NGINX Load Balancing – TCP and UDP Load Balancer

This chapter describes how to use NGINX Plus and open source NGINX to proxy and load balance TCP and UDP traffic.

Introduction

Load balancing refers to efficiently distributing network traffic across multiple backend servers.

In Release 5 and later, NGINX Plus can proxy and load balance Transmission Control Protocol) (TCP) traffic. TCP is the protocol for many popular applications and services, such as LDAP, MySQL, and RTMP.

In Release 9 and later, NGINX Plus can proxy and load balance UDP traffic. UDP (User Datagram Protocol) is the protocol for many popular nontransactional applications, such as DNS, syslog, and RADIUS.

To load balance HTTP traffic, refer to the HTTP Load Balancing article.

Prerequisites

• Latest open source NGINX built with the --with-stream configuration flag, or latest NGINX Plus (no extra build steps required)
• An application, database, or service that communicates over TCP or UDP
• Upstream servers, each running the same instance of the application, database, or service

Configuring Reverse Proxy

First, you will need to configure reverse proxy so that NGINX can forward TCP connections or UDP datagrams from clients to an upstream group or a proxied server.

Open the NGINX configuration file and perform the following steps:

1. Create a top‑level stream {} block:

   stream {
       # ...
   }
   

2. Define one or more server {} configuration blocks for each virtual server in the top‑level stream {} context.

3. Within the server {} configuration block for each server, include the listen directive to define the IP address and/or port on which the server listens. For UDP traffic, also include the udp parameter. TCP is the default protocol for the stream context, so there is no tcp parameter to the listen directive:

   stream {
       server {
           listen 12345;
           # ...
       }
   
       server {
           listen 53 udp;
           # ...
       }
       # ...
   }
   

4. Include the proxy_pass directive to define the proxied server or an upstream group to which the server forwards traffic:

   stream {
       server {
           listen     12345;
           #TCP traffic will be forwarded to the "stream_backend" upstream group
           proxy_pass stream_backend;
       }
   
       server {
           listen     12346;
           #TCP traffic will be forwarded to the specified server
           proxy_pass backend.example.com:12346;
       }
   
       server {
           listen     53 udp;
           #UDP traffic will be forwarded to the "dns_servers" upstream group
           proxy_pass dns_servers;
       }
       # ...
   }
   

5. If the proxy server has several network interfaces, you can optionally configure NGINX to use a particular source IP address when connecting to an upstream server. This may be useful if a proxied server behind NGINX is configured to accept connections from particular IP networks or IP address ranges.

   Include the proxy_bind directive and the IP address of the appropriate network interface:

   stream {
       # ...
       server {
           listen     127.0.0.1:12345;
           proxy_pass backend.example.com:12345;
           proxy_bind 127.0.0.1:12345;
       }
   }
   

6. Optionally, you can tune the size of two in‑memory buffers where NGINX can put data from both the client and upstream connections. If there is a small volume of data, the buffers can be reduced which may save memory resources. If there is a large volume of data, the buffer size can be increased to reduce the number of socket read/write operations. As soon as data is received on one connection, NGINX reads it and forwards it over the other connection. The buffers are controlled with the proxy_buffer_sizedirective:

   stream {
       # ...
       server {
           listen            127.0.0.1:12345;
           proxy_pass        backend.example.com:12345;
           proxy_buffer_size 16k;
       }
   }
   

Configuring TCP or UDP Load Balancing

To configure load balancing:

1. Create a group of servers, or an upstream group whose traffic will be load balanced. Define one or more upstream {} configuration blocks in the top‑level stream {}context and set the name for the upstream group, for example, stream_backend for TCP servers and dns_servers for UDP servers:

   stream {
       upstream stream_backend {
           # ...
       }
   
       upstream dns_servers {
           # ...
       }
       # ...
   }
   

   Make sure that the name of the upstream group is referenced by a proxy_pass directive, like those configured above for reverse proxy.

2. Populate the upstream group with upstream servers. Within the upstream {} block, add a server directive for each upstream server, specifying its IP address or hostname (which can resolve to multiple IP addresses) and an obligatory port number. Note that you do not define the protocol for each server, because that is defined for the entire upstream group by the parameter you include on the listen directive in the server block, which you have created earlier.

   stream {
       upstream stream_backend {
           server backend1.example.com:12345;
           server backend2.example.com:12345;
           server backend3.example.com:12346;
           # ...
       }
   
       upstream dns_servers {
           server 192.168.136.130:53;
           server 192.168.136.131:53;
           # ...
       }
       # ...
   }
   

3. Configure the load‑balancing method used by the upstream group. You can specify one of the following methods:

   • Round Robin – By default, NGINX uses the Round Robin algorithm to load balance traffic, directing it sequentially to the servers in the configured upstream group. Because it is the default method, there is no round‑robin directive; simply create an upstream configuration block in the top‑level stream context and add server directives as described in the previous step.

