README

Name
    ngx_lua - Embed the power of Lua into Nginx

    *This module is not distributed with the Nginx source.* See the
    installation instructions.

Status
    This module is under active development and is production ready.

Version
    This document describes ngx_lua v0.8.3
    (<https://github.com/chaoslawful/lua-nginx-module/tags>) released on 20
    June 2013.

Synopsis
        # set search paths for pure Lua external libraries (';;' is the default path):
        lua_package_path '/foo/bar/?.lua;/blah/?.lua;;';

        # set search paths for Lua external libraries written in C (can also use ';;'):
        lua_package_cpath '/bar/baz/?.so;/blah/blah/?.so;;';

        server {
            location /inline_concat {
                # MIME type determined by default_type:
                default_type 'text/plain';

                set $a "hello";
                set $b "world";
                # inline Lua script
                set_by_lua $res "return ngx.arg[1]..ngx.arg[2]" $a $b;
                echo $res;
            }

            location /rel_file_concat {
                set $a "foo";
                set $b "bar";
                # script path relative to nginx prefix
                # $ngx_prefix/conf/concat.lua contents:
                #
                #    return ngx.arg[1]..ngx.arg[2]
                #
                set_by_lua_file $res conf/concat.lua $a $b;
                echo $res;
            }

            location /abs_file_concat {
                set $a "fee";
                set $b "baz";
                # absolute script path not modified
                set_by_lua_file $res /usr/nginx/conf/concat.lua $a $b;
                echo $res;
            }

            location /lua_content {
                # MIME type determined by default_type:
                default_type 'text/plain';

                content_by_lua "ngx.say('Hello,world!')";
            }

             location /nginx_var {
                # MIME type determined by default_type:
                default_type 'text/plain';

                # try access /nginx_var?a=hello,world
                content_by_lua "ngx.print(ngx.var['arg_a'], '\\n')";
            }

            location /request_body {
                 # force reading request body (default off)
                 lua_need_request_body on;
                 client_max_body_size 50k;
                 client_body_buffer_size 50k;

                 content_by_lua 'ngx.print(ngx.var.request_body)';
            }

            # transparent non-blocking I/O in Lua via subrequests
            location /lua {
                # MIME type determined by default_type:
                default_type 'text/plain';

                content_by_lua '
                    local res = ngx.location.capture("/some_other_location")
                    if res.status == 200 then
                        ngx.print(res.body)
                    end';
            }

            # GET /recur?num=5
            location /recur {
                # MIME type determined by default_type:
                default_type 'text/plain';

                content_by_lua '
                   local num = tonumber(ngx.var.arg_num) or 0

                   if num > 50 then
                       ngx.say("num too big")
                       return
                   end

                   ngx.say("num is: ", num)

                   if num > 0 then
                       res = ngx.location.capture("/recur?num=" .. tostring(num - 1))
                       ngx.print("status=", res.status, " ")
                       ngx.print("body=", res.body)
                   else
                       ngx.say("end")
                   end
                   ';
            }

            location /foo {
                rewrite_by_lua '
                    res = ngx.location.capture("/memc",
                        { args = { cmd = "incr", key = ngx.var.uri } }
                    )
                ';

                proxy_pass http://blah.blah.com;
            }

            location /blah {
                access_by_lua '
                    local res = ngx.location.capture("/auth")

                    if res.status == ngx.HTTP_OK then
                        return
                    end

                    if res.status == ngx.HTTP_FORBIDDEN then
                        ngx.exit(res.status)
                    end

                    ngx.exit(ngx.HTTP_INTERNAL_SERVER_ERROR)
                ';

                # proxy_pass/fastcgi_pass/postgres_pass/...
            }

            location /mixed {
                rewrite_by_lua_file /path/to/rewrite.lua;
                access_by_lua_file /path/to/access.lua;
                content_by_lua_file /path/to/content.lua;
            }

            # use nginx var in code path
            # WARN: contents in nginx var must be carefully filtered,
            # otherwise there'll be great security risk!
            location ~ ^/app/(.+) {
                    content_by_lua_file /path/to/lua/app/root/$1.lua;
            }

            location / {
               lua_need_request_body on;

               client_max_body_size 100k;
               client_body_buffer_size 100k;

               access_by_lua '
                   -- check the client IP address is in our black list
                   if ngx.var.remote_addr == "132.5.72.3" then
                       ngx.exit(ngx.HTTP_FORBIDDEN)
                   end

                   -- check if the request body contains bad words
                   if ngx.var.request_body and
                            string.match(ngx.var.request_body, "fsck")
                   then
                       return ngx.redirect("/terms_of_use.html")
                   end

                   -- tests passed
               ';

               # proxy_pass/fastcgi_pass/etc settings
            }
        }

Description
    This module embeds Lua, via the standard Lua 5.1 interpreter or LuaJIT
    2.0 (<http://luajit.org/luajit.html>), into Nginx and by leveraging
    Nginx's subrequests, allows the integration of the powerful Lua threads
    (Lua coroutines) into the Nginx event model.

    Unlike Apache's mod_lua
    (<http://httpd.apache.org/docs/2.3/mod/mod_lua.html>) and Lighttpd's
    mod_magnet (<http://redmine.lighttpd.net/wiki/1/Docs:ModMagnet>), Lua
    code executed using this module can be *100% non-blocking* on network
    traffic as long as the Nginx API for Lua provided by this module is used
    to handle requests to upstream services such as MySQL, PostgreSQL,
    Memcached, Redis, or upstream HTTP web services.

    At least the following Lua libraries and Nginx modules can be used with
    this ngx_lua module:

    *   lua-resty-memcached
        (<https://github.com/agentzh/lua-resty-memcached>)

    *   lua-resty-mysql (<https://github.com/agentzh/lua-resty-mysql>)

    *   lua-resty-redis (<https://github.com/agentzh/lua-resty-redis>)

    *   lua-resty-dns (<https://github.com/agentzh/lua-resty-dns>)

    *   lua-resty-upload (<https://github.com/agentzh/lua-resty-upload>)

    *   ngx_memc

    *   ngx_postgres (<https://github.com/FRiCKLE/ngx_postgres>)

    *   ngx_redis2

    *   ngx_redis

    *   ngx_proxy

    *   ngx_fastcgi

    Almost all the Nginx modules can be used with this ngx_lua module by
    means of ngx.location.capture or ngx.location.capture_multi but it is
    recommended to use those "lua-resty-*" libraries instead of creating
    subrequests to access the Nginx upstream modules because the former is
    usually much more flexible and memory-efficient.

    The Lua interpreter or LuaJIT instance is shared across all the requests
    in a single nginx worker process but request contexts are segregated
    using lightweight Lua coroutines.

    Loaded Lua modules persist in the nginx worker process level resulting
    in a small memory footprint in Lua even when under heavy loads.

Directives
  lua_code_cache
    syntax: *lua_code_cache on | off*

    default: *lua_code_cache on*

    context: *main, server, location, location if*

    Enables or disables the Lua code cache for set_by_lua_file,
    content_by_lua_file, rewrite_by_lua_file, and access_by_lua_file, and
    also force Lua module reloading on a per-request basis.

    The Lua files referenced in set_by_lua_file, content_by_lua_file,
    access_by_lua_file, and rewrite_by_lua_file will not be cached and the
    Lua "package.loaded" table will be cleared at the entry point of every
    request (such that Lua modules will not be cached either). With this in
    place, developers can adopt an edit-and-refresh approach.

    Please note however, that Lua code written inlined within nginx.conf
    such as those specified by set_by_lua, content_by_lua, access_by_lua,
    and rewrite_by_lua will *always* be cached because only the Nginx config
    file parser can correctly parse the "nginx.conf" file and the only ways
    to to reload the config file are to send a "HUP" signal or to restart
    Nginx.

    Also, Lua files which are loaded by "dofile" or "loadfile" in
    *_by_lua_file will never be cached. To ensure code caching, you can
    either use the init_by_lua or init_by_lua_file directives to load all
    such files or just make these Lua files true Lua modules and load them
    via "require".

    The ngx_lua module does not currently support the "stat" mode available
    with the Apache "mod_lua" module but this is planned for implementation
    in the future.

    Disabling the Lua code cache is strongly discouraged for production use
    and should only be used during development as it has a significant
    negative impact on overall performance. In addition, race conditions
    when reloading Lua modules are common for concurrent requests when the
    code cache is disabled.

  lua_regex_cache_max_entries
    syntax: *lua_regex_cache_max_entries <num>*

    default: *lua_regex_cache_max_entries 1024*

    context: *http*

    Specifies the maximum number of entries allowed in the worker process
    level compiled regex cache.

    The regular expressions used in ngx.re.match, ngx.re.gmatch, ngx.re.sub,
    and ngx.re.gsub will be cached within this cache if the regex option "o"
    (i.e., compile-once flag) is specified.

    The default number of entries allowed is 1024 and when this limit is
    reached, new regular expressions will not be cached (as if the "o"
    option was not specified) and there will be one, and only one, warning
    in the "error.log" file:

        2011/08/27 23:18:26 [warn] 31997#0: *1 lua exceeding regex cache max entries (1024), ...

    Do not activate the "o" option for regular expressions (and/or "replace"
    string arguments for ngx.re.sub and ngx.re.gsub) that are generated *on
    the fly* and give rise to infinite variations to avoid hitting the
    specified limit.

  lua_package_path
    syntax: *lua_package_path <lua-style-path-str>*

    default: *The content of LUA_PATH environ variable or Lua's compiled-in
    defaults.*

    context: *main*

    Sets the Lua module search path used by scripts specified by set_by_lua,
    content_by_lua and others. The path string is in standard Lua path form,
    and ";;" can be used to stand for the original search paths.

    As from the "v0.5.0rc29" release, the special notation $prefix or
    "${prefix}" can be used in the search path string to indicate the path
    of the "server prefix" usually determined by the "-p PATH" command-line
    option while starting the Nginx server.

  lua_package_cpath
    syntax: *lua_package_cpath <lua-style-cpath-str>*

    default: *The content of LUA_CPATH environment variable or Lua's
    compiled-in defaults.*

    context: *main*

    Sets the Lua C-module search path used by scripts specified by
    set_by_lua, content_by_lua and others. The cpath string is in standard
    Lua cpath form, and ";;" can be used to stand for the original cpath.

    As from the "v0.5.0rc29" release, the special notation $prefix or
    "${prefix}" can be used in the search path string to indicate the path
    of the "server prefix" usually determined by the "-p PATH" command-line
    option while starting the Nginx server.

  init_by_lua
    syntax: *init_by_lua <lua-script-str>*

    context: *http*

    phase: *loading-config*

    Runs the Lua code specified by the argument "<lua-script-str>" on the
    global Lua VM level when the Nginx master process (if any) is loading
    the Nginx config file.

    When Nginx receives the "HUP" signal and starts reloading the config
    file, the Lua VM will also be re-created and "init_by_lua" will run
    again on the new Lua VM.

    Usually you can register (true) Lua global variables or pre-load Lua
    modules at server start-up by means of this hook. Here is an example for
    pre-loading Lua modules:

        init_by_lua 'cjson = require "cjson"';

        server {
            location = /api {
                content_by_lua '
                    ngx.say(cjson.encode({dog = 5, cat = 6}))
                ';
            }
        }

    You can also initialize the lua_shared_dict shm storage at this phase.
    Here is an example for this:

        lua_shared_dict dogs 1m;

        init_by_lua '
            local dogs = ngx.shared.dogs;
            dogs:set("Tom", 56)
        ';

        server {
            location = /api {
                content_by_lua '
                    local dogs = ngx.shared.dogs;
                    ngx.say(dogs:get("Tom"))
                ';
            }
        }

    But note that, the lua_shared_dict's shm storage will not be cleared
    through a config reload (via the "HUP" signal, for example). So if you
    do *not* want to re-initialize the shm storage in your "init_by_lua"
    code in this case, then you just need to set a custom flag in the shm
    storage and always check the flag in your "init_by_lua" code.

    Because the Lua code in this context runs before Nginx forks its worker
    processes (if any), data or code loaded here will enjoy the
    Copy-on-write (COW) (<http://en.wikipedia.org/wiki/Copy-on-write>)
    feature provided by many operating systems among all the worker
    processes, thus saving a lot of memory.

    Only a small set of the Nginx API for Lua is supported in this context:

    *   Logging APIs: ngx.log and print,

    *   Shared Dictionary API: ngx.shared.DICT.

    More Nginx APIs for Lua may be supported in this context upon future
    user requests.

    Basically you can safely use Lua libraries that do blocking I/O in this
    very context because blocking the master process during server start-up
    is completely okay. Even the Nginx core does blocking I/O (at least on
    resolving upstream's host names) at the configure-loading phase.

    You should be very careful about potential security vulnerabilities in
    your Lua code registered in this context because the Nginx master
    process is often run under the "root" account.

    This directive was first introduced in the "v0.5.5" release.

  init_by_lua_file
    syntax: *init_by_lua_file <path-to-lua-script-file>*

    context: *http*

    phase: *loading-config*

    Equivalent to init_by_lua, except that the file specified by
    "<path-to-lua-script-file>" contains the Lua code or Lua/LuaJIT bytecode
    to be executed.

    When a relative path like "foo/bar.lua" is given, they will be turned
    into the absolute path relative to the "server prefix" path determined
    by the "-p PATH" command-line option while starting the Nginx server.

    This directive was first introduced in the "v0.5.5" release.

  set_by_lua
    syntax: *set_by_lua $res <lua-script-str> [$arg1 $arg2 ...]*

    context: *server, server if, location, location if*

    phase: *server-rewrite, rewrite*

    Executes code specified in "<lua-script-str>" with optional input
    arguments "$arg1 $arg2 ...", and returns string output to $res. The code
    in "<lua-script-str>" can make API calls and can retrieve input
    arguments from the "ngx.arg" table (index starts from 1 and increases
    sequentially).

    This directive is designed to execute short, fast running code blocks as
    the Nginx event loop is blocked during code execution. Time consuming
    code sequences should therefore be avoided.

    Note that the following API functions are currently disabled within this
    context:

    *   Output API functions (e.g., ngx.say and ngx.send_headers)

    *   Control API functions (e.g., ngx.exit)

    *   Subrequest API functions (e.g., ngx.location.capture and
        ngx.location.capture_multi)

    *   Cosocket API functions (e.g., ngx.socket.tcp and ngx.req.socket).

    In addition, note that this directive can only write out a value to a
    single Nginx variable at a time. However, a workaround is possible using
    the ngx.var.VARIABLE interface.

        location /foo {
            set $diff ''; # we have to predefine the $diff variable here

            set_by_lua $sum '
                local a = 32
                local b = 56

                ngx.var.diff = a - b;  -- write to $diff directly
                return a + b;          -- return the $sum value normally
            ';

            echo "sum = $sum, diff = $diff";
        }

    This directive can be freely mixed with all directives of the
    [[HttpRewriteModule]], [[HttpSetMiscModule]], and [[HttpArrayVarModule]]
    modules. All of these directives will run in the same order as they
    appear in the config file.

        set $foo 32;
        set_by_lua $bar 'tonumber(ngx.var.foo) + 1';
        set $baz "bar: $bar";  # $baz == "bar: 33"

    As from the "v0.5.0rc29" release, Nginx variable interpolation is
    disabled in the "<lua-script-str>" argument of this directive and
    therefore, the dollar sign character ("$") can be used directly.

    This directive requires the ngx_devel_kit
    (<https://github.com/simpl/ngx_devel_kit>) module.

  set_by_lua_file
    syntax: *set_by_lua_file $res <path-to-lua-script-file> [$arg1 $arg2
    ...]*

    context: *server, server if, location, location if*

    phase: *server-rewrite, rewrite*

    Equivalent to set_by_lua, except that the file specified by
    "<path-to-lua-script-file>" contains the Lua code, or, as from the
    "v0.5.0rc32" release, the Lua/LuaJIT bytecode to be executed.

    Nginx variable interpolation is supported in the
    "<path-to-lua-script-file>" argument string of this directive. But
    special care must be taken for injection attacks.

    When a relative path like "foo/bar.lua" is given, they will be turned
    into the absolute path relative to the "server prefix" path determined
    by the "-p PATH" command-line option while starting the Nginx server.

    When the Lua code cache is turned on (by default), the user code is
    loaded once at the first request and cached and the Nginx config must be
    reloaded each time the Lua source file is modified. The Lua code cache
    can be temporarily disabled during development by switching
    lua_code_cache "off" in "nginx.conf" to avoid reloading Nginx.

    This directive requires the ngx_devel_kit
    (<https://github.com/simpl/ngx_devel_kit>) module.

  content_by_lua
    syntax: *content_by_lua <lua-script-str>*

    context: *location, location if*

    phase: *content*

    Acts as a "content handler" and executes Lua code string specified in
    "<lua-script-str>" for every request. The Lua code may make API calls
    and is executed as a new spawned coroutine in an independent global
    environment (i.e. a sandbox).

    Do not use this directive and other content handler directives in the
    same location. For example, this directive and the proxy_pass directive
    should not be used in the same location.

  content_by_lua_file
    syntax: *content_by_lua_file <path-to-lua-script-file>*

    context: *location, location if*

    phase: *content*

    Equivalent to content_by_lua, except that the file specified by
    "<path-to-lua-script-file>" contains the Lua code, or, as from the
    "v0.5.0rc32" release, the Lua/LuaJIT bytecode to be executed.

    Nginx variables can be used in the "<path-to-lua-script-file>" string to
    provide flexibility. This however carries some risks and is not
    ordinarily recommended.

    When a relative path like "foo/bar.lua" is given, they will be turned
    into the absolute path relative to the "server prefix" path determined
    by the "-p PATH" command-line option while starting the Nginx server.

    When the Lua code cache is turned on (by default), the user code is
    loaded once at the first request and cached and the Nginx config must be
    reloaded each time the Lua source file is modified. The Lua code cache
    can be temporarily disabled during development by switching
    lua_code_cache "off" in "nginx.conf" to avoid reloading Nginx.

  rewrite_by_lua
    syntax: *rewrite_by_lua <lua-script-str>*

    context: *http, server, location, location if*

    phase: *rewrite tail*

    Acts as a rewrite phase handler and executes Lua code string specified
    in "<lua-script-str>" for every request. The Lua code may make API calls
    and is executed as a new spawned coroutine in an independent global
    environment (i.e. a sandbox).

    Note that this handler always runs *after* the standard
    [[HttpRewriteModule]]. So the following will work as expected:

       location /foo {
           set $a 12; # create and initialize $a
           set $b ""; # create and initialize $b
           rewrite_by_lua 'ngx.var.b = tonumber(ngx.var.a) + 1';
           echo "res = $b";
       }

    because "set $a 12" and "set $b """ run *before* rewrite_by_lua.

    On the other hand, the following will not work as expected:

        ?  location /foo {
        ?      set $a 12; # create and initialize $a
        ?      set $b ''; # create and initialize $b
        ?      rewrite_by_lua 'ngx.var.b = tonumber(ngx.var.a) + 1';
        ?      if ($b = '13') {
        ?         rewrite ^ /bar redirect;
        ?         break;
        ?      }
        ?
        ?      echo "res = $b";
        ?  }

    because "if" runs *before* rewrite_by_lua even if it is placed after
    rewrite_by_lua in the config.

    The right way of doing this is as follows:

        location /foo {
            set $a 12; # create and initialize $a
            set $b ''; # create and initialize $b
            rewrite_by_lua '
                ngx.var.b = tonumber(ngx.var.a) + 1
                if tonumber(ngx.var.b) == 13 then
                    return ngx.redirect("/bar");
                end
            ';

            echo "res = $b";
        }

    Note that the ngx_eval (<http://www.grid.net.ru/nginx/eval.en.html>)
    module can be approximated by using rewrite_by_lua. For example,

        location / {
            eval $res {
                proxy_pass http://foo.com/check-spam;
            }

            if ($res = 'spam') {
                rewrite ^ /terms-of-use.html redirect;
            }

            fastcgi_pass ...;
        }

    can be implemented in ngx_lua as:

        location = /check-spam {
            internal;
            proxy_pass http://foo.com/check-spam;
        }

        location / {
            rewrite_by_lua '
                local res = ngx.location.capture("/check-spam")
                if res.body == "spam" then
                    return ngx.redirect("/terms-of-use.html")
                end
            ';

            fastcgi_pass ...;
        }

    Just as any other rewrite phase handlers, rewrite_by_lua also runs in
    subrequests.

    Note that when calling "ngx.exit(ngx.OK)" within a rewrite_by_lua
    handler, the nginx request processing control flow will still continue
    to the content handler. To terminate the current request from within a
    rewrite_by_lua handler, calling ngx.exit with status >= 200
    ("ngx.HTTP_OK") and status < 300 ("ngx.HTTP_SPECIAL_RESPONSE") for
    successful quits and "ngx.exit(ngx.HTTP_INTERNAL_SERVER_ERROR)" (or its
    friends) for failures.

    If the [[HttpRewriteModule]]'s rewrite directive is used to change the
    URI and initiate location re-lookups (internal redirections), then any
    rewrite_by_lua or rewrite_by_lua_file code sequences within the current
    location will not be executed. For example,

        location /foo {
            rewrite ^ /bar;
            rewrite_by_lua 'ngx.exit(503)';
        }
        location /bar {
            ...
        }

    Here the Lua code "ngx.exit(503)" will never run. This will be the case
    if "rewrite ^ /bar last" is used as this will similarly initiate an
    internal redirection. If the "break" modifier is used instead, there
    will be no internal redirection and the "rewrite_by_lua" code will be
    executed.

    The "rewrite_by_lua" code will always run at the end of the "rewrite"
    request-processing phase unless rewrite_by_lua_no_postpone is turned on.

  rewrite_by_lua_file
    syntax: *rewrite_by_lua_file <path-to-lua-script-file>*

    context: *http, server, location, location if*

    phase: *rewrite tail*

    Equivalent to rewrite_by_lua, except that the file specified by
    "<path-to-lua-script-file>" contains the Lua code, or, as from the
    "v0.5.0rc32" release, the Lua/LuaJIT bytecode to be executed.

    Nginx variables can be used in the "<path-to-lua-script-file>" string to
    provide flexibility. This however carries some risks and is not
    ordinarily recommended.

    When a relative path like "foo/bar.lua" is given, they will be turned
    into the absolute path relative to the "server prefix" path determined
    by the "-p PATH" command-line option while starting the Nginx server.

    When the Lua code cache is turned on (by default), the user code is
    loaded once at the first request and cached and the Nginx config must be
    reloaded each time the Lua source file is modified. The Lua code cache
    can be temporarily disabled during development by switching
    lua_code_cache "off" in "nginx.conf" to avoid reloading Nginx.

    The "rewrite_by_lua_file" code will always run at the end of the
    "rewrite" request-processing phase unless rewrite_by_lua_no_postpone is
    turned on.

  access_by_lua
    syntax: *access_by_lua <lua-script-str>*

    context: *http, server, location, location if*

    phase: *access tail*

    Acts as an access phase handler and executes Lua code string specified
    in "<lua-script-str>" for every request. The Lua code may make API calls
    and is executed as a new spawned coroutine in an independent global
    environment (i.e. a sandbox).

    Note that this handler always runs *after* the standard
    [[HttpAccessModule]]. So the following will work as expected:

        location / {
            deny    192.168.1.1;
            allow   192.168.1.0/24;
            allow   10.1.1.0/16;
            deny    all;

            access_by_lua '
                local res = ngx.location.capture("/mysql", { ... })
                ...
            ';

            # proxy_pass/fastcgi_pass/...
        }

    That is, if a client IP address is in the blacklist, it will be denied
    before the MySQL query for more complex authentication is executed by
    access_by_lua.

