Skip to main content

HA Deployment

Linkiir uses a true active-passive service-tier design.

Normal state​

  • Active Grid serves users and supervises Runtime and Archiver.
  • Standby Grid monitors the active lease but does not run Runtime or Archiver.
  • The queue cluster operates independently.

Failover​

When the active server fails, the standby acquires the lease, starts Runtime and Archiver, and becomes healthy for the load balancer. The Archiver resumes from committed queue offsets; Kafka/Redpanda retains records during the gap.

Requirements​

  • Avoid split brain with a reliable lease or fencing mechanism.
  • Share or replicate the Linkiir working directory.
  • Use the same master encryption key on both nodes.
  • Use PostgreSQL or MS SQL, not SQLite, for shared production logging.
  • Test failover, failback, and recovery under load.
  • Ensure broker replication is independently resilient.