| vSphere object for role | When required | Required privileges in vSphere API |
|---|---|---|
vSphere vCenter |
Always |
|
vSphere vCenter Cluster |
If VMs will be created in the cluster root |
|
vSphere vCenter Resource Pool |
If an existing resource pool is provided |
|
vSphere Datastore |
Always |
|
vSphere Port Group |
Always |
|
Virtual Machine Folder |
Always |
|
vSphere vCenter Datacenter |
If the installation program creates the virtual machine folder. For UPI, |
|
vSphere installation requirements for user-provisioned infrastructure
- VMware vSphere infrastructure requirements
- VMware vSphere CSI Driver Operator requirements
- Requirements for a cluster with user-provisioned infrastructure
- vCenter requirements
- Required machines for cluster installation
- Minimum resource requirements for cluster installation
- Requirements for encrypting virtual machines
- Certificate signing requests management
- Networking requirements for user-provisioned infrastructure
- User-provisioned DNS requirements
- Load balancing requirements for user-provisioned infrastructure
Before you begin an installation on infrastructure that you provision, be sure that your vSphere environment meets the following installation requirements.
VMware vSphere infrastructure requirements
You must install an OKD cluster on one of the following versions of a VMware vSphere instance that meets the requirements for the components that you use:
-
Version 7.0 Update 2 or later, or VMware Cloud Foundation 4.3 or later
-
Version 8.0 Update 1 or later, or VMware Cloud Foundation 5.0 or later
Both of these releases support Container Storage Interface (CSI) migration, which is enabled by default on OKD 4.14.
You can host the VMware vSphere infrastructure on-premise or on a VMware Cloud Verified provider that meets the requirements outlined in the following tables:
| Virtual environment product | Required version |
|---|---|
VMware virtual hardware |
15 or later |
vSphere ESXi hosts |
7.0 Update 2 or later, or VMware Cloud Foundation 4.3 or later; 8.0 Update 1 or later, or VMware Cloud Foundation 5.0 or later |
vCenter host |
7.0 Update 2 or later, or VMware Cloud Foundation 4.3 or later; 8.0 Update 1 or later, or VMware Cloud Foundation 5.0 or later |
|
You must ensure that the time on your ESXi hosts is synchronized before you install OKD. See Edit Time Configuration for a Host in the VMware documentation. |
| Component | Minimum supported versions | Description |
|---|---|---|
Hypervisor |
vSphere 7.0 Update 2 or later, or VMware Cloud Foundation 4.3 or later; vSphere 8.0 Update 1 or later, or VMware Cloud Foundation 5.0 or later with virtual hardware version 15 |
This hypervisor version is the minimum version that Fedora CoreOS (FCOS) supports. For more information about supported hardware on the latest version of Fedora that is compatible with FCOS, see Hardware on the Red Hat Customer Portal. |
Optional: Networking (NSX-T) |
vSphere 7.0 Update 2 or later, or VMware Cloud Foundation 4.3 or later; vSphere 8.0 Update 1 or later, or VMware Cloud Foundation 5.0 or later |
For more information about the compatibility of NSX and OKD, see the Release Notes section of VMware’s NSX container plugin documentation. |
CPU micro-architecture |
x86-64-v2 or higher |
OpenShift 4.13 and later are based on RHEL 9.2 host operating system which raised the microarchitecture requirements to x86-64-v2. See the RHEL Microarchitecture requirements documentation. You can verify compatibility by following the procedures outlined in this KCS article. |
|
To ensure the best performance conditions for your cluster workloads that operate on Oracle® Cloud Infrastructure (OCI) and on the Oracle® Cloud VMware Solution (OCVS) service, ensure volume performance units (VPUs) for your block volume are sized for your workloads. The following list provides some guidance in selecting the VPUs needed for specific performance needs:
Consider allocating additional VPUs to give enough capacity for updates and scaling activities. See Block Volume Performance Levels (Oracle documentation). |
VMware vSphere CSI Driver Operator requirements
To install the vSphere CSI Driver Operator, the following requirements must be met:
-
VMware vSphere version: 7.0 Update 2 or later, or VMware Cloud Foundation 4.3 or later; 8.0 Update 1 or later, or VMware Cloud Foundation 5.0 or later
-
vCenter version: 7.0 Update 2 or later, or VMware Cloud Foundation 4.3 or later; 8.0 Update 1 or later, or VMware Cloud Foundation 5.0 or later
-
Virtual machines of hardware version 15 or later
-
No third-party vSphere CSI driver already installed in the cluster
If a third-party vSphere CSI driver is present in the cluster, OKD does not overwrite it. The presence of a third-party vSphere CSI driver prevents OKD from updating to OKD 4.13 or later.