    Note that the ngx_auth_request
    (<http://mdounin.ru/hg/ngx_http_auth_request_module/>) module can be
    approximated by using access_by_lua:

        location / {
            auth_request /auth;

            # proxy_pass/fastcgi_pass/postgres_pass/...
        }

    can be implemented in ngx_lua as:

        location / {
            access_by_lua '
                local res = ngx.location.capture("/auth")

                if res.status == ngx.HTTP_OK then
                    return
                end

                if res.status == ngx.HTTP_FORBIDDEN then
                    ngx.exit(res.status)
                end

                ngx.exit(ngx.HTTP_INTERNAL_SERVER_ERROR)
            ';

            # proxy_pass/fastcgi_pass/postgres_pass/...
        }

    As with other access phase handlers, access_by_lua will *not* run in
    subrequests.

    Note that when calling "ngx.exit(ngx.OK)" within a access_by_lua
    handler, the nginx request processing control flow will still continue
    to the content handler. To terminate the current request from within a
    access_by_lua handler, calling ngx.exit with status >= 200
    ("ngx.HTTP_OK") and status < 300 ("ngx.HTTP_SPECIAL_RESPONSE") for
    successful quits and "ngx.exit(ngx.HTTP_INTERNAL_SERVER_ERROR)" (or its
    friends) for failures.

  access_by_lua_file
    syntax: *access_by_lua_file <path-to-lua-script-file>*

    context: *http, server, location, location if*

    phase: *access tail*

    Equivalent to access_by_lua, except that the file specified by
    "<path-to-lua-script-file>" contains the Lua code, or, as from the
    "v0.5.0rc32" release, the Lua/LuaJIT bytecode to be executed.

    Nginx variables can be used in the "<path-to-lua-script-file>" string to
    provide flexibility. This however carries some risks and is not
    ordinarily recommended.

    When a relative path like "foo/bar.lua" is given, they will be turned
    into the absolute path relative to the "server prefix" path determined
    by the "-p PATH" command-line option while starting the Nginx server.

    When the Lua code cache is turned on (by default), the user code is
    loaded once at the first request and cached and the Nginx config must be
    reloaded each time the Lua source file is modified. The Lua code cache
    can be temporarily disabled during development by switching
    lua_code_cache "off" in "nginx.conf" to avoid repeatedly reloading
    Nginx.

  header_filter_by_lua
    syntax: *header_filter_by_lua <lua-script-str>*

    context: *http, server, location, location if*

    phase: *output-header-filter*

    Uses Lua code specified in "<lua-script-str>" to define an output header
    filter.

    Note that the following API functions are currently disabled within this
    context:

    *   Output API functions (e.g., ngx.say and ngx.send_headers)

    *   Control API functions (e.g., ngx.exit and ngx.exec)

    *   Subrequest API functions (e.g., ngx.location.capture and
        ngx.location.capture_multi)

    *   Cosocket API functions (e.g., ngx.socket.tcp and ngx.req.socket).

    Here is an example of overriding a response header (or adding one if
    absent) in our Lua header filter:

        location / {
            proxy_pass http://mybackend;
            header_filter_by_lua 'ngx.header.Foo = "blah"';
        }

    This directive was first introduced in the "v0.2.1rc20" release.

  header_filter_by_lua_file
    syntax: *header_filter_by_lua_file <path-to-lua-script-file>*

    context: *http, server, location, location if*

    phase: *output-header-filter*

    Equivalent to header_filter_by_lua, except that the file specified by
    "<path-to-lua-script-file>" contains the Lua code, or as from the
    "v0.5.0rc32" release, the Lua/LuaJIT bytecode to be executed.

    When a relative path like "foo/bar.lua" is given, they will be turned
    into the absolute path relative to the "server prefix" path determined
    by the "-p PATH" command-line option while starting the Nginx server.

    This directive was first introduced in the "v0.2.1rc20" release.

  body_filter_by_lua
    syntax: *body_filter_by_lua <lua-script-str>*

    context: *http, server, location, location if*

    phase: *output-body-filter*

    Uses Lua code specified in "<lua-script-str>" to define an output body
    filter.

    The input data chunk is passed via ngx.arg[1] (as a Lua string value)
    and the "eof" flag indicating the end of the response body data stream
    is passed via ngx.arg[2] (as a Lua boolean value).

    Behind the scene, the "eof" flag is just the "last_buf" (for main
    requests) or "last_in_chain" (for subrequests) flag of the Nginx chain
    link buffers. (Before the "v0.7.14" release, the "eof" flag does not
    work at all in subrequests.)

    The output data stream can be aborted immediately by running the
    following Lua statement:

        return ngx.ERROR

    This will truncate the response body and usually result in incomplete
    and also invalid responses.

    The Lua code can pass its own modified version of the input data chunk
    to the downstream Nginx output body filters by overriding ngx.arg[1]
    with a Lua string or a Lua table of strings. For example, to transform
    all the lowercase letters in the response body, we can just write:

        location / {
            proxy_pass http://mybackend;
            body_filter_by_lua 'ngx.arg[1] = string.upper(ngx.arg[1])';
        }

    When setting "nil" or an empty Lua string value to "ngx.arg[1]", no data
    chunk will be passed to the downstream Nginx output filters at all.

    Likewise, new "eof" flag can also be specified by setting a boolean
    value to ngx.arg[2]. For example,

        location /t {
            echo hello world;
            echo hiya globe;

            body_filter_by_lua '
                local chunk = ngx.arg[1]
                if string.match(chunk, "hello") then
                    ngx.arg[2] = true  -- new eof
                    return
                end

                -- just throw away any remaining chunk data
                ngx.arg[1] = nil
            ';
        }

    Then "GET /t" will just return the output

        hello world

    That is, when the body filter sees a chunk containing the word "hello",
    then it will set the "eof" flag to true immediately, resulting in
    truncated but still valid responses.

    When the Lua code may change the length of the response body, then it is
    required to always clear out the "Content-Length" response header (if
    any) in a header filter to enforce streaming output, as in

        location /foo {
            # fastcgi_pass/proxy_pass/...

            header_filter_by_lua 'ngx.header.content_length = nil';
            body_filter_by_lua 'ngx.arg[1] = {string.len(arg[1]), "\n"}'
        }

    Note that the following API functions are currently disabled within this
    context:

    *   Output API functions (e.g., ngx.say and ngx.send_headers)

    *   Control API functions (e.g., ngx.exit and ngx.exec)

    *   Subrequest API functions (e.g., ngx.location.capture and
        ngx.location.capture_multi)

    *   Cosocket API functions (e.g., ngx.socket.tcp and ngx.req.socket).

    Nginx output filters may be called multiple times for a single request
    because response body may be delivered in chunks. Thus, the Lua code
    specified by in this directive may also run multiple times in the
    lifetime of a single HTTP request.

    This directive was first introduced in the "v0.5.0rc32" release.

  body_filter_by_lua_file
    syntax: *body_filter_by_lua_file <path-to-lua-script-file>*

    context: *http, server, location, location if*

    phase: *output-body-filter*

    Equivalent to body_filter_by_lua, except that the file specified by
    "<path-to-lua-script-file>" contains the Lua code, or, as from the
    "v0.5.0rc32" release, the Lua/LuaJIT bytecode to be executed.

    When a relative path like "foo/bar.lua" is given, they will be turned
    into the absolute path relative to the "server prefix" path determined
    by the "-p PATH" command-line option while starting the Nginx server.

    This directive was first introduced in the "v0.5.0rc32" release.

  log_by_lua
    syntax: *log_by_lua <lua-script-str>*

    context: *http, server, location, location if*

    phase: *log*

    Run the Lua source code inlined as the "<lua-script-str>" at the "log"
    request processing phase. This does not replace the current access logs,
    but runs after.

    Note that the following API functions are currently disabled within this
    context:

    *   Output API functions (e.g., ngx.say and ngx.send_headers)

    *   Control API functions (e.g., ngx.exit)

    *   Subrequest API functions (e.g., ngx.location.capture and
        ngx.location.capture_multi)

    *   Cosocket API functions (e.g., ngx.socket.tcp and ngx.req.socket).

    Here is an example of gathering average data for
    $upstream_response_time:

        lua_shared_dict log_dict 5M;

        server {
            location / {
                proxy_pass http://mybackend;

                log_by_lua '
                    local log_dict = ngx.shared.log_dict
                    local upstream_time = tonumber(ngx.var.upstream_response_time)

                    local sum = log_dict:get("upstream_time-sum") or 0
                    sum = sum + upstream_time
                    log_dict:set("upstream_time-sum", sum)

                    local newval, err = log_dict:incr("upstream_time-nb", 1)
                    if not newval and err == "not found" then
                        log_dict:add("upstream_time-nb", 0)
                        log_dict:incr("upstream_time-nb", 1)
                    end
                ';
            }

            location = /status {
                content_by_lua '
                    local log_dict = ngx.shared.log_dict
                    local sum = log_dict:get("upstream_time-sum")
                    local nb = log_dict:get("upstream_time-nb")

                    if nb and sum then
                        ngx.say("average upstream response time: ", sum / nb,
                                " (", nb, " reqs)")
                    else
                        ngx.say("no data yet")
                    end
                ';
            }
        }

    This directive was first introduced in the "v0.5.0rc31" release.

  log_by_lua_file
    syntax: *log_by_lua_file <path-to-lua-script-file>*

    context: *http, server, location, location if*

    phase: *log*

    Equivalent to log_by_lua, except that the file specified by
    "<path-to-lua-script-file>" contains the Lua code, or, as from the
    "v0.5.0rc32" release, the Lua/LuaJIT bytecode to be executed.

    When a relative path like "foo/bar.lua" is given, they will be turned
    into the absolute path relative to the "server prefix" path determined
    by the "-p PATH" command-line option while starting the Nginx server.

    This directive was first introduced in the "v0.5.0rc31" release.

  lua_need_request_body
    syntax: *lua_need_request_body <on|off>*

    default: *off*

    context: *main | server | location*

    phase: *depends on usage*

    Determines whether to force the request body data to be read before
    running rewrite/access/access_by_lua* or not. The Nginx core does not
    read the client request body by default and if request body data is
    required, then this directive should be turned "on" or the
    ngx.req.read_body function should be called within the Lua code.

    To read the request body data within the $request_body variable,
    client_body_buffer_size must have the same value as
    client_max_body_size. Because when the content length exceeds
    client_body_buffer_size but less than client_max_body_size, Nginx will
    buffer the data into a temporary file on the disk, which will lead to
    empty value in the $request_body variable.

    If the current location includes rewrite_by_lua or rewrite_by_lua_file
    directives, then the request body will be read just before the
    rewrite_by_lua or rewrite_by_lua_file code is run (and also at the
    "rewrite" phase). Similarly, if only content_by_lua is specified, the
    request body will not be read until the content handler's Lua code is
    about to run (i.e., the request body will be read during the content
    phase).

    It is recommended however, to use the ngx.req.read_body and
    ngx.req.discard_body functions for finer control over the request body
    reading process instead.

    This also applies to access_by_lua and access_by_lua_file.

  lua_shared_dict
    syntax: *lua_shared_dict <name> <size>*

    default: *no*

    context: *http*

    phase: *depends on usage*

    Declares a shared memory zone, "<name>", to serve as storage for the shm
    based Lua dictionary "ngx.shared.<name>".

    The "<size>" argument accepts size units such as "k" and "m":

        http {
            lua_shared_dict dogs 10m;
            ...
        }

    See ngx.shared.DICT for details.

    This directive was first introduced in the "v0.3.1rc22" release.

  lua_socket_connect_timeout
    syntax: *lua_socket_connect_timeout <time>*

    default: *lua_socket_connect_timeout 60s*

    context: *http, server, location*

    This directive controls the default timeout value used in
    TCP/unix-domain socket object's connect method and can be overridden by
    the settimeout method.

    The "<time>" argument can be an integer, with an optional time unit,
    like "s" (second), "ms" (millisecond), "m" (minute). The default time
    unit is "s", i.e., "second". The default setting is "60s".

    This directive was first introduced in the "v0.5.0rc1" release.

  lua_socket_send_timeout
    syntax: *lua_socket_send_timeout <time>*

    default: *lua_socket_send_timeout 60s*

    context: *http, server, location*

    Controls the default timeout value used in TCP/unix-domain socket
    object's send method and can be overridden by the settimeout method.

    The "<time>" argument can be an integer, with an optional time unit,
    like "s" (second), "ms" (millisecond), "m" (minute). The default time
    unit is "s", i.e., "second". The default setting is "60s".

    This directive was first introduced in the "v0.5.0rc1" release.

  lua_socket_send_lowat
    syntax: *lua_socket_send_lowat <size>*

    default: *lua_socket_send_lowat 0*

    context: *http, server, location*

    Controls the "lowat" (low water) value for the cosocket send buffer.

  lua_socket_read_timeout
    syntax: *lua_socket_read_timeout <time>*

    default: *lua_socket_read_timeout 60s*

    context: *http, server, location*

    phase: *depends on usage*

    This directive controls the default timeout value used in
    TCP/unix-domain socket object's receive method and iterator functions
    returned by the receiveuntil method. This setting can be overridden by
    the settimeout method.

    The "<time>" argument can be an integer, with an optional time unit,
    like "s" (second), "ms" (millisecond), "m" (minute). The default time
    unit is "s", i.e., "second". The default setting is "60s".

    This directive was first introduced in the "v0.5.0rc1" release.

  lua_socket_buffer_size
    syntax: *lua_socket_buffer_size <size>*

    default: *lua_socket_buffer_size 4k/8k*

    context: *http, server, location*

    Specifies the buffer size used by cosocket reading operations.

    This buffer does not have to be that big to hold everything at the same
    time because cosocket supports 100% non-buffered reading and parsing. So
    even 1 byte buffer size should still work everywhere but the performance
    could be terrible.

    This directive was first introduced in the "v0.5.0rc1" release.

  lua_socket_pool_size
    syntax: *lua_socket_pool_size <size>*

    default: *lua_socket_pool_size 30*

    context: *http, server, location*

    Specifies the size limit (in terms of connection count) for every
    cosocket connection pool associated with every remote server (i.e.,
    identified by either the host-port pair or the unix domain socket file
    path).

    Default to 30 connections for every pool.

    When the connection pool exceeds the available size limit, the least
    recently used (idle) connection already in the pool will be closed to
    make room for the current connection.

    Note that the cosocket connection pool is per nginx worker process
    rather than per nginx server instance, so so size limit specified here
    also applies to every single nginx worker process.

    This directive was first introduced in the "v0.5.0rc1" release.

  lua_socket_keepalive_timeout
    syntax: *lua_socket_keepalive_timeout <time>*

    default: *lua_socket_keepalive_timeout 60s*

    context: *http, server, location*

    This directive controls the default maximal idle time of the connections
    in the cosocket built-in connection pool. When this timeout reaches,
    idle connections will be closed and removed from the pool. This setting
    can be overridden by cosocket objects' setkeepalive method.

    The "<time>" argument can be an integer, with an optional time unit,
    like "s" (second), "ms" (millisecond), "m" (minute). The default time
    unit is "s", i.e., "second". The default setting is "60s".

    This directive was first introduced in the "v0.5.0rc1" release.

  lua_socket_log_errors
    syntax: *lua_socket_log_errors on|off*

    default: *lua_socket_log_errors on*

    context: *http, server, location*

    This directive can be used to toggle error logging when a failure occurs
    for the TCP or UDP cosockets. If you are already doing proper error
    handling and logging in your Lua code, then it is recommended to turn
    this directive off to prevent data flushing in your nginx error log
    files (which is usually rather expensive).

    This directive was first introduced in the "v0.5.13" release.

  lua_http10_buffering
    syntax: *lua_http10_buffering on|off*

    default: *lua_http10_buffering on*

    context: *http, server, location, location-if*

    Enables or disables automatic response buffering for HTTP 1.0 (or older)
    requests. This buffering mechanism is mainly used for HTTP 1.0
    keep-alive which replies on a proper "Content-Length" response header.

    If the Lua code explicitly sets a "Content-Length" response header
    before sending the headers (either explicitly via ngx.send_headers or
    implicitly via the first ngx.say or ngx.print call), then the HTTP 1.0
    response buffering will be disabled even when this directive is turned
    on.

    To output very large response data in a streaming fashion (via the
    ngx.flush call, for example), this directive MUST be turned off to
    minimize memory usage.

    This directive is turned "on" by default.

    This directive was first introduced in the "v0.5.0rc19" release.

  rewrite_by_lua_no_postpone
    syntax: *rewrite_by_lua_no_postpone on|off*

    default: *rewrite_by_lua_no_postpone off*

    context: *http*

    Controls whether or not to disable postponing rewrite_by_lua and
    rewrite_by_lua_file directives to run at the end of the "rewrite"
    request-processing phase. By default, this directive is turned off and
    the Lua code is postponed to run at the end of the "rewrite" phase.

    This directive was first introduced in the "v0.5.0rc29" release.

  lua_transform_underscores_in_response_headers
    syntax: *lua_transform_underscores_in_response_headers on|off*

    default: *lua_transform_underscores_in_response_headers on*

    context: *http, server, location, location-if*

    Controls whether to transform underscores ("_") in the response header
    names specified in the ngx.header.HEADER API to hypens ("-").

    This directive was first introduced in the "v0.5.0rc32" release.

  lua_check_client_abort
    syntax: *lua_check_client_abort on|off*

    default: *lua_check_client_abort off*

    context: *http, server, location, location-if*

    This directive controls whether to check for premature client connection
    abortion.

    When this directive is turned on, the ngx_lua module will monitor the
    premature connection close event on the downstream connections. And when
    there is such an event, it will call the user Lua function callback
    (registered by ngx.on_abort) or just stop and clean up all the Lua
    "light threads" running in the current request's request handler when
    there is no user callback function registered.

    According to the current implementation, however, if the client closes
    the connection before the Lua code finishes reading the request body
    data via ngx.req.socket, then ngx_lua will neither stop all the running
    "light threads" nor call the user callback (if ngx.on_abort has been
    called). Instead, the reading operation on ngx.req.socket will just
    return the error message "client aborted" as the second return value
    (the first return value is surely "nil").

    When TCP keepalive is disabled, it is relying on the client side to
    close the socket gracefully (by sending a "FIN" packet or something like
    that). For (soft) real-time web applications, it is highly recommended
    to configure the TCP keepalive
    (<http://tldp.org/HOWTO/TCP-Keepalive-HOWTO/overview.html>) support in
    your system's TCP stack implementation in order to detect "half-open"
    TCP connections in time.

    For example, on Linux, you can configure the standard listen directive
    in your "nginx.conf" file like this:

        listen 80 so_keepalive=2s:2s:8;

    On FreeBSD, you can only tune the system-wide configuration for TCP
    keepalive, for example:

        # sysctl net.inet.tcp.keepintvl=2000
        # sysctl net.inet.tcp.keepidle=2000

    This directive was first introduced in the "v0.7.4" release.

    See also ngx.on_abort.

  lua_max_pending_timers
    syntax: *lua_max_pending_timers <count>*

    default: *lua_max_pending_timers 1024*

    context: *http*

    Controls the maximum number of pending timers allowed.

    Pending timers are those timers that have not expired yet.

    When exceeding this limit, the ngx.timer.at call will immediately return
    "nil" and the error string "too many pending timers".

    This directive was first introduced in the "v0.8.0" release.

  lua_max_running_timers
    syntax: *lua_max_running_timers <count>*

    default: *lua_max_running_timers 256*

    context: *http*

    Controls the maximum number of "running timers" allowed.

    Running timers are those timers whose user callback functions are still
    running.

    When exceeding this limit, Nginx will stop running the callbacks of
    newly expired timers and log an error message "N lua_max_running_timers
    are not enough" where "N" is the current value of this directive.

    This directive was first introduced in the "v0.8.0" release.

Nginx API for Lua
  Introduction
    The various *_by_lua and *_by_lua_file configuration directives serve as
    gateways to the Lua API within the "nginx.conf" file. The Nginx Lua API
    described below can only be called within the user Lua code run in the
    context of these configuration directives.

    The API is exposed to Lua in the form of two standard packages "ngx" and
    "ndk". These packages are in the default global scope within ngx_lua and
    are always available within ngx_lua directives.

    The packages can be introduced into external Lua modules like this:

        local say = ngx.say

        module(...)

        function foo(a) 
            say(a) 
        end

    Use of the package.seeall
    (<http://www.lua.org/manual/5.1/manual.html#pdf-package.seeall>) flag is
    strongly discouraged due to its various bad side-effects.

    It is also possible to directly require the packages in external Lua
    modules:

        local ngx = require "ngx"
        local ndk = require "ndk"

    The ability to require these packages was introduced in the "v0.2.1rc19"
    release.

    Network I/O operations in user code should only be done through the
    Nginx Lua API calls as the Nginx event loop may be blocked and
    performance drop off dramatically otherwise. Disk operations with
    relatively small amount of data can be done using the standard Lua "io"
    library but huge file reading and writing should be avoided wherever
    possible as they may block the Nginx process significantly. Delegating
    all network and disk I/O operations to Nginx's subrequests (via the
    ngx.location.capture method and similar) is strongly recommended for
    maximum performance.

  ngx.arg
    syntax: *val = ngx.arg[index]*

    context: *set_by_lua*, body_filter_by_lua**

    When this is used in the context of the set_by_lua or set_by_lua_file
    directives, this table is read-only and holds the input arguments to the
    config directives:

        value = ngx.arg[n]

    Here is an example

        location /foo {
            set $a 32;
            set $b 56;

            set_by_lua $res
                'return tonumber(ngx.arg[1]) + tonumber(ngx.arg[2])'
                $a $b;

            echo $sum;
        }

    that writes out 88, the sum of 32 and 56.

    When this table is used in the context of body_filter_by_lua or
    body_filter_by_lua_file, the first element holds the input data chunk to
    the output filter code and the second element holds the boolean flag for
    the "eof" flag indicating the end of the whole output data stream.

    The data chunk and "eof" flag passed to the downstream Nginx output
    filters can also be overridden by assigning values directly to the
    corresponding table elements. When setting "nil" or an empty Lua string
    value to "ngx.arg[1]", no data chunk will be passed to the downstream
    Nginx output filters at all.

  ngx.var.VARIABLE
    syntax: *ngx.var.VAR_NAME*

    context: *set_by_lua*, rewrite_by_lua*, access_by_lua*, content_by_lua*,
    header_filter_by_lua*, body_filter_by_lua*, log_by_lua**

    Read and write Nginx variable values.

        value = ngx.var.some_nginx_variable_name
        ngx.var.some_nginx_variable_name = value

    Note that only already defined nginx variables can be written to. For
    example:

        location /foo {
            set $my_var ''; # this line is required to create $my_var at config time
            content_by_lua '
                ngx.var.my_var = 123;
                ...
            ';
        }

    That is, nginx variables cannot be created on-the-fly.

    Some special nginx variables like $args and $limit_rate can be assigned
    a value, some are not, like $arg_PARAMETER.

    Nginx regex group capturing variables $1, $2, $3, and etc, can be read
    by this interface as well, by writing "ngx.var[1]", "ngx.var[2]",
    "ngx.var[3]", and etc.

    Setting "ngx.var.Foo" to a "nil" value will unset the $Foo Nginx
    variable.

        ngx.var.args = nil

    WARNING When reading from an Nginx variable, Nginx will allocate memory
    in the per-request memory pool which is freed only at request
    termination. So when you need to read from an Nginx variable repeatedly
    in your Lua code, cache the Nginx variable value to your own Lua
    variable, for example,

        local val = ngx.var.some_var
        --- use the val repeatedly later

    to prevent (temporary) memory leaking within the current request's
    lifetime.

