High Availability Concepts

High Availability Concepts

High availability (HA) means designing systems so that a single component failure does not cause the entire service to become unavailable. HA is achieved through redundancy: no single point of failure. Understanding HA concepts is prerequisite to designing reliable infrastructure — you need to know what failure modes exist, what the recovery options are, and what the trade-offs are between availability, consistency, and cost.

HA terminology

Common HA patterns

Active-Passive (simple, low cost):
  Primary ──→ serves traffic
  Standby ──→ idle, ready to take over
  Failover time: 30s-5min (VIP/DNS change)
  Cost: 2x infrastructure

Active-Active (complex, better performance):
  Node 1 ──→ serves requests
  Node 2 ──→ serves requests simultaneously
  Failover time: seconds (load balancer removes failed node)
  Challenge: data consistency between nodes

  Examples: Nginx/Apache (stateless web servers — easy)
            MySQL Galera Cluster (synchronous multi-master)
            PostgreSQL BDR (bi-directional replication)

Database HA

# MySQL HA with Orchestrator (automated failover):
# Orchestrator monitors replication topology and promotes a replica
# when the primary fails, updating all application connection strings

# MySQL Galera Cluster (active-active, synchronous):
# All nodes can accept writes, changes synchronized before commit
# Requires minimum 3 nodes (quorum)
# Add to mysqld.cnf:
# wsrep_on = ON
# wsrep_cluster_address = gcomm://node1,node2,node3

# PostgreSQL HA with Patroni:
# Patroni uses etcd/Consul/ZooKeeper for distributed consensus
# Automatic failover with <30s downtime
sudo apt install -y patroni
# Configuration in /etc/patroni/config.yml

# Simple active-passive with keepalived (VIP failover):
sudo apt install -y keepalived
sudo nano /etc/keepalived/keepalived.conf

keepalived.conf for primary node

vrrp_instance VI_1 {
    state MASTER
    interface eth0
    virtual_router_id 51
    priority 100             # Higher priority = preferred master
    advert_int 1
    authentication {
        auth_type PASS
        auth_pass secretpass
    }
    virtual_ipaddress {
        192.168.1.100        # Virtual IP that follows the active node
    }
}

Measuring availability

# Monitor uptime continuously:
sudo apt install -y nagios4    # OR use Prometheus + Alertmanager

# Simple uptime check with systemd watchdog in a service unit:
# WatchdogSec=30s
# Restart=on-failure
# RestartSec=5s

Conclusion

Start by identifying every SPOF in your architecture: single database server, single web server, single network switch. Each one is a component whose failure causes an outage. Fix the most impactful SPOFs first based on your RTO and RPO requirements. For databases, replication + automated failover (Orchestrator for MySQL, Patroni for PostgreSQL) is the practical path to 99.9%+ availability. For stateless web servers, a load balancer with health checks provides automatic failover in seconds.