Skip to content
Guides & References

High availability for Vector Server, Central Manager and Prism

Make each Vector component highly available as an active/passive pair, cluster RabbitMQ only if you need to, and keep agents connected through failover.

High AvailabilityUpdated
On this page

High availability for Vector Server, Central Manager and Prism

The rConfig V8 Vector Server, Vector Central Manager and Vector Prism are built on the same platform as rConfig itself. Each one is made highly available the same way: an active/passive pair behind a VIP, with a replicated database, Redis or Valkey failover, and shared or replicated storage. This page covers what’s specific to Vector: which components matter most, RabbitMQ between the Central Manager and its servers, and what agents do during a failover.

Figure HA-9: a Vector Central Manager active/passive pair linked to RabbitMQ, two regional rConfig V8 Vector Server pairs each with its own VIP and data tier, a Vector Prism pair with its own VIP, and Vector agents connecting outbound to each server's VIP
Figure HA-9: Vector high availability. Each component is its own active/passive pair, and RabbitMQ is optional to cluster. Click the diagram to open the interactive version.

Each component is its own active/passive pair

Component What it does If it’s down HA design
Vector Server Stores and processes configs collected by its agents; runs schedules and compliance Its agents can’t deliver configs and no schedules run for its devices Active/passive pair
Vector Central Manager Orchestrates inventory across several Vector Servers Changes made centrally don’t reach the servers; each server keeps collecting on its own Active/passive pair, plus RabbitMQ (below)
Vector Prism Read-only customer portal over a Vector deployment Customers can’t see their configs; collection is unaffected Active/passive pair, or a tested rebuild if the portal’s RTO allows
Vector agents Collect configs close to the devices That site’s devices aren’t collected A replacement agent, or a second agent where the site justifies it

The one-active-node rule applies to every one of them, for the same reason as rConfig: they do scheduled, outbound work that the nodes don’t share. See why one node is always passive.

Put the effort where an outage costs most. The Vector Server is usually the priority, because it’s where collection stops. The Central Manager and Prism can often accept a longer RTO.


RabbitMQ is different

RabbitMQ carries messages between the Central Manager and its Vector Servers. It’s the one component in a Vector deployment that isn’t built on the rConfig platform, and it behaves differently:

  • It holds little that matters for long. Messages are processed shortly after they arrive. Stored messages and their audit trail are rarely needed beyond a few days.
  • It doesn’t hold configurations. Each Vector Server stores its own devices’ configs. A RabbitMQ outage delays messages between the Central Manager and its servers; it doesn’t lose backups.
  • Clustering is supported, for deployments whose targets need RabbitMQ to keep running through a node failure.

So there are two sound choices:

Choice When it fits
A RabbitMQ cluster (three nodes) The Central Manager’s RTO is short, and messages must keep flowing through a node failure
A single RabbitMQ node, monitored, with a tested rebuild A delay of a rebuild’s length in central changes is acceptable

Either way, keep message retention short. There’s no need to keep messages for weeks for rConfig’s sake. See Message Queuing with Vector CM.


Agents during a failover

Agents connect to their Vector Server’s address. Make that address the server pair’s VIP:

  • During failover, agents lose their connection and reconnect when the VIP moves to the passive node.
  • Schedules that fell inside the outage run at their next interval.
  • Never point agents at one node’s own address. After failover they’d keep trying the failed node.

For agents connecting over the internet, the VIP (or the load balancer in front of the pair) is the address published in DNS and in the agents’ configuration.


HA across the Vector architectures

Each Vector architecture adds HA in the same places:

Architecture Make highly available first Then
Single server The Vector Server Agents at the most important customer sites
Multi-country Each country’s Vector Server The Central Manager and RabbitMQ
Access nodes The Vector Server The shared access nodes, which carry many customers’ agents
Prism The Vector Server Prism, if customers rely on it continuously
ISP PoP The Vector Server in the NOC Agents in the largest PoPs
Security zones The Vector Server Agents in zones where collection gaps matter most (PCI, OT)

Type to search.