deployments. We will briefly explain how to set up access to the OpenShift API with and without floating IP addresses.

Using Floating IPs

This method allows you to attach two floating IP (FIP) addresses to endpoints in OpenShift.

A standard deployment uses three floating IP addresses in total:

External access to the OpenShift API
External access to the workloads (apps) running on the OpenShift cluster
Temporary IP address for bootstrap log collection

The first two addresses (API and Ingress) are generally created up-front and have the corresponding DNS records resolve to them.

The third floating IP is created automatically by the installer and will be destroyed along with all the other bootstrap resources. If the bootstrapping process fails, the installer will try to SSH into the bootstrap node and collect the logs.

Create API and Ingress Floating IP Addresses

The deployed OpenShift cluster will need two floating IP addresses, one to attach to the API load balancer (lb FIP), and one for the OpenShift applications (apps FIP). Note that the LB FIP is the IP address you will add to your install-config.yaml or select in the interactive installer prompt.

You can create them like so:

openstack floating ip create --description "API <cluster name>.<base domain>" <external network>
# => <lb FIP>
openstack floating ip create --description "Ingress <cluster name>.<base domain>" <external network>
# => <apps FIP>

NOTE: These IP addresses will not show up attached to any particular server (e.g. when running openstack server list). Similarly, the API and Ingress ports will always be in the DOWN state.

This is because the ports are not attached to the servers directly. Instead, their fixed IP addresses are managed by keepalived. This has no record in Neutron's database and as such, is not visible to OpenStack.

The network traffic will flow through even though the IPs and ports do not show up in the servers.

For more details, read the OpenShift on OpenStack networking infrastructure design document.

Create API and Ingress DNS Records

You will also need to add the following records to your DNS:

api.<cluster name>.<base domain>.  IN  A  <lb FIP>
*.apps.<cluster name>.<base domain>.  IN  A  <apps FIP>

If you're unable to create and publish these DNS records, you can add them to your /etc/hosts file.

<lb FIP> api.<cluster name>.<base domain>
<apps FIP> console-openshift-console.apps.<cluster name>.<base domain>
<apps FIP> integrated-oauth-server-openshift-authentication.apps.<cluster name>.<base domain>
<apps FIP> oauth-openshift.apps.<cluster name>.<base domain>
<apps FIP> prometheus-k8s-openshift-monitoring.apps.<cluster name>.<base domain>
<apps FIP> grafana-openshift-monitoring.apps.<cluster name>.<base domain>
<apps FIP> <app name>.apps.<cluster name>.<base domain>

WARNING: this workaround will make the API accessible only to the computer with these /etc/hosts entries. This is fine for your own testing (and it is enough for the installation to succeed), but it is not enough for a production deployment. In addition, if you create new OpenShift apps or routes, you will have to add their entries too, because /etc/hosts does not support wildcard entries.

External API Access

If you have specified the API floating IP (either via the installer prompt or by adding the apiFloatingIP entry in your install-config.yaml) the installer will attach the Floating IP address to the api-port automatically.

If you have created the API DNS record, you should be able access the OpenShift API.

External Ingress (apps) Access

In the same manner, you may have specified an Ingress floating IP by adding the ingressFloatingIP entry in your install-config.yaml, in which case the installer attaches the Floating IP address to the ingress-port automatically.

If ingressFloatingIP is empty or absent in install-config.yaml, the Ingress port will be created but not attached to any floating IP. You can manually attach the Ingress floating IP to the ingress-port after the cluster is created.

That can be done in the following steps:

openstack port show <cluster name>-<clusterID>-ingress-port

Then attach the FIP to it:

openstack floating ip set --port <cluster name>-<clusterID>-ingress-port <apps FIP>

This assumes the floating IP and corresponding *.apps DNS record exists.

Without Floating IPs

If you cannot or don't want to pre-create a floating IP address, the installation should still succeed, however the installer will fail waiting for the API.

WARNING: The installer will fail if it can't reach the bootstrap OpenShift API in 20 minutes.

Even if the installer times out, the OpenShift cluster should still come up. Once the bootstrapping process is in place, it should all run to completion. So you should be able to deploy OpenShift without any floating IP addresses and DNS records and create everything yourself after the cluster is up.

