Module ngx_http_upstream_module

• Example Configuration
• Directives
• upstream
• server
• zone
• state
• hash
• ip_hash
• keepalive
• keepalive_requests
• keepalive_timeout
• ntlm
• least_conn
• least_time
• queue
• random
• sticky
• sticky_cookie_insert
• Embedded Variables

The ngx_http_upstream_module module is used to define groups of servers that can be referenced by the proxy_pass, fastcgi_pass, uwsgi_pass, scgi_pass, memcached_pass, and grpc_pass directives.

Example Configuration

upstream backend {
    server backend1.example.com       weight=5;
    server backend2.example.com:8080;
    server unix:/tmp/backend3;

    server backup1.example.com:8080   backup;
    server backup2.example.com:8080   backup;
}

server {
    location / {
        proxy_pass http://backend;
    }
}

Dynamically configurable group with periodic health checks is available as part of our commercial subscription:

resolver 10.0.0.1;

upstream dynamic {
    zone upstream_dynamic 64k;

    server backend1.example.com      weight=5;
    server backend2.example.com:8080 fail_timeout=5s slow_start=30s;
    server 192.0.2.1                 max_fails=3;
    server backend3.example.com      resolve;
    server backend4.example.com      service=http resolve;

    server backup1.example.com:8080  backup;
    server backup2.example.com:8080  backup;
}

server {
    location / {
        proxy_pass http://dynamic;
        health_check;
    }
}

Directives

Defines a group of servers. Servers can listen on different ports. In addition, servers listening on TCP and UNIX-domain sockets can be mixed.

Example:

upstream backend {
    server backend1.example.com weight=5;
    server 127.0.0.1:8080       max_fails=3 fail_timeout=30s;
    server unix:/tmp/backend3;

    server backup1.example.com  backup;
}

By default, requests are distributed between the servers using a weighted round-robin balancing method. In the above example, each 7 requests will be distributed as follows: 5 requests go to backend1.example.com and one request to each of the second and third servers. If an error occurs during communication with a server, the request will be passed to the next server, and so on until all of the functioning servers will be tried. If a successful response could not be obtained from any of the servers, the client will receive the result of the communication with the last server.

Defines the address and other parameters of a server. The address can be specified as a domain name or IP address, with an optional port, or as a UNIX-domain socket path specified after the “unix:” prefix. If a port is not specified, the port 80 is used. A domain name that resolves to several IP addresses defines multiple servers at once.

The following parameters can be defined:

weight=number

sets the weight of the server, by default, 1.

max_conns=number

limits the maximum number of simultaneous active connections to the proxied server (1.11.5). Default value is zero, meaning there is no limit. If the server group does not reside in the shared memory, the limitation works per each worker process.

If idle keepalive connections, multiple workers, and the shared memory are enabled, the total number of active and idle connections to the proxied server may exceed the max_conns value.

Since version 1.5.9 and prior to version 1.11.5, this parameter was available as part of our commercial subscription.

max_fails=number

sets the number of unsuccessful attempts to communicate with the server that should happen in the duration set by the fail_timeout parameter to consider the server unavailable for a duration also set by the fail_timeout parameter. By default, the number of unsuccessful attempts is set to 1. The zero value disables the accounting of attempts. What is considered an unsuccessful attempt is defined by the proxy_next_upstream, fastcgi_next_upstream, uwsgi_next_upstream, scgi_next_upstream, memcached_next_upstream, and grpc_next_upstream directives.

fail_timeout=time

sets

• the time during which the specified number of unsuccessful attempts to communicate with the server should happen to consider the server unavailable;
• and the period of time the server will be considered unavailable.

By default, the parameter is set to 10 seconds.

backup

marks the server as a backup server. It will be passed requests when the primary servers are unavailable.

down

marks the server as permanently unavailable.

Additionally, the following parameters are available as part of our commercial subscription:

resolve

monitors changes of the IP addresses that correspond to a domain name of the server, and automatically modifies the upstream configuration without the need of restarting nginx (1.5.12). The server group must reside in the shared memory.

