This is what Claude tells me:

global
    log /dev/log local0
    maxconn 4096

defaults
    log     global
    mode    http
    option  httplog
    timeout connect 5000
    timeout client  50000
    timeout server  50000

frontend http_front
    bind *:80
    
    # Generate X-Request-ID if not present
    http-request set-header X-Request-ID %[rand(0,1000000000)] if !{ req.hdr(X-Request-ID) -m found }
    
    # Use consistent hashing based on X-Request-ID
    default_backend http_back

backend http_back
    balance consistent
    hash-type consistent sdbm
    
    # Use X-Request-ID for hash key
    hash-header X-Request-ID
    
    server server1 server1.example.com:80 check
    server server2 server2.example.com:80 check
    server server3 server3.example.com:80 check
    
    # Preserve the X-Request-ID header
    http-response set-header X-Request-ID %[req.hdr(X-Request-ID)]

And, here's how Claude explains it:

This configuration:

1. Generates a random X-Request-ID if one isn't present in the incoming request
2. Uses consistent hashing based on the X-Request-ID header
3. Ensures requests with the same X-Request-ID always go to the same backend server
4. Preserves the X-Request-ID header in the response

Key features:

• Uses hash-type consistent to minimize redistribution when servers are added/removed
• sdbm hash algorithm provides good distribution
• Automatically generates IDs for requests that don't have them
• Maintains session stickiness based on the request ID

This is particularly useful when you need:

• Consistent routing for distributed tracing
• Debug capability across services
• Load balancing while maintaining request affinity
• Correlation of requests across multiple services

balance hdr(X-Request-ID)
hash-type consistent