|
The VMware vSphere CSI Driver Operator is supported only on clusters deployed with |
-
To remove a third-party vSphere CSI driver, see Removing a third-party vSphere CSI Driver.
-
To update the hardware version for your vSphere nodes, see Updating hardware on nodes running in vSphere.
Requirements for a cluster with user-provisioned infrastructure
For a cluster that contains user-provisioned infrastructure, you must deploy all of the required machines.
This section describes the requirements for deploying OKD on user-provisioned infrastructure.
vCenter requirements
Before you install an OKD cluster on your vCenter that uses infrastructure that you provided, you must prepare your environment.
Required vCenter account privileges
To install an OKD cluster in a vCenter, your vSphere account must include privileges for reading and creating the required resources. Using an account that has global administrative privileges is the simplest way to access all of the necessary permissions.
Roles and privileges required for installation in vSphere API
Roles and privileges required for installation in vCenter graphical user interface (GUI)
| vSphere object for role | When required | Required privileges in vCenter GUI |
|---|---|---|
vSphere vCenter |
Always |
|
vSphere vCenter Cluster |
If VMs will be created in the cluster root |
|
vSphere vCenter Resource Pool |
If an existing resource pool is provided |
|
vSphere Datastore |
Always |
|
vSphere Port Group |
Always |
|
Virtual Machine Folder |
Always |
|
vSphere vCenter Datacenter |
If the installation program creates the virtual machine folder. For UPI, |
|
Additionally, the user requires some ReadOnly permissions, and some of the roles require permission to propogate the permissions to child objects. These settings vary depending on whether or not you install the cluster into an existing folder.
Required permissions and propagation settings
| vSphere object | When required | Propagate to children | Permissions required |
|---|---|---|---|
vSphere vCenter |
Always |
False |
Listed required privileges |
vSphere vCenter Datacenter |
Existing folder |
False |
|
Installation program creates the folder |
True |
Listed required privileges |
|
vSphere vCenter Cluster |
Existing resource pool |
False |
|
VMs in cluster root |
True |
Listed required privileges |
|
vSphere vCenter Datastore |
Always |
False |
Listed required privileges |
vSphere Switch |
Always |
False |
|
vSphere Port Group |
Always |
False |
Listed required privileges |
vSphere vCenter Virtual Machine Folder |
Existing folder |
True |
Listed required privileges |
vSphere vCenter Resource Pool |
Existing resource pool |
True |
Listed required privileges |
For more information about creating an account with only the required privileges, see vSphere Permissions and User Management Tasks in the vSphere documentation.
Using OKD with vMotion
If you intend on using vMotion in your vSphere environment, consider the following before installing an OKD cluster.
-
Using Storage vMotion can cause issues and is not supported.
-
Using VMware compute vMotion to migrate the workloads for both OKD compute machines and control plane machines is generally supported, where generally implies that you meet all VMware best practices for vMotion.
To help ensure the uptime of your compute and control plane nodes, ensure that you follow the VMware best practices for vMotion, and use VMware anti-affinity rules to improve the availability of OKD during maintenance or hardware issues.
