Part 192: High Availability Topologies: Galera, Patroni & Failover

Achieving 99.999% uptime with automated failover and zero split-brain scenarios.

1. The Split-Brain Dilemma

If the network link between two database nodes fails: - Node 2 might assume Node 1 died and promote itself to Primary. - Clients write to both nodes independently, causing irrecoverable data divergence! - Solution: Always maintain an odd number of nodes (3, 5, 7) using Raft or Paxos consensus.

2. PostgreSQL High Availability with Patroni & etcd

Patroni orchestrates PostgreSQL nodes using etcd distributed consensus: - Leader Key: Stored in etcd with a TTL heartbeat. - If the Primary node misses heartbeats, the cluster automatically elects the replica with the lowest replication lag as the new Primary within 10 seconds.

3. MySQL Galera Cluster (Synchronous Multi-Primary)

Galera replicates changes using certification-based replication: - Any node can accept writes. - Transactions are certified across all cluster nodes before commit.

Key Takeaways

• Always deploy HA clusters across 3 distinct Availability Zones with automated health-check routing.

In Part 193, we explore Zero-Downtime Schema Migrations: The Expand & Contract Pattern!