   • Least Connections (least_conn) – NGINX selects the server with the smaller number of current active connections.

   • Least Time (least_time; NGINX Plus only) – NGINX Plus selects the server with the lowest average latency and the least number of active connections. The method used to calculate lowest average latency depends on which of the following parameters is included on the least_time directive:

     • connect – Time to connect to the upstream server
     • first_byte – Time to receive the first byte of data
     • last_byte – Time to receive the full response from the server

     upstream stream_backend {
         least_time first_byte;
         server backend1.example.com:12345;
         server backend2.example.com:12345;
         server backend3.example.com:12346;
     }
     

   • Hash (hash) – NGINX selects the server based on a user‑defined key, for example, the source IP address ($remote_addr):

     upstream stream_backend {
         hash $remote_addr;
         server backend1.example.com:12345;
         server backend2.example.com:12345;
         server backend3.example.com:12346;
     }
     

     The Hash load‑balancing method is also used to configure session persistence. As the hash function is based on client IP address, connections from a given client are always passed to the same server unless the server is down or otherwise unavailable. Specify an optional consistent parameter to apply the ketama consistent hashing method:

     hash $remote_addr consistent;
     

4. Optionally, for each upstream server specify server‑specific parameters including maximum number of connections, server weight, and so on:

   upstream stream_backend {
       hash   $remote_addr consistent;
       server backend1.example.com:12345 weight=5;
       server backend2.example.com:12345;
       server backend3.example.com:12346 max_conns=3;
   }
   
   upstream dns_servers {
       least_conn;
       server 192.168.136.130:53;
       server 192.168.136.131:53;
       # ...
   }
   

An alternative approach is to proxy traffic to a single server instead of an upstream group. If you identify the server by hostname, and configure the hostname to resolve to multiple IP addresses, then NGINX load balances traffic across the IP addresses using the Round Robin algorithm. In this case, you must specify the server’s port number in the proxy_passdirective and must not specify the protocol before IP address or hostname:

stream {
    # ...
    server {
        listen     12345;
        proxy_pass backend.example.com:12345;
    }
}

Passive Health Monitoring

If an attempt to connect to an upstream server times out or results in an error, open source NGINX or NGINX Plus can mark the server as unavailable and stop sending requests to it for a defined amount of time. To define the conditions under which NGINX considers an upstream server unavailable, include the following parameters to the server directive:

• fail_timeout – The amount of time within which a specified number of connection attempts must fail for the server to be considered unavailable. Also, the amount of time that NGINX considers the server unavailable after marking it so.
• max_fails – The number of failed attempts that happen during the specified time for NGINX to consider the server unavailable.

The default values are 10 seconds and 1 attempt. So if a connection attempt times out or fails at least once in a 10‑second period, NGINX marks the server as unavailable for 10 seconds. The example shows how to set these parameters to 2 failures within 30 seconds:

upstream stream_backend {
    server backend1.example.com:12345 weight=5;
    server backend2.example.com:12345 max_fails=2 fail_timeout=30s;
    server backend3.example.com:12346 max_conns=3;
}

Active Health Monitoring

Health checks can be configured to test a wide range of failure types. For example, NGINX Plus can continually test upstream servers for responsiveness and avoid servers that have failed.

NGINX Plus sends special health‑check requests to each upstream server and checks for a response that satisfies certain conditions. If a connection to the server cannot be established, the health check fails, and the server is considered unhealthy. NGINX Plus does not proxy client connections to unhealthy servers. If several health checks are defined for a group of servers, the failure of any one check is enough for the corresponding server be considered unhealthy.

Prerequisites

• You have configured an upstream group of servers in the stream context, for example:

  stream {
      upstream stream_backend {
          server backend1.example.com:12345;
          server backend2.example.com:12345;
          server backend3.example.com:12345;
     }
  }
  

• You have configured a server that passes traffic (in this case, TCP connections) to the server group:

  server {
      listen     12345;
      proxy_pass stream_backend;
  }
  

Basic Configuration

1. Specify a shared memory zone, a special area where the NGINX Plus worker processes share state information about counters and connections. Include the zone directive int the upstream server group and specify the zone name (here, stream_backend) and the amount of memory (64 KB):

   stream {
       upstream stream_backend {
           zone   stream_backend 64k;
           server backend1.example.com:12345;
           server backend2.example.com:12345;
           server backend3.example.com:12345;
       }
   }
   

2. Enable health checks for the servers in the upstream group. Include the health_check and health_check_timeout directives in the server block that proxies connections to the upstream group:

   server {
       listen        12345;
       proxy_pass    stream_backend;
       health_check;
       health_check_timeout 5s;
   }
   

   The health_check directive enables the health‑check functionality, while the health_check_timeout directive overrides the proxy_timeout value for health checks, as for health checks this timeout needs to be significantly shorter.

