FAQ

Page Discussion Edit History

ChsHttpUpstreamRequestHashModule

Contents

ngx_http_upstream_hash_module

鏈ā鍧楃敱绗笁鏂规彁渚涳紝涓嶅寘鍚湪 Nginx 鐨勬簮鐮佸彂甯冪増涓傚畨瑁呬粙缁嶇瓑璇风湅 [#installation 杩欓噷] .

The upstream_hash module provides simple upstream load distribution by hashing a configurable variable (e.g., the request URI, incoming HTTP headers, or some combination). Example usage:

upstream backend {
: server server1;
: server server2;
: hash   $request_uri;
}

Here, Nginx will choose server1 or server2 by hashing the request URI ($request_uri).

鎸囦护

  • [#hash hash]
  • [#hash_method hash_method]
  • [#hash_again hash_again]

Template:Anchor

hash

璇硶锛 hash $variable

浣滅敤鍩燂細 upstream

Enables upstream hashing of $variable.

When present, the "server" directives cannot take any arguments ("weight", "max_fails", etc.).

Template:Anchor

hash_method

璇硶锛 hash_method [ crc32 | simple ]

榛樿鍊硷細 simple

浣滅敤鍩燂細 upstream

The hash algorithm to use. "crc32" takes the last 15 bits of the crc32 value and modulo's by the number of servers. (This behavior is compatible with libmemcache.) "simple" is described [#hash-algorithm below] .

Template:Anchor

hash_again

璇硶锛 hash_again number

榛樿鍊硷細 0

浣滅敤鍩燂細 upstream

Number of times to rehash the value and choose a different server if the backend connection fails. Increase this number to provide high availability.

Template:Anchor

Installation

This module is not distributed with the Nginx source. You can download the request_hash module here: nginx_upstream_hash-0.2.tar.gz

After extracting, you will need to patch the latest Nginx source (0.5.21 as of this writing). Run patch like this:

: cd nginx-0.5.21
: patch -p0 < /path/to/upstream/hash/directory/nginx-0.5.21.patch

Then add the following option to your Nginx ./configure command:

: --add-module=path/to/upstream/hash/directory

Then "make" and "make install" as usual.

Template:Anchor

The "simple" hash algorithm

The "simple" hash algorithm is the same as that used for Nginx's internal hash tables. The function is:


u_int ngx_hash_key(u_char *data, size_t len)
{
: u_int  i, key;

: key = 0;

: for (i = 0; i < len; i++) {
: key *= 31; 
: key += data[i] ;
: }

: return key;
}

The hash value is then modulo'd by the number of servers, and thus a server is picked. A server's index is determined by the order in which it appears in the upstream block.