Notice: Our URL has changed! Please update your bookmarks. Dismiss

External Load Balancer Integrations with OpenShift Enterprise 3

Load Balancing For HA Applications

Simple Integration (External LB as a Passthrough)

This is the most lightweight integration possible between OpenShift and an external load balancer. The goal here is to satisfy common requirements that application traffic originating outside of an organization go through a DMZ or public network layer before hitting applications behind a firewall.

In order to keep the implementation as simple as possible, we use the load balancer as a simple passthrough to OpenShift’s routing layer. The OpenShift routers then handle things like SSL termination and making decisions on where to send traffic for particular applications.

Required components for a simple integration include:

  • External load balancer

    • Single VIP for OpenShift applications

    • SSL passthrough

    • Round robin load balancing

  • OpenShift router

    • Will support application specific routes

    • Handles SSL via a combination of wildcard and unique certificates

      • Any application can create a route with a unique certificate

      • Applications that want SSL but don’t provide unique certificates will use the wildcard certificate provided for the router

      • Non-Secure HTTP routes are also supported and are the default behavior out of the box

Simple Integration

Hybrid Integration (External LB Termination)

Note
Hybrid Integration is recommended only in the case that security policy or network structure inhibit the ability to implement the Simple Integration above.

This scenario is very similar to the Simple Integration with one simple, but important difference. In this scenario we are offloading SSL termination to the Load Balancer, instead of to OpenShift. The advantage here is that we have the ability to host SSL Certificates and unique public IP addresses at the Load Balancer level, which some organizations prefer. The drawback is that, in order to do that, we now have some extra administration overhead on the Load Balancer in order to onboard each new application. This process can generally be automated without much trouble, but it does significantly increase the level of effort required to integrate.

Note
For organizations that use an F5 BIG-IP as their External Load Balancer, OpenShift provides a built-in plugin for integrating that as OpenShift’s router, thus removing the overhead of building this custom automation.

Required components for the hybrid approach include:

  • External Load Balancer

    • One Default VIP, hosting a wildcard certificate for OpenShift applications

    • One additional VIP per application that requires one of:

      • Unique Certificate

      • Vanity URL that doesn’t match the OpenShift Wildcard domain name

    • Round Robin config Responsible for

      • LB algorithm (sticky, round robin, etc.)

    • OpenShift Router Responsible for

Hybrid Integration

Full Integration (Integrating F5 as the OpenShift Router)

OpenShift provides an out of the box plugin that allows an administrator to configure OpenShift to use an external F5 BIG-IP appliance as its router. This feature works by providing OpenShift with access to the F5 BIG-IP™’s API in order to dynamically configure new unique virtual hosts as new applications come online. While this feature makes for a much more dynamic environment with Enterprise-ready features, it has more stringent requirements than the Simple or Hybrid Integration in order to be successful.

Requirements for a Full Integration include:

  • F5 BIG-IP™ version 11.6 or greater

  • Connectivity and Authentication Access to the F5 BIG-IP™ REST API from OpenShift nodes

    • Specifically the OpenShift nodes running your router pods

    • SSH Access to the BIG-IP for transferring files like certificates and keys

  • Authorization allowances for OpenShift to Control the F5 BIG-IP™ to do things like:

    • Create virtual hosts

    • Delete virtual hosts

    • Add and configure certificates

  • One or more OpenShift Ramp Nodes by which access to the internal SDN is provided to the BIG-IP™

In this way, the OpenShift router pods work as configuration agents for the F5. Rather than handling incoming web traffic, the pods just watch the OpenShift API for new routes to be created and pass that configuration information up to the F5 via its API to configure routes and handle traffic at the load balancer level.

Through the use of tunnels provided by ramp nodes, the F5 BIG-IP™ is given the ability to directly send traffic to OpenShift pods.

Tip
By default, the use of a Ramp Node creates a single point of failure. To avoid this, you must configure a highly available ramp node.
Full Integration

For more information on configuration, view the routing section of the official OpenShift Enterprise docs: https://docs.openshift.com/container-platform/3.7/install_config/router/f5_router.html

Load Balancing For HA Master Infrastructure

OpenShift Enterprise 3.1 introduced a Native HA mode that provides high availability of the Master without the need for Pacemaker/RHEL Clustering. When installed in Native HA, the stateless API component (atomic-openshift-master-api) of the master is split out from the stateful Controller component (atomic-openshift-master-controllers). This is so that the API service can be setup active-active, while the Controller service runs active-passive. Once we have an active-active master API (which includes the Web Console as well), all that’s needed to have a fully HA orchestration infrastructure is a proper load balancer.

Out of the box, the OpenShift has the ability to install an HAProxy instance on a host you designate as a lightweight load balancer between masters in Native HA mode. However, this only creates another single point of failure. It is much preferred to integrate an enterprise load balancer (LB) such as an F5 Big-IP™ or a Citrix Netscaler™ appliance. This integration does add some complexity to the install process. We attempt to explain those options below.