For more information about vMotion and anti-affinity rules, see the VMware vSphere documentation for vMotion networking requirements and VM anti-affinity rules.
-
If you are using VMware vSphere volumes in your pods, migrating a VM across datastores, either manually or through Storage vMotion, causes invalid references within OKD persistent volume (PV) objects that can result in data loss.
-
OKD does not support selective migration of VMDKs across datastores, using datastore clusters for VM provisioning or for dynamic or static provisioning of PVs, or using a datastore that is part of a datastore cluster for dynamic or static provisioning of PVs.
You can specify the path of any datastore that exists in a datastore cluster. By default, Storage Distributed Resource Scheduler (SDRS), which uses Storage vMotion, is automatically enabled for a datastore cluster. Red Hat does not support Storage vMotion, so you must disable Storage DRS to avoid data loss issues for your OKD cluster.
If you must specify VMs across multiple datastores, use a
datastoreobject to specify a failure domain in your cluster’sinstall-config.yamlconfiguration file. For more information, see "VMware vSphere region and zone enablement".
Cluster resources
When you deploy an OKD cluster that uses infrastructure that you provided, you must create the following resources in your vCenter instance:
-
1 Folder
-
1 Tag category
-
1 Tag
-
Virtual machines:
-
1 template
-
1 temporary bootstrap node
-
3 control plane nodes
-
3 compute machines
-
Although these resources use 856 GB of storage, the bootstrap node is destroyed during the cluster installation process. A minimum of 800 GB of storage is required to use a standard cluster.
If you deploy more compute machines, the OKD cluster will use more storage.
Cluster limits
Available resources vary between clusters. The number of possible clusters within a vCenter is limited primarily by available storage space and any limitations on the number of required resources. Be sure to consider both limitations to the vCenter resources that the cluster creates and the resources that you require to deploy a cluster, such as IP addresses and networks.
Networking requirements
You can use Dynamic Host Configuration Protocol (DHCP) for the network and configure the DHCP server to set persistent IP addresses to machines in your cluster. In the DHCP lease, you must configure the DHCP to use the default gateway.
|
You do not need to use the DHCP for the network if you want to provision nodes with static IP addresses. |
If you specify nodes or groups of nodes on different VLANs for a cluster that you want to install on user-provisioned infrastructure, you must ensure that machines in your cluster meet the requirements outlined in the "Network connectivity requirements" section of the Networking requirements for user-provisioned infrastructure document.
If you are installing to a restricted environment, the VM in your restricted network must have access to vCenter so that it can provision and manage nodes, persistent volume claims (PVCs), and other resources.
|
Ensure that each OKD node in the cluster has access to a Network Time Protocol (NTP) server that is discoverable by DHCP. Installation is possible without an NTP server. However, asynchronous server clocks can cause errors, which the NTP server prevents. |
Additionally, you must create the following networking resources before you install the OKD cluster:
Required IP Addresses
DNS records
You must create DNS records for two static IP addresses in the appropriate DNS server for the vCenter instance that hosts your OKD cluster. In each record, <cluster_name> is the cluster name and <base_domain> is the cluster base domain that you specify when you install the cluster. A complete DNS record takes the form: <component>.<cluster_name>.<base_domain>..
| Component | Record | Description |
|---|---|---|
API VIP |
|
This DNS A/AAAA or CNAME record must point to the load balancer for the control plane machines. This record must be resolvable by both clients external to the cluster and from all the nodes within the cluster. |
ingress VIP |
|
A wildcard DNS A/AAAA or CNAME record that points to the load balancer that targets the machines that run the ingress router pods, which are the worker nodes by default. This record must be resolvable by both clients external to the cluster and from all the nodes within the cluster. |
Required machines for cluster installation
The smallest OKD clusters require the following hosts:
| Hosts | Description |
|---|---|
One temporary bootstrap machine |
The cluster requires the bootstrap machine to deploy the OKD cluster on the three control plane machines. You can remove the bootstrap machine after you install the cluster. |
Three control plane machines |
The control plane machines run the Kubernetes and OKD services that form the control plane. |
At least two compute machines, which are also known as worker machines. |
The workloads requested by OKD users run on the compute machines. |
|
To maintain high availability of your cluster, use separate physical hosts for these cluster machines. |
The bootstrap and control plane machines must use Fedora CoreOS (FCOS) as the operating system. However, the compute machines can choose between Fedora CoreOS (FCOS), Fedora 8.6 and later.
