Using a load balancer, you can have multiple Graylog servers ingesting logs and providing additional interfaces for the analysts. You can configure the MongoDB and Elasticsearch databases to be redundant to ensure no data loss. Additionally, all processing pipelines utilize a message journal, allowing quick collection and storage to disk in case of a power loss, so no messages are lost.
Fault tolerance is also built into many of the agents we support via Sidecar. This approach allows hosts to locally spool their logs in case of a network outage, and once the network connection is established, the logs are sent.