  Core constants
    context: *init_by_lua*, set_by_lua*, rewrite_by_lua*, access_by_lua*,
    content_by_lua*, header_filter_by_lua*, body_filter_by_lua,
    *log_by_lua*, ngx.timer.**

      ngx.OK (0)
      ngx.ERROR (-1)
      ngx.AGAIN (-2)
      ngx.DONE (-4)
      ngx.DECLINED (-5)

    Note that only three of these constants are utilized by the Nginx API
    for Lua (i.e., ngx.exit accepts "NGX_OK", "NGX_ERROR", and
    "NGX_DECLINED" as input).

      ngx.null

    The "ngx.null" constant is a "NULL" light userdata usually used to
    represent nil values in Lua tables etc and is similar to the lua-cjson
    (<http://www.kyne.com.au/~mark/software/lua-cjson.php>) library's
    "cjson.null" constant. This constant was first introduced in the
    "v0.5.0rc5" release.

    The "ngx.DECLINED" constant was first introduced in the "v0.5.0rc19"
    release.

  HTTP method constants
    context: *init_by_lua*, set_by_lua*, rewrite_by_lua*, access_by_lua*,
    content_by_lua*, header_filter_by_lua*, body_filter_by_lua, log_by_lua*,
    ngx.timer.**

      ngx.HTTP_GET
      ngx.HTTP_HEAD
      ngx.HTTP_PUT
      ngx.HTTP_POST
      ngx.HTTP_DELETE
      ngx.HTTP_OPTIONS   (added in the v0.5.0rc24 release)
      ngx.HTTP_MKCOL     (added in the v0.8.2 release)
      ngx.HTTP_COPY      (added in the v0.8.2 release)
      ngx.HTTP_MOVE      (added in the v0.8.2 release)
      ngx.HTTP_PROPFIND  (added in the v0.8.2 release)
      ngx.HTTP_PROPPATCH (added in the v0.8.2 release)
      ngx.HTTP_LOCK      (added in the v0.8.2 release)
      ngx.HTTP_UNLOCK    (added in the v0.8.2 release)
      ngx.HTTP_PATCH     (added in the v0.8.2 release)
      ngx.HTTP_TRACE     (added in the v0.8.2 release)

    These constants are usually used in ngx.location.capture and
    ngx.location.capture_multi method calls.

  HTTP status constants
    context: *init_by_lua*, set_by_lua*, rewrite_by_lua*, access_by_lua*,
    content_by_lua*, header_filter_by_lua*, body_filter_by_lua, log_by_lua*,
    ngx.timer.**

      value = ngx.HTTP_OK (200)
      value = ngx.HTTP_CREATED (201)
      value = ngx.HTTP_SPECIAL_RESPONSE (300)
      value = ngx.HTTP_MOVED_PERMANENTLY (301)
      value = ngx.HTTP_MOVED_TEMPORARILY (302)
      value = ngx.HTTP_SEE_OTHER (303)
      value = ngx.HTTP_NOT_MODIFIED (304)
      value = ngx.HTTP_BAD_REQUEST (400)
      value = ngx.HTTP_UNAUTHORIZED (401)
      value = ngx.HTTP_FORBIDDEN (403)
      value = ngx.HTTP_NOT_FOUND (404)
      value = ngx.HTTP_NOT_ALLOWED (405)
      value = ngx.HTTP_GONE (410)
      value = ngx.HTTP_INTERNAL_SERVER_ERROR (500)
      value = ngx.HTTP_METHOD_NOT_IMPLEMENTED (501)
      value = ngx.HTTP_SERVICE_UNAVAILABLE (503)
      value = ngx.HTTP_GATEWAY_TIMEOUT (504) (first added in the v0.3.1rc38 release)

  Nginx log level constants
    context: *set_by_lua*, rewrite_by_lua*, access_by_lua*, content_by_lua*,
    header_filter_by_lua*, body_filter_by_lua, log_by_lua*, ngx.timer.**

      ngx.STDERR
      ngx.EMERG
      ngx.ALERT
      ngx.CRIT
      ngx.ERR
      ngx.WARN
      ngx.NOTICE
      ngx.INFO
      ngx.DEBUG

    These constants are usually used by the ngx.log method.

  print
    syntax: *print(...)*

    context: *init_by_lua*, set_by_lua*, rewrite_by_lua*, access_by_lua*,
    content_by_lua*, header_filter_by_lua*, body_filter_by_lua, log_by_lua*,
    ngx.timer.**

    Writes argument values into the nginx "error.log" file with the
    "ngx.NOTICE" log level.

    It is equivalent to

        ngx.log(ngx.NOTICE, ...)

    Lua "nil" arguments are accepted and result in literal "nil" strings
    while Lua booleans result in literal "true" or "false" strings. And the
    "ngx.null" constant will yield the "null" string output.

    There is a hard coded 2048 byte limitation on error message lengths in
    the Nginx core. This limit includes trailing newlines and leading time
    stamps. If the message size exceeds this limit, Nginx will truncate the
    message text accordingly. This limit can be manually modified by editing
    the "NGX_MAX_ERROR_STR" macro definition in the "src/core/ngx_log.h"
    file in the Nginx source tree.

  ngx.ctx
    context: *set_by_lua*, rewrite_by_lua*, access_by_lua*, content_by_lua*,
    header_filter_by_lua*, body_filter_by_lua, log_by_lua*, ngx.timer.**

    This table can be used to store per-request Lua context data and has a
    life time identical to the current request (as with the Nginx
    variables).

    Consider the following example,

        location /test {
            rewrite_by_lua '
                ngx.say("foo = ", ngx.ctx.foo)
                ngx.ctx.foo = 76
            ';
            access_by_lua '
                ngx.ctx.foo = ngx.ctx.foo + 3
            ';
            content_by_lua '
                ngx.say(ngx.ctx.foo)
            ';
        }

    Then "GET /test" will yield the output

        foo = nil
        79

    That is, the "ngx.ctx.foo" entry persists across the rewrite, access,
    and content phases of a request.

    Every request, including subrequests, has its own copy of the table. For
    example:

        location /sub {
            content_by_lua '
                ngx.say("sub pre: ", ngx.ctx.blah)
                ngx.ctx.blah = 32
                ngx.say("sub post: ", ngx.ctx.blah)
            ';
        }

        location /main {
            content_by_lua '
                ngx.ctx.blah = 73
                ngx.say("main pre: ", ngx.ctx.blah)
                local res = ngx.location.capture("/sub")
                ngx.print(res.body)
                ngx.say("main post: ", ngx.ctx.blah)
            ';
        }

    Then "GET /main" will give the output

        main pre: 73
        sub pre: nil
        sub post: 32
        main post: 73

    Here, modification of the "ngx.ctx.blah" entry in the subrequest does
    not affect the one in the parent request. This is because they have two
    separate versions of "ngx.ctx.blah".

    Internal redirection will destroy the original request "ngx.ctx" data
    (if any) and the new request will have an empty "ngx.ctx" table. For
    instance,

        location /new {
            content_by_lua '
                ngx.say(ngx.ctx.foo)
            ';
        }

        location /orig {
            content_by_lua '
                ngx.ctx.foo = "hello"
                ngx.exec("/new")
            ';
        }

    Then "GET /orig" will give

        nil

    rather than the original "hello" value.

    Arbitrary data values, including Lua closures and nested tables, can be
    inserted into this "magic" table. It also allows the registration of
    custom meta methods.

    Overriding "ngx.ctx" with a new Lua table is also supported, for
    example,

        ngx.ctx = { foo = 32, bar = 54 }

  ngx.location.capture
    syntax: *res = ngx.location.capture(uri, options?)*

    context: *rewrite_by_lua*, access_by_lua*, content_by_lua**

    Issue a synchronous but still non-blocking *Nginx Subrequest* using
    "uri".

    Nginx's subrequests provide a powerful way to make non-blocking internal
    requests to other locations configured with disk file directory or *any*
    other nginx C modules like "ngx_proxy", "ngx_fastcgi", "ngx_memc",
    "ngx_postgres", "ngx_drizzle", and even ngx_lua itself and etc etc etc.

    Also note that subrequests just mimic the HTTP interface but there is
    *no* extra HTTP/TCP traffic *nor* IPC involved. Everything works
    internally, efficiently, on the C level.

    Subrequests are completely different from HTTP 301/302 redirection (via
    ngx.redirect) and internal redirection (via ngx.exec).

    Here is a basic example:

        res = ngx.location.capture(uri)

    Returns a Lua table with three slots ("res.status", "res.header", and
    "res.body").

    "res.header" holds all the response headers of the subrequest and it is
    a normal Lua table. For multi-value response headers, the value is a Lua
    (array) table that holds all the values in the order that they appear.
    For instance, if the subrequest response headers contain the following
    lines:

        Set-Cookie: a=3
        Set-Cookie: foo=bar
        Set-Cookie: baz=blah

    Then "res.header["Set-Cookie"]" will be evaluated to the table value
    "{"a=3", "foo=bar", "baz=blah"}".

    URI query strings can be concatenated to URI itself, for instance,

        res = ngx.location.capture('/foo/bar?a=3&b=4')

    Named locations like @foo are not allowed due to a limitation in the
    nginx core. Use normal locations combined with the "internal" directive
    to prepare internal-only locations.

    An optional option table can be fed as the second argument, which
    supports the options:

    *   "method" specify the subrequest's request method, which only accepts
        constants like "ngx.HTTP_POST". =item *

        "body" specify the subrequest's request body (string value only).
        =item *

        "args" specify the subrequest's URI query arguments (both string
        value and Lua tables are accepted) =item *

        "ctx" specify a Lua table to be the ngx.ctx table for the
        subrequest. It can be the current request's ngx.ctx table, which
        effectively makes the parent and its subrequest to share exactly the
        same context table. This option was first introduced in the
        "v0.3.1rc25" release. =item *

        "vars" take a Lua table which holds the values to set the specified
        Nginx variables in the subrequest as this option's value. This
        option was first introduced in the "v0.3.1rc31" release. =item *

        "copy_all_vars" specify whether to copy over all the Nginx variable
        values of the current request to the subrequest in question.
        modifications of the nginx variables in the subrequest will not
        affect the current (parent) request. This option was first
        introduced in the "v0.3.1rc31" release. =item *

        "share_all_vars" specify whether to share all the Nginx variables of
        the subrequest with the current (parent) request. modifications of
        the Nginx variables in the subrequest will affect the current
        (parent) request.

    Issuing a POST subrequest, for example, can be done as follows

        res = ngx.location.capture(
            '/foo/bar',
            { method = ngx.HTTP_POST, body = 'hello, world' }
        )

    See HTTP method constants methods other than POST. The "method" option
    is "ngx.HTTP_GET" by default.

    The "args" option can specify extra URI arguments, for instance,

        ngx.location.capture('/foo?a=1',
            { args = { b = 3, c = ':' } }
        )

    is equivalent to

        ngx.location.capture('/foo?a=1&b=3&c=%3a')

    that is, this method will escape argument keys and values according to
    URI rules and concatenate them together into a complete query string.
    The format for the Lua table passed as the "args" argument is identical
    to the format used in the ngx.encode_args method.

    The "args" option can also take plain query strings:

        ngx.location.capture('/foo?a=1',
            { args = 'b=3&c=%3a' } }
        )

    This is functionally identical to the previous examples.

    The "share_all_vars" option controls whether to share nginx variables
    among the current request and its subrequests. If this option is set to
    "true", then the current request and associated subrequests will share
    the same Nginx variable scope. Hence, changes to Nginx variables made by
    a subrequest will affect the current request.

    Care should be taken in using this option as variable scope sharing can
    have unexpected side effects. The "args", "vars", or "copy_all_vars"
    options are generally preferable instead.

    This option is set to "false" by default

        location /other {
            set $dog "$dog world";
            echo "$uri dog: $dog";
        }

        location /lua {
            set $dog 'hello';
            content_by_lua '
                res = ngx.location.capture("/other",
                    { share_all_vars = true });

                ngx.print(res.body)
                ngx.say(ngx.var.uri, ": ", ngx.var.dog)
            ';
        }

    Accessing location "/lua" gives

        /other dog: hello world
        /lua: hello world

    The "copy_all_vars" option provides a copy of the parent request's Nginx
    variables to subrequests when such subrequests are issued. Changes made
    to these variables by such subrequests will not affect the parent
    request or any other subrequests sharing the parent request's variables.

        location /other {
            set $dog "$dog world";
            echo "$uri dog: $dog";
        }

        location /lua {
            set $dog 'hello';
            content_by_lua '
                res = ngx.location.capture("/other",
                    { copy_all_vars = true });

                ngx.print(res.body)
                ngx.say(ngx.var.uri, ": ", ngx.var.dog)
            ';
        }

    Request "GET /lua" will give the output

        /other dog: hello world
        /lua: hello

    Note that if both "share_all_vars" and "copy_all_vars" are set to true,
    then "share_all_vars" takes precedence.

    In addition to the two settings above, it is possible to specify values
    for variables in the subrequest using the "vars" option. These variables
    are set after the sharing or copying of variables has been evaluated,
    and provides a more efficient method of passing specific values to a
    subrequest over encoding them as URL arguments and unescaping them in
    the Nginx config file.

        location /other {
            content_by_lua '
                ngx.say("dog = ", ngx.var.dog)
                ngx.say("cat = ", ngx.var.cat)
            ';
        }

        location /lua {
            set $dog '';
            set $cat '';
            content_by_lua '
                res = ngx.location.capture("/other",
                    { vars = { dog = "hello", cat = 32 }});

                ngx.print(res.body)
            ';
        }

    Accessing "/lua" will yield the output

        dog = hello
        cat = 32

    The "ctx" option can be used to specify a custom Lua table to serve as
    the ngx.ctx table for the subrequest.

        location /sub {
            content_by_lua '
                ngx.ctx.foo = "bar";
            ';
        }
        location /lua {
            content_by_lua '
                local ctx = {}
                res = ngx.location.capture("/sub", { ctx = ctx })

                ngx.say(ctx.foo);
                ngx.say(ngx.ctx.foo);
            ';
        }

    Then request "GET /lua" gives

        bar
        nil

    It is also possible to use this "ctx" option to share the same ngx.ctx
    table between the current (parent) request and the subrequest:

        location /sub {
            content_by_lua '
                ngx.ctx.foo = "bar";
            ';
        }
        location /lua {
            content_by_lua '
                res = ngx.location.capture("/sub", { ctx = ngx.ctx })
                ngx.say(ngx.ctx.foo);
            ';
        }

    Request "GET /lua" yields the output

        bar

    Note that subrequests issued by ngx.location.capture inherit all the
    request headers of the current request by default and that this may have
    unexpected side effects on the subrequest responses. For example, when
    using the standard "ngx_proxy" module to serve subrequests, an
    "Accept-Encoding: gzip" header in the main request may result in gzipped
    responses that cannot be handled properly in Lua code. Original request
    headers should be ignored by setting proxy_pass_request_headers to "off"
    in subrequest locations.

    When the "body" option is not specified, the "POST" and "PUT"
    subrequests will inherit the request bodies of the parent request (if
    any).

    There is a hard-coded upper limit on the number of concurrent
    subrequests possible for every main request. In older versions of Nginx,
    the limit was 50 concurrent subrequests and in more recent versions,
    Nginx "1.1.x" onwards, this was increased to 200 concurrent subrequests.
    When this limit is exceeded, the following error message is added to the
    "error.log" file:

        [error] 13983#0: *1 subrequests cycle while processing "/uri"

    The limit can be manually modified if required by editing the definition
    of the "NGX_HTTP_MAX_SUBREQUESTS" macro in the
    "nginx/src/http/ngx_http_request.h" file in the Nginx source tree.

    Please also refer to restrictions on capturing locations configured by
    subrequest directives of other modules.

  ngx.location.capture_multi
    syntax: *res1, res2, ... = ngx.location.capture_multi({ {uri, options?},
    {uri, options?}, ... })*

    context: *rewrite_by_lua*, access_by_lua*, content_by_lua**

    Just like ngx.location.capture, but supports multiple subrequests
    running in parallel.

    This function issues several parallel subrequests specified by the input
    table and returns their results in the same order. For example,

        res1, res2, res3 = ngx.location.capture_multi{
            { "/foo", { args = "a=3&b=4" } },
            { "/bar" },
            { "/baz", { method = ngx.HTTP_POST, body = "hello" } },
        }

        if res1.status == ngx.HTTP_OK then
            ...
        end

        if res2.body == "BLAH" then
            ...
        end

    This function will not return until all the subrequests terminate. The
    total latency is the longest latency of the individual subrequests
    rather than the sum.

    Lua tables can be used for both requests and responses when the number
    of subrequests to be issued is not known in advance:

        -- construct the requests table
        local reqs = {}
        table.insert(reqs, { "/mysql" })
        table.insert(reqs, { "/postgres" })
        table.insert(reqs, { "/redis" })
        table.insert(reqs, { "/memcached" })

        -- issue all the requests at once and wait until they all return
        local resps = { ngx.location.capture_multi(reqs) }

        -- loop over the responses table
        for i, resp in ipairs(resps) do
            -- process the response table "resp"
        end

    The ngx.location.capture function is just a special form of this
    function. Logically speaking, the ngx.location.capture can be
    implemented like this

        ngx.location.capture =
            function (uri, args)
                return ngx.location.capture_multi({ {uri, args} })
            end

    Please also refer to restrictions on capturing locations configured by
    subrequest directives of other modules.

  ngx.status
    context: *set_by_lua*, rewrite_by_lua*, access_by_lua*, content_by_lua*,
    header_filter_by_lua*, body_filter_by_lua, log_by_lua**

    Read and write the current request's response status. This should be
    called before sending out the response headers.

        ngx.status = ngx.HTTP_CREATED
        status = ngx.status

    Setting "ngx.status" after the response header is sent out has no effect
    but leaving an error message in your nginx's error log file:

        attempt to set ngx.status after sending out response headers

  ngx.header.HEADER
    syntax: *ngx.header.HEADER = VALUE*

    syntax: *value = ngx.header.HEADER*

    context: *rewrite_by_lua*, access_by_lua*, content_by_lua*,
    header_filter_by_lua*, body_filter_by_lua, log_by_lua**

    Set, add to, or clear the current request's "HEADER" response header
    that is to be sent.

    Underscores ("_") in the header names will be replaced by hyphens ("-")
    by default. This transformation can be turned off via the
    lua_transform_underscores_in_response_headers directive.

    The header names are matched case-insensitively.

        -- equivalent to ngx.header["Content-Type"] = 'text/plain'
        ngx.header.content_type = 'text/plain';

        ngx.header["X-My-Header"] = 'blah blah';

    Multi-value headers can be set this way:

        ngx.header['Set-Cookie'] = {'a=32; path=/', 'b=4; path=/'}

    will yield

        Set-Cookie: a=32; path=/
        Set-Cookie: b=4; path=/

    in the response headers.

    Only Lua tables are accepted (Only the last element in the table will
    take effect for standard headers such as "Content-Type" that only accept
    a single value).

        ngx.header.content_type = {'a', 'b'}

    is equivalent to

        ngx.header.content_type = 'b'

    Setting a slot to "nil" effectively removes it from the response
    headers:

        ngx.header["X-My-Header"] = nil;

    The same applies to assigning an empty table:

        ngx.header["X-My-Header"] = {};

    Setting "ngx.header.HEADER" after sending out response headers (either
    explicitly with ngx.send_headers or implicitly with ngx.print and
    similar) will throw out a Lua exception.

    Reading "ngx.header.HEADER" will return the value of the response header
    named "HEADER".

    Underscores ("_") in the header names will also be replaced by dashes
    ("-") and the header names will be matched case-insensitively. If the
    response header is not present at all, "nil" will be returned.

    This is particularly useful in the context of header_filter_by_lua and
    header_filter_by_lua_file, for example,

        location /test {
            set $footer '';

            proxy_pass http://some-backend;

            header_filter_by_lua '
                if ngx.header["X-My-Header"] == "blah" then
                    ngx.var.footer = "some value"
                end
            ';

            echo_after_body $footer;
        }

    For multi-value headers, all of the values of header will be collected
    in order and returned as a Lua table. For example, response headers

        Foo: bar
        Foo: baz

    will result in

        {"bar", "baz"}

    to be returned when reading "ngx.header.Foo".

    Note that "ngx.header" is not a normal Lua table and as such, it is not
    possible to iterate through it using the Lua "ipairs" function.

    For reading *request* headers, use the ngx.req.get_headers function
    instead.

  ngx.req.start_time
    syntax: *secs = ngx.req.start_time()*

    context: *set_by_lua*, rewrite_by_lua*, access_by_lua*, content_by_lua*,
    header_filter_by_lua*, body_filter_by_lua*, log_by_lua**

    Returns a floating-point number representing the timestamp (including
    milliseconds as the decimal part) when the current request was created.

    The following example emulates the $request_time variable value
    (provided by [[HttpLogModule]]) in pure Lua:

        local request_time = ngx.now() - ngx.req.start_time()

    This function was first introduced in the "v0.7.7" release.

    See also ngx.now and ngx.update_time.

  ngx.req.http_version
    syntax: *num = ngx.req.http_version()*

    context: *set_by_lua*, rewrite_by_lua*, access_by_lua*, content_by_lua*,
    header_filter_by_lua**

    Returns the HTTP version number for the current request as a Lua number.

    Current possible values are 1.0, 1.1, and 0.9. Returns "nil" for
    unrecognized values.

    This method was first introduced in the "v0.7.17" release.

  ngx.req.raw_header
    syntax: *str = ngx.req.raw_header(no_request_line?)*

    context: *set_by_lua*, rewrite_by_lua*, access_by_lua*, content_by_lua*,
    header_filter_by_lua**

    Returns the original raw HTTP protocol header received by the Nginx
    server.

    By default, the request line and trailing "CR LF" terminator will also
    be included. For example,

        ngx.print(ngx.req.raw_header())

    gives something like this:

        GET /t HTTP/1.1
        Host: localhost
        Connection: close
        Foo: bar

    You can specify the optional "no_request_line" argument as a "true"
    value to exclude the request line from the result. For example,

        ngx.print(ngx.req.raw_header(true))

    outputs something like this:

        Host: localhost
        Connection: close
        Foo: bar

    This method was first introduced in the "v0.7.17" release.

  ngx.req.get_method
    syntax: *method_name = ngx.req.get_method()*

    context: *set_by_lua*, rewrite_by_lua*, access_by_lua*, content_by_lua*,
    header_filter_by_lua**

    Retrieves the current request's request method name. Strings like "GET"
    and "POST" are returned instead of numerical method constants.

    If the current request is an Nginx subrequest, then the subrequest's
    method name will be returned.

    This method was first introduced in the "v0.5.6" release.

    See also ngx.req.set_method.

  ngx.req.set_method
    syntax: *ngx.req.set_method(method_id)*

    context: *set_by_lua*, rewrite_by_lua*, access_by_lua*, content_by_lua*,
    header_filter_by_lua**

    Overrides the current request's request method with the "request_id"
    argument. Currently only numerical method constants are supported, like
    "ngx.HTTP_POST" and "ngx.HTTP_GET".

    If the current request is an Nginx subrequest, then the subrequest's
    method will be overridden.

    This method was first introduced in the "v0.5.6" release.

    See also ngx.req.get_method.

  ngx.req.set_uri
    syntax: *ngx.req.set_uri(uri, jump?)*

    context: *set_by_lua*, rewrite_by_lua*, access_by_lua*, content_by_lua*,
    header_filter_by_lua*, body_filter_by_lua**

    Rewrite the current request's (parsed) URI by the "uri" argument. The
    "uri" argument must be a Lua string and cannot be of zero length, or a
    Lua exception will be thrown.

    The optional boolean "jump" argument can trigger location rematch (or
    location jump) as [[HttpRewriteModule]]'s rewrite directive, that is,
    when "jump" is "true" (default to "false"), this function will never
    return and it will tell Nginx to try re-searching locations with the new
    URI value at the later "post-rewrite" phase and jumping to the new
    location.

