Skip to content

High Availability

Overview

High availability (HA) in LibreNMS keeps the operation continuous and makes the downtime as short as possible. HA gives redundancy to two critical components:

  • Polling: the data collection process
  • WebUI: the web interface for the users

For high availability, these components must be redundant:

  1. Database: MySQL or MariaDB with clustering
  2. Redis with Redis Sentinel: for session management and caching
  3. RRD Files: for the storage of the polled data

The poller must use a distributed setup. This setup must be the LibreNMS Dispatcher service.

For a simpler design, the web interface and the poller can use the same Redis Sentinel cluster.

Note about RRD Files

A poller writes RRD data to files on a disk. We recommend RRDCached, which accepts RRD data over TCP/IP. RRDCached does not give HA for the RRD data. It lets many pollers write to the same RRD files over a network connection. For more information, read RRDCached.md.

Shared storage over NFS with GlusterFS is one way to add HA for RRD.

WebUI High Availability

For HA, many LibreNMS instances share these backend services:

  • Clustered Database
  • Redis with Sentinel
  • Centralized RRD Storage

Implementation

  1. Configure Database HA:
  2. Build a Galera Cluster

  3. For the full instructions, read Galera-Cluster.md

  4. Configure Redis HA:

  5. Install Redis Sentinel

  6. For the configuration details, read Redis-Sentinel.md

  7. Deploy multiple LibreNMS instances:

  8. Install LibreNMS on more than one server
  9. Configure each instance for the same database and the same Redis Sentinel cluster
  10. Make the .env configuration identical on all instances. Set APP_KEY to the same value on all instances.

  11. Give each install a unique NODE_ID in the .env file.

  12. Configure RRD Access: Use RRDCached, which gives all instances access to the same RRD files. You can also use shared storage for the RRD files over NFS.

Polling High Availability

With distributed polling, many pollers work together. This design distributes the load and gives failover capability.

Warning

The poller does not support MySQL Galera clustering. Put a TCP load balancer, such as Nginx or HAProxy, in front of the cluster. The load balancer then points to the cluster nodes.

Implementation

  1. Configure distributed polling:
  2. Obey the instructions in Distributed-Poller.md
  3. Connect all pollers to the clustered database and to Redis Sentinel. All pollers must reach the same RRD files.