Simple SSL Passthrough (Non-Prod only)

Overview

One option is to configure a VIP on a load balancer as SSL Passthrough. This means that the LB does not terminate SSL, but simply proxies encrypted traffic through to the masters, which then handle termination. This has the advantage of being a fairly simple implementation on the LB side, and a slightly simpler setup process on the OpenShift installation than terminating on the LB. The drawback of this method is that we are presenting a self-signed certificate, so users of the Web Console or API will see untrusted or unknown certificate errors.

Simple Passthrough for OpenShift Masters

Example Configuration

1. Pre-requisites
  • Load Balancer VIP pre-created

    • Configured for SSL Passthrough

    • VIP must listen on port 8443, and proxy back to all master hosts on port 8443

    • WebSockets Enabled

  • Domain Name for VIP registered in DNS

    • Domain name will become value of both openshift_master_cluster_public_hostname and openshift_master_cluster_hostname in OpenShift Installer

    • For this example, we will give our VIP an FQDN: paas.myorg.com

  • Master hosts created and prepped per Install Prerequisites

Warning
The OpenShift Web Console heavily utilizes WebSockets for many of the rich interface features like tailing pod logs, watching a deployment, or getting a terminal of a pod. While most enterprise load balancers support Web Sockets, they often need to be explicitly enabled for a particular VIP/Profile/etc.
Warning
The OpenShift Web Console heavily utilizes WebSockets for many of the rich interface features like tailing pod logs, watching a deployment, or getting a terminal of a pod. While most enterprise load balancers support Web Sockets, they often need to be explicitly enabled for a particular VIP/Profile/etc.

For the purposes of this example, we will use the hosts specified in the table below

Host Name Infrastructure Component

paas.myorg.com

Pre-Configured LB VIP

master01.myorg.com

Master (clustered using native HA) and node

master02.myorg.com

master03.myorg.com

etcd01.myorg.com

etcd Data Store

etcd02.myorg.com

etcd03.myorg.com

node01.myorg.com

Node

node02.myorg.com

2. Setup the Installer

To set up the install, we need to create a host inventory file for the ansible-based installer.

# Create an OSEv3 group that contains the master, nodes, etcd, and lb groups.
# The lb group lets Ansible configure HAProxy as the load balancing solution.
# Comment lb out if your load balancer is pre-configured.
[OSEv3:children]
masters
nodes
etcd
# Since we are providing a pre-configured LB VIP, no need for this group
#lb

# Set variables common for all OSEv3 hosts
[OSEv3:vars]
ansible_ssh_user=root
deployment_type=openshift-enterprise

# Uncomment the following to enable htpasswd authentication; defaults to
# DenyAllPasswordIdentityProvider.
openshift_master_identity_providers=[{'name': 'htpasswd_auth', 'login': 'true', 'challenge': 'true', 'kind': 'HTPasswdPasswordIdentityProvider', 'filename': '/etc/origin/htpasswd'}]

# Native HA with External LB VIPs
openshift_master_cluster_method=native
openshift_master_cluster_hostname=paas.myorg.com
openshift_master_cluster_public_hostname=paas.myorg.com

# host group for masters
[masters]
master01.myorg.com
master02.myorg.com
master03.myorg.com

# host group for etcd
[etcd]
etcd01.myorg.com
etcd02.myorg.com
etcd03.myorg.com

# Since we are providing a pre-configured LB VIP, no need for this group
#[lb]
#lb.example.com

# host group for nodes, includes region info
[nodes]
master[01:03].myorg.com openshift_node_labels="{'region': 'infra', 'zone': 'default'}"
node01.myorg.com openshift_node_labels="{'region': 'primary', 'zone': 'east'}"
node02.myorg.com openshift_node_labels="{'region': 'primary', 'zone': 'west'}"
3. Run the Advanced Installer

Run the installer per the instructions in the Advanced Insallation Guide

Custom Certificate SSL Termination (Production)

Overview

The other option is to have the LB terminate SSL connections for incoming . The LB would need to re-encrpyt and send traffic on to the masters. The advantage here is that we now have the ability to make the Web Console & API available externally using a publicly signed certificate, or one signed by your organization’s PKI, making for a better user experience.

Production Ready Load Balancing for OpenShift Masters

Example Configuration

1. Pre-requisites
  • Two Load Balancer VIPs Pre-created

    • External VIP

      • Configured for SSL Termination, using either:

        • A certificate signed by a Public Certificate Authority

        • A trusted certificate from your org’s PKI

      • VIP must listen on port 8443, and proxy back to all master hosts on port 8443

      • For this example, we will give our External VIP a domain name: paas.myorg.com

      • WebSockets Enabled

    • Internal VIP (will be the same configuration as the Simple integration option above)

      • Configured for SSL Passthrough (see step 1 below)

      • VIP must listen on port 8443, and proxy back to all master hosts on port 8443

      • For this example, we will give our Internal VIP a domain name: paas-internal.myorg.com

  • FQDN for each VIP registered in DNS

    • FQDN for external VIP will become value of openshift_master_cluster_public_hostname in OpenShift Installer (i.e. paas.myorg.com)

    • FQDN for internal VIP will become value of openshift_master_cluster_hostname in OpenShift Installer (i.e. paas-internal.myorg.com)

  • Master hosts created and prepped per Install Prerequisites

Warning
The OpenShift Web Console heavily utilizes WebSockets for many of the rich interface features like tailing pod logs, watching a deployment, or getting a terminal of a pod. While most enterprise load balancers support Web Sockets, they often need to be explicitly enabled for a particular VIP/Profile/etc.
Warning
The OpenShift Web Console heavily utilizes WebSockets for many of the rich interface features like tailing pod logs, watching a deployment, or getting a terminal of a pod. While most enterprise load balancers support Web Sockets, they often need to be explicitly enabled for a particular VIP/Profile/etc.

