HAProxy的TCP负载均衡是否原生支持持久化TCP连接？

Absolutely, HAProxy natively supports persistent TCP connections for both client-to-load balancer and load balancer-to-backend server links—this is a standard use case for long-lived TCP services, so you’re in good hands. Let’s walk through how to set this up properly:

Core Configuration Components #

1. Client-Side Persistent Connections #

To keep clients connected to HAProxy for extended periods, focus on two key settings in your frontend block:

• option tcpka: Enables TCP keepalive packets, which prevent intermediate network devices (like firewalls) from dropping idle connections.
• timeout client: Sets the maximum idle time before HAProxy closes a client connection. Adjust this to match your service’s needs (e.g., 8h for all-day persistent links).

2. Backend-Side Persistent Connections #

HAProxy will maintain long-lived connections to your backend servers as long as you configure these settings in the backend block:

• option tcpka: Applies keepalive logic to backend connections too, ensuring links to your servers stay alive.
• timeout server: Defines how long an idle backend connection stays open. Use a value aligned with your client timeout (e.g., 8h).
• Health checks: Add check to your server definitions to ensure HAProxy only maintains connections to healthy backend instances.

Full Configuration Example #

Here’s a working HAProxy config that implements persistent TCP connections end-to-end:

global
    log /dev/log local0
    maxconn 10000  # Adjust based on your expected concurrent connections
    daemon

defaults
    log global
    mode tcp
    timeout connect 10s  # Timeout for initial backend connection attempts
    timeout client 8h     # Keep client connections alive for 8 hours
    timeout server 8h     # Keep backend connections alive for 8 hours

frontend tcp_service_frontend
    bind *:8000  # Your public-facing TCP port
    option tcpka  # Enable keepalive for client connections
    default_backend tcp_service_backend

backend tcp_service_backend
    option tcpka  # Enable keepalive for backend connections
    balance roundrobin  # Use your preferred balancing algorithm (e.g., source, leastconn)
    # Define backend servers with health checks
    server backend1 192.168.1.10:9000 check inter 5s rise 2 fall 3
    server backend2 192.168.1.11:9000 check inter 5s rise 2 fall 3

Optional: Session Sticky (If Needed) #

If you need the same client to always use the same backend server (common for stateful services), add session sticking logic to the backend:

backend tcp_service_backend
    option tcpka
    balance roundrobin
    # Create a stick table to track client IPs for 8 hours
    stick-table type ip size 100k expire 8h
    stick on src  # Bind clients to backends based on their source IP
    server backend1 192.168.1.10:9000 check
    server backend2 192.168.1.11:9000 check

Key Notes #

• Ensure your backend servers are configured to accept and maintain long-lived TCP connections (avoid auto-closing idle connections prematurely).
• Adjust maxconn in the global section to handle your expected number of concurrent persistent connections.
• If you’re working with strict firewalls, tweak the keepalive interval using tcpka-int <seconds> (e.g., tcpka-int 30 to send keepalives every 30 seconds).

内容的提问来源于stack exchange，提问作者ks2bmallik