In order for this parameter to work, the resolver directive must be specified in the http block. Example:

http {
    resolver 10.0.0.1;

    upstream u {
        zone ...;
        ...
        server example.com resolve;
    }
}

route=string

sets the server route name.

service=name

enables resolving of DNS SRV records and sets the service name (1.9.13). In order for this parameter to work, it is necessary to specify the resolve parameter for the server and specify a hostname without a port number.

If the service name does not contain a dot (“.”), then the RFC-compliant name is constructed and the TCP protocol is added to the service prefix. For example, to look up the _http._tcp.backend.example.com SRV record, it is necessary to specify the directive:

server backend.example.com service=http resolve;

If the service name contains one or more dots, then the name is constructed by joining the service prefix and the server name. For example, to look up the _http._tcp.backend.example.com and server1.backend.example.com SRV records, it is necessary to specify the directives:

server backend.example.com service=_http._tcp resolve;
server example.com service=server1.backend resolve;

Highest-priority SRV records (records with the same lowest-number priority value) are resolved as primary servers, the rest of SRV records are resolved as backup servers. If the backup parameter is specified for the server, high-priority SRV records are resolved as backup servers, the rest of SRV records are ignored.

slow_start=time

sets the time during which the server will recover its weight from zero to a nominal value, when unhealthy server becomes healthy, or when the server becomes available after a period of time it was considered unavailable. Default value is zero, i.e. slow start is disabled.

The parameter cannot be used along with the hash and ip_hash load balancing methods.

drain

puts the server into the “draining” mode (1.13.6). In this mode, only requests bound to the server will be proxied to it.

Prior to version 1.13.6, the parameter could be changed only with the API module.

If there is only a single server in a group, max_fails, fail_timeout and slow_start parameters are ignored, and such a server will never be considered unavailable.

Defines the name and size of the shared memory zone that keeps the group’s configuration and run-time state that are shared between worker processes. Several groups may share the same zone. In this case, it is enough to specify the size only once.

Additionally, as part of our commercial subscription, such groups allow changing the group membership or modifying the settings of a particular server without the need of restarting nginx. The configuration is accessible via the API module (1.13.3).

Prior to version 1.13.3, the configuration was accessible only via a special location handled by upstream_conf.

Specifies a file that keeps the state of the dynamically configurable group.

Examples:

state /var/lib/nginx/state/servers.conf; # path for Linux
state /var/db/nginx/state/servers.conf;  # path for FreeBSD

The state is currently limited to the list of servers with their parameters. The file is read when parsing the configuration and is updated each time the upstream configuration is changed. Changing the file content directly should be avoided. The directive cannot be used along with the server directive.

Changes made during configuration reload or binary upgrade can be lost.

This directive is available as part of our commercial subscription.

Specifies a load balancing method for a server group where the client-server mapping is based on the hashed key value. The key can contain text, variables, and their combinations. Note that adding or removing a server from the group may result in remapping most of the keys to different servers. The method is compatible with the Cache::Memcached Perl library.

If the consistent parameter is specified, the ketama consistent hashing method will be used instead. The method ensures that only a few keys will be remapped to different servers when a server is added to or removed from the group. This helps to achieve a higher cache hit ratio for caching servers. The method is compatible with the Cache::Memcached::Fast Perl library with the ketama_points parameter set to 160.

Specifies that a group should use a load balancing method where requests are distributed between servers based on client IP addresses. The first three octets of the client IPv4 address, or the entire IPv6 address, are used as a hashing key. The method ensures that requests from the same client will always be passed to the same server except when this server is unavailable. In the latter case client requests will be passed to another server. Most probably, it will always be the same server as well.

IPv6 addresses are supported starting from versions 1.3.2 and 1.2.2.

If one of the servers needs to be temporarily removed, it should be marked with the down parameter in order to preserve the current hashing of client IP addresses.