Running a Deployment

To run the installer, you have the option of using the interactive wizard, or providing your own install-config.yaml file for it. The wizard is the easier way to run the installer, but passing your own install-config.yaml enables you to use more fine grained customizations. If you are going to create your own install-config.yaml, read through the available OpenStack customizations. For information on running the installer with Kuryr, see the Kuryr docs.

./openshift-install create cluster --dir ostest

If you want to create an install config without deploying a cluster, you can use the command:

./openshift-install create install-config --dir ostest

Current Expected Behavior

Currently:

Deploys an isolated tenant network
Deploys a bootstrap instance to bootstrap the OpenShift cluster
Deploys 3 master nodes
Once the masters are deployed, the bootstrap instance is destroyed
Deploys 3 worker nodes

Look for a message like this to verify that your install succeeded:

INFO Install complete!
INFO To access the cluster as the system:admin user when using 'oc', run 'export KUBECONFIG=/home/stack/ostest/auth/kubeconfig'
INFO Access the OpenShift web-console here: https://console-openshift-console.apps.ostest.shiftstack.com
INFO Login to the console with user: kubeadmin, password: xxx

Checking Cluster Status

If you want to see the status of the apps and services in your cluster during, or after a deployment, first export your administrator's kubeconfig:

export KUBECONFIG=ostest/auth/kubeconfig

After a finished deployment, there should be a node for each master and worker server created. You can check this with the command:

oc get nodes

To see the version of your OpenShift cluster, do:

oc get clusterversion

To see the status of you operators, do:

oc get clusteroperator

Finally, to see all the running pods in your cluster, you can do:

oc get pods -A

Destroying The Cluster

To destroy the cluster, point it to your cluster with this command:

./openshift-install --log-level debug destroy cluster --dir ostest

Then, you can delete the folder containing the cluster metadata:

rm -rf ostest/

Post Install Operations

Adding a MachineSet

Groups of Compute nodes are managed using the MachineSet resource. It is possible to create additional MachineSets post-install, for example to assign workloads to specific machines.

When running on OpenStack, the MachineSet has platform-specific fields under spec.template.spec.providerSpec.value. For more information about the values that you can set in the providerSpec, see the API definition.

apiVersion: machine.openshift.io/v1beta1
kind: MachineSet
metadata:
  labels:
    machine.openshift.io/cluster-api-cluster: <infrastructure_ID>
    machine.openshift.io/cluster-api-machine-role: <node_role>
    machine.openshift.io/cluster-api-machine-type: <node_role>
  name: <infrastructure_ID>-<node_role>
  namespace: openshift-machine-api
spec:
  replicas: <number_of_replicas>
  selector:
    matchLabels:
      machine.openshift.io/cluster-api-cluster: <infrastructure_ID>
      machine.openshift.io/cluster-api-machineset: <infrastructure_ID>-<node_role>
  template:
    metadata:
      labels:
        machine.openshift.io/cluster-api-cluster: <infrastructure_ID>
        machine.openshift.io/cluster-api-machine-role: <node_role>
        machine.openshift.io/cluster-api-machine-type: <node_role>
        machine.openshift.io/cluster-api-machineset: <infrastructure_ID>-<node_role>
    spec:
      providerSpec:
        value:
          apiVersion: openstackproviderconfig.openshift.io/v1alpha1
          cloudName: openstack
          cloudsSecret:
            name: openstack-cloud-credentials
            namespace: openshift-machine-api
          flavor: <nova_flavor>
          image: <glance_image_name_or_location>
          serverGroupID: <UUID of the pre-created Nova server group (optional)>
          kind: OpenstackProviderSpec
          networks:
          - filter: {}
            subnets:
            - filter:
                name: <subnet_name>
                tags: openshiftClusterID=<infrastructure_ID>
          securityGroups:
          - filter: {}
            name: <infrastructure_ID>-<node_role>
          serverMetadata:
            Name: <infrastructure_ID>-<node_role>
            openshiftClusterID: <infrastructure_ID>
          tags:
          - openshiftClusterID=<infrastructure_ID>
          trunk: true
          userDataSecret:
            name: <node_role>-user-data
          availabilityZone: <optional_openstack_availability_zone>

Defining a MachineSet That Uses Multiple Networks

To define a MachineSet with multiple networks, the primarySubnet value in the providerSpec must be set to the OpenStack subnet that you want the Kubernetes endpoints of the nodes to be published on. For most use cases, this is the same subnet as the machinesSubnet in the install-config.yaml.

