Skip to content

High availability options for rConfig V8

A highly available rConfig is an active/passive pair: one node does all the work, a second identical node stands by, and every component behind them (database, Redis or Valkey, storage) is made redundant in its own way. This page sets out who supports which HA deployments, why rConfig is always active/passive, and the options for each component. The same design applies to rConfig Pro, Enterprise, the Vector Server, Vector Central Manager and Vector Prism.

Read this page when one rConfig server going down would cost you more than the time it takes to restore it from backup. If a restore within your recovery targets is acceptable, a single server with good backups is simpler and is the right answer.

  • An agreed Recovery Time Objective (RTO) and Recovery Point Objective (RPO) for rConfig.
  • A second server sized identically to the first.
  • Infrastructure for a virtual IP address (VIP), or a load balancer that can send all traffic to a single node.
  • For Enterprise and Vector customers: contact rConfig support before you design the deployment.

High availability is documented for every edition. What differs is who builds it and who supports the HA layer.

Edition Who designs and builds the HA deployment What rConfig supports
Enterprise and Vector Your team, or rConfig with you through Professional Services The application and the HA design and its operation, directly
Pro Your team rConfig on any HA design you build. The HA layer itself (replication, VIP, storage, failover tooling) is yours to run
Core Your team As for Core today. The patterns on this page and in the High Availability guides still apply

rConfig is not a typical web application that can run on several nodes behind a load balancer. Most of its work is outbound: it opens SSH, Telnet, API and TL1 sessions to devices, on schedules and on demand. The nodes don’t coordinate that outbound work with each other. Two active nodes would both run the scheduler and both run queue workers, so every scheduled backup, compliance check and snippet would run twice, against the same devices.

So in every rConfig HA design:

  • Exactly one node is active. It serves the web interface and API, runs the scheduler and runs the Horizon queue workers.
  • The other node is passive. Its web server, scheduler and queue workers are stopped. It holds the same release, the same .env and the same keys, ready to take over.
  • There is no active/active option, and no load balancing across nodes. A load balancer in front of rConfig must send all traffic to one node, using health checks only to choose which.

The same applies to the Vector Server, Central Manager and Prism: each is an active/passive pair.


Layer What lives there How it’s made highly available
Traffic steering The address users, integrations and Vector agents connect to A VIP (keepalived/VRRP or Pacemaker), or a load balancer that sends all traffic to the active node
Application nodes Web interface, API, scheduler, Horizon workers Active/passive pair with identical releases, .env and keys
Database Devices, tasks, policies, users, compliance history, audit trail Replication with failover behind a VIP or proxy (see below)
Redis or Valkey The job queue, cache, user sessions and locks Replication with Sentinel, behind a VIP
Storage The storage directory: downloaded configs, compliance results, logs, uploads Shared storage, or replication from the active node to the passive node
Keys and .env APP_KEY, which encrypts device credentials and, optionally, configs Identical on both nodes, kept in step
RabbitMQ (Vector Central Manager only) Messages between the Central Manager and Vector Servers Optional RabbitMQ clustering; messages are short-lived

rConfig itself connects to one database host, one Redis host and one storage path, named in .env. Failover happens underneath: the VIP or proxy moves to the new primary, and rConfig keeps using the same address.


Virtual IP with keepalived (recommended). Both nodes run keepalived, and the active node holds the VIP. If the active node or its health check fails, the passive node takes the VIP. Health check scripts should test the application (the web server responds, Horizon is running), not only that the host is up. Both nodes need a shared Layer 2 segment, and the firewall must allow VRRP.

Corosync/Pacemaker. Adds service ordering (start the database before rConfig, mount storage before Horizon) and fencing (STONITH) to prevent split-brain. It suits teams that already run Pacemaker.

Load balancers, including cloud load balancers. Allowed only in active/passive mode: configure the passive node as a backup member that receives traffic only when the active node fails its health check. Never configure round-robin or least-connections across both nodes.


rConfig connects to one database endpoint. Point DB_HOST at a VIP or proxy that always reaches the current primary.