    Location jump will not be triggered otherwise, and only the current
    request's URI will be modified, which is also the default behavior. This
    function will return but with no returned values when the "jump"
    argument is "false" or absent altogether.

    For example, the following nginx config snippet

        rewrite ^ /foo last;

    can be coded in Lua like this:

        ngx.req.set_uri("/foo", true)

    Similarly, Nginx config

        rewrite ^ /foo break;

    can be coded in Lua as

        ngx.req.set_uri("/foo", false)

    or equivalently,

        ngx.req.set_uri("/foo")

    The "jump" can only be set to "true" in rewrite_by_lua and
    rewrite_by_lua_file. Use of jump in other contexts is prohibited and
    will throw out a Lua exception.

    A more sophisticated example involving regex substitutions is as follows

        location /test {
            rewrite_by_lua '
                local uri = ngx.re.sub(ngx.var.uri, "^/test/(.*)", "$1", "o")
                ngx.req.set_uri(uri)
            ';
            proxy_pass http://my_backend;
        }

    which is functionally equivalent to

        location /test {
            rewrite ^/test/(.*) /$1 break;
            proxy_pass http://my_backend;
        }

    Note that it is not possible to use this interface to rewrite URI
    arguments and that ngx.req.set_uri_args should be used for this instead.
    For instance, Nginx config

        rewrite ^ /foo?a=3? last;

    can be coded as

        ngx.req.set_uri_args("a=3")
        ngx.req.set_uri("/foo", true)

    or

        ngx.req.set_uri_args({a = 3})
        ngx.req.set_uri("/foo", true)

    This interface was first introduced in the "v0.3.1rc14" release.

  ngx.req.set_uri_args
    syntax: *ngx.req.set_uri_args(args)*

    context: *set_by_lua*, rewrite_by_lua*, access_by_lua*, content_by_lua*,
    header_filter_by_lua*, body_filter_by_lua**

    Rewrite the current request's URI query arguments by the "args"
    argument. The "args" argument can be either a Lua string, as in

        ngx.req.set_uri_args("a=3&b=hello%20world")

    or a Lua table holding the query arguments' key-value pairs, as in

        ngx.req.set_uri_args({ a = 3, b = "hello world" })

    where in the latter case, this method will escape argument keys and
    values according to the URI escaping rule.

    Multi-value arguments are also supported:

        ngx.req.set_uri_args({ a = 3, b = {5, 6} })

    which will result in a query string like "a=3&b=5&b=6".

    This interface was first introduced in the "v0.3.1rc13" release.

    See also ngx.req.set_uri.

  ngx.req.get_uri_args
    syntax: *args = ngx.req.get_uri_args(max_args?)*

    context: *set_by_lua*, rewrite_by_lua*, access_by_lua*, content_by_lua*,
    header_filter_by_lua*, body_filter_by_lua, log_by_lua**

    Returns a Lua table holding all the current request URL query arguments.

        location = /test {
            content_by_lua '
                local args = ngx.req.get_uri_args()
                for key, val in pairs(args) do
                    if type(val) == "table" then
                        ngx.say(key, ": ", table.concat(val, ", "))
                    else
                        ngx.say(key, ": ", val)
                    end
                end
            ';
        }

    Then "GET /test?foo=bar&bar=baz&bar=blah" will yield the response body

        foo: bar
        bar: baz, blah

    Multiple occurrences of an argument key will result in a table value
    holding all the values for that key in order.

    Keys and values are unescaped according to URI escaping rules. In the
    settings above, "GET /test?a%20b=1%61+2" will yield:

        a b: 1a 2

    Arguments without the "=<value>" parts are treated as boolean arguments.
    "GET /test?foo&bar" will yield:

        foo: true
        bar: true

    That is, they will take Lua boolean values "true". However, they are
    different from arguments taking empty string values. "GET
    /test?foo=&bar=" will give something like

        foo: 
        bar:

    Empty key arguments are discarded. "GET /test?=hello&=world" will yield
    an empty output for instance.

    Updating query arguments via the nginx variable $args (or "ngx.var.args"
    in Lua) at runtime is also supported:

        ngx.var.args = "a=3&b=42"
        local args = ngx.req.get_uri_args()

    Here the "args" table will always look like

        {a = 3, b = 42}

    regardless of the actual request query string.

    Note that a maximum of 100 request arguments are parsed by default
    (including those with the same name) and that additional request
    arguments are silently discarded to guard against potential denial of
    service attacks.

    However, the optional "max_args" function argument can be used to
    override this limit:

        local args = ngx.req.get_uri_args(10)

    This argument can be set to zero to remove the limit and to process all
    request arguments received:

        local args = ngx.req.get_uri_args(0)

    Removing the "max_args" cap is strongly discouraged.

  ngx.req.get_post_args
    syntax: *args, err = ngx.req.get_post_args(max_args?)*

    context: *rewrite_by_lua*, access_by_lua*, content_by_lua*,
    header_filter_by_lua*, body_filter_by_lua, log_by_lua**

    Returns a Lua table holding all the current request POST query arguments
    (of the MIME type "application/x-www-form-urlencoded"). Call
    ngx.req.read_body to read the request body first or turn on the
    lua_need_request_body directive to avoid errors.

        location = /test {
            content_by_lua '
                ngx.req.read_body()
                local args = ngx.req.get_post_args()
                if not args then
                    ngx.say("failed to get post args: ", err)
                    return
                end
                for key, val in pairs(args) do
                    if type(val) == "table" then
                        ngx.say(key, ": ", table.concat(val, ", "))
                    else
                        ngx.say(key, ": ", val)
                    end
                end
            ';
        }

    Then

        # Post request with the body 'foo=bar&bar=baz&bar=blah'
        $ curl --data 'foo=bar&bar=baz&bar=blah' localhost/test

    will yield the response body like

        foo: bar
        bar: baz, blah

    Multiple occurrences of an argument key will result in a table value
    holding all of the values for that key in order.

    Keys and values will be unescaped according to URI escaping rules.

    With the settings above,

        # POST request with body 'a%20b=1%61+2'
        $ curl -d 'a%20b=1%61+2' localhost/test

    will yield:

        a b: 1a 2

    Arguments without the "=<value>" parts are treated as boolean arguments.
    "GET /test?foo&bar" will yield:

        foo: true
        bar: true

    That is, they will take Lua boolean values "true". However, they are
    different from arguments taking empty string values. "POST /test" with
    request body "foo=&bar=" will return something like

        foo: 
        bar:

    Empty key arguments are discarded. "POST /test" with body
    "=hello&=world" will yield empty outputs for instance.

    Note that a maximum of 100 request arguments are parsed by default
    (including those with the same name) and that additional request
    arguments are silently discarded to guard against potential denial of
    service attacks.

    However, the optional "max_args" function argument can be used to
    override this limit:

        local args = ngx.req.get_post_args(10)

    This argument can be set to zero to remove the limit and to process all
    request arguments received:

        local args = ngx.req.get_post_args(0)

    Removing the "max_args" cap is strongly discouraged.

  ngx.req.get_headers
    syntax: *headers = ngx.req.get_headers(max_headers?, raw?)*

    context: *set_by_lua*, rewrite_by_lua*, access_by_lua*, content_by_lua*,
    header_filter_by_lua*, body_filter_by_lua, log_by_lua**

    Returns a Lua table holding all the current request headers.

        local h = ngx.req.get_headers()
        for k, v in pairs(h) do
            ...
        end

    To read an individual header:

        ngx.say("Host: ", ngx.req.get_headers()["Host"])

    Note that the ngx.var.HEADER API call, which uses core $http_HEADER
    variables, may be more preferable for reading individual request
    headers.

    For multiple instances of request headers such as:

        Foo: foo
        Foo: bar
        Foo: baz

    the value of "ngx.req.get_headers()["Foo"]" will be a Lua (array) table
    such as:

        {"foo", "bar", "baz"}

    Note that a maximum of 100 request headers are parsed by default
    (including those with the same name) and that additional request headers
    are silently discarded to guard against potential denial of service
    attacks.

    However, the optional "max_headers" function argument can be used to
    override this limit:

        local args = ngx.req.get_headers(10)

    This argument can be set to zero to remove the limit and to process all
    request headers received:

        local args = ngx.req.get_headers(0)

    Removing the "max_headers" cap is strongly discouraged.

    Since the 0.6.9 release, all the header names in the Lua table returned
    are converted to the pure lower-case form by default, unless the "raw"
    argument is set to "true" (default to "false").

    Also, by default, an "__index" metamethod is added to the resulting Lua
    table and will normalize the keys to a pure lowercase form with all
    underscores converted to dashes in case of a lookup miss. For example,
    if a request header "My-Foo-Header" is present, then the following
    invocations will all pick up the value of this header correctly:

        ngx.say(headers.my_foo_header)
        ngx.say(headers["My-Foo-Header"])
        ngx.say(headers["my-foo-header"])

    The "__index" metamethod will not be added when the "raw" argument is
    set to "true".

  ngx.req.set_header
    syntax: *ngx.req.set_header(header_name, header_value)*

    context: *set_by_lua*, rewrite_by_lua*, access_by_lua*, content_by_lua*,
    header_filter_by_lua*, body_filter_by_lua*

    Set the current request's request header named "header_name" to value
    "header_value", overriding any existing ones.

    By default, all the subrequests subsequently initiated by
    ngx.location.capture and ngx.location.capture_multi will inherit the new
    header.

    Here is an example of setting the "Content-Length" header:

        ngx.req.set_header("Content-Type", "text/css")

    The "header_value" can take an array list of values, for example,

        ngx.req.set_header("Foo", {"a", "abc"})

    will produce two new request headers:

        Foo: a
        Foo: abc

    and old "Foo" headers will be overridden if there is any.

    When the "header_value" argument is "nil", the request header will be
    removed. So

        ngx.req.set_header("X-Foo", nil)

    is equivalent to

        ngx.req.clear_header("X-Foo")

  ngx.req.clear_header
    syntax: *ngx.req.clear_header(header_name)*

    context: *set_by_lua*, rewrite_by_lua*, access_by_lua*, content_by_lua*,
    header_filter_by_lua*, body_filter_by_lua**

    Clear the current request's request header named "header_name". None of
    the current request's subrequests will be affected.

  ngx.req.read_body
    syntax: *ngx.req.read_body()*

    context: *rewrite_by_lua*, access_by_lua*, content_by_lua**

    Reads the client request body synchronously without blocking the Nginx
    event loop.

        ngx.req.read_body()
        local args = ngx.req.get_post_args()

    If the request body is already read previously by turning on
    lua_need_request_body or by using other modules, then this function does
    not run and returns immediately.

    If the request body has already been explicitly discarded, either by the
    ngx.req.discard_body function or other modules, this function does not
    run and returns immediately.

    In case of errors, such as connection errors while reading the data,
    this method will throw out a Lua exception *or* terminate the current
    request with a 500 status code immediately.

    The request body data read using this function can be retrieved later
    via ngx.req.get_body_data or, alternatively, the temporary file name for
    the body data cached to disk using ngx.req.get_body_file. This depends
    on

    1.  whether the current request body is already larger than the
        client_body_buffer_size,

    2.  and whether client_body_in_file_only has been switched on.

    In cases where current request may have a request body and the request
    body data is not required, The ngx.req.discard_body function must be
    used to explicitly discard the request body to avoid breaking things
    under HTTP 1.1 keepalive or HTTP 1.1 pipelining.

    This function was first introduced in the "v0.3.1rc17" release.

  ngx.req.discard_body
    syntax: *ngx.req.discard_body()*

    context: *rewrite_by_lua*, access_by_lua*, content_by_lua**

    Explicitly discard the request body, i.e., read the data on the
    connection and throw it away immediately. Please note that ignoring
    request body is not the right way to discard it, and that this function
    must be called to avoid breaking things under HTTP 1.1 keepalive or HTTP
    1.1 pipelining.

    This function is an asynchronous call and returns immediately.

    If the request body has already been read, this function does nothing
    and returns immediately.

    This function was first introduced in the "v0.3.1rc17" release.

    See also ngx.req.read_body.

  ngx.req.get_body_data
    syntax: *data = ngx.req.get_body_data()*

    context: *rewrite_by_lua*, access_by_lua*, content_by_lua**

    Retrieves in-memory request body data. It returns a Lua string rather
    than a Lua table holding all the parsed query arguments. Use the
    ngx.req.get_post_args function instead if a Lua table is required.

    This function returns "nil" if

    1.  the request body has not been read,

    2.  the request body has been read into disk temporary files,

    3.  or the request body has zero size.

    If the request body has not been read yet, call ngx.req.read_body first
    (or turned on lua_need_request_body to force this module to read the
    request body. This is not recommended however).

    If the request body has been read into disk files, try calling the
    ngx.req.get_body_file function instead.

    To force in-memory request bodies, try setting client_body_buffer_size
    to the same size value in client_max_body_size.

    Note that calling this function instead of using "ngx.var.request_body"
    or "ngx.var.echo_request_body" is more efficient because it can save one
    dynamic memory allocation and one data copy.

    This function was first introduced in the "v0.3.1rc17" release.

    See also ngx.req.get_body_file.

  ngx.req.get_body_file
    syntax: *file_name = ngx.req.get_body_file()*

    context: *rewrite_by_lua*, access_by_lua*, content_by_lua**

    Retrieves the file name for the in-file request body data. Returns "nil"
    if the request body has not been read or has been read into memory.

    The returned file is read only and is usually cleaned up by Nginx's
    memory pool. It should not be manually modified, renamed, or removed in
    Lua code.

    If the request body has not been read yet, call ngx.req.read_body first
    (or turned on lua_need_request_body to force this module to read the
    request body. This is not recommended however).

    If the request body has been read into memory, try calling the
    ngx.req.get_body_data function instead.

    To force in-file request bodies, try turning on
    client_body_in_file_only.

    This function was first introduced in the "v0.3.1rc17" release.

    See also ngx.req.get_body_data.

  ngx.req.set_body_data
    syntax: *ngx.req.set_body_data(data)*

    context: *rewrite_by_lua*, access_by_lua*, content_by_lua**

    Set the current request's request body using the in-memory data
    specified by the "data" argument.

    If the current request's request body has not been read, then it will be
    properly discarded. When the current request's request body has been
    read into memory or buffered into a disk file, then the old request
    body's memory will be freed or the disk file will be cleaned up
    immediately, respectively.

    This function was first introduced in the "v0.3.1rc18" release.

    See also ngx.req.set_body_file.

  ngx.req.set_body_file
    syntax: *ngx.req.set_body_file(file_name, auto_clean?)*

    context: *rewrite_by_lua*, access_by_lua*, content_by_lua**

    Set the current request's request body using the in-file data specified
    by the "file_name" argument.

    If the optional "auto_clean" argument is given a "true" value, then this
    file will be removed at request completion or the next time this
    function or ngx.req.set_body_data are called in the same request. The
    "auto_clean" is default to "false".

    Please ensure that the file specified by the "file_name" argument exists
    and is readable by an Nginx worker process by setting its permission
    properly to avoid Lua exception errors.

    If the current request's request body has not been read, then it will be
    properly discarded. When the current request's request body has been
    read into memory or buffered into a disk file, then the old request
    body's memory will be freed or the disk file will be cleaned up
    immediately, respectively.

    This function was first introduced in the "v0.3.1rc18" release.

    See also ngx.req.set_body_data.

  ngx.req.init_body
    syntax: *ngx.req.init_body(buffer_size?)*

    context: *set_by_lua*, rewrite_by_lua*, access_by_lua*, content_by_lua**

    Creates a new blank request body for the current request and inializes
    the buffer for later request body data writing via the
    ngx.req.append_body and ngx.req.finish_body APIs.

    If the "buffer_size" argument is specified, then its value will be used
    for the size of the memory buffer for body writing with
    ngx.req.append_body. If the argument is omitted, then the value
    specified by the standard client_body_buffer_size directive will be used
    instead.

    When the data can no longer be hold in the memory buffer for the request
    body, then the data will be flushed onto a temporary file just like the
    standard request body reader in the Nginx core.

    It is important to always call the ngx.req.finish_body after all the
    data has been appended onto the current request body. Also, when this
    function is used together with ngx.req.socket, it is required to call
    ngx.req.socket *before* this function, or you will get the "request body
    already exists" error message.

    The usage of this function is often like this:

        ngx.req.init_body(128 * 1024)  -- buffer is 128KB
        for chunk in next_data_chunk() do
            ngx.req.append_body(chunk) -- each chunk can be 4KB
        end
        ngx.req.finish_body()

    This function can be used with ngx.req.append_body, ngx.req.finish_body,
    and ngx.req.socket to implement efficient input filters in pure Lua (in
    the context of rewrite_by_lua* or access_by_lua*), which can be used
    with other Nginx content handler or upstream modules like
    [[HttpProxyModule]] and [[HttpFastcgiModule]].

    This function was first introduced in the "v0.5.11" release.

  ngx.req.append_body
    syntax: *ngx.req.append_body(data_chunk)*

    context: *set_by_lua*, rewrite_by_lua*, access_by_lua*, content_by_lua**

    Append new data chunk specified by the "data_chunk" argument onto the
    existing request body created by the ngx.req.init_body call.

    When the data can no longer be hold in the memory buffer for the request
    body, then the data will be flushed onto a temporary file just like the
    standard request body reader in the Nginx core.

    It is important to always call the ngx.req.finish_body after all the
    data has been appended onto the current request body.

    This function can be used with ngx.req.init_body, ngx.req.finish_body,
    and ngx.req.socket to implement efficient input filters in pure Lua (in
    the context of rewrite_by_lua* or access_by_lua*), which can be used
    with other Nginx content handler or upstream modules like
    [[HttpProxyModule]] and [[HttpFastcgiModule]].

    This function was first introduced in the "v0.5.11" release.

    See also ngx.req.init_body.

  ngx.req.finish_body
    syntax: *ngx.req.finish_body()*

    context: *set_by_lua*, rewrite_by_lua*, access_by_lua*, content_by_lua**

    Completes the construction process of the new request body created by
    the ngx.req.init_body and ngx.req.append_body calls.

    This function can be used with ngx.req.init_body, ngx.req.append_body,
    and ngx.req.socket to implement efficient input filters in pure Lua (in
    the context of rewrite_by_lua* or access_by_lua*), which can be used
    with other Nginx content handler or upstream modules like
    [[HttpProxyModule]] and [[HttpFastcgiModule]].

    This function was first introduced in the "v0.5.11" release.

    See also ngx.req.init_body.

  ngx.req.socket
    syntax: *tcpsock, err = ngx.req.socket()*

    context: *rewrite_by_lua*, access_by_lua*, content_by_lua**

    Returns a read-only cosocket object that wraps the downstream
    connection. Only receive and receiveuntil methods are supported on this
    object.

    In case of error, "nil" will be returned as well as a string describing
    the error.

    The socket object returned by this method is usually used to read the
    current request's body in a streaming fashion. Do not turn on the
    lua_need_request_body directive, and do not mix this call with
    ngx.req.read_body and ngx.req.discard_body.

    If any request body data has been pre-read into the Nginx core request
    header buffer, the resulting cosocket object will take care of this to
    avoid potential data loss resulting from such pre-reading.

    This function was first introduced in the "v0.5.0rc1" release.

  ngx.exec
    syntax: *ngx.exec(uri, args?)*

    context: *rewrite_by_lua*, access_by_lua*, content_by_lua**

    Does an internal redirect to "uri" with "args".

        ngx.exec('/some-location');
        ngx.exec('/some-location', 'a=3&b=5&c=6');
        ngx.exec('/some-location?a=3&b=5', 'c=6');

    Named locations are also supported, but query strings are ignored. For
    example,

        location /foo {
            content_by_lua '
                ngx.exec("@bar");
            ';
        }

        location @bar {
            ...
        }

    The optional second "args" can be used to specify extra URI query
    arguments, for example:

        ngx.exec("/foo", "a=3&b=hello%20world")

    Alternatively, a Lua table can be passed for the "args" argument for
    ngx_lua to carry out URI escaping and string concatenation.

        ngx.exec("/foo", { a = 3, b = "hello world" })

    The result is exactly the same as the previous example. The format for
    the Lua table passed as the "args" argument is identical to the format
    used in the ngx.encode_args method.

    Note that this is very different from ngx.redirect in that it is just an
    internal redirect and no new HTTP traffic is involved.

    This method never returns.

    This method *must* be called before ngx.send_headers or explicit
    response body outputs by either ngx.print or ngx.say.

    It is strongly recommended to combine the "return" statement with this
    call, i.e., "return ngx.exec(...)".

    This method is similar to the echo_exec directive of the
    [[HttpEchoModule]].

  ngx.redirect
    syntax: *ngx.redirect(uri, status?)*

    context: *rewrite_by_lua*, access_by_lua*, content_by_lua**

    Issue an "HTTP 301" or 302 redirection to "uri".

    The optional "status" parameter specifies whether 301 or 302 to be used.
    It is 302 ("ngx.HTTP_MOVED_TEMPORARILY") by default.

    Here is an example assuming the current server name is "localhost" and
    that it is listening on Port 1984:

        return ngx.redirect("/foo")

    which is equivalent to

        return ngx.redirect("http://localhost:1984/foo", ngx.HTTP_MOVED_TEMPORARILY)

    Redirecting arbitrary external URLs is also supported, for example:

        return ngx.redirect("http://www.google.com")

    We can also use the numerical code directly as the second "status"
    argument:

        return ngx.redirect("/foo", 301)

    This method *must* be called before ngx.send_headers or explicit
    response body outputs by either ngx.print or ngx.say.

    This method is very much like the rewrite directive with the "redirect"
    modifier in the standard [[HttpRewriteModule]], for example, this
    "nginx.conf" snippet

        rewrite ^ /foo? redirect;  # nginx config

    is equivalent to the following Lua code

        return ngx.redirect('/foo');  -- Lua code

    while

        rewrite ^ /foo? permanent;  # nginx config

    is equivalent to

        return ngx.redirect('/foo', ngx.HTTP_MOVED_PERMANENTLY)  -- Lua code

    URI arguments can be specified as well, for example:

        return ngx.redirect('/foo?a=3&b=4')

    This method call terminates the current request's processing and never
    returns. It is recommended to combine the "return" statement with this
    call, i.e., "return ngx.redirect(...)", so as to be more explicit.

  ngx.send_headers
    syntax: *ok, err = ngx.send_headers()*

    context: *rewrite_by_lua*, access_by_lua*, content_by_lua**

    Explicitly send out the response headers.

    Since "v0.8.3" this function returns 1 on success, or returns "nil" and
    a string describing the error otherwise.

    Note that there is normally no need to manually send out response
    headers as ngx_lua will automatically send headers out before content is
    output with ngx.say or ngx.print or when content_by_lua exits normally.

  ngx.headers_sent
    syntax: *value = ngx.headers_sent*

    context: *set_by_lua*, rewrite_by_lua*, access_by_lua*, content_by_lua**

    Returns "true" if the response headers have been sent (by ngx_lua), and
    "false" otherwise.

    This API was first introduced in ngx_lua v0.3.1rc6.

  ngx.print
    syntax: *ok, err = ngx.print(...)*

    context: *rewrite_by_lua*, access_by_lua*, content_by_lua**

    Emits arguments concatenated to the HTTP client (as response body). If
    response headers have not been sent, this function will send headers out
    first and then output body data.

    Since "v0.8.3" this function returns 1 on success, or returns "nil" and
    a string describing the error otherwise.

    Lua "nil" values will output "nil" strings and Lua boolean values will
    output "true" and "false" literal strings respectively.

    Nested arrays of strings are permitted and the elements in the arrays
    will be sent one by one:

        local table = {
            "hello, ",
            {"world: ", true, " or ", false,
                {": ", nil}}
        }
        ngx.print(table)

    will yield the output

        hello, world: true or false: nil

    Non-array table arguments will cause a Lua exception to be thrown.