After you set the subnet, add all of the networks that you want to attach to your machines to the Networks list in providerSpec. You must also add the network that the primary subnet is part of to this list.

Using a Server Group

In order to hint the Nova scheduler to spread the Machines across different hosts, first create a Server Group with the desired policy:

openstack server group create --policy=anti-affinity <server-group-name>
## OR ##
openstack --os-compute-api-version=2.15 server group create --policy=soft-anti-affinity <server-group-name>

If the command is successful, the OpenStack CLI will return the ID of the newly created Server Group. Paste it in the optional serverGroupID property of the MachineSet.

Setting Nova Availability Zones

In order to use Availability Zones, create one MachineSet per target Availability Zone, and set the Availability Zone in the availabilityZone property of the MachineSet.

NOTE: Note when deploying with Kuryr there is an Octavia API loadbalancer VM that will not fulfill the Availability Zones restrictions due to Octavia lack of support for it. In addition, if Octavia only has the amphora provider instead of also the OVN-Octavia provider, all the OpenShift services will be backed up by Octavia Load Balancer VMs which will not fulfill the Availability Zone restrictions either.

Using a Custom External Load Balancer

You can shift ingress/egress traffic from the default OpenShift on OpenStack load balancer to a load balancer that you provide. To do so, the instance that it runs from must be able to access every machine in your cluster. You might ensure this access by creating the instance on a subnet that is within your cluster's network, and then attaching a router interface to that subnet from the OpenShift-external-router [object/instance/whatever]. This can also be accomplished by attaching floating ips to the machines you want to add to your load balancer.

External Facing OpenShift Services

Add the following external facing services to your new load balancer:

The master nodes serve the OpenShift API on port 6443 using TCP.
The apps hosted on the worker nodes are served on ports 80, and 443. They are both served using TCP.

Note: Make sure the instance that your new load balancer is running on has security group rules that allow TCP traffic over these ports.

HAProxy Example Load Balancer Config

The following HAProxy config file demonstrates a basic configuration for an external load balancer:

listen <cluster-name>-api-6443
        bind 0.0.0.0:6443
        mode tcp
        balance roundrobin
        server <cluster-name>-master-0 192.168.0.154:6443 check
        server <cluster-name>-master-1 192.168.0.15:6443 check
        server <cluster-name>-master-2 192.168.3.128:6443 check
listen <cluster-name>-apps-443
        bind 0.0.0.0:443
        mode tcp
        balance roundrobin
        server <cluster-name>-worker-0 192.168.3.18:443 check
        server <cluster-name>-worker-1 192.168.2.228:443 check
        server <cluster-name>-worker-2 192.168.1.253:443 check
listen <cluster-name>-apps-80
        bind 0.0.0.0:80
        mode tcp
        balance roundrobin
        server <cluster-name>-worker-0 192.168.3.18:80 check
        server <cluster-name>-worker-1 192.168.2.228:80 check
        server <cluster-name>-worker-2 192.168.1.253:80 check

DNS Lookups

To ensure that your API and apps are accessible through your load balancer, create or update your DNS entries for those endpoints. To use your new load balancing service for external traffic, make sure the IP address for these DNS entries is the IP address your load balancer is reachable at.

<load balancer ip> api.<cluster-name>.<base domain>
<load balancer ip> apps.<cluster-name>.base domain>

Verifying that the API is Reachable

One good way to test whether or not you can reach the API is to run the oc command. If you can't do that easily, you can use this curl command:

curl https://api.<cluster-name>.<base domain>:6443/version --insecure

Result:

{
  "major": "1",
  "minor": "19",
  "gitVersion": "v1.19.2+4abb4a7",
  "gitCommit": "4abb4a77838037b8dbb8e4ca34e63c4a129654c8",
  "gitTreeState": "clean",
  "buildDate": "2020-11-12T05:46:36Z",
  "goVersion": "go1.15.2",
  "compiler": "gc",
  "platform": "linux/amd64"
}

Note: The versions in the sample output may differ from your own. As long as you get a JSON payload response, the API is accessible.

Verifying that Apps Reachable

The simplest way to verify that apps are reachable is to open the OpenShift console in a web browser. If you don't have access to a web browser, query the console with the following curl command:

curl http://console-openshift-console.apps.<cluster-name>.<base domain> -I -L --insecure