Minimum resource requirements for cluster installation
Each cluster machine must meet the following minimum requirements:
| Machine | Operating System | vCPU [1] | Virtual RAM | Storage | Input/Output Per Second (IOPS)[2] |
|---|---|---|---|---|---|
Bootstrap |
FCOS |
4 |
16 GB |
100 GB |
300 |
Control plane |
FCOS |
4 |
16 GB |
100 GB |
300 |
Compute |
FCOS |
2 |
8 GB |
100 GB |
300 |
-
One vCPU is equivalent to one physical core when simultaneous multithreading (SMT), or Hyper-Threading, is not enabled. When enabled, use the following formula to calculate the corresponding ratio: (threads per core × cores) × sockets = vCPUs.
-
OKD and Kubernetes are sensitive to disk performance, and faster storage is recommended, particularly for etcd on the control plane nodes which require a 10 ms p99 fsync duration. Note that on many cloud platforms, storage size and IOPS scale together, so you might need to over-allocate storage volume to obtain sufficient performance.
-
As with all user-provisioned installations, if you choose to use Fedora compute machines in your cluster, you take responsibility for all operating system life cycle management and maintenance, including performing system updates, applying patches, and completing all other required tasks. Use of Fedora 7 compute machines is deprecated and has been removed in OKD 4.10 and later.
|
As of OKD version 4.13, RHCOS is based on RHEL version 9.2, which updates the micro-architecture requirements. The following list contains the minimum instruction set architectures (ISA) that each architecture requires:
For more information, see RHEL Architectures. |
If an instance type for your platform meets the minimum requirements for cluster machines, it is supported to use in OKD.
Requirements for encrypting virtual machines
You can encrypt your virtual machines prior to installing OKD 4.14 by meeting the following requirements.
-
You have configured a Standard key provider in vSphere. For more information, see Adding a KMS to vCenter Server.
The Native key provider in vCenter is not supported. For more information, see vSphere Native Key Provider Overview.
-
You have enabled host encryption mode on all of the ESXi hosts that are hosting the cluster. For more information, see Enabling host encryption mode.
-
You have a vSphere account which has all cryptographic privileges enabled. For more information, see Cryptographic Operations Privileges.
When you deploy the OVF template in the section titled "Installing RHCOS and starting the OpenShift Container Platform bootstrap process", select the option to "Encrypt this virtual machine" when you are selecting storage for the OVF template. After completing cluster installation, create a storage class that uses the encryption storage policy you used to encrypt the virtual machines.
Certificate signing requests management
Because your cluster has limited access to automatic machine management when you use infrastructure that you provision, you must provide a mechanism for approving cluster certificate signing requests (CSRs) after installation. The kube-controller-manager only approves the kubelet client CSRs. The machine-approver cannot guarantee the validity of a serving certificate that is requested by using kubelet credentials because it cannot confirm that the correct machine issued the request. You must determine and implement a method of verifying the validity of the kubelet serving certificate requests and approving them.
Networking requirements for user-provisioned infrastructure
All the Fedora CoreOS (FCOS) machines require networking to be configured in initramfs during boot
to fetch their Ignition config files.
During the initial boot, the machines require an IP address configuration that is set either through a DHCP server or statically by providing the required boot options. After a network connection is established, the machines download their Ignition config files from an HTTP or HTTPS server. The Ignition config files are then used to set the exact state of each machine. The Machine Config Operator completes more changes to the machines, such as the application of new certificates or keys, after installation.