    The "ngx.null" constant will yield the "null" string output.

    This is an asynchronous call and will return immediately without waiting
    for all the data to be written into the system send buffer. To run in
    synchronous mode, call "ngx.flush(true)" after calling "ngx.print". This
    can be particularly useful for streaming output. See ngx.flush for more
    details.

    Please note that both "ngx.print" and ngx.say will always invoke the
    whole Nginx output body filter chain, which is an expensive operation.
    So be careful when calling either of these two in a tight loop; buffer
    the data yourself in Lua and save the calls.

  ngx.say
    syntax: *ok, err = ngx.say(...)*

    context: *rewrite_by_lua*, access_by_lua*, content_by_lua**

    Just as ngx.print but also emit a trailing newline.

  ngx.log
    syntax: *ngx.log(log_level, ...)*

    context: *init_by_lua*, set_by_lua*, rewrite_by_lua*, access_by_lua*,
    content_by_lua*, header_filter_by_lua*, body_filter_by_lua*,
    log_by_lua*, ngx.timer.**

    Log arguments concatenated to error.log with the given logging level.

    Lua "nil" arguments are accepted and result in literal "nil" string
    while Lua booleans result in literal "true" or "false" string outputs.
    And the "ngx.null" constant will yield the "null" string output.

    The "log_level" argument can take constants like "ngx.ERR" and
    "ngx.WARN". Check out Nginx log level constants for details.

    There is a hard coded 2048 byte limitation on error message lengths in
    the Nginx core. This limit includes trailing newlines and leading time
    stamps. If the message size exceeds this limit, Nginx will truncate the
    message text accordingly. This limit can be manually modified by editing
    the "NGX_MAX_ERROR_STR" macro definition in the "src/core/ngx_log.h"
    file in the Nginx source tree.

  ngx.flush
    syntax: *ok, err = ngx.flush(wait?)*

    context: *rewrite_by_lua*, access_by_lua*, content_by_lua**

    Flushes response output to the client.

    "ngx.flush" accepts an optional boolean "wait" argument (Default:
    "false") first introduced in the "v0.3.1rc34" release. When called with
    the default argument, it issues an asynchronous call (Returns
    immediately without waiting for output data to be written into the
    system send buffer). Calling the function with the "wait" argument set
    to "true" switches to synchronous mode.

    In synchronous mode, the function will not return until all output data
    has been written into the system send buffer or until the send_timeout
    setting has expired. Note that using the Lua coroutine mechanism means
    that this function does not block the Nginx event loop even in the
    synchronous mode.

    When "ngx.flush(true)" is called immediately after ngx.print or ngx.say,
    it causes the latter functions to run in synchronous mode. This can be
    particularly useful for streaming output.

    Note that "ngx.flush" is non functional when in the HTTP 1.0 output
    buffering mode. See HTTP 1.0 support.

    Since "v0.8.3" this function returns 1 on success, or returns "nil" and
    a string describing the error otherwise.

  ngx.exit
    syntax: *ngx.exit(status)*

    context: *rewrite_by_lua*, access_by_lua*, content_by_lua**

    When "status >= 200" (i.e., "ngx.HTTP_OK" and above), it will interrupt
    the execution of the current request and return status code to nginx.

    When "status == 0" (i.e., "ngx.OK"), it will only quit the current phase
    handler (or the content handler if the content_by_lua directive is used)
    and continue to run later phases (if any) for the current request.

    The "status" argument can be "ngx.OK", "ngx.ERROR",
    "ngx.HTTP_NOT_FOUND", "ngx.HTTP_MOVED_TEMPORARILY", or other HTTP status
    constants.

    To return an error page with custom contents, use code snippets like
    this:

        ngx.status = ngx.HTTP_GONE
        ngx.say("This is our own content")
        -- to cause quit the whole request rather than the current phase handler
        ngx.exit(ngx.HTTP_OK)

    The effect in action:

        $ curl -i http://localhost/test
        HTTP/1.1 410 Gone
        Server: nginx/1.0.6
        Date: Thu, 15 Sep 2011 00:51:48 GMT
        Content-Type: text/plain
        Transfer-Encoding: chunked
        Connection: keep-alive

        This is our own content

    Number literals can be used directly as the argument, for instance,

        ngx.exit(501)

    Note that while this method accepts all HTTP status constants as input,
    it only accepts "NGX_OK" and "NGX_ERROR" of the core constants.

    It is recommended, though not necessary, to combine the "return"
    statement with this call, i.e., "return ngx.exit(...)", to give a visual
    hint to others reading the code.

  ngx.eof
    syntax: *ok, err = ngx.eof()*

    context: *rewrite_by_lua*, access_by_lua*, content_by_lua**

    Explicitly specify the end of the response output stream. In the case of
    HTTP 1.1 chunked encoded output, it will just trigger the Nginx core to
    send out the "last chunk".

    When you disable the HTTP 1.1 keep-alive feature for your downstream
    connections, you can rely on descent HTTP clients to close the
    connection actively for you when you call this method. This trick can be
    used do back-ground jobs without letting the HTTP clients to wait on the
    connection, as in the following example:

        location = /async {
            keepalive_timeout 0;
            content_by_lua '
                ngx.say("got the task!")
                ngx.eof()  -- descent HTTP client will close the connection at this point
                -- access MySQL, PostgreSQL, Redis, Memcached, and etc here...
            ';
        }

    But if you create subrequests to access other locations configured by
    Nginx upstream modules, then you should configure those upstream modules
    to ignore client connection abortions if they are not by default. For
    example, by default the standard [[HttpProxyModule]] will terminate both
    the subrequest and the main request as soon as the client closes the
    connection, so it is important to turn on the proxy_ignore_client_abort
    directive in your location block configured by [[HttpProxyModule]]:

        proxy_ignore_client_abort on;

    Since "v0.8.3" this function returns 1 on success, or returns "nil" and
    a string describing the error otherwise.

  ngx.sleep
    syntax: *ngx.sleep(seconds)*

    context: *rewrite_by_lua*, access_by_lua*, content_by_lua*, ngx.timer.**

    Sleeps for the specified seconds without blocking. One can specify time
    resolution up to 0.001 seconds (i.e., one milliseconds).

    Behind the scene, this method makes use of the Nginx timers.

    Since the 0.7.20 release, The 0 time argument can also be specified.

    This method was introduced in the "0.5.0rc30" release.

  ngx.escape_uri
    syntax: *newstr = ngx.escape_uri(str)*

    context: *set_by_lua*, rewrite_by_lua*, access_by_lua*, content_by_lua*,
    header_filter_by_lua*, body_filter_by_lua*, log_by_lua*, ngx.timer.**

    Escape "str" as a URI component.

  ngx.unescape_uri
    syntax: *newstr = ngx.unescape_uri(str)*

    context: *set_by_lua*, rewrite_by_lua*, access_by_lua*, content_by_lua*,
    header_filter_by_lua*, body_filter_by_lua*, log_by_lua*, ngx.timer.**

    Unescape "str" as an escaped URI component.

    For example,

        ngx.say(ngx.unescape_uri("b%20r56+7"))

    gives the output

        b r56 7

  ngx.encode_args
    syntax: *str = ngx.encode_args(table)*

    context: *set_by_lua*, rewrite_by_lua*, access_by_lua*, content_by_lua*,
    header_filter_by_lua*, body_filter_by_lua*, log_by_lua*, ngx.timer.**

    Encode the Lua table to a query args string according to the URI encoded
    rules.

    For example,

        ngx.encode_args({foo = 3, ["b r"] = "hello world"})

    yields

        foo=3&b%20r=hello%20world

    The table keys must be Lua strings.

    Multi-value query args are also supported. Just use a Lua table for the
    argument's value, for example:

        ngx.encode_args({baz = {32, "hello"}})

    gives

        baz=32&baz=hello

    If the value table is empty and the effect is equivalent to the "nil"
    value.

    Boolean argument values are also supported, for instance,

        ngx.encode_args({a = true, b = 1})

    yields

        a&b=1

    If the argument value is "false", then the effect is equivalent to the
    "nil" value.

    This method was first introduced in the "v0.3.1rc27" release.

  ngx.decode_args
    syntax: *table = ngx.decode_args(str, max_args?)*

    context: *set_by_lua*, rewrite_by_lua*, access_by_lua*, content_by_lua*,
    header_filter_by_lua*, body_filter_by_lua*, log_by_lua*, ngx.timer.**

    Decodes a URI encoded query-string into a Lua table. This is the inverse
    function of ngx.encode_args.

    The optional "max_args" argument can be used to specify the maximum
    number of arguments parsed from the "str" argument. By default, a
    maximum of 100 request arguments are parsed (including those with the
    same name) and that additional URI arguments are silently discarded to
    guard against potential denial of service attacks.

    This argument can be set to zero to remove the limit and to process all
    request arguments received:

        local args = ngx.decode_args(str, 0)

    Removing the "max_args" cap is strongly discouraged.

    This method was introduced in the "v0.5.0rc29".

  ngx.encode_base64
    syntax: *newstr = ngx.encode_base64(str)*

    context: *set_by_lua*, rewrite_by_lua*, access_by_lua*, content_by_lua*,
    header_filter_by_lua*, body_filter_by_lua*, log_by_lua*, ngx.timer.**

    Encode "str" to a base64 digest.

  ngx.decode_base64
    syntax: *newstr = ngx.decode_base64(str)*

    context: *set_by_lua*, rewrite_by_lua*, access_by_lua*, content_by_lua*,
    header_filter_by_lua*, body_filter_by_lua*, log_by_lua*, ngx.timer.**

    Decodes the "str" argument as a base64 digest to the raw form. Returns
    "nil" if "str" is not well formed.

  ngx.crc32_short
    syntax: *intval = ngx.crc32_short(str)*

    context: *set_by_lua*, rewrite_by_lua*, access_by_lua*, content_by_lua*,
    header_filter_by_lua*, body_filter_by_lua*, log_by_lua*, ngx.timer.**

    Calculates the CRC-32 (Cyclic Redundancy Code) digest for the "str"
    argument.

    This method performs better on relatively short "str" inputs (i.e., less
    than 30 ~ 60 bytes), as compared to ngx.crc32_long. The result is
    exactly the same as ngx.crc32_long.

    Behind the scene, it is just a thin wrapper around the "ngx_crc32_short"
    function defined in the Nginx core.

    This API was first introduced in the "v0.3.1rc8" release.

  ngx.crc32_long
    syntax: *intval = ngx.crc32_long(str)*

    context: *set_by_lua*, rewrite_by_lua*, access_by_lua*, content_by_lua*,
    header_filter_by_lua*, body_filter_by_lua*, log_by_lua*, ngx.timer.**

    Calculates the CRC-32 (Cyclic Redundancy Code) digest for the "str"
    argument.

    This method performs better on relatively long "str" inputs (i.e.,
    longer than 30 ~ 60 bytes), as compared to ngx.crc32_short. The result
    is exactly the same as ngx.crc32_short.

    Behind the scene, it is just a thin wrapper around the "ngx_crc32_long"
    function defined in the Nginx core.

    This API was first introduced in the "v0.3.1rc8" release.

  ngx.hmac_sha1
    syntax: *digest = ngx.hmac_sha1(secret_key, str)*

    context: *set_by_lua*, rewrite_by_lua*, access_by_lua*, content_by_lua*,
    header_filter_by_lua*, body_filter_by_lua*, log_by_lua*, ngx.timer.**

    Computes the HMAC-SHA1 (<http://en.wikipedia.org/wiki/HMAC>) digest of
    the argument "str" and turns the result using the secret key
    "<secret_key>".

    The raw binary form of the "HMAC-SHA1" digest will be generated, use
    ngx.encode_base64, for example, to encode the result to a textual
    representation if desired.

    For example,

        local key = "thisisverysecretstuff"
        local src = "some string we want to sign"
        local digest = ngx.hmac_sha1(key, src)
        ngx.say(ngx.encode_base64(digest))

    yields the output

        R/pvxzHC4NLtj7S+kXFg/NePTmk=

    This API requires the OpenSSL library enabled in the Nginx build
    (usually by passing the "--with-http_ssl_module" option to the
    "./configure" script).

    This function was first introduced in the "v0.3.1rc29" release.

  ngx.md5
    syntax: *digest = ngx.md5(str)*

    context: *set_by_lua*, rewrite_by_lua*, access_by_lua*, content_by_lua*,
    header_filter_by_lua*, body_filter_by_lua*, log_by_lua*, ngx.timer.**

    Returns the hexadecimal representation of the MD5 digest of the "str"
    argument.

    For example,

        location = /md5 {
            content_by_lua 'ngx.say(ngx.md5("hello"))';
        }

    yields the output

        5d41402abc4b2a76b9719d911017c592

    See ngx.md5_bin if the raw binary MD5 digest is required.

  ngx.md5_bin
    syntax: *digest = ngx.md5_bin(str)*

    context: *set_by_lua*, rewrite_by_lua*, access_by_lua*, content_by_lua*,
    header_filter_by_lua*, body_filter_by_lua*, log_by_lua*, ngx.timer.**

    Returns the binary form of the MD5 digest of the "str" argument.

    See ngx.md5 if the hexadecimal form of the MD5 digest is required.

  ngx.sha1_bin
    syntax: *digest = ngx.sha1_bin(str)*

    context: *set_by_lua*, rewrite_by_lua*, access_by_lua*, content_by_lua*,
    header_filter_by_lua*, body_filter_by_lua*, log_by_lua*, ngx.timer.**

    Returns the binary form of the SHA-1 digest of the "str" argument.

    This function requires SHA-1 support in the Nginx build. (This usually
    just means OpenSSL should be installed while building Nginx).

    This function was first introduced in the "v0.5.0rc6".

  ngx.quote_sql_str
    syntax: *quoted_value = ngx.quote_sql_str(raw_value)*

    context: *set_by_lua*, rewrite_by_lua*, access_by_lua*, content_by_lua*,
    header_filter_by_lua*, body_filter_by_lua*, log_by_lua*, ngx.timer.**

    Returns a quoted SQL string literal according to the MySQL quoting
    rules.

  ngx.today
    syntax: *str = ngx.today()*

    context: *set_by_lua*, rewrite_by_lua*, access_by_lua*, content_by_lua*,
    header_filter_by_lua*, body_filter_by_lua*, log_by_lua*, ngx.timer.**

    Returns current date (in the format "yyyy-mm-dd") from the nginx cached
    time (no syscall involved unlike Lua's date library).

    This is the local time.

  ngx.time
    syntax: *secs = ngx.time()*

    context: *set_by_lua*, rewrite_by_lua*, access_by_lua*, content_by_lua*,
    header_filter_by_lua*, body_filter_by_lua*, log_by_lua*, ngx.timer.**

    Returns the elapsed seconds from the epoch for the current time stamp
    from the nginx cached time (no syscall involved unlike Lua's date
    library).

    Updates of the Nginx time cache an be forced by calling ngx.update_time
    first.

  ngx.now
    syntax: *secs = ngx.now()*

    context: *set_by_lua*, rewrite_by_lua*, access_by_lua*, content_by_lua*,
    header_filter_by_lua*, body_filter_by_lua*, log_by_lua*, ngx.timer.**

    Returns a floating-point number for the elapsed time in seconds
    (including milliseconds as the decimal part) from the epoch for the
    current time stamp from the nginx cached time (no syscall involved
    unlike Lua's date library).

    You can forcibly update the Nginx time cache by calling ngx.update_time
    first.

    This API was first introduced in "v0.3.1rc32".

  ngx.update_time
    syntax: *ngx.update_time()*

    context: *set_by_lua*, rewrite_by_lua*, access_by_lua*, content_by_lua*,
    header_filter_by_lua*, body_filter_by_lua*, log_by_lua*, ngx.timer.**

    Forcibly updates the Nginx current time cache. This call involves a
    syscall and thus has some overhead, so do not abuse it.

    This API was first introduced in "v0.3.1rc32".

  ngx.localtime
    syntax: *str = ngx.localtime()*

    context: *set_by_lua*, rewrite_by_lua*, access_by_lua*, content_by_lua*,
    header_filter_by_lua*, body_filter_by_lua*, log_by_lua*, ngx.timer.**

    Returns the current time stamp (in the format "yyyy-mm-dd hh:mm:ss") of
    the nginx cached time (no syscall involved unlike Lua's os.date
    (<http://www.lua.org/manual/5.1/manual.html#pdf-os.date>) function).

    This is the local time.

  ngx.utctime
    syntax: *str = ngx.utctime()*

    context: *set_by_lua*, rewrite_by_lua*, access_by_lua*, content_by_lua*,
    header_filter_by_lua*, body_filter_by_lua*, log_by_lua*, ngx.timer.**

    Returns the current time stamp (in the format "yyyy-mm-dd hh:mm:ss") of
    the nginx cached time (no syscall involved unlike Lua's os.date
    (<http://www.lua.org/manual/5.1/manual.html#pdf-os.date>) function).

    This is the UTC time.

  ngx.cookie_time
    syntax: *str = ngx.cookie_time(sec)*

    context: *set_by_lua*, rewrite_by_lua*, access_by_lua*, content_by_lua*,
    header_filter_by_lua*, body_filter_by_lua*, log_by_lua*, ngx.timer.**

    Returns a formated string can be used as the cookie expiration time. The
    parameter "sec" is the time stamp in seconds (like those returned from
    ngx.time).

        ngx.say(ngx.cookie_time(1290079655))
            -- yields "Thu, 18-Nov-10 11:27:35 GMT"

  ngx.http_time
    syntax: *str = ngx.http_time(sec)*

    context: *set_by_lua*, rewrite_by_lua*, access_by_lua*, content_by_lua*,
    header_filter_by_lua*, body_filter_by_lua*, log_by_lua*, ngx.timer.**

    Returns a formated string can be used as the http header time (for
    example, being used in "Last-Modified" header). The parameter "sec" is
    the time stamp in seconds (like those returned from ngx.time).

        ngx.say(ngx.http_time(1290079655))
            -- yields "Thu, 18 Nov 2010 11:27:35 GMT"

  ngx.parse_http_time
    syntax: *sec = ngx.parse_http_time(str)*

    context: *set_by_lua*, rewrite_by_lua*, access_by_lua*, content_by_lua*,
    header_filter_by_lua*, body_filter_by_lua*, log_by_lua*, ngx.timer.**

    Parse the http time string (as returned by ngx.http_time) into seconds.
    Returns the seconds or "nil" if the input string is in bad forms.

        local time = ngx.parse_http_time("Thu, 18 Nov 2010 11:27:35 GMT")
        if time == nil then
            ...
        end

  ngx.is_subrequest
    syntax: *value = ngx.is_subrequest*

    context: *set_by_lua*, rewrite_by_lua*, access_by_lua*, content_by_lua*,
    header_filter_by_lua*, body_filter_by_lua*, log_by_lua**

    Returns "true" if the current request is an nginx subrequest, or "false"
    otherwise.

  ngx.re.match
    syntax: *captures, err = ngx.re.match(subject, regex, options?, ctx?)*

    context: *set_by_lua*, rewrite_by_lua*, access_by_lua*, content_by_lua*,
    header_filter_by_lua*, body_filter_by_lua*, log_by_lua*, ngx.timer.**

    Matches the "subject" string using the Perl compatible regular
    expression "regex" with the optional "options".

    Only the first occurrence of the match is returned, or "nil" if no match
    is found. In case of errors, like seeing a bad regular expression or
    exceeding the PCRE stack limit, "nil" and a string describing the error
    will be returned.

    When a match is found, a Lua table "captures" is returned, where
    "captures[0]" holds the whole substring being matched, and "captures[1]"
    holds the first parenthesized sub-pattern's capturing, "captures[2]" the
    second, and so on.

        local m, err = ngx.re.match("hello, 1234", "[0-9]+")
        if m then
            -- m[0] == "1234"

        else
            if err then
                ngx.log(ngx.ERR, "error: ", err)
                return
            end

            ngx.say("match not found")
        end

        local m, err = ngx.re.match("hello, 1234", "([0-9])[0-9]+")
        -- m[0] == "1234"
        -- m[1] == "1"

    Named captures are also supported since the "v0.7.14" release and are
    returned in the same Lua table as key-value pairs as the numbered
    captures.

        local m, err = ngx.re.match("hello, 1234", "([0-9])(?<remaining>[0-9]+)")
        -- m[0] == "1234"
        -- m[1] == "1"
        -- m[2] == "234"
        -- m["remaining"] == "234"

    Unmatched subpatterns will have "nil" values in their "captures" table
    fields.

        local m, err = ngx.re.match("hello, world", "(world)|(hello)|(?<named>howdy)")
        -- m[0] == "hello"
        -- m[1] == nil
        -- m[2] == "hello"
        -- m[3] == nil
        -- m["named"] == nil

    Specify "options" to control how the match operation will be performed.
    The following option characters are supported:

        a             anchored mode (only match from the beginning)

        d             enable the DFA mode (or the longest token match semantics).
                      this requires PCRE 6.0+ or else a Lua exception will be thrown.
                      first introduced in ngx_lua v0.3.1rc30.

        D             enable duplicate named pattern support. This allows named
                      subpattern names to be repeated, returning the captures in
                      an array-like Lua table. for example,
                        local m = ngx.re.match("hello, world",
                                               "(?<named>\w+), (?<named>\w+)",
                                               "D")
                        -- m["named"] == {"hello", "world"}
                      this option was first introduced in the v0.7.14 release.
                      this option requires at least PCRE 8.12.

        i             case insensitive mode (similar to Perl's /i modifier)

        j             enable PCRE JIT compilation, this requires PCRE 8.21+ which
                      must be built with the --enable-jit option. for optimum performance,
                      this option should always be used together with the 'o' option.
                      first introduced in ngx_lua v0.3.1rc30.

        J             enable the PCRE Javascript compatible mode. this option was
                      first introduced in the v0.7.14 release. this option requires
                      at least PCRE 8.12.

        m             multi-line mode (similar to Perl's /m modifier)

        o             compile-once mode (similar to Perl's /o modifier),
                      to enable the worker-process-level compiled-regex cache

        s             single-line mode (similar to Perl's /s modifier)

        u             UTF-8 mode. this requires PCRE to be built with
                      the --enable-utf8 option or else a Lua exception will be thrown.

        U             similar to "u" but disables PCRE's UTF-8 validity check on
                      the subject string. first introduced in ngx_lua v0.8.1.

        x             extended mode (similar to Perl's /x modifier)

    These options can be combined:

        local m, err = ngx.re.match("hello, world", "HEL LO", "ix")
        -- m[0] == "hello"

        local m, err = ngx.re.match("hello, 美好生活", "HELLO, (.{2})", "iu")
        -- m[0] == "hello, 美好"
        -- m[1] == "美好"

    The "o" option is useful for performance tuning, because the regex
    pattern in question will only be compiled once, cached in the
    worker-process level, and shared among all requests in the current Nginx
    worker process. The upper limit of the regex cache can be tuned via the
    lua_regex_cache_max_entries directive.

    The optional fourth argument, "ctx", can be a Lua table holding an
    optional "pos" field. When the "pos" field in the "ctx" table argument
    is specified, "ngx.re.match" will start matching from that offset.
    Regardless of the presence of the "pos" field in the "ctx" table,
    "ngx.re.match" will always set this "pos" field to the position *after*
    the substring matched by the whole pattern in case of a successful
    match. When match fails, the "ctx" table will be left intact.

        local ctx = {}
        local m, err = ngx.re.match("1234, hello", "[0-9]+", "", ctx)
             -- m[0] = "1234"
             -- ctx.pos == 4

        local ctx = { pos = 2 }
        local m, err = ngx.re.match("1234, hello", "[0-9]+", "", ctx)
             -- m[0] = "34"
             -- ctx.pos == 4

    The "ctx" table argument combined with the "a" regex modifier can be
    used to construct a lexer atop "ngx.re.match".