Example:

upstream backend {
    ip_hash;

    server backend1.example.com;
    server backend2.example.com;
    server backend3.example.com down;
    server backend4.example.com;
}

Until versions 1.3.1 and 1.2.2, it was not possible to specify a weight for servers using the ip_hash load balancing method.

Activates the cache for connections to upstream servers.

The connections parameter sets the maximum number of idle keepalive connections to upstream servers that are preserved in the cache of each worker process. When this number is exceeded, the least recently used connections are closed.

It should be particularly noted that the keepalive directive does not limit the total number of connections to upstream servers that an nginx worker process can open. The connections parameter should be set to a number small enough to let upstream servers process new incoming connections as well.

Example configuration of memcached upstream with keepalive connections:

upstream memcached_backend {
    server 127.0.0.1:11211;
    server 10.0.0.2:11211;

    keepalive 32;
}

server {
    ...

    location /memcached/ {
        set $memcached_key $uri;
        memcached_pass memcached_backend;
    }

}

For HTTP, the proxy_http_version directive should be set to “1.1” and the “Connection” header field should be cleared:

upstream http_backend {
    server 127.0.0.1:8080;

    keepalive 16;
}

server {
    ...

    location /http/ {
        proxy_pass http://http_backend;
        proxy_http_version 1.1;
        proxy_set_header Connection "";
        ...
    }
}

Alternatively, HTTP/1.0 persistent connections can be used by passing the “Connection: Keep-Alive” header field to an upstream server, though this method is not recommended.

For FastCGI servers, it is required to set fastcgi_keep_conn for keepalive connections to work:

upstream fastcgi_backend {
    server 127.0.0.1:9000;

    keepalive 8;
}

server {
    ...

    location /fastcgi/ {
        fastcgi_pass fastcgi_backend;
        fastcgi_keep_conn on;
        ...
    }
}

When using load balancer methods other than the default round-robin method, it is necessary to activate them before the keepalive directive.

SCGI and uwsgi protocols do not have a notion of keepalive connections.

Sets the maximum number of requests that can be served through one keepalive connection. After the maximum number of requests is made, the connection is closed.

Sets a timeout during which an idle keepalive connection to an upstream server will stay open.

Allows proxying requests with NTLM Authentication. The upstream connection is bound to the client connection once the client sends a request with the “Authorization” header field value starting with “Negotiate” or “NTLM”. Further client requests will be proxied through the same upstream connection, keeping the authentication context.

In order for NTLM authentication to work, it is necessary to enable keepalive connections to upstream servers. The proxy_http_version directive should be set to “1.1” and the “Connection” header field should be cleared:

upstream http_backend {
    server 127.0.0.1:8080;

    ntlm;
}

server {
    ...

    location /http/ {
        proxy_pass http://http_backend;
        proxy_http_version 1.1;
        proxy_set_header Connection "";
        ...
    }
}

When using load balancer methods other than the default round-robin method, it is necessary to activate them before the ntlm directive.

This directive is available as part of our commercial subscription.

Specifies that a group should use a load balancing method where a request is passed to the server with the least number of active connections, taking into account weights of servers. If there are several such servers, they are tried in turn using a weighted round-robin balancing method.

Specifies that a group should use a load balancing method where a request is passed to the server with the least average response time and least number of active connections, taking into account weights of servers. If there are several such servers, they are tried in turn using a weighted round-robin balancing method.

If the header parameter is specified, time to receive the response header is used. If the last_byte parameter is specified, time to receive the full response is used. If the inflight parameter is specified (1.11.6), incomplete requests are also taken into account.

Prior to version 1.11.6, incomplete requests were taken into account by default.

This directive is available as part of our commercial subscription.

If an upstream server cannot be selected immediately while processing a request, the request will be placed into the queue. The directive specifies the maximum number of requests that can be in the queue at the same time. If the queue is filled up, or the server to pass the request to cannot be selected within the time period specified in the timeout parameter, the 502 (Bad Gateway) error will be returned to the client.