It is recommended to use a DHCP server for long-term management of the cluster machines. Ensure that the DHCP server is configured to provide persistent IP addresses, DNS server information, and hostnames to the cluster machines.
|
If a DHCP service is not available for your user-provisioned infrastructure, you can instead provide the IP networking configuration and the address of the DNS server to the nodes at FCOS install time. These can be passed as boot arguments if you are installing from an ISO image. See the Installing FCOS and starting the OKD bootstrap process section for more information about static IP provisioning and advanced networking options. |
The Kubernetes API server must be able to resolve the node names of the cluster machines. If the API servers and worker nodes are in different zones, you can configure a default DNS search zone to allow the API server to resolve the node names. Another supported approach is to always refer to hosts by their fully-qualified domain names in both the node objects and all DNS requests.
Setting the cluster node hostnames through DHCP
On Fedora CoreOS (FCOS) machines, the hostname is set through NetworkManager. By default, the machines obtain their hostname through DHCP. If the hostname is not provided by DHCP, set statically through kernel arguments, or another method, it is obtained through a reverse DNS lookup. Reverse DNS lookup occurs after the network has been initialized on a node and can take time to resolve. Other system services can start prior to this and detect the hostname as localhost or similar. You can avoid this by using DHCP to provide the hostname for each cluster node.
Additionally, setting the hostnames through DHCP can bypass any manual DNS record name configuration errors in environments that have a DNS split-horizon implementation.
Network connectivity requirements
You must configure the network connectivity between machines to allow OKD cluster components to communicate. Each machine must be able to resolve the hostnames of all other machines in the cluster.
This section provides details about the ports that are required.
|
In connected OKD environments, all nodes are required to have internet access to pull images for platform containers and provide telemetry data to Red Hat. |
| Protocol | Port | Description |
|---|---|---|
ICMP |
N/A |
Network reachability tests |
TCP |
|
Metrics |
|
Host level services, including the node exporter on ports |
|
|
The default ports that Kubernetes reserves |
|
|
openshift-sdn |
|
UDP |
|
VXLAN |
|
Geneve |
|
|
Host level services, including the node exporter on ports |
|
|
IPsec IKE packets |
|
|
IPsec NAT-T packets |
|
|
Network Time Protocol (NTP) on UDP port If an external NTP time server is configured, you must open UDP port |
|
TCP/UDP |
|
Kubernetes node port |
ESP |
N/A |
IPsec Encapsulating Security Payload (ESP) |
| Protocol | Port | Description |
|---|---|---|
TCP |
|
Kubernetes API |
| Protocol | Port | Description |
|---|---|---|
TCP |
|
etcd server and peer ports |
NTP configuration for user-provisioned infrastructure
OKD clusters are configured to use a public Network Time Protocol (NTP) server by default. If you want to use a local enterprise NTP server, or if your cluster is being deployed in a disconnected network, you can configure the cluster to use a specific time server. For more information, see the documentation for Configuring chrony time service.
If a DHCP server provides NTP server information, the chrony time service on the Fedora CoreOS (FCOS) machines read the information and can sync the clock with the NTP servers.
User-provisioned DNS requirements
In OKD deployments, DNS name resolution is required for the following components:
-
The Kubernetes API
-
The OKD application wildcard
-
The bootstrap, control plane, and compute machines
Reverse DNS resolution is also required for the Kubernetes API, the bootstrap machine, the control plane machines, and the compute machines.