    Note that, the "options" argument is not optional when the "ctx"
    argument is specified and that the empty Lua string ("") must be used as
    placeholder for "options" if no meaningful regex options are required.

    This method requires the PCRE library enabled in Nginx. (Known Issue
    With Special PCRE Sequences).

    To confirm that PCRE JIT is enabled, activate the Nginx debug log by
    adding the "--with-debug" option to Nginx or ngx_openresty's
    "./configure" script. Then, enable the "debug" error log level in
    "error_log" directive. The following message will be generated if PCRE
    JIT is enabled:

        pcre JIT compiling result: 1

    This feature was introduced in the "v0.2.1rc11" release.

  ngx.re.gmatch
    syntax: *iterator, err = ngx.re.gmatch(subject, regex, options?)*

    context: *set_by_lua*, rewrite_by_lua*, access_by_lua*, content_by_lua*,
    header_filter_by_lua*, body_filter_by_lua*, log_by_lua*, ngx.timer.**

    Similar to ngx.re.match, but returns a Lua iterator instead, so as to
    let the user programmer iterate all the matches over the "<subject>"
    string argument with the PCRE "regex".

    In case of errors, like seeing an ill-formed regular expression, "nil"
    and a string describing the error will be returned.

    Here is a small example to demonstrate its basic usage:

        local iterator, err = ngx.re.gmatch("hello, world!", "([a-z]+)", "i")
        if not iterator then
            ngx.log(ngx.ERR, "error: ", err)
            return
        end

        local m
        m, err = iterator()    -- m[0] == m[1] == "hello"
        if err then
            ngx.log(ngx.ERR, "error: ", err)
            return
        end

        m, err = iterator()    -- m[0] == m[1] == "world"
        if err then
            ngx.log(ngx.ERR, "error: ", err)
            return
        end

        m, err = iterator()    -- m == nil
        if err then
            ngx.log(ngx.ERR, "error: ", err)
            return
        end

    More often we just put it into a Lua loop:

        local it, err = ngx.re.gmatch("hello, world!", "([a-z]+)", "i")
        if not it then
            ngx.log(ngx.ERR, "error: ", err)
            return
        end

        while true do
            local m, err = it()
            if err then
                ngx.log(ngx.ERR, "error: ", err)
                return
            end

            if not m then
                -- no match found (any more)
                break
            end

            -- found a match
            ngx.say(m[0])
            ngx.say(m[1])
        end

    The optional "options" argument takes exactly the same semantics as the
    ngx.re.match method.

    The current implementation requires that the iterator returned should
    only be used in a single request. That is, one should *not* assign it to
    a variable belonging to persistent namespace like a Lua package.

    This method requires the PCRE library enabled in Nginx. (Known Issue
    With Special PCRE Sequences).

    This feature was first introduced in the "v0.2.1rc12" release.

  ngx.re.sub
    syntax: *newstr, n, err = ngx.re.sub(subject, regex, replace, options?)*

    context: *set_by_lua*, rewrite_by_lua*, access_by_lua*, content_by_lua*,
    header_filter_by_lua*, body_filter_by_lua*, log_by_lua*, ngx.timer.**

    Substitutes the first match of the Perl compatible regular expression
    "regex" on the "subject" argument string with the string or function
    argument "replace". The optional "options" argument has exactly the same
    meaning as in ngx.re.match.

    This method returns the resulting new string as well as the number of
    successful substitutions. In case of failures, like syntax errors in the
    regular expressions or the "<replace>" string argument, it will return
    "nil" and a string describing the error.

    When the "replace" is a string, then it is treated as a special template
    for string replacement. For example,

        local newstr, n, err = ngx.re.sub("hello, 1234", "([0-9])[0-9]", "[$0][$1]")
        if newstr then
            -- newstr == "hello, [12][1]34"
            -- n == 1
        else
            ngx.log(ngx.ERR, "error: ", err)
            return
        end

    where $0 referring to the whole substring matched by the pattern and $1
    referring to the first parenthesized capturing substring.

    Curly braces can also be used to disambiguate variable names from the
    background string literals:

        local newstr, n, err = ngx.re.sub("hello, 1234", "[0-9]", "${0}00")
            -- newstr == "hello, 10034"
            -- n == 1

    Literal dollar sign characters ("$") in the "replace" string argument
    can be escaped by another dollar sign, for instance,

        local newstr, n, err = ngx.re.sub("hello, 1234", "[0-9]", "$$")
            -- newstr == "hello, $234"
            -- n == 1

    Do not use backlashes to escape dollar signs; it will not work as
    expected.

    When the "replace" argument is of type "function", then it will be
    invoked with the "match table" as the argument to generate the replace
    string literal for substitution. The "match table" fed into the
    "replace" function is exactly the same as the return value of
    ngx.re.match. Here is an example:

        local func = function (m)
            return "[" .. m[0] .. "][" .. m[1] .. "]"
        end
        local newstr, n, err = ngx.re.sub("hello, 1234", "( [0-9] ) [0-9]", func, "x")
            -- newstr == "hello, [12][1]34"
            -- n == 1

    The dollar sign characters in the return value of the "replace" function
    argument are not special at all.

    This method requires the PCRE library enabled in Nginx. (Known Issue
    With Special PCRE Sequences).

    This feature was first introduced in the "v0.2.1rc13" release.

  ngx.re.gsub
    syntax: *newstr, n, err = ngx.re.gsub(subject, regex, replace,
    options?)*

    context: *set_by_lua*, rewrite_by_lua*, access_by_lua*, content_by_lua*,
    header_filter_by_lua*, body_filter_by_lua*, log_by_lua*, ngx.timer.**

    Just like ngx.re.sub, but does global substitution.

    Here is some examples:

        local newstr, n, err = ngx.re.gsub("hello, world", "([a-z])[a-z]+", "[$0,$1]", "i")
        if newstr then
            -- newstr == "[hello,h], [world,w]"
            -- n == 2
        else
            ngx.log(ngx.ERR, "error: ", err)
            return
        end

        local func = function (m)
            return "[" .. m[0] .. "," .. m[1] .. "]"
        end
        local newstr, n, err = ngx.re.gsub("hello, world", "([a-z])[a-z]+", func, "i")
            -- newstr == "[hello,h], [world,w]"
            -- n == 2

    This method requires the PCRE library enabled in Nginx. (Known Issue
    With Special PCRE Sequences).

    This feature was first introduced in the "v0.2.1rc15" release.

  ngx.shared.DICT
    syntax: *dict = ngx.shared.DICT*

    context: *init_by_lua*, set_by_lua*, rewrite_by_lua*, access_by_lua*,
    content_by_lua*, header_filter_by_lua*, body_filter_by_lua*,
    log_by_lua*, ngx.timer.**

    Fetching the shm-based Lua dictionary object for the shared memory zone
    named "DICT" defined by the lua_shared_dict directive.

    The resulting object "dict" has the following methods:

    *   get

    *   set

    *   safe_set

    *   add

    *   safe_add

    *   replace

    *   incr

    *   delete

    *   flush_all

    *   flush_expired

    Here is an example:

        http {
            lua_shared_dict dogs 10m;
            server {
                location /set {
                    content_by_lua '
                        local dogs = ngx.shared.dogs
                        dogs:set("Jim", 8)
                        ngx.say("STORED")
                    ';
                }
                location /get {
                    content_by_lua '
                        local dogs = ngx.shared.dogs
                        ngx.say(dogs:get("Jim"))
                    ';
                }
            }
        }

    Let us test it:

        $ curl localhost/set
        STORED

        $ curl localhost/get
        8

        $ curl localhost/get
        8

    The number 8 will be consistently output when accessing "/get"
    regardless of how many Nginx workers there are because the "dogs"
    dictionary resides in the shared memory and visible to *all* of the
    worker processes.

    The shared dictionary will retain its contents through a server config
    reload (either by sending the "HUP" signal to the Nginx process or by
    using the "-s reload" command-line option).

    The contents in the dictionary storage will be lost, however, when the
    Nginx server quits.

    This feature was first introduced in the "v0.3.1rc22" release.

  ngx.shared.DICT.get
    syntax: *value, flags = ngx.shared.DICT:get(key)*

    context: *set_by_lua*, rewrite_by_lua*, access_by_lua*, content_by_lua*,
    header_filter_by_lua*, body_filter_by_lua*, log_by_lua*, ngx.timer.**

    Retrieving the value in the dictionary ngx.shared.DICT for the key
    "key". If the key does not exist or has been expired, then "nil" will be
    returned.

    The value returned will have the original data type when they were
    inserted into the dictionary, for example, Lua booleans, numbers, or
    strings.

    The first argument to this method must be the dictionary object itself,
    for example,

        local cats = ngx.shared.cats
        local value, flags = cats.get(cats, "Marry")

    or use Lua's syntactic sugar for method calls:

        local cats = ngx.shared.cats
        local value, flags = cats:get("Marry")

    These two forms are fundamentally equivalent.

    If the user flags is 0 (the default), then no flags value will be
    returned.

    This feature was first introduced in the "v0.3.1rc22" release.

    See also ngx.shared.DICT.

  ngx.shared.DICT.set
    syntax: *success, err, forcible = ngx.shared.DICT:set(key, value,
    exptime?, flags?)*

    context: *init_by_lua*, set_by_lua*, rewrite_by_lua*, access_by_lua*,
    content_by_lua*, header_filter_by_lua*, body_filter_by_lua*,
    log_by_lua*, ngx.timer.**

    Unconditionally sets a key-value pair into the shm-based dictionary
    ngx.shared.DICT. Returns three values:

    *   "success": boolean value to indicate whether the key-value pair is
        stored or not.

    *   "err": textual error message, can be "no memory".

    *   "forcible": a boolean value to indicate whether other valid items
        have been removed forcibly when out of storage in the shared memory
        zone.

    The "value" argument inserted can be Lua booleans, numbers, strings, or
    "nil". Their value type will also be stored into the dictionary and the
    same data type can be retrieved later via the get method.

    The optional "exptime" argument specifies expiration time (in seconds)
    for the inserted key-value pair. The time resolution is 0.001 seconds.
    If the "exptime" takes the value 0 (which is the default), then the item
    will never be expired.

    The optional "flags" argument specifies a user flags value associated
    with the entry to be stored. It can also be retrieved later with the
    value. The user flags is stored as an unsigned 32-bit integer
    internally. Defaults to 0. The user flags argument was first introduced
    in the "v0.5.0rc2" release.

    When it fails to allocate memory for the current key-value item, then
    "set" will try removing existing items in the storage according to the
    Least-Recently Used (LRU) algorithm. Note that, LRU takes priority over
    expiration time here. If up to tens of existing items have been removed
    and the storage left is still insufficient (either due to the total
    capacity limit specified by lua_shared_dict or memory segmentation),
    then the "err" return value will be "no memory" and "success" will be
    "false".

    If this method succeeds in storing the current item by forcibly removing
    other not-yet-expired items in the dictionary via LRU, the "forcible"
    return value will be "true". If it stores the item without forcibly
    removing other valid items, then the return value "forcible" will be
    "false".

    The first argument to this method must be the dictionary object itself,
    for example,

        local cats = ngx.shared.cats
        local succ, err, forcible = cats.set(cats, "Marry", "it is a nice cat!")

    or use Lua's syntactic sugar for method calls:

        local cats = ngx.shared.cats
        local succ, err, forcible = cats:set("Marry", "it is a nice cat!")

    These two forms are fundamentally equivalent.

    This feature was first introduced in the "v0.3.1rc22" release.

    Please note that while internally the key-value pair is set atomically,
    the atomicity does not go across the method call boundary.

    See also ngx.shared.DICT.

  ngx.shared.DICT.safe_set
    syntax: *ok, err = ngx.shared.DICT:safe_set(key, value, exptime?,
    flags?)*

    context: *init_by_lua*, set_by_lua*, rewrite_by_lua*, access_by_lua*,
    content_by_lua*, header_filter_by_lua*, body_filter_by_lua*,
    log_by_lua*, ngx.timer.**

    Similar to the set method, but never overrides the (least recently used)
    unexpired items in the store when running out of storage in the shared
    memory zone. In this case, it will immediately return "nil" and the
    string "no memory".

    This feature was first introduced in the "v0.7.18" release.

    See also ngx.shared.DICT.

  ngx.shared.DICT.add
    syntax: *success, err, forcible = ngx.shared.DICT:add(key, value,
    exptime?, flags?)*

    context: *init_by_lua*, set_by_lua*, rewrite_by_lua*, access_by_lua*,
    content_by_lua*, header_filter_by_lua*, body_filter_by_lua*,
    log_by_lua*, ngx.timer.**

    Just like the set method, but only stores the key-value pair into the
    dictionary ngx.shared.DICT if the key does *not* exist.

    If the "key" argument already exists in the dictionary (and not expired
    for sure), the "success" return value will be "false" and the "err"
    return value will be "exists".

    This feature was first introduced in the "v0.3.1rc22" release.

    See also ngx.shared.DICT.

  ngx.shared.DICT.safe_add
    syntax: *ok, err = ngx.shared.DICT:safe_add(key, value, exptime?,
    flags?)*

    context: *init_by_lua*, set_by_lua*, rewrite_by_lua*, access_by_lua*,
    content_by_lua*, header_filter_by_lua*, body_filter_by_lua*,
    log_by_lua*, ngx.timer.**

    Similar to the add method, but never overrides the (least recently used)
    unexpired items in the store when running out of storage in the shared
    memory zone. In this case, it will immediately return "nil" and the
    string "no memory".

    This feature was first introduced in the "v0.7.18" release.

    See also ngx.shared.DICT.

  ngx.shared.DICT.replace
    syntax: *success, err, forcible = ngx.shared.DICT:replace(key, value,
    exptime?, flags?)*

    context: *init_by_lua*, set_by_lua*, rewrite_by_lua*, access_by_lua*,
    content_by_lua*, header_filter_by_lua*, body_filter_by_lua*,
    log_by_lua*, ngx.timer.**

    Just like the set method, but only stores the key-value pair into the
    dictionary ngx.shared.DICT if the key *does* exist.

    If the "key" argument does *not* exist in the dictionary (or expired
    already), the "success" return value will be "false" and the "err"
    return value will be "not found".

    This feature was first introduced in the "v0.3.1rc22" release.

    See also ngx.shared.DICT.

  ngx.shared.DICT.delete
    syntax: *ngx.shared.DICT:delete(key)*

    context: *init_by_lua*, set_by_lua*, rewrite_by_lua*, access_by_lua*,
    content_by_lua*, header_filter_by_lua*, body_filter_by_lua*,
    log_by_lua*, ngx.timer.**

    Unconditionally removes the key-value pair from the shm-based dictionary
    ngx.shared.DICT.

    It is equivalent to "ngx.shared.DICT:set(key, nil)".

    This feature was first introduced in the "v0.3.1rc22" release.

    See also ngx.shared.DICT.

  ngx.shared.DICT.incr
    syntax: *newval, err = ngx.shared.DICT:incr(key, value)*

    context: *init_by_lua*, set_by_lua*, rewrite_by_lua*, access_by_lua*,
    content_by_lua*, header_filter_by_lua*, body_filter_by_lua*,
    log_by_lua*, ngx.timer.**

    Increments the (numerical) value for "key" in the shm-based dictionary
    ngx.shared.DICT by the step value "value". Returns the new resulting
    number if the operation is successfully completed or "nil" and an error
    message otherwise.

    The key must already exist in the dictionary, otherwise it will return
    "nil" and "not found".

    If the original value is not a valid Lua number in the dictionary, it
    will return "nil" and "not a number".

    The "value" argument can be any valid Lua numbers, like negative numbers
    or floating-point numbers.

    This feature was first introduced in the "v0.3.1rc22" release.

    See also ngx.shared.DICT.

  ngx.shared.DICT.flush_all
    syntax: *ngx.shared.DICT:flush_all()*

    context: *init_by_lua*, set_by_lua*, rewrite_by_lua*, access_by_lua*,
    content_by_lua*, header_filter_by_lua*, body_filter_by_lua*,
    log_by_lua*, ngx.timer.**

    Flushes out all the items in the dictionary.

    This feature was first introduced in the "v0.5.0rc17" release.

    See also ngx.shared.DICT.flush_expired and ngx.shared.DICT.

  ngx.shared.DICT.flush_expired
    syntax: *flushed = ngx.shared.DICT:flush_expired(max_count?)*

    context: *init_by_lua*, set_by_lua*, rewrite_by_lua*, access_by_lua*,
    content_by_lua*, header_filter_by_lua*, body_filter_by_lua*,
    log_by_lua*, ngx.timer.**

    Flushes out the expired items in the dictionary, up to the maximal
    number specified by the optional "max_count" argument. When the
    "max_count" argument is given 0 or not given at all, then it means
    unlimited. Returns the number of items that have actually been flushed.

    This feature was first introduced in the "v0.6.3" release.

    See also ngx.shared.DICT.flush_all and ngx.shared.DICT.

  ngx.shared.DICT.get_keys
    syntax: *keys = ngx.shared.DICT:get_keys(max_count?)*

    context: *init_by_lua*, set_by_lua*, rewrite_by_lua*, access_by_lua*,
    content_by_lua*, header_filter_by_lua*, body_filter_by_lua*,
    log_by_lua*, ngx.timer.**

    Fetch a list of the keys from the dictionary, up to "<max_count>".

    By default, only the first 1024 keys (if any) are returned. When the
    "<max_count>" argument is given the value 0, then all the keys will be
    returned even there is more than 1024 keys in the dictionary.

    WARNING Be careful when calling this method on dictionaries with a
    really huge number of keys. This method may lock the dictionary for
    quite a while and block all the nginx worker processes that are trying
    to access the dictionary.

    This feature was first introduced in the "v0.7.3" release.

  ngx.socket.udp
    syntax: *udpsock = ngx.socket.udp()*

    context: *rewrite_by_lua*, access_by_lua*, content_by_lua*, ngx.timer.**

    Creates and returns a UDP or datagram-oriented unix domain socket object
    (also known as one type of the "cosocket" objects). The following
    methods are supported on this object:

    *   setpeername

    *   send

    *   receive

    *   close

    *   settimeout

    It is intended to be compatible with the UDP API of the LuaSocket
    (<http://w3.impa.br/~diego/software/luasocket/udp.html>) library but is
    100% nonblocking out of the box.

    This feature was first introduced in the "v0.5.7" release.

    See also ngx.socket.tcp.

  udpsock:setpeername
    syntax: *ok, err = udpsock:setpeername(host, port)*

    syntax: *ok, err =
    udpsock:setpeername("unix:/path/to/unix-domain.socket")*

    context: *rewrite_by_lua*, access_by_lua*, content_by_lua*, ngx.timer.**

    Attempts to connect a UDP socket object to a remote server or to a
    datagram unix domain socket file. Because the datagram protocol is
    actually connection-less, this method does not really establish a
    "connection", but only just set the name of the remote peer for
    subsequent read/write operations.

    Both IP addresses and domain names can be specified as the "host"
    argument. In case of domain names, this method will use Nginx core's
    dynamic resolver to parse the domain name without blocking and it is
    required to configure the resolver directive in the "nginx.conf" file
    like this:

        resolver 8.8.8.8;  # use Google's public DNS nameserver

    If the nameserver returns multiple IP addresses for the host name, this
    method will pick up one randomly.

    In case of error, the method returns "nil" followed by a string
    describing the error. In case of success, the method returns 1.

    Here is an example for connecting to a UDP (memcached) server:

        location /test {
            resolver 8.8.8.8;

            content_by_lua '
                local sock = ngx.socket.udp()
                local ok, err = sock:setpeername("my.memcached.server.domain", 11211)
                if not ok then
                    ngx.say("failed to connect to memcached: ", err)
                    return
                end
                ngx.say("successfully connected to memcached!")
                sock:close()
            ';
        }

    Since the "v0.7.18" release, connecting to a datagram unix domain socket
    file is also possible on Linux:

        local sock = ngx.socket.udp()
        local ok, err = sock:setpeername("unix:/tmp/some-datagram-service.sock")
        if not ok then
            ngx.say("failed to connect to the datagram unix domain socket: ", err)
            return
        end

    assuming the datagram service is listening on the unix domain socket
    file "/tmp/some-datagram-service.sock" and the client socket will use
    the "autobind" feature on Linux.

    Calling this method on an already connected socket object will cause the
    original connection to be closed first.

    This method was first introduced in the "v0.5.7" release.

  udpsock:send
    syntax: *ok, err = udpsock:send(data)*

    context: *rewrite_by_lua*, access_by_lua*, content_by_lua*, ngx.timer.**

    Sends data on the current UDP or datagram unix domain socket object.

    In case of success, it returns 1. Otherwise, it returns "nil" and a
    string describing the error.

    The input argument "data" can either be a Lua string or a (nested) Lua
    table holding string fragments. In case of table arguments, this method
    will copy all the string elements piece by piece to the underlying Nginx
    socket send buffers, which is usually optimal than doing string
    concatenation operations on the Lua land.

    This feature was first introduced in the "v0.5.7" release.

  udpsock:receive
    syntax: *data, err = udpsock:receive(size?)*

    context: *rewrite_by_lua*, access_by_lua*, content_by_lua*, ngx.timer.**

    Receives data from the UDP or datagram unix domain socket object with an
    optional receive buffer size argument, "size".

    This method is a synchronous operation and is 100% nonblocking.

    In case of success, it returns the data received; in case of error, it
    returns "nil" with a string describing the error.

    If the "size" argument is specified, then this method will use this size
    as the receive buffer size. But when this size is greater than 8192,
    then 8192 will be used instead.

    If no argument is specified, then the maximal buffer size, 8192 is
    assumed.

    Timeout for the reading operation is controlled by the
    lua_socket_read_timeout config directive and the settimeout method. And
    the latter takes priority. For example:

        sock:settimeout(1000)  -- one second timeout
        local data, err = sock:receive()
        if not data then
            ngx.say("failed to read a packet: ", data)
            return
        end
        ngx.say("successfully read a packet: ", data)

    It is important here to call the settimeout method *before* calling this
    method.

    This feature was first introduced in the "v0.5.7" release.

  udpsock:close
    syntax: *ok, err = udpsock:close()*

    context: *rewrite_by_lua*, access_by_lua*, content_by_lua*, ngx.timer.**

    Closes the current UDP or datagram unix domain socket. It returns the 1
    in case of success and returns "nil" with a string describing the error
    otherwise.

    Socket objects that have not invoked this method (and associated
    connections) will be closed when the socket object is released by the
    Lua GC (Garbage Collector) or the current client HTTP request finishes
    processing.

    This feature was first introduced in the "v0.5.7" release.

  udpsock:settimeout
    syntax: *udpsock:settimeout(time)*

    context: *rewrite_by_lua*, access_by_lua*, content_by_lua*, ngx.timer.**

    Set the timeout value in milliseconds for subsequent socket operations
    (like receive).

    Settings done by this method takes priority over those config
    directives, like lua_socket_read_timeout.

    This feature was first introduced in the "v0.5.7" release.

  ngx.socket.tcp
    syntax: *tcpsock = ngx.socket.tcp()*

    context: *rewrite_by_lua*, access_by_lua*, content_by_lua*, ngx.timer.**

    Creates and returns a TCP or stream-oriented unix domain socket object
    (also known as one type of the "cosocket" objects). The following
    methods are supported on this object:

    *   connect

    *   send

    *   receive

    *   close

    *   settimeout

    *   setoption

    *   receiveuntil

    *   setkeepalive

    *   getreusedtimes

    It is intended to be compatible with the TCP API of the LuaSocket
    (<http://w3.impa.br/~diego/software/luasocket/tcp.html>) library but is
    100% nonblocking out of the box. Also, we introduce some new APIs to
    provide more functionalities.

    This feature was first introduced in the "v0.5.0rc1" release.