The default value of the timeout parameter is 60 seconds.

When using load balancer methods other than the default round-robin method, it is necessary to activate them before the queue directive.

This directive is available as part of our commercial subscription.

Specifies that a group should use a load balancing method where a request is passed to a randomly selected server, taking into account weights of servers.

The optional two parameter instructs nginx to randomly select two servers and then choose a server using the specified method. The default method is least_conn which passes a request to a server with the least number of active connections.

The least_time method passes a request to a server with the least average response time and least number of active connections. If least_time=header is specified, the time to receive the response header is used. If least_time=last_byte is specified, the time to receive the full response is used.

The least_time method is available as a part of our commercial subscription.

Enables session affinity, which causes requests from the same client to be passed to the same server in a group of servers. Three methods are available:

cookie

When the cookie method is used, information about the designated server is passed in an HTTP cookie generated by nginx:

upstream backend {
    server backend1.example.com;
    server backend2.example.com;

    sticky cookie srv_id expires=1h domain=.example.com path=/;
}

A request that comes from a client not yet bound to a particular server is passed to the server selected by the configured balancing method. Further requests with this cookie will be passed to the designated server. If the designated server cannot process a request, the new server is selected as if the client has not been bound yet.

The first parameter sets the name of the cookie to be set or inspected. The cookie value is a hexadecimal representation of the MD5 hash of the IP address and port, or of the UNIX-domain socket path. However, if the “route” parameter of the server directive is specified, the cookie value will be the value of the “route” parameter:

upstream backend {
    server backend1.example.com route=a;
    server backend2.example.com route=b;

    sticky cookie srv_id expires=1h domain=.example.com path=/;
}

In this case, the value of the “srv_id” cookie will be either a or b.

Additional parameters may be as follows:

expires=time

Sets the time for which a browser should keep the cookie. The special value max will cause the cookie to expire on “31 Dec 2037 23:55:55 GMT”. If the parameter is not specified, it will cause the cookie to expire at the end of a browser session.

domain=domain

Defines the domain for which the cookie is set. Parameter value can contain variables (1.11.5).

httponly

Adds the HttpOnly attribute to the cookie (1.7.11).

secure

Adds the Secure attribute to the cookie (1.7.11).

path=path

Defines the path for which the cookie is set.

If any parameters are omitted, the corresponding cookie fields are not set.

route

When the route method is used, proxied server assigns client a route on receipt of the first request. All subsequent requests from this client will carry routing information in a cookie or URI. This information is compared with the “route” parameter of the server directive to identify the server to which the request should be proxied. If the “route” parameter is not specified, the route name will be a hexadecimal representation of the MD5 hash of the IP address and port, or of the UNIX-domain socket path. If the designated server cannot process a request, the new server is selected by the configured balancing method as if there is no routing information in the request.

The parameters of the route method specify variables that may contain routing information. The first non-empty variable is used to find the matching server.

Example:

map $cookie_jsessionid $route_cookie {
    ~.+\.(?P<route>\w+)$ $route;
}

map $request_uri $route_uri {
    ~jsessionid=.+\.(?P<route>\w+)$ $route;
}

upstream backend {
    server backend1.example.com route=a;
    server backend2.example.com route=b;

    sticky route $route_cookie $route_uri;
}

Here, the route is taken from the “JSESSIONID” cookie if present in a request. Otherwise, the route from the URI is used.

learn

When the learn method (1.7.1) is used, nginx analyzes upstream server responses and learns server-initiated sessions usually passed in an HTTP cookie.

upstream backend {
   server backend1.example.com:8080;
   server backend2.example.com:8081;

   sticky learn
          create=$upstream_cookie_examplecookie
          lookup=$cookie_examplecookie
          zone=client_sessions:1m;
}

In the example, the upstream server creates a session by setting the cookie “EXAMPLECOOKIE” in the response. Further requests with this cookie will be passed to the same server. If the server cannot process the request, the new server is selected as if the client has not been bound yet.

