PCE High Availability and Disaster Recovery Requirements
Learn how to make sure that your underlying systems are sufficient to provide high availability (HA) and disaster recovery (DR) features. Check all of the following system requirements.
PCE Cluster Front End Load Balancing
In order for a PCE cluster to provide high availability, it requires a front-end load balancer to manage traffic distribution and system health checking for the PCE.
The load balancer must be customer-provided and managed and is not included as part of the PCE software distribution. You have the option of using a traffic load balancer or DNS load balancer.
Important
The load balancer must be able to run application level health checks on each of the core nodes in the PCE cluster so it can be aware at all times whether each node is available to service requests.
Traffic Load Balancer Requirements
The PCE requires the following traffic load balancer configuration.
Layer 4 with Secure Network Address Translation (SNAT)
Least connection (recommended) or round robin load balancing to core nodes
HTTP health checks from load balancer to core nodes
High availability capabilities
A virtual IP (VIP) configured in the
runtime_env.yml
parametercluster_public_ips
Note
Using a traffic load balancer is recommended over DNS, because it provides a quicker failure response, while DNS load balancing typically has a longer failover time.
DNS Load Balancing
Another option for load balancing the PCE cluster is using DNS where traffic is load balanced to the core nodes based on DNS rather than connection-based load balancing.
When you plan to use DNS for load balancing the PCE software, the PCE requires the following DNS load balancer configuration:
Round robin load balancing to core nodes
30 to 60 second TTL to allow for quick failover
PCE core node IP addresses configured in the
runtime_env.yml
parameter namedcluster_public_ips
HTTP health checks from the load balancer to core nodes
The DNS must be able to run health checks against the PCE
node_available
API, and the DNS load balancer should only serve IP addresses for the cluster FQDN of those nodes that respond to thenode_available
API.
Network Latency Between Nodes
Important
Make sure that network latency between and among the nodes of the clusters does not exceed 10ms.
Proper operation of Illumination and Explorer is assured when latency is 10ms or less.