    See also ngx.socket.udp.

  tcpsock:connect
    syntax: *ok, err = tcpsock:connect(host, port, options_table?)*

    syntax: *ok, err = tcpsock:connect("unix:/path/to/unix-domain.socket",
    options_table?)*

    context: *rewrite_by_lua*, access_by_lua*, content_by_lua*, ngx.timer.**

    Attempts to connect a TCP socket object to a remote server or to a
    stream unix domain socket file without blocking.

    Before actually resolving the host name and connecting to the remote
    backend, this method will always look up the connection pool for matched
    idle connections created by previous calls of this method (or the
    ngx.socket.connect function).

    Both IP addresses and domain names can be specified as the "host"
    argument. In case of domain names, this method will use Nginx core's
    dynamic resolver to parse the domain name without blocking and it is
    required to configure the resolver directive in the "nginx.conf" file
    like this:

        resolver 8.8.8.8;  # use Google's public DNS nameserver

    If the nameserver returns multiple IP addresses for the host name, this
    method will pick up one randomly.

    In case of error, the method returns "nil" followed by a string
    describing the error. In case of success, the method returns 1.

    Here is an example for connecting to a TCP server:

        location /test {
            resolver 8.8.8.8;

            content_by_lua '
                local sock = ngx.socket.tcp()
                local ok, err = sock:connect("www.google.com", 80)
                if not ok then
                    ngx.say("failed to connect to google: ", err)
                    return
                end
                ngx.say("successfully connected to google!")
                sock:close()
            ';
        }

    Connecting to a Unix Domain Socket file is also possible:

        local sock = ngx.socket.tcp()
        local ok, err = sock:connect("unix:/tmp/memcached.sock")
        if not ok then
            ngx.say("failed to connect to the memcached unix domain socket: ", err)
            return
        end

    assuming memcached (or something else) is listening on the unix domain
    socket file "/tmp/memcached.sock".

    Timeout for the connecting operation is controlled by the
    lua_socket_connect_timeout config directive and the settimeout method.
    And the latter takes priority. For example:

        local sock = ngx.socket.tcp()
        sock:settimeout(1000)  -- one second timeout
        local ok, err = sock:connect(host, port)

    It is important here to call the settimeout method *before* calling this
    method.

    Calling this method on an already connected socket object will cause the
    original connection to be closed first.

    An optional Lua table can be specified as the last argument to this
    method to specify various connect options:

    *   "pool" specify a custom name for the connection pool being used. If
        omitted, then the connection pool name will be generated from the
        string template "<host>:<port>" or "<unix-socket-path>".

    The support for the options table argument was first introduced in the
    "v0.5.7" release.

    This method was first introduced in the "v0.5.0rc1" release.

  tcpsock:send
    syntax: *bytes, err = tcpsock:send(data)*

    context: *rewrite_by_lua*, access_by_lua*, content_by_lua*, ngx.timer.**

    Sends data without blocking on the current TCP or Unix Domain Socket
    connection.

    This method is a synchronous operation that will not return until *all*
    the data has been flushed into the system socket send buffer or an error
    occurs.

    In case of success, it returns the total number of bytes that have been
    sent. Otherwise, it returns "nil" and a string describing the error.

    The input argument "data" can either be a Lua string or a (nested) Lua
    table holding string fragments. In case of table arguments, this method
    will copy all the string elements piece by piece to the underlying Nginx
    socket send buffers, which is usually optimal than doing string
    concatenation operations on the Lua land.

    Timeout for the sending operation is controlled by the
    lua_socket_send_timeout config directive and the settimeout method. And
    the latter takes priority. For example:

        sock:settimeout(1000)  -- one second timeout
        local bytes, err = sock:send(request)

    It is important here to call the settimeout method *before* calling this
    method.

    This feature was first introduced in the "v0.5.0rc1" release.

  tcpsock:receive
    syntax: *data, err, partial = tcpsock:receive(size)*

    syntax: *data, err, partial = tcpsock:receive(pattern?)*

    context: *rewrite_by_lua*, access_by_lua*, content_by_lua*, ngx.timer.**

    Receives data from the connected socket according to the reading pattern
    or size.

    This method is a synchronous operation just like the send method and is
    100% nonblocking.

    In case of success, it returns the data received; in case of error, it
    returns "nil" with a string describing the error and the partial data
    received so far.

    If a number-like argument is specified (including strings that look like
    numbers), then it is interpreted as a size. This method will not return
    until it reads exactly this size of data or an error occurs.

    If a non-number-like string argument is specified, then it is
    interpreted as a "pattern". The following patterns are supported:

    *   '*a': reads from the socket until the connection is closed. No
        end-of-line translation is performed;

    *   '*l': reads a line of text from the socket. The line is terminated
        by a "Line Feed" (LF) character (ASCII 10), optionally preceded by a
        "Carriage Return" (CR) character (ASCII 13). The CR and LF
        characters are not included in the returned line. In fact, all CR
        characters are ignored by the pattern.

    If no argument is specified, then it is assumed to be the pattern '*l',
    that is, the line reading pattern.

    Timeout for the reading operation is controlled by the
    lua_socket_read_timeout config directive and the settimeout method. And
    the latter takes priority. For example:

        sock:settimeout(1000)  -- one second timeout
        local line, err, partial = sock:receive()
        if not line then
            ngx.say("failed to read a line: ", err)
            return
        end
        ngx.say("successfully read a line: ", line)

    It is important here to call the settimeout method *before* calling this
    method.

    This feature was first introduced in the "v0.5.0rc1" release.

  tcpsock:receiveuntil
    syntax: *iterator = tcpsock:receiveuntil(pattern, options?)*

    context: *rewrite_by_lua*, access_by_lua*, content_by_lua*, ngx.timer.**

    This method returns an iterator Lua function that can be called to read
    the data stream until it sees the specified pattern or an error occurs.

    Here is an example for using this method to read a data stream with the
    boundary sequence "--abcedhb":

        local reader = sock:receiveuntil("\r\n--abcedhb")
        local data, err, partial = reader()
        if not data then
            ngx.say("failed to read the data stream: ", err)
        end
        ngx.say("read the data stream: ", data)

    When called without any argument, the iterator function returns the
    received data right *before* the specified pattern string in the
    incoming data stream. So for the example above, if the incoming data
    stream is 'hello, world! -agentzh\r\n--abcedhb blah blah', then the
    string 'hello, world! -agentzh' will be returned.

    In case of error, the iterator function will return "nil" along with a
    string describing the error and the partial data bytes that have been
    read so far.

    The iterator function can be called multiple times and can be mixed
    safely with other cosocket method calls or other iterator function
    calls.

    The iterator function behaves differently (i.e., like a real iterator)
    when it is called with a "size" argument. That is, it will read that
    "size" of data on each invocation and will return "nil" at the last
    invocation (either sees the boundary pattern or meets an error). For the
    last successful invocation of the iterator function, the "err" return
    value will be "nil" too. The iterator function will be reset after the
    last successful invocation that returns "nil" data and "nil" error.
    Consider the following example:

        local reader = sock:receiveuntil("\r\n--abcedhb")

        while true do
            local data, err, partial = reader(4)
            if not data then
                if err then
                    ngx.say("failed to read the data stream: ", err)
                    break
                end

                ngx.say("read done")
                break
            end
            ngx.say("read chunk: [", data, "]")
        end

    Then for the incoming data stream 'hello, world! -agentzh\r\n--abcedhb
    blah blah', we shall get the following output from the sample code
    above:

        read chunk: [hell]
        read chunk: [o, w]
        read chunk: [orld]
        read chunk: [! -a]
        read chunk: [gent]
        read chunk: [zh]
        read done

    Note that, the actual data returned *might* be a little longer than the
    size limit specified by the "size" argument when the boundary pattern
    has ambiguity for streaming parsing. Near the boundary of the data
    stream, the data string actually returned could also be shorter than the
    size limit.

    Timeout for the iterator function's reading operation is controlled by
    the lua_socket_read_timeout config directive and the settimeout method.
    And the latter takes priority. For example:

        local readline = sock:receiveuntil("\r\n")

        sock:settimeout(1000)  -- one second timeout
        line, err, partial = readline()
        if not line then
            ngx.say("failed to read a line: ", err)
            return
        end
        ngx.say("successfully read a line: ", line)

    It is important here to call the settimeout method *before* calling the
    iterator function (note that the "receiveuntil" call is irrelevant
    here).

    As from the "v0.5.1" release, this method also takes an optional
    "options" table argument to control the behavior. The following options
    are supported:

    *   "inclusive"

    The "inclusive" takes a boolean value to control whether to include the
    pattern string in the returned data string. Default to "false". For
    example,

        local reader = tcpsock:receiveuntil("_END_", { inclusive = true })
        local data = reader()
        ngx.say(data)

    Then for the input data stream "hello world _END_ blah blah blah", then
    the example above will output "hello world _END_", including the pattern
    string "_END_" itself.

    This method was first introduced in the "v0.5.0rc1" release.

  tcpsock:close
    syntax: *ok, err = tcpsock:close()*

    context: *rewrite_by_lua*, access_by_lua*, content_by_lua*, ngx.timer.**

    Closes the current TCP or stream unix domain socket. It returns the 1 in
    case of success and returns "nil" with a string describing the error
    otherwise.

    Note that there is no need to call this method on socket objects that
    have invoked the setkeepalive method because the socket object is
    already closed (and the current connection is saved into the built-in
    connection pool).

    Socket objects that have not invoked this method (and associated
    connections) will be closed when the socket object is released by the
    Lua GC (Garbage Collector) or the current client HTTP request finishes
    processing.

    This feature was first introduced in the "v0.5.0rc1" release.

  tcpsock:settimeout
    syntax: *tcpsock:settimeout(time)*

    context: *rewrite_by_lua*, access_by_lua*, content_by_lua*, ngx.timer.**

    Set the timeout value in milliseconds for subsequent socket operations
    (connect, receive, and iterators returned from receiveuntil).

    Settings done by this method takes priority over those config
    directives, i.e., lua_socket_connect_timeout, lua_socket_send_timeout,
    and lua_socket_read_timeout.

    Note that this method does *not* affect the lua_socket_keepalive_timeout
    setting; the "timeout" argument to the setkeepalive method should be
    used for this purpose instead.

    This feature was first introduced in the "v0.5.0rc1" release.

  tcpsock:setoption
    syntax: *tcpsock:setoption(option, value?)*

    context: *rewrite_by_lua*, access_by_lua*, content_by_lua*, ngx.timer.**

    This function is added for LuaSocket
    (<http://w3.impa.br/~diego/software/luasocket/tcp.html>) API
    compatibility and does nothing for now. Its functionality will be
    implemented in future.

    This feature was first introduced in the "v0.5.0rc1" release.

  tcpsock:setkeepalive
    syntax: *ok, err = tcpsock:setkeepalive(timeout?, size?)*

    context: *rewrite_by_lua*, access_by_lua*, content_by_lua*, ngx.timer.**

    Puts the current socket's connection immediately into the cosocket
    built-in connection pool and keep it alive until other connect method
    calls request it or the associated maximal idle timeout is expired.

    The first optional argument, "timeout", can be used to specify the
    maximal idle timeout (in milliseconds) for the current connection. If
    omitted, the default setting in the lua_socket_keepalive_timeout config
    directive will be used. If the 0 value is given, then the timeout
    interval is unlimited.

    The second optional argument, "size", can be used to specify the maximal
    number of connections allowed in the connection pool for the current
    server (i.e., the current host-port pair or the unix domain socket file
    path). Note that the size of the connection pool cannot be changed once
    the pool is created. When this argument is omitted, the default setting
    in the lua_socket_pool_size config directive will be used.

    When the connection pool exceeds the available size limit, the least
    recently used (idle) connection already in the pool will be closed to
    make room for the current connection.

    Note that the cosocket connection pool is per Nginx worker process
    rather than per Nginx server instance, so the size limit specified here
    also applies to every single Nginx worker process.

    Idle connections in the pool will be monitored for any exceptional
    events like connection abortion or unexpected incoming data on the line,
    in which cases the connection in question will be closed and removed
    from the pool.

    In case of success, this method returns 1; otherwise, it returns "nil"
    and a string describing the error.

    This method also makes the current cosocket object enter the "closed"
    state, so there is no need to manually call the close method on it
    afterwards.

    This feature was first introduced in the "v0.5.0rc1" release.

  tcpsock:getreusedtimes
    syntax: *count, err = tcpsock:getreusedtimes()*

    context: *rewrite_by_lua*, access_by_lua*, content_by_lua*, ngx.timer.**

    This method returns the (successfully) reused times for the current
    connection. In case of error, it returns "nil" and a string describing
    the error.

    If the current connection does not come from the built-in connection
    pool, then this method always returns 0, that is, the connection has
    never been reused (yet). If the connection comes from the connection
    pool, then the return value is always non-zero. So this method can also
    be used to determine if the current connection comes from the pool.

    This feature was first introduced in the "v0.5.0rc1" release.

  ngx.socket.connect
    syntax: *tcpsock, err = ngx.socket.connect(host, port)*

    syntax: *tcpsock, err =
    ngx.socket.connect("unix:/path/to/unix-domain.socket")*

    context: *rewrite_by_lua*, access_by_lua*, content_by_lua*, ngx.timer.**

    This function is a shortcut for combining ngx.socket.tcp() and the
    connect() method call in a single operation. It is actually implemented
    like this:

        local sock = ngx.socket.tcp()
        local ok, err = sock:connect(...)
        if not ok then
            return nil, err
        end
        return sock

    There is no way to use the settimeout method to specify connecting
    timeout for this method and the lua_socket_connect_timeout directive
    must be set at configure time instead.

    This feature was first introduced in the "v0.5.0rc1" release.

  ngx.get_phase
    syntax: *str = ngx.get_phase()*

    context: *init_by_lua*, set_by_lua*, rewrite_by_lua*, access_by_lua*,
    content_by_lua*, header_filter_by_lua*, body_filter_by_lua*,
    log_by_lua*, ngx.timer.**

    Retrieves the current running phase name. Possible return values are

    *   "init" for the context of init_by_lua or init_by_lua_file. =item *

        "set" for the context of set_by_lua or set_by_lua_file. =item *

        "rewrite" for the context of rewrite_by_lua or rewrite_by_lua_file.
        =item *

        "access" for the context of access_by_lua or access_by_lua_file.
        =item *

        "content" for the context of content_by_lua or content_by_lua_file.
        =item *

        "header_filter" for the context of header_filter_by_lua or
        header_filter_by_lua_file. =item *

        "body_filter" for the context of body_filter_by_lua or
        body_filter_by_lua_file. =item *

        "log" for the context of log_by_lua or log_by_lua_file. =item *

        "timer" for the context of user callback functions for ngx.timer.*.

    This API was first introduced in the "v0.5.10" release.

  ngx.thread.spawn
    syntax: *co = ngx.thread.spawn(func, arg1, arg2, ...)*

    context: *rewrite_by_lua*, access_by_lua*, content_by_lua*, ngx.timer.**

    Spawns a new user "light thread" with the Lua function "func" as well as
    those optional arguments "arg1", "arg2", and etc. Returns a Lua thread
    (or Lua coroutine) object represents this "light thread".

    "Light threads" are just a special kind of Lua coroutines that are
    scheduled by the ngx_lua module.

    Before "ngx.thread.spawn" returns, the "func" will be called with those
    optional arguments until it returns, aborts with an error, or gets
    yielded due to I/O operations via the Nginx API for Lua (like
    <tcpsock:receive|/"tcpsock:receive">).

    After "ngx.thread.spawn" returns, the newly-created "light thread" will
    keep running asynchronously usually at various I/O events.

    All the Lua code chunks running by rewrite_by_lua, access_by_lua, and
    content_by_lua are in a boilerplate "light thread" created automatically
    by ngx_lua. Such boilerplate "light thread" are also called "entry
    threads".

    By default, the corresponding Nginx handler (e.g., rewrite_by_lua
    handler) will not terminate until

    1.  both the "entry thread" and all the user "light threads" terminates,

    2.  a "light thread" (either the "entry thread" or a user "light thread"
        aborts by calling ngx.exit, ngx.exec, ngx.redirect, or
        ngx.req.set_uri(uri, true), or

    3.  the "entry thread" terminates with a Lua error.

    When the user "light thread" terminates with a Lua error, however, it
    will not abort other running "light threads" like the "entry thread"
    does.

    Due to the limitation in the Nginx subrequest model, it is not allowed
    to abort a running Nginx subrequest in general. So it is also prohibited
    to abort a running "light thread" that is pending on one ore more Nginx
    subrequests. You must call ngx.thread.wait to wait for those "light
    thread" to terminate before quitting the "world". A notable exception
    here is that you can abort pending subrequests by calling ngx.exit with
    and only with the status code "ngx.ERROR" (-1), 408, 444, or 499.

    The "light threads" are not scheduled in a pre-emptive way. In other
    words, no time-slicing is performed automatically. A "light thread" will
    keep running exclusively on the CPU until

    1.  a (nonblocking) I/O operation cannot be completed in a single run,

    2.  it calls coroutine.yield to actively give up execution, or

    3.  it is aborted by a Lua error or an invocation of ngx.exit, ngx.exec,
        ngx.redirect, or ngx.req.set_uri(uri, true).

    For the first two cases, the "light thread" will usually be resumed
    later by the ngx_lua scheduler unless a "stop-the-world" event happens.

    User "light threads" can create "light threads" themselves and normal
    user coroutiens created by coroutine.create can also create "light
    threads". The coroutine (be it a normal Lua coroutine or a "light
    thread") that directly spawns the "light thread" is called the "parent
    coroutine" for the "light thread" newly spawned.

    The "parent coroutine" can call ngx.thread.wait to wait on the
    termination of its child "light thread".

    You can call coroutine.status() and coroutine.yield() on the "light
    thread" coroutines.

    The status of the "light thread" coroutine can be "zombie" if

    1.  the current "light thread" already terminates (either successfully
        or with an error),

    2.  its parent coroutine is still alive, and

    3.  its parent coroutine is not waiting on it with ngx.thread.wait.

    The following example demonstrates the use of coroutine.yield() in the
    "light thread" coroutines to do manual time-slicing:

        local yield = coroutine.yield

        function f()
            local self = coroutine.running()
            ngx.say("f 1")
            yield(self)
            ngx.say("f 2")
            yield(self)
            ngx.say("f 3")
        end

        local self = coroutine.running()
        ngx.say("0")
        yield(self)

        ngx.say("1")
        ngx.thread.spawn(f)

        ngx.say("2")
        yield(self)

        ngx.say("3")
        yield(self)

        ngx.say("4")

    Then it will generate the output

        0
        1
        f 1
        2
        f 2
        3
        f 3
        4

    "Light threads" are mostly useful for doing concurrent upstream requests
    in a single Nginx request handler, kinda like a generalized version of
    ngx.location.capture_multi that can work with all the Nginx API for Lua.
    The following example demonstrates parallel requests to MySQL,
    Memcached, and upstream HTTP services in a single Lua handler, and
    outputting the results in the order that they actually return (very much
    like the Facebook BigPipe model):

        -- query mysql, memcached, and a remote http service at the same time,
        -- output the results in the order that they
        -- actually return the results.

        local mysql = require "resty.mysql"
        local memcached = require "resty.memcached"

        local function query_mysql()
            local db = mysql:new()
            db:connect{
                        host = "127.0.0.1",
                        port = 3306,
                        database = "test",
                        user = "monty",
                        password = "mypass"
                      }
            local res, err, errno, sqlstate =
                    db:query("select * from cats order by id asc")
            db:set_keepalive(0, 100)
            ngx.say("mysql done: ", cjson.encode(res))
        end

        local function query_memcached()
            local memc = memcached:new()
            memc:connect("127.0.0.1", 11211)
            local res, err = memc:get("some_key")
            ngx.say("memcached done: ", res)
        end

        local function query_http()
            local res = ngx.location.capture("/my-http-proxy")
            ngx.say("http done: ", res.body)
        end

        ngx.thread.spawn(query_mysql)      -- create thread 1
        ngx.thread.spawn(query_memcached)  -- create thread 2
        ngx.thread.spawn(query_http)       -- create thread 3

    This API was first enabled in the "v0.7.0" release.

  ngx.thread.wait
    syntax: *ok, res1, res2, ... = ngx.thread.wait(thread1, thread2, ...)*

    context: *rewrite_by_lua*, access_by_lua*, content_by_lua*, ngx.timer.**

    Waits on one or more child "light threads" and returns the results of
    the first "light thread" that terminates (either successfully or with an
    error).

    The arguments "thread1", "thread2", and etc are the Lua thread objects
    returned by earlier calls of ngx.thread.spawn.

    The return values have exactly the same meaning as coroutine.resume,
    that is, the first value returned is a boolean value indicating whether
    the "light thread" terminates successfully or not, and subsequent values
    returned are the return values of the user Lua function that was used to
    spawn the "light thread" (in case of success) or the error object (in
    case of failure).

    Only the direct "parent coroutine" can wait on its child "light thread",
    otherwise a Lua exception will be raised.

    The following example demonstrates the use of "ngx.thread.wait" and
    ngx.location.capture to emulate ngx.location.capture_multi:

        local capture = ngx.location.capture
        local spawn = ngx.thread.spawn
        local wait = ngx.thread.wait
        local say = ngx.say

        local function fetch(uri)
            return capture(uri)
        end

        local threads = {
            spawn(fetch, "/foo"),
            spawn(fetch, "/bar"),
            spawn(fetch, "/baz")
        }

        for i = 1, #threads do
            local ok, res = wait(threads[i])
            if not ok then
                say(i, ": failed to run: ", res)
            else
                say(i, ": status: ", res.status)
                say(i, ": body: ", res.body)
            end
        end

    Here it essentially implements the "wait all" model.

    And below is an example demonstrating the "wait any" model:

        function f()
            ngx.sleep(0.2)
            ngx.say("f: hello")
            return "f done"
        end

        function g()
            ngx.sleep(0.1)
            ngx.say("g: hello")
            return "g done"
        end

        local tf, err = ngx.thread.spawn(f)
        if not tf then
            ngx.say("failed to spawn thread f: ", err)
            return
        end

        ngx.say("f thread created: ", coroutine.status(tf))

        local tg, err = ngx.thread.spawn(g)
        if not tg then
            ngx.say("failed to spawn thread g: ", err)
            return
        end

        ngx.say("g thread created: ", coroutine.status(tg))

        ok, res = ngx.thread.wait(tf, tg)
        if not ok then
            ngx.say("failed to wait: ", res)
            return
        end

        ngx.say("res: ", res)

        -- stop the "world", aborting other running threads
        ngx.exit(ngx.OK)

    And it will generate the following output:

        f thread created: running
        g thread created: running
        g: hello
        res: g done

    This API was first enabled in the "v0.7.0" release.

  ngx.on_abort
    syntax: *ok, err = ngx.on_abort(callback)*

    context: *rewrite_by_lua*, access_by_lua*, content_by_lua**

    Registers a user Lua function as the callback which gets called
    automatically when the client closes the (downstream) connection
    prematurely.

    Returns 1 if the callback is registered successfully or returns "nil"
    and a string describing the error otherwise.

    All the Nginx API for Lua can be used in the callback function because
    the function is run in a special "light thread", just as those "light
    threads" created by ngx.thread.spawn.

    The callback function can decide what to do with the client abortion
    event all by itself. For example, it can simply ignore the event by
    doing nothing and the current Lua request handler will continue
    executing without interruptions. And the callback function can also
    decide to terminate everything by calling ngx.exit, for example,

        local function my_cleanup()
            -- custom cleanup work goes here, like cancelling a pending DB transaction

            -- now abort all the "light threads" running in the current request handler
            ngx.exit(499)
        end

        local ok, err = ngx.on_abort(my_cleanup)
        if not ok then
            ngx.log(ngx.ERR, "failed to register the on_abort callback: ", err)
            ngx.exit(500)
        end

    When lua_check_client_abort is set to "off" (which is the default), then
    this function call will always return the error message
    "lua_check_client_abort is off".