DNS A/AAAA or CNAME records are used for name resolution and PTR records are used for reverse name resolution. The reverse records are important because Fedora CoreOS (FCOS) uses the reverse records to set the hostnames for all the nodes, unless the hostnames are provided by DHCP. Additionally, the reverse records are used to generate the certificate signing requests (CSR) that OKD needs to operate.
|
It is recommended to use a DHCP server to provide the hostnames to each cluster node. See the DHCP recommendations for user-provisioned infrastructure section for more information. |
The following DNS records are required for a user-provisioned OKD cluster and they must be in place before installation. In each record, <cluster_name> is the cluster name and <base_domain> is the base domain that you specify in the install-config.yaml file. A complete DNS record takes the form: <component>.<cluster_name>.<base_domain>..
| Component | Record | Description | |
|---|---|---|---|
Kubernetes API |
|
A DNS A/AAAA or CNAME record, and a DNS PTR record, to identify the API load balancer. These records must be resolvable by both clients external to the cluster and from all the nodes within the cluster. |
|
|
A DNS A/AAAA or CNAME record, and a DNS PTR record, to internally identify the API load balancer. These records must be resolvable from all the nodes within the cluster.
|
||
Routes |
|
A wildcard DNS A/AAAA or CNAME record that refers to the application ingress load balancer. The application ingress load balancer targets the machines that run the ingress Controller pods. The ingress Controller pods run on the compute machines by default. These records must be resolvable by both clients external to the cluster and from all the nodes within the cluster. For example, |
|
Bootstrap machine |
|
A DNS A/AAAA or CNAME record, and a DNS PTR record, to identify the bootstrap machine. These records must be resolvable by the nodes within the cluster. |
|
Control plane machines |
|
DNS A/AAAA or CNAME records and DNS PTR records to identify each machine for the control plane nodes. These records must be resolvable by the nodes within the cluster. |
|
Compute machines |
|
DNS A/AAAA or CNAME records and DNS PTR records to identify each machine for the worker nodes. These records must be resolvable by the nodes within the cluster. |
|
In OKD 4.4 and later, you do not need to specify etcd host and SRV records in your DNS configuration. |
|
You can use the |
Example DNS configuration for user-provisioned clusters
This section provides A and PTR record configuration samples that meet the DNS requirements for deploying OKD on user-provisioned infrastructure. The samples are not meant to provide advice for choosing one DNS solution over another.
In the examples, the cluster name is ocp4 and the base domain is example.com.
The following example is a BIND zone file that shows sample A records for name resolution in a user-provisioned cluster.
Sample DNS zone database
$TTL 1W
@ IN SOA ns1.example.com. root (
2019070700 ; serial
3H ; refresh (3 hours)
30M ; retry (30 minutes)
2W ; expiry (2 weeks)
1W ) ; minimum (1 week)
IN NS ns1.example.com.
IN MX 10 smtp.example.com.
;
;
ns1.example.com. IN A 192.168.1.5
smtp.example.com. IN A 192.168.1.5
;
helper.example.com. IN A 192.168.1.5
helper.ocp4.example.com. IN A 192.168.1.5
;
api.ocp4.example.com. IN A 192.168.1.5 (1)
api-int.ocp4.example.com. IN A 192.168.1.5 (2)
;
*.apps.ocp4.example.com. IN A 192.168.1.5 (3)
;
bootstrap.ocp4.example.com. IN A 192.168.1.96 (4)
;
control-plane0.ocp4.example.com. IN A 192.168.1.97 (5)
control-plane1.ocp4.example.com. IN A 192.168.1.98 (5)
control-plane2.ocp4.example.com. IN A 192.168.1.99 (5)
;
compute0.ocp4.example.com. IN A 192.168.1.11 (6)
compute1.ocp4.example.com. IN A 192.168.1.7 (6)
;
;EOF| 1 | Provides name resolution for the Kubernetes API. The record refers to the IP address of the API load balancer. | ||
| 2 | Provides name resolution for the Kubernetes API. The record refers to the IP address of the API load balancer and is used for internal cluster communications. | ||
| 3 | Provides name resolution for the wildcard routes. The record refers to the IP address of the application ingress load balancer. The application ingress load balancer targets the machines that run the ingress Controller pods. The ingress Controller pods run on the compute machines by default.
|
||
| 4 | Provides name resolution for the bootstrap machine. | ||
| 5 | Provides name resolution for the control plane machines. | ||
| 6 | Provides name resolution for the compute machines. |
The following example BIND zone file shows sample PTR records for reverse name resolution in a user-provisioned cluster.