   To enable health checks for UDP traffic, in the health_check directive specify the udp parameter and optionally the match parameter with the name of a corresponding match block of tests for verifying server responses (see Fine‑Tuning UDP Health Checks):

   server {
       listen       5053;
       proxy_pass   dns_servers;
       health_check udp match=dns;
       health_check_timeout 5s;
   }
   

Fine‑Tuning Health Checks

By default, NGINX Plus tries to connect to each server in an upstream server group every 5 seconds. If the connection cannot be established, NGINX Plus considers the health check failed, marks the server as unhealthy, and stops forwarding client connections to the server.

To change the default behavior, include these parameters to the health_check directive:

• interval – How often (in seconds) NGINX Plus sends health check requests (default is 5 seconds)
• passes – Number of consecutive health checks the server must respond to to be considered healthy (default is 1)
• fails – Number of consecutive health checks the server must fail to respond to to be considered unhealthy (default is 1)

server {
    listen       12345;
    proxy_pass   stream_backend;
    health_check interval=10 passes=2 fails=3;
}

In the example, the time between TCP health checks is increased to 10 seconds, the server is considered unhealthy after 3 consecutive failed health checks, and the server needs to pass 2 consecutive checks to be considered healthy again.

By default, NGINX Plus sends health checks to the port specified by the server directive in the upstream block. You can specify another port for health checks, which is particularly helpful when monitoring the health of many services on the same host. To override the port, include the port parameter to the health_check directive:

server {
    listen       12345;
    proxy_pass   stream_backend;
    health_check port=8080;
}

Fine‑Tuning Health Checks with the match Configuration Block

You can verify server responses to health checks by configuring a number of tests. These tests are defined with the match configuration block placed in the stream context. Specify the match block and set its name (here, tcp_test):

stream {
    # ...
    match  tcp_test {
        # ...
    }
}

Then refer to the block from the health_check directive by including the match parameter and the name of the match block:

stream {
    server {
        listen       12345;
        health_check match=tcp_test;
        proxy_pass   stream_backend;
    }
}

The conditions or tests under which a health check succeeds are set with send and expect parameters:

• send – The text string or hexadecimal literals (/x followed by two hex digits) to send to the server
• expect – Literal string or regular expression that the data returned by the server needs to match

These parameters can be used in different combinations, but no more than one send and one expect parameter can be specified at a time. The syntax for the parameters also depends on the protocol (TCP or UDP).

Fine-Tuning TCP Health Checks

To fine‑tune health checks for TCP, you can combine send and expect parameters as follows:

• If no send or expect parameters are specified, the ability to connect to the server is tested.

• If the expect parameter is specified, the server is expected to unconditionally send data first:

  match pop3 {
      expect ~* "\+OK";
  }
  

• If the send parameter is specified, it is expected that the connection will be successfully established and the specified string will be sent to the server:

  match pop_quit {
      send QUIT;
  }
  

• If both the send and expect parameters are specified, then the string from the send parameter must match the regular expression from the expect parameter:

  stream {
      upstream stream_backend {
          zone   upstream_backend 64k;
          server backend1.example.com:12345;
      }
  
      match http {
          send      "GET / HTTP/1.0\r\nHost: localhost\r\n\r\n";
          expect ~* "200 OK";
      }
  
      server {
          listen       12345;
          health_check match=http;
          proxy_pass   stream_backend;
      }
  }
  

  In this example, for the health check to pass the indicated HTTP request must be sent to the server, and the expected result from the server contains 200 OK to indicate a successful HTTP response.

Fine-Tuning UDP Health Checks

To fine‑tune health checks for UDP, specify both send and expect parameters:

Example for NTP:

match ntp {
    send \xe3\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00;
    expect ~* \x24;
}

Example for DNS:

match dns {
    send \x00\x2a\x00\x00\x00\x01\x00\x00\x00\x00\x00\x00\x03\x73\x74\x6c\x04\x75\x6d\x73\x6c\x03\x65\x64\x75\x00\x00\x01\x00\x01;
    expect ~* "health.is.good";
}

On-the-Fly Configuration

Upstream server groups can be easily reconfigured on-the-fly using a simple HTTP interface. Using this interface, you can view all servers in an upstream group or a particular server, modify server parameters, and add or remove upstream servers.