Database Common HA designs Notes
PostgreSQL (Enterprise) Patroni with etcd, or repmgr, with a VIP (for example vip-manager) or HAProxy in front Synchronous replication avoids losing committed writes on failover, at some cost to write speed
MariaDB Primary/replica with automatic failover (MaxScale), or a three-node Galera cluster with writes sent to one node (MaxScale, ProxySQL or HAProxy) Send all writes to a single Galera node to avoid certification conflicts
MySQL InnoDB Cluster (Group Replication) with MySQL Router, or primary/replica with orchestrated failover Single-primary mode
Managed cloud databases Amazon RDS or Aurora, Azure Database, Google Cloud SQL, with multi-zone failover The provider moves the endpoint; rConfig keeps the same DB_HOST

Whatever you choose:

  • Size the replica like the primary. After failover it carries the full load.
  • Watch replication lag. With asynchronous replication, lag is the data you lose on failover.
  • Keep independent backups. Replication copies mistakes and corruption to every replica.

Redis (or Valkey, which replaces it on EL10) holds the job queue, the cache and user sessions. rConfig connects to one host, so make that host a VIP that always points at the current primary:

  1. Run Redis or Valkey on at least two nodes with replication, and Sentinel on at least three nodes (Sentinel needs a majority to promote a replica).
  2. Put a VIP in front of the primary. Move it when Sentinel promotes a new primary, using keepalived with a check that the local instance reports role:master, or Sentinel’s client-reconfig-script.
  3. Set REDIS_HOST in rConfig’s .env to the VIP, and use the same password on every Redis or Valkey node. See Redis shared with other nodes.
  4. Set maxmemory with noeviction on every node. See Redis memory optimization.

Jobs queued but not yet replicated when the primary fails can be lost. Scheduled tasks run again at their next interval; re-run any manual download that was in flight.


The /var/www/html/rconfig8/current/storage directory holds downloaded configurations, compliance results, system backups, logs and uploads. The passive node needs the same data when it takes over.

Option How it works Watch out for
Shared storage (NFS, GlusterFS, Amazon EFS, Azure Files) Both nodes mount the same storage at the same path The storage service must itself be highly available. Needs working file locking and low latency (under 10 ms)
Replication from the active node (for example rsync or lsyncd) The active node copies changes to the passive node continuously Anything not yet copied when the active node fails is lost. Reverse the direction after failover

Object storage such as S3 can’t be used for the storage directory: rConfig needs a normal filesystem.


Both nodes need the same .env, including APP_KEY. APP_KEY encrypts device credentials (and configs, when configuration encryption is on). A passive node with a different key can’t read any credential, so every backup fails after failover. See System Encryption Key.

  • Copy .env from the active node to the passive node when you build the pair, and again after any change to it.
  • Never run key generation on the passive node.
  • After an upgrade, check both nodes run the same release and the same .env.

The .env and keys guide covers keeping the copies in step and checking they match.


The Vector Server, Vector Central Manager and Vector Prism are built on the same platform as rConfig. Each is made highly available the same way: an active/passive pair with a VIP, a replicated database, Redis or Valkey failover, and shared or replicated storage.

RabbitMQ carries messages between the Central Manager and its Vector Servers, and is the one component built differently. RabbitMQ clustering is supported, but messages are short-lived from rConfig’s point of view: stored messages and their audit trail are rarely needed beyond a few days. A RabbitMQ outage delays messages between the Central Manager and its servers; device configurations are stored by each Vector Server, not in RabbitMQ. A single RabbitMQ node with good monitoring and a tested rebuild procedure is a reasonable alternative to a cluster. See Message Queuing with Vector CM.

Vector agents connect to their Vector Server’s address. Point agents at the VIP, so they reconnect to the new active node after failover.


Several independent rConfig servers, each owning part of the inventory (by region, customer or device type), limit how much of the estate one outage affects. They are not failover: if one server fails, its devices aren’t backed up until it’s restored. Each server can itself be an active/passive pair. For distributed collection under one view, see Vector.


  1. Agree the targets. Record the RTO and RPO for rConfig with the business.
  2. Choose each component’s design from the tables above, so that no component’s failover is slower or loses more than the targets allow.
  3. Build both nodes identically: same OS, same rConfig release, same .env, same keys.
  4. Wire failover: VIP or load balancer, service start and stop on the active node only, database and Redis or Valkey endpoints behind VIPs or proxies.
  5. Monitor every layer: VIP ownership, web and Horizon on the active node, replication lag, storage mounts or sync, Sentinel and RabbitMQ health. See System Monitoring.
  6. Test failover before go-live and on a schedule. Measure the real recovery time and data loss against the targets.

The High Availability guides explain the designs in depth, with architecture diagrams for each layer:

Uplink guideHigh Availability guidesHow a highly available rConfig fits together: one active node, one passive node, and redundant database, Redis or Valkey, storage and Vector components.Read the guide