HAProxy TCP Load Balancing for Databases

TCP Load Balancing for Database Workloads

While HAProxy is best known for HTTP load balancing, its Layer 4 (TCP) capabilities make it an excellent database proxy. TCP mode enables read/write splitting for PostgreSQL, Redis Sentinel routing, and MySQL cluster load balancing — all without application changes.

PostgreSQL Read/Write Splitting

Route writes to the primary and reads to replicas using HAProxy's TCP health checks:

listen postgres_write
    bind *:5432
    mode tcp
    option pgsql-check user haproxy
    default-server inter 3s fall 3 rise 2
    server pg-primary 10.0.1.10:5432 check

listen postgres_read
    bind *:5433
    mode tcp
    balance leastconn
    option pgsql-check user haproxy
    default-server inter 3s fall 3 rise 2
    server pg-replica1 10.0.1.11:5432 check
    server pg-replica2 10.0.1.12:5432 check
    server pg-primary 10.0.1.10:5432 check backup

The option pgsql-check directive performs actual PostgreSQL protocol health checks, detecting not just TCP connectivity but database readiness.

Redis Sentinel Routing

Route Redis traffic to the current master using an external check script:

listen redis_master
    bind *:6379
    mode tcp
    balance first
    option tcp-check
    tcp-check connect
    tcp-check send "PING\r\n"
    tcp-check expect string +PONG
    tcp-check send "INFO replication\r\n"
    tcp-check expect string role:master
    tcp-check send "QUIT\r\n"
    tcp-check expect string +OK
    server redis1 10.0.1.20:6379 check inter 2s
    server redis2 10.0.1.21:6379 check inter 2s
    server redis3 10.0.1.22:6379 check inter 2s

This configuration actively checks each Redis node's replication role and only routes to the current master.

MySQL Group Replication

listen mysql_cluster
    bind *:3306
    mode tcp
    balance roundrobin
    option mysql-check user haproxy
    server mysql1 10.0.1.30:3306 check
    server mysql2 10.0.1.31:3306 check
    server mysql3 10.0.1.32:3306 check

Connection Limits and Timeouts

Database connections are long-lived compared to HTTP. Tune HAProxy accordingly:

defaults
    mode tcp
    timeout connect 5s
    timeout client 30m
    timeout server 30m
    timeout check 5s

listen postgres_write
    bind *:5432
    maxconn 200
    default-server maxconn 100

The 30-minute client/server timeouts accommodate long-running queries, while maxconn prevents connection exhaustion on database backends.

Monitoring Database Backends

Key metrics to track via the HAProxy stats page:

• Current sessions — Monitor connection pool utilization
• Backend health — Detect replica lag or failures
• Connection queue — Identify when backends are saturated
• Response time — Track TCP connection establishment time

Eazy SaaS Tip: We combine HAProxy TCP load balancing with PgBouncer connection pooling for PostgreSQL deployments. This dual-layer approach provides both high availability and efficient connection management for applications with thousands of concurrent users.