To enable on-the-fly configuration:

1. Create the top-level http {} block or make sure it is present in your configuration:

   http {
       # ...
   }
   

2. Create a location for configuration requests, for example, upstream_conf:

   http {
       server {
           location /upstream_conf {
               # ...
           }
       }
   }
   

3. In this location specify the upstream_conf directive – a special handler that can be used to inspect and reconfigure upstream groups in NGINX Plus:

   http {
       server {
           location /upstream_conf {
               upstream_conf;
               # ...
           }
       }
   

} ```

1. Limit access to this location with allow and deny directives:

   http {
       server {
           location /upstream_conf {
               upstream_conf;
               allow 127.0.0.1; # permit access from localhost
               deny  all;       # deny access from everywhere else
           }
       }
   }
   

2. Create a shared memory zone for the group of upstream servers so that all worker processes can use the same configuration. To do this, in the top-level stream {} block, find the target upsteam group, add the zone directive to the upstream server group and specify the zone name and the amount of memory:

   stream {
       upstream stream_backend {
           zone backend 64k;
           # ...
       }
   }
   

On-the-Fly Configuration Example

stream {
    # ...
    # Configuration of an upstream server group
    upstream appservers {
        zone appservers 64k;
        server appserv1.example.com:12345 weight=5;
        server appserv2.example.com:12345 fail_timeout=5s;
        server backup1.example.com:12345 backup;
        server backup2.example.com:12345 backup;
    }

    server {
        # Server that proxies connections to the upstream group
        proxy_pass appservers;
        health_check;
    }
}
http {
    # ...
    server {
        # Location for configuration requests
        location /upstream_conf {
            upstream_conf;
            allow 127.0.0.1;
            deny  all;
        }
    }
}

Here, access to the location is allowed only from the localhost address (127.0.0.1). Access from all other IP addresses is denied.

To pass a configuration command to NGINX, send an HTTP request by any method. The request must have an appropriate URI to get into the location that includes the upstream_conf directive. The request must include the upstream argument set to the name of the server group.

For example, to view all backup servers (indicated by the backup argument) in the server group, send:

http://127.0.0.1/upstream_conf?stream=&upstream=appservers&backup=

To add a new server to the server group, send a request with add and server arguments:

http://127.0.0.1/upstream_conf?stream=&add=&upstream=appservers&server=appserv3.example.com:12345&weight=2&max_fails=3

To remove a server from the server group, send a request with the remove argument and the id argument identifying the server:

http://127.0.0.1/upstream_conf?stream=&remove=&upstream=appservers&id=2

To modify a parameter for a specific server, send a request with the id argument identifying the server and the name of the parameter:

http://127.0.0.1/upstream_conf?stream=&upstream=appservers&id=2&down=

Example of TCP and UDP Load-Balancing Configuration

This is a configuration example of TCP and UDP load balancing with NGINX:

stream {
    upstream stream_backend {
        least_conn;
        server backend1.example.com:12345 weight=5;
        server backend2.example.com:12345 max_fails=2 fail_timeout=30s;
        server backend3.example.com:12345 max_conns=3;
    }

    upstream dns_servers {
        least_conn;
        server 192.168.136.130:53;
        server 192.168.136.131:53;
        server 192.168.136.132:53;
    }

    server {
        listen        12345;
        proxy_pass    stream_backend;
        proxy_timeout 3s;
        proxy_connect_timeout 1s;
    }

    server {
        listen     53 udp;
        proxy_pass dns_servers;
    }

    server {
        listen     12346;
        proxy_pass backend4.example.com:12346;
    }
}

In this example, all TCP and UDP proxy‑related functionality is configured inside the stream block, just as settings for HTTP requests are configured in the http block.

There are two named upstream blocks, each containing three servers that host the same content as one another. In the server for eadch server, the server name is followed by the obligatory port number. Connections are distributed among the servers according to the Least Connections load‑balancing method: a connection goes to the server with the fewest number of active connections.

The three server blocks define three virtual servers:

• The first server listens on port 12345 and proxies all TCP connections to the stream_backend group of upstream servers. Note that the proxy_pass directive defined in the context of the stream module must not contain a protocol.

  Two optional timeout parameters are specified: the proxy_connect_timeout directive sets the timeout required for establishing a connection with a server in the stream_backend group. The proxy_timeout directive sets a timeout used after proxying to one of the servers in the stream_backend group has started.

• The second server listens on port 53 and proxies all UDP datagrams (the udp parameter to the listen directive) to an upstream group called dns_servers. If the udpparameter is not specified, the socket listens for TCP connections.

• The third virtual server listens on port 12346 and proxies TCP connections to backend4.example.com, which can resolve to several IP addresses that are load balanced with the Round Robin method.