For the purposes of this example, we will use the hosts specified in the table below

Host Name Infrastructure Component

paas.myorg.com

Pre-Configured External LB VIP with Public SSL Cert

paas-internal.myorg.com

Pre-Configured Internal LB VIP configured for SSL Passthrough

master01.myorg.com

Master (clustered using native HA) and node

master02.myorg.com

master03.myorg.com

etcd01.myorg.com

etcd Data Store

etcd02.myorg.com

etcd03.myorg.com

node01.myorg.com

Node

node02.myorg.com

2. Setup the Installer

To set up the install, we need to create a host inventory file for the ansible-based installer.

# Create an OSEv3 group that contains the master, nodes, etcd, and lb groups.
# The lb group lets Ansible configure HAProxy as the load balancing solution.
# Comment lb out if your load balancer is pre-configured.
[OSEv3:children]
masters
nodes
etcd
# Since we are providing pre-configured LB VIPs, no need for this group
#lb

# Set variables common for all OSEv3 hosts
[OSEv3:vars]
ansible_ssh_user=root
deployment_type=openshift-enterprise

# Uncomment the following to enable htpasswd authentication; defaults to
# DenyAllPasswordIdentityProvider.
openshift_master_identity_providers=[{'name': 'htpasswd_auth', 'login': 'true', 'challenge': 'true', 'kind': 'HTPasswdPasswordIdentityProvider', 'filename': '/etc/origin/htpasswd'}]

# Native HA with an Internal & External LB VIPs
openshift_master_cluster_method=native
openshift_master_cluster_hostname=paas-internal.myorg.com
openshift_master_cluster_public_hostname=paas.myorg.com

# host group for masters
[masters]
master01.myorg.com
master02.myorg.com
master03.myorg.com

# host group for etcd
[etcd]
etcd01.myorg.com
etcd02.myorg.com
etcd03.myorg.com

# Since we are providing pre-configured LB VIPs, no need for this group
#[lb]
#lb.example.com

# host group for nodes, includes region info
[nodes]
master[01:03].myorg.com openshift_node_labels="{'region': 'infra', 'zone': 'default'}"
node01.myorg.com openshift_node_labels="{'region': 'primary', 'zone': 'east'}"
node02.myorg.com openshift_node_labels="{'region': 'primary', 'zone': 'west'}"
3. Run the Advanced Installer

-Run the installer per the instructions in the Advanced Installation Guide

Guides for Specific Load Balancer Implementations

1. Load Balancing in Amazon Web Services

When deploying your HA OpenShift cluster into AWS, there are a couple of things to consider. Amazon provides its own Elastic Load Balancer objects you can associate with a group of instances. They work by providing an enterprise-level load balancer that sits outside of your instances. In this case, it would load balance across the OSE masters. Theoretically, this should work seamlessly but in reality there are issues with how AWS handles load balancing.

  1. Create ELB and associate it with your instances. Follow the directions on Amazon to define and register your instances with the load balancer.

    http://docs.aws.amazon.com/ElasticLoadBalancing/latest/DeveloperGuide/elb-getting-started.html

    In a typical setup, you would configure an ELB with a public-facing interface. This generates a public DNS name that clients can see to send requests.

  2. Configure SSL/TLS termination at the ELB

    Because communication to the master and API takes place over secured port 8443, the ELB would need to be aware of this.

    Note
    Because of the way ELB distributes loads, there is no guarentee of session persistance between masters. Every single web request can be re-directed to a different OSE master, completely destroying any stateful sessions. Because of this, you will need to configure TLS/SSL termination in the ELB

    The ELB needs to be aware of the of the SSL certificate in use by the OSE master in order to inspect the HTTP session for session persistance. Follow the instructions on http://docs.aws.amazon.com/ElasticLoadBalancing/latest/DeveloperGuide/elb-update-ssl-cert.html to upload your OSE master certificate into AWS.

  3. Configure ELB for "sticky-sessions"

    We will need to configure AWS ELB to persist HTTPS sessions via the "ssn" cookie. See SessionName under master-config.yaml https://docs.openshift.com/enterprise/3.2/install_config/configuring_authentication.html

    Follow the "Application-Controlled Session Stickiness" procedure in http://docs.aws.amazon.com/ElasticLoadBalancing/latest/DeveloperGuide/elb-sticky-sessions.html to set your ELB to be aware of the OpenShift session.