Sample DNS zone database for reverse records
$TTL 1W
@ IN SOA ns1.example.com. root (
2019070700 ; serial
3H ; refresh (3 hours)
30M ; retry (30 minutes)
2W ; expiry (2 weeks)
1W ) ; minimum (1 week)
IN NS ns1.example.com.
;
5.1.168.192.in-addr.arpa. IN PTR api.ocp4.example.com. (1)
5.1.168.192.in-addr.arpa. IN PTR api-int.ocp4.example.com. (2)
;
96.1.168.192.in-addr.arpa. IN PTR bootstrap.ocp4.example.com. (3)
;
97.1.168.192.in-addr.arpa. IN PTR control-plane0.ocp4.example.com. (4)
98.1.168.192.in-addr.arpa. IN PTR control-plane1.ocp4.example.com. (4)
99.1.168.192.in-addr.arpa. IN PTR control-plane2.ocp4.example.com. (4)
;
11.1.168.192.in-addr.arpa. IN PTR compute0.ocp4.example.com. (5)
7.1.168.192.in-addr.arpa. IN PTR compute1.ocp4.example.com. (5)
;
;EOF| 1 | Provides reverse DNS resolution for the Kubernetes API. The PTR record refers to the record name of the API load balancer. |
| 2 | Provides reverse DNS resolution for the Kubernetes API. The PTR record refers to the record name of the API load balancer and is used for internal cluster communications. |
| 3 | Provides reverse DNS resolution for the bootstrap machine. |
| 4 | Provides reverse DNS resolution for the control plane machines. |
| 5 | Provides reverse DNS resolution for the compute machines. |
|
A PTR record is not required for the OKD application wildcard. |
Load balancing requirements for user-provisioned infrastructure
Before you install OKD, you must provision the API and application ingress load balancing infrastructure. In production scenarios, you can deploy the API and application ingress load balancers separately so that you can scale the load balancer infrastructure for each in isolation.
|
If you want to deploy the API and application ingress load balancers with a Fedora instance, you must purchase the Fedora subscription separately. |
The load balancing infrastructure must meet the following requirements:
-
API load balancer: Provides a common endpoint for users, both human and machine, to interact with and configure the platform. Configure the following conditions:
-
Layer 4 load balancing only. This can be referred to as Raw TCP or SSL Passthrough mode.
-
A stateless load balancing algorithm. The options vary based on the load balancer implementation.
Do not configure session persistence for an API load balancer. Configuring session persistence for a Kubernetes API server might cause performance issues from excess application traffic for your OKD cluster and the Kubernetes API that runs inside the cluster.
Configure the following ports on both the front and back of the load balancers:
Table 10. API load balancer Port Back-end machines (pool members) Internal External Description 6443Bootstrap and control plane. You remove the bootstrap machine from the load balancer after the bootstrap machine initializes the cluster control plane. You must configure the
/readyzendpoint for the API server health check probe.X
X
Kubernetes API server
22623Bootstrap and control plane. You remove the bootstrap machine from the load balancer after the bootstrap machine initializes the cluster control plane.
X
Machine config server
The load balancer must be configured to take a maximum of 30 seconds from the time the API server turns off the
/readyzendpoint to the removal of the API server instance from the pool. Within the time frame after/readyzreturns an error or becomes healthy, the endpoint must have been removed or added. Probing every 5 or 10 seconds, with two successful requests to become healthy and three to become unhealthy, are well-tested values. -
-
Application ingress load balancer: Provides an ingress point for application traffic flowing in from outside the cluster. A working configuration for the ingress router is required for an OKD cluster.
Configure the following conditions:
-
Layer 4 load balancing only. This can be referred to as Raw TCP or SSL Passthrough mode.