The parameters create and lookup specify variables that indicate how new sessions are created and existing sessions are searched, respectively. Both parameters may be specified more than once, in which case the first non-empty variable is used.

Sessions are stored in a shared memory zone, whose name and size are configured by the zone parameter. One megabyte zone can store about 4000 sessions on the 64-bit platform. The sessions that are not accessed during the time specified by the timeout parameter get removed from the zone. By default, timeout is set to 10 minutes.

The header parameter (1.13.1) allows creating a session right after receiving response headers from the upstream server.

The sync parameter (1.13.8) enables synchronization of the shared memory zone.

This directive is available as part of our commercial subscription.

This directive is obsolete since version 1.5.7. An equivalent sticky directive with a new syntax should be used instead:

sticky cookie name [expires=time] [domain=domain] [path=path];

Embedded Variables

The ngx_http_upstream_module module supports the following embedded variables:

$upstream_addr

keeps the IP address and port, or the path to the UNIX-domain socket of the upstream server. If several servers were contacted during request processing, their addresses are separated by commas, e.g. “192.168.1.1:80, 192.168.1.2:80, unix:/tmp/sock”. If an internal redirect from one server group to another happens, initiated by “X-Accel-Redirect” or error_page, then the server addresses from different groups are separated by colons, e.g. “192.168.1.1:80, 192.168.1.2:80, unix:/tmp/sock : 192.168.10.1:80, 192.168.10.2:80”. If a server cannot be selected, the variable keeps the name of the server group.

$upstream_bytes_received

number of bytes received from an upstream server (1.11.4). Values from several connections are separated by commas and colons like addresses in the $upstream_addr variable.

$upstream_bytes_sent

number of bytes sent to an upstream server (1.15.8). Values from several connections are separated by commas and colons like addresses in the $upstream_addr variable.

$upstream_cache_status

keeps the status of accessing a response cache (0.8.3). The status can be either “MISS”, “BYPASS”, “EXPIRED”, “STALE”, “UPDATING”, “REVALIDATED”, or “HIT”.

$upstream_connect_time

keeps time spent on establishing a connection with the upstream server (1.9.1); the time is kept in seconds with millisecond resolution. In case of SSL, includes time spent on handshake. Times of several connections are separated by commas and colons like addresses in the $upstream_addr variable.

$upstream_cookie_name

cookie with the specified name sent by the upstream server in the “Set-Cookie” response header field (1.7.1). Only the cookies from the response of the last server are saved.

$upstream_header_time

keeps time spent on receiving the response header from the upstream server (1.7.10); the time is kept in seconds with millisecond resolution. Times of several responses are separated by commas and colons like addresses in the $upstream_addr variable.

$upstream_http_name

keep server response header fields. For example, the “Server” response header field is available through the $upstream_http_server variable. The rules of converting header field names to variable names are the same as for the variables that start with the “$http_” prefix. Only the header fields from the response of the last server are saved.

$upstream_queue_time

keeps time the request spent in the upstream queue (1.13.9); the time is kept in seconds with millisecond resolution. Times of several responses are separated by commas and colons like addresses in the $upstream_addr variable.

$upstream_response_length

keeps the length of the response obtained from the upstream server (0.7.27); the length is kept in bytes. Lengths of several responses are separated by commas and colons like addresses in the $upstream_addr variable.

$upstream_response_time

keeps time spent on receiving the response from the upstream server; the time is kept in seconds with millisecond resolution. Times of several responses are separated by commas and colons like addresses in the $upstream_addr variable.

$upstream_status

keeps status code of the response obtained from the upstream server. Status codes of several responses are separated by commas and colons like addresses in the $upstream_addr variable. If a server cannot be selected, the variable keeps the 502 (Bad Gateway) status code.

$upstream_trailer_name

keeps fields from the end of the response obtained from the upstream server (1.13.10).