nginx go around 429

nginx.conf

http {
    upstream azure_ai_foundry {
        zone azure_ai 64k;
        
        server foundry-endpoint-1.azure.com:443 max_fails=3 fail_timeout=60s;
        server foundry-endpoint-2.azure.com:443 max_fails=3 fail_timeout=60s;
        server foundry-endpoint-3.azure.com:443 max_fails=3 fail_timeout=60s;
        
        least_conn;
        keepalive 32;
        keepalive_requests 100;
        keepalive_timeout 60s;
    }

    server {
        listen 80;
        server_name your-domain.com;

        location / {
            proxy_pass https://azure_ai_foundry;
            proxy_http_version 1.1;
            
            # Preserve headers
            proxy_set_header Host $host;
            proxy_set_header X-Real-IP $remote_addr;
            proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
            proxy_set_header X-Forwarded-Proto $scheme;
            proxy_set_header Connection "";
            
            # SSL settings
            proxy_ssl_server_name on;
            proxy_ssl_protocols TLSv1.2 TLSv1.3;
            
            # Streaming configuration
            proxy_buffering off;                    # Disable response buffering
            proxy_cache off;                        # Disable caching
            proxy_read_timeout 300s;                # Extended timeout for long streams
            proxy_send_timeout 300s;
            chunked_transfer_encoding on;           # Enable chunked encoding
            
            # Disable X-Accel-Buffering for streaming
            proxy_set_header X-Accel-Buffering no;
            
            # Disable gzip for streaming responses
            gzip off;
            
            # Handle 429 and other errors
            # First attempt: Nginx sends the request to an Azure AI Foundry endpoint
            # If 429 received: Due to proxy_next_upstream http_429, nginx immediately tries
            # the second endpoint
            # Second attempt fails: Nginx tries the third endpoint
            # (still within proxy_next_upstream_tries 3)
            # If 90 seconds elapsed: Even if tries remain, nginx stops due to
            # proxy_next_upstream_timeout 90s and returns the error to the client
            proxy_next_upstream error timeout http_429 http_502 http_503 http_504;
            proxy_next_upstream_tries 3;
            proxy_next_upstream_timeout 90s;
            
            # Custom error handling
            proxy_intercept_errors on;
            error_page 429 = @rate_limit_error;
        }
        
        location @rate_limit_error {
            default_type application/json;
            return 429 '{"error": "Rate limit exceeded on all endpoints", "status": 429}';
        }
    }
}