-
A connection-based or session-based persistence is recommended, based on the options available and types of applications that will be hosted on the platform.
If the true IP address of the client can be seen by the application ingress load balancer, enabling source IP-based session persistence can improve performance for applications that use end-to-end TLS encryption.
Configure the following ports on both the front and back of the load balancers:
Table 11. Application ingress load balancer Port Back-end machines (pool members) Internal External Description 443The machines that run the ingress Controller pods, compute, or worker, by default.
X
X
HTTPS traffic
80The machines that run the ingress Controller pods, compute, or worker, by default.
X
X
HTTP traffic
If you are deploying a three-node cluster with zero compute nodes, the ingress Controller pods run on the control plane nodes. In three-node cluster deployments, you must configure your application ingress load balancer to route HTTP and HTTPS traffic to the control plane nodes.
-
Example load balancer configuration for user-provisioned clusters
This section provides an example API and application ingress load balancer configuration that meets the load balancing requirements for user-provisioned clusters. The sample is an /etc/haproxy/haproxy.cfg configuration for an HAProxy load balancer. The example is not meant to provide advice for choosing one load balancing solution over another.
In the example, the same load balancer is used for the Kubernetes API and application ingress traffic. In production scenarios, you can deploy the API and application ingress load balancers separately so that you can scale the load balancer infrastructure for each in isolation.
|
If you are using HAProxy as a load balancer and SELinux is set to |
Sample API and application ingress load balancer configuration
global
log 127.0.0.1 local2
pidfile /var/run/haproxy.pid
maxconn 4000
daemon
defaults
mode http
log global
option dontlognull
option http-server-close
option redispatch
retries 3
timeout http-request 10s
timeout queue 1m
timeout connect 10s
timeout client 1m
timeout server 1m
timeout http-keep-alive 10s
timeout check 10s
maxconn 3000
listen api-server-6443 (1)
bind *:6443
mode tcp
option httpchk GET /readyz HTTP/1.0
option log-health-checks
balance roundrobin
server bootstrap bootstrap.ocp4.example.com:6443 verify none check check-ssl inter 10s fall 2 rise 3 backup (2)
server master0 master0.ocp4.example.com:6443 weight 1 verify none check check-ssl inter 10s fall 2 rise 3
server master1 master1.ocp4.example.com:6443 weight 1 verify none check check-ssl inter 10s fall 2 rise 3
server master2 master2.ocp4.example.com:6443 weight 1 verify none check check-ssl inter 10s fall 2 rise 3
listen machine-config-server-22623 (3)
bind *:22623
mode tcp
server bootstrap bootstrap.ocp4.example.com:22623 check inter 1s backup (2)
server master0 master0.ocp4.example.com:22623 check inter 1s
server master1 master1.ocp4.example.com:22623 check inter 1s
server master2 master2.ocp4.example.com:22623 check inter 1s
listen ingress-router-443 (4)
bind *:443
mode tcp
balance source
server worker0 worker0.ocp4.example.com:443 check inter 1s
server worker1 worker1.ocp4.example.com:443 check inter 1s
listen ingress-router-80 (5)
bind *:80
mode tcp
balance source
server worker0 worker0.ocp4.example.com:80 check inter 1s
server worker1 worker1.ocp4.example.com:80 check inter 1s| 1 | Port 6443 handles the Kubernetes API traffic and points to the control plane machines. |
||
| 2 | The bootstrap entries must be in place before the OKD cluster installation and they must be removed after the bootstrap process is complete. | ||
| 3 | Port 22623 handles the machine config server traffic and points to the control plane machines. |
||
| 4 | Port 443 handles the HTTPS traffic and points to the machines that run the ingress Controller pods. The ingress Controller pods run on the compute machines by default. |
||
| 5 | Port 80 handles the HTTP traffic and points to the machines that run the ingress Controller pods. The ingress Controller pods run on the compute machines by default.
|
|
If you are using HAProxy as a load balancer, you can check that the |