    According to the current implementation, this function can only be
    called once in a single request handler; subsequent calls will return
    the error message "duplicate call".

    This API was first introduced in the "v0.7.4" release.

    See also lua_check_client_abort.

  ngx.timer.at
    syntax: *ok, err = ngx.timer.at(delay, callback, user_arg1, user_arg2,
    ...)*

    context: *set_by_lua*, rewrite_by_lua*, access_by_lua*, content_by_lua*,
    header_filter_by_lua*, body_filter_by_lua*, log_by_lua*, ngx.timer.**

    Creates an Nginx timer with a user callback function as well as optional
    user arguments.

    The first argument, "delay", specifies the delay for the timer, in
    seconds. One can specify fractional seconds like 0.001 to mean 1
    millisecond here. 0 delay can also be specified, in which case the timer
    will immediately expire when the current handler yields execution.

    The second argument, "callback", can be any Lua function, which will be
    invoked later in a background "light thread" after the delay specified.
    The user callback will be called automatically by the Nginx core with
    the arguments "premature", "user_arg1", "user_arg2", and etc, where the
    "premature" argument takes a boolean value indicating whether it is a
    premature timer expiration or not, and "user_arg1", "user_arg2", and
    etc, are those (extra) user arguments specified when calling
    "ngx.timer.at" as the remaining arguments.

    Premature timer expiration happens when the Nginx worker process is
    trying to shut down, as in an Nginx configuration reload triggered by
    the "HUP" signal or in an Nginx server shutdown. When the Nginx worker
    is trying to shut down, one can no longer call "ngx.timer.at" to create
    new timers and in that case "ngx.timer.at" will return "nil" and a
    string describing the error, that is, "process exiting".

    When a timer expires, the user Lua code in the timer callback is running
    in a "light thread" detached completely from the original request
    creating the timer. So objects with the same lifetime as the request
    creating them, like cosockets, cannot be shared between the original
    request and the timer user callback function.

    Here is a simple example:

        location / {
            ...
            log_by_lua '
                local function push_data(premature, uri, args, status)
                    -- push the data uri, args, and status to the remote
                    -- via ngx.socket.tcp or ngx.socket.udp
                    -- (one may want to buffer the data in Lua a bit to
                    -- save I/O operations)
                end
                local ok, err = ngx.timer.at(0, push_data,
                                             ngx.var.uri, ngx.var.args, ngx.header.status)
                if not ok then
                    ngx.log(ngx.ERR, "failed to create timer: ", err)
                    return
                end
            ';
        }

    One can also create infinite re-occuring timers, for instance, a timer
    getting triggered every 5 seconds, by calling "ngx.timer.at" recursively
    in the timer callback function. Here is such an example,

        local delay = 5
        local handler
        handler = function (premature)
            -- do some routine job in Lua just like a cron job
            if premature then
                return
            end
            local ok, err = ngx.timer.at(delay, handler)
            if not ok then
                ngx.log(ngx.ERR, "failed to create the timer: ", err)
                return
            end
        end

        local ok, err = ngx.timer.at(delay, handler)
        if not ok then
            ngx.log(ngx.ERR, "failed to create the timer: ", err)
            return
        end

    Because timer callbacks run in the background and their running time
    will not add to any client request's response time, they can easily
    accumulate in the server and exhaust system resources due to either Lua
    programming mistakes or just too much client traffic. To prevent extreme
    consequences like crashing the Nginx server, there are built-in
    limitations on both the number of "pending timers" and the number of
    "running timers" in an Nginx worker process. The "pending timers" here
    mean timers that have not yet been expired and "running timers" are
    those whose user callbacks are currently running.

    The maximal number of pending timers allowed in an Nginx worker is
    constrolled by the lua_max_pending_timers directive. The maximal number
    of running timers is controlled by the lua_max_running_timers directive.

    According to the current implementation, each "running timer" will take
    one (fake) connection record from the global connection record list
    configured by the standard worker_connections directive in "nginx.conf".
    So ensure that the worker_connections directive is set to a large enough
    value that takes into account both the real connections and fake
    connections required by timer callbacks (as limited by the
    lua_max_running_timers directive).

    A lot of the Lua APIs for Nginx are enabled in the context of the timer
    callbacks, like stream/datagram cosockets (ngx.socket.tcp and
    ngx.socket.udp), shared memory dictionaries (ngx.shared.DICT), user
    coroutines (coroutine.*), user "light threads" (ngx.thread.*), ngx.exit,
    ngx.now/ngx.time, ngx.md5/ngx.sha1_bin, are all allowed. But the
    subrequest API (like ngx.location.capture), the ngx.req.* API, the
    downstream output API (like ngx.say, ngx.print, and ngx.flush) are
    explicitly disabled in this context.

    This API was first introduced in the "v0.8.0" release.

  ndk.set_var.DIRECTIVE
    syntax: *res = ndk.set_var.DIRECTIVE_NAME*

    context: *set_by_lua*, rewrite_by_lua*, access_by_lua*, content_by_lua*,
    header_filter_by_lua*, body_filter_by_lua*, log_by_lua*, ngx.timer.**

    This mechanism allows calling other nginx C modules' directives that are
    implemented by Nginx Devel Kit
    (<https://github.com/simpl/ngx_devel_kit>) (NDK)'s set_var submodule's
    "ndk_set_var_value".

    For example, the following [[HttpSetMiscModule]] directives can be
    invoked this way:

    *   set_quote_sql_str

    *   set_quote_pgsql_str

    *   set_quote_json_str

    *   set_unescape_uri

    *   set_escape_uri

    *   set_encode_base32

    *   set_decode_base32

    *   set_encode_base64

    *   set_decode_base64

    *   set_encode_hex

    *   set_decode_hex

    *   set_sha1

    *   set_md5

    For instance,

        local res = ndk.set_var.set_escape_uri('a/b');
        -- now res == 'a%2fb'

    Similarly, the following directives provided by
    [[HttpEncryptedSessionModule]] can be invoked from within Lua too:

    *   set_encrypt_session

    *   set_decrypt_session

    This feature requires the ngx_devel_kit
    (<https://github.com/simpl/ngx_devel_kit>) module.

  coroutine.create
    syntax: *co = coroutine.create(f)*

    context: *rewrite_by_lua*, access_by_lua*, content_by_lua*, ngx.timer.**

    Creates a user Lua coroutines with a Lua function, and returns a
    coroutine object.

    Similar to the standard Lua coroutine.create
    (<http://www.lua.org/manual/5.1/manual.html#pdf-coroutine.create>) API,
    but works in the context of the Lua coroutines created by ngx_lua.

    This API was first introduced in the "v0.6.0" release.

  coroutine.resume
    syntax: *ok, ... = coroutine.resume(co, ...)*

    context: *rewrite_by_lua*, access_by_lua*, content_by_lua*, ngx.timer.**

    Resumes the executation of a user Lua coroutine object previously
    yielded or just created.

    Similar to the standard Lua coroutine.resume
    (<http://www.lua.org/manual/5.1/manual.html#pdf-coroutine.resume>) API,
    but works in the context of the Lua coroutines created by ngx_lua.

    This API was first introduced in the "v0.6.0" release.

  coroutine.yield
    syntax: *... = coroutine.yield(co, ...)*

    context: *rewrite_by_lua*, access_by_lua*, content_by_lua*, ngx.timer.**

    Yields the executation of the current user Lua coroutine.

    Similar to the standard Lua coroutine.yield
    (<http://www.lua.org/manual/5.1/manual.html#pdf-coroutine.yield>) API,
    but works in the context of the Lua coroutines created by ngx_lua.

    This API was first introduced in the "v0.6.0" release.

  coroutine.wrap
    syntax: *co = coroutine.wrap(f)*

    context: *rewrite_by_lua*, access_by_lua*, content_by_lua*, ngx.timer.**

    Similar to the standard Lua coroutine.wrap
    (<http://www.lua.org/manual/5.1/manual.html#pdf-coroutine.wrap>) API,
    but works in the context of the Lua coroutines created by ngx_lua.

    This API was first introduced in the "v0.6.0" release.

  coroutine.running
    syntax: *co = coroutine.running()*

    context: *rewrite_by_lua*, access_by_lua*, content_by_lua*, ngx.timer.**

    Identical to the standard Lua coroutine.running
    (<http://www.lua.org/manual/5.1/manual.html#pdf-coroutine.running>) API.

    This API was first enabled in the "v0.6.0" release.

  coroutine.status
    syntax: *status = coroutine.status(co)*

    context: *rewrite_by_lua*, access_by_lua*, content_by_lua*, ngx.timer.**

    Identical to the standard Lua coroutine.status
    (<http://www.lua.org/manual/5.1/manual.html#pdf-coroutine.status>) API.

    This API was first enabled in the "v0.6.0" release.

Lua/LuaJIT bytecode support
    As from the "v0.5.0rc32" release, all *_by_lua_file configure directives
    (such as content_by_lua_file) support loading Lua 5.1 and LuaJIT 2.0 raw
    bytecode files directly.

    Please note that the bytecode format used by LuaJIT 2.0 is not
    compatible with that used by the standard Lua 5.1 interpreter. So if
    using LuaJIT 2.0 with ngx_lua, LuaJIT compatible bytecode files must be
    generated as shown:

        /path/to/luajit/bin/luajit -b /path/to/input_file.lua /path/to/output_file.luac

    The "-bg" option can be used to include debug information in the LuaJIT
    bytecode file:

        /path/to/luajit/bin/luajit -bg /path/to/input_file.lua /path/to/output_file.luac

    Please refer to the official LuaJIT documentation on the "-b" option for
    more details:

    http://luajit.org/running.html#opt_b

    Similarly, if using the standard Lua 5.1 interpreter with ngx_lua, Lua
    compatible bytecode files must be generated using the "luac" commandline
    utility as shown:

        luac -o /path/to/output_file.luac /path/to/input_file.lua

    Unlike as with LuaJIT, debug information is included in standard Lua 5.1
    bytecode files by default. This can be striped out by specifying the
    "-s" option as shown:

        luac -s -o /path/to/output_file.luac /path/to/input_file.lua

    Attempts to load standard Lua 5.1 bytecode files into ngx_lua instances
    linked to LuaJIT 2.0 or vice versa, will result in an error message,
    such as that below, being logged into the Nginx "error.log" file:

        [error] 13909#0: *1 failed to load Lua inlined code: bad byte-code header in /path/to/test_file.luac

    Loading bytecode files via the Lua primitives like "require" and
    "dofile" should always work as expected.

HTTP 1.0 support
    The HTTP 1.0 protocol does not support chunked output and requires an
    explicit "Content-Length" header when the response body is not empty in
    order to support the HTTP 1.0 keep-alive. So when a HTTP 1.0 request is
    made and the lua_http10_buffering directive is turned "on", ngx_lua will
    buffer the output of ngx.say and ngx.print calls and also postpone
    sending response headers until all the response body output is received.
    At that time ngx_lua can calculate the total length of the body and
    construct a proper "Content-Length" header to return to the HTTP 1.0
    client. If the "Content-Length" response header is set in the running
    Lua code, however, this buffering will be disabled even if the
    lua_http10_buffering directive is turned "on".

    For large streaming output responses, it is important to disable the
    lua_http10_buffering directive to minimise memory usage.

    Note that common HTTP benchmark tools such as "ab" and "http_load" issue
    HTTP 1.0 requests by default. To force "curl" to send HTTP 1.0 requests,
    use the -0 option.

Data Sharing within an Nginx Worker
    To globally share data among all the requests handled by the same nginx
    worker process, encapsulate the shared data into a Lua module, use the
    Lua "require" builtin to import the module, and then manipulate the
    shared data in Lua. This works because required Lua modules are loaded
    only once and all coroutines will share the same copy of the module
    (both its code and data). Note however that Lua global variables (note,
    not module-level variables) WILL NOT persist between requests because of
    the one-coroutine-per-request isolation design.

    Here is a complete small example:

        -- mydata.lua
        module(...)

        local data = {
            dog = 3,
            cat = 4,
            pig = 5,
        }

        function get_age(name)
            return data[name]
        end

    and then accessing it from "nginx.conf":

        location /lua {
            content_by_lua '
                local mydata = require "mydata"
                ngx.say(mydata.get_age("dog"))
            ';
        }

    The "mydata" module in this example will only be loaded and run on the
    first request to the location "/lua", and all subsequent requests to the
    same nginx worker process will use the reloaded instance of the module
    as well as the same copy of the data in it, until a "HUP" signal is sent
    to the Nginx master process to force a reload. This data sharing
    technique is essential for high performance Lua applications based on
    this module.

    Note that this data sharing is on a *per-worker* basis and not on a
    *per-server* basis. That is, when there are multiple nginx worker
    processes under an Nginx master, data sharing cannot cross the process
    boundary between these workers.

    If server-wide data sharing is required, then use one or more of the
    following approaches:

    1.  Use the ngx.shared.DICT API provided by this module.

    2.  Use only a single nginx worker and a single server (this is however
        not recommended when there is a multi core CPU or multiple CPUs in a
        single machine).

    3.  Use data storage mechanisms such as "memcached", "redis", "MySQL" or
        "PostgreSQL". The ngx_openresty bundle (<http://openresty.org>)
        associated with this module comes with a set of companion Nginx
        modules and Lua libraries that provide interfaces with these data
        storage mechanisms.

Known Issues
  TCP socket connect operation issues
    The <tcpsock:connect|/"tcpsock:connect"> method may indicate "success"
    despite connection failures such as with "Connection Refused" errors.

    However, later attempts to manipulate the cosocket object will fail and
    return the actual error status message generated by the failed connect
    operation.

    This issue is due to limitations in the Nginx event model and only
    appears to affect Mac OS X.

  Lua Coroutine Yielding/Resuming
    *   Lua's "dofile" builtin is implemented as a C function in both Lua
        5.1 and LuaJIT 2.0 and when ngx.location.capture is called,
        ngx.exec, ngx.exit or ngx.req.read_body or similar in the file to be
        loaded by "dofile", a coroutine yield across the C function boundary
        will be initiated. This however is not normally allowed within
        ngx_lua and will usually result in error messages like "lua handler
        aborted: runtime error: attempt to yield across C-call boundary". To
        avoid this, define a real Lua module and use the Lua "require"
        builtin instead.

    *   As the standard Lua 5.1 interpreter's VM is not fully resumable, the
        methods ngx.location.capture, ngx.location.capture_multi,
        ngx.redirect, ngx.exec, and ngx.exit cannot be used within the
        context of a Lua pcall()
        (<http://www.lua.org/manual/5.1/manual.html#pdf-pcall>) or xpcall()
        (<http://www.lua.org/manual/5.1/manual.html#pdf-xpcall>) or even the
        first line of the "for ... in ..." statement when the standard Lua
        5.1 interpreter is used and the "attempt to yield across
        metamethod/C-call boundary" error will be produced. Please use
        LuaJIT 2.0, which supports a fully resumable VM, to avoid this.

  Lua Variable Scope
    Care must be taken when importing modules and this form should be used:

        local xxx = require('xxx')

    instead of the old deprecated form: require('xxx')

    Here is the reason: by design, the global environment has exactly the
    same lifetime as the Nginx request handler associated with it. Each
    request handler has its own set of Lua global variables and that is the
    idea of request isolation. The Lua module is actually loaded by the
    first Nginx request handler and is cached by the "require()" built-in in
    the package.loaded table for later reference, and "require()" has the
    side effect of setting a global variable to the loaded module table. But
    this global variable will be cleared at the end of the request handler,
    and every subsequent request handler all has its own (clean) global
    environment. So one will get Lua exception for accessing the "nil"
    value.

    It is recommended to always place the following piece of code at the end
    of Lua modules that use the I/O operations to prevent casual use of
    module-level global variables that are shared among *all* requests:

        local class_mt = {
            -- to prevent use of casual module global variables
            __newindex = function (table, key, val)
                error('attempt to write to undeclared variable "' .. key .. '"')
            end
        }
        setmetatable(_M, class_mt)

    This will guarantee that local variables in the Lua module functions are
    all declared with the "local" keyword, otherwise a runtime exception
    will be thrown. It prevents undesirable race conditions while accessing
    such variables. See Data Sharing within an Nginx Worker for the reasons
    behind this.

  Locations Configured by Subrequest Directives of Other Modules
    The ngx.location.capture and ngx.location.capture_multi directives
    cannot capture locations that include the echo_location,
    echo_location_async, echo_subrequest, or echo_subrequest_async
    directives.

        location /foo {
            content_by_lua '
                res = ngx.location.capture("/bar")
            ';
        }
        location /bar {
            echo_location /blah;
        }
        location /blah {
            echo "Success!";
        }

        $ curl -i http://example.com/foo

    will not work as expected.

  Special PCRE Sequences
    PCRE sequences such as "\d", "\s", or "\w", require special attention
    because in string literals, the backslash character, "\", is stripped
    out by both the Lua language parser and by the Nginx config file parser
    before processing. So the following snippet will not work as expected:

        # nginx.conf
        ? location /test {
        ?     content_by_lua '
        ?         local regex = "\d+"  -- THIS IS WRONG!!
        ?         local m = ngx.re.match("hello, 1234", regex)
        ?         if m then ngx.say(m[0]) else ngx.say("not matched!") end
        ?     ';
        ? }
        # evaluates to "not matched!"

    To avoid this, *double* escape the backslash:

        # nginx.conf
        location /test {
            content_by_lua '
                local regex = "\\\\d+"
                local m = ngx.re.match("hello, 1234", regex)
                if m then ngx.say(m[0]) else ngx.say("not matched!") end
            ';
        }
        # evaluates to "1234"

    Here, "\\\\d+" is stripped down to "\\d+" by the Nginx config file
    parser and this is further stripped down to "\d+" by the Lua language
    parser before running.

    Alternatively, the regex pattern can be presented as a long-bracketed
    Lua string literal by encasing it in "long brackets", "[[...]]", in
    which case backslashes have to only be escaped once for the Nginx config
    file parser.

        # nginx.conf
        location /test {
            content_by_lua '
                local regex = [[\\d+]]
                local m = ngx.re.match("hello, 1234", regex)
                if m then ngx.say(m[0]) else ngx.say("not matched!") end
            ';
        }
        # evaluates to "1234"

    Here, "[[\\d+]]" is stripped down to "[[\d+]]" by the Nginx config file
    parser and this is processed correctly.

    Note that a longer from of the long bracket, "[=[...]=]", may be
    required if the regex pattern contains "[...]" sequences. The
    "[=[...]=]" form may be used as the default form if desired.

        # nginx.conf
        location /test {
            content_by_lua '
                local regex = [=[[0-9]+]=]
                local m = ngx.re.match("hello, 1234", regex)
                if m then ngx.say(m[0]) else ngx.say("not matched!") end
            ';
        }
        # evaluates to "1234"

    An alternative approach to escaping PCRE sequences is to ensure that Lua
    code is placed in external script files and executed using the various
    *_by_lua_file directives. With this approach, the backslashes are only
    stripped by the Lua language parser and therefore only need to be
    escaped once each.

        -- test.lua
        local regex = "\\d+"
        local m = ngx.re.match("hello, 1234", regex)
        if m then ngx.say(m[0]) else ngx.say("not matched!") end
        -- evaluates to "1234"

    Within external script files, PCRE sequences presented as long-bracketed
    Lua string literals do not require modification.

        -- test.lua
        local regex = [[\d+]]
        local m = ngx.re.match("hello, 1234", regex)
        if m then ngx.say(m[0]) else ngx.say("not matched!") end
        -- evaluates to "1234"

Typical Uses
    Just to name a few:

    *   Mashup'ing and processing outputs of various nginx upstream outputs
        (proxy, drizzle, postgres, redis, memcached, and etc) in Lua,

    *   doing arbitrarily complex access control and security checks in Lua
        before requests actually reach the upstream backends,

    *   manipulating response headers in an arbitrary way (by Lua)

    *   fetching backend information from external storage backends (like
        redis, memcached, mysql, postgresql) and use that information to
        choose which upstream backend to access on-the-fly,

    *   coding up arbitrarily complex web applications in a content handler
        using synchronous but still non-blocking access to the database
        backends and other storage,

    *   doing very complex URL dispatch in Lua at rewrite phase,

    *   using Lua to implement advanced caching mechanism for Nginx's
        subrequests and arbitrary locations.

    The possibilities are unlimited as the module allows bringing together
    various elements within Nginx as well as exposing the power of the Lua
    language to the user. The module provides the full flexibility of
    scripting while offering performance levels comparable with native C
    language programs both in terms of CPU time as well as memory footprint.
    This is particularly the case when LuaJIT 2.0 is enabled.

    Other scripting language implementations typically struggle to match
    this performance level.

    The Lua state (Lua VM instance) is shared across all the requests
    handled by a single nginx worker process to minimize memory use.

    On a ThinkPad T400 2.80 GHz laptop, the Hello World example readily
    achieves 28k req/sec using "http_load -p 10". By contrast, Nginx +
    php-fpm 5.2.8 + Unix Domain Socket yields 6k req/sec and Node.js
    (<http://nodejs.org/>) v0.6.1 yields 10.2k req/sec for their Hello World
    equivalents.

Nginx Compatibility
    The latest module is compatible with the following versions of Nginx:

    *   1.4.x (last tested: 1.4.1)

    *   1.3.x (last tested: 1.3.11)

    *   1.2.x (last tested: 1.2.9)

    *   1.1.x (last tested: 1.1.5)

    *   1.0.x (last tested: 1.0.15)

    *   0.9.x (last tested: 0.9.4)

    *   0.8.x >= 0.8.54 (last tested: 0.8.54)

Code Repository
    The code repository of this project is hosted on github at
    chaoslawful/lua-nginx-module
    (<http://github.com/chaoslawful/lua-nginx-module>).

Installation
    The ngx_openresty bundle (<http://openresty.org>) can be used to install
    Nginx, ngx_lua, either one of the standard Lua 5.1 interpreter or LuaJIT
    2.0, as well as a package of powerful companion Nginx modules. The basic
    installation step is a simple "./configure --with-luajit && make && make
    install".

    Alternatively, ngx_lua can be manually compiled into Nginx:

    1.  Install LuaJIT 2.0 (recommended) or Lua 5.1 (Lua 5.2 is *not*
        supported yet). LuajIT can be downloaded from the the LuaJIT project
        website (<http://luajit.org/download.html>) and Lua 5.1, from the
        Lua project website (<http://www.lua.org/>). Some distribution
        package managers also distribute LuajIT and/or Lua.

    2.  Download the latest version of the ngx_devel_kit (NDK) module HERE
        (<http://github.com/simpl/ngx_devel_kit/tags>).

    3.  Download the latest version of ngx_lua HERE
        (<http://github.com/chaoslawful/lua-nginx-module/tags>).

    4.  Download the latest version of Nginx HERE (<http://nginx.org/>) (See
        Nginx Compatibility)

    Build the source with this module:

        wget 'http://nginx.org/download/nginx-1.4.1.tar.gz'
        tar -xzvf nginx-1.4.1.tar.gz
        cd nginx-1.4.1/

        # tell nginx's build system where to find LuaJIT:
        export LUAJIT_LIB=/path/to/luajit/lib
        export LUAJIT_INC=/path/to/luajit/include/luajit-2.0

        # or tell where to find Lua if using Lua instead:
        #export LUA_LIB=/path/to/lua/lib
        #export LUA_INC=/path/to/lua/include

        # Here we assume Nginx is to be installed under /opt/nginx/.
        ./configure --prefix=/opt/nginx \
                --add-module=/path/to/ngx_devel_kit \
                --add-module=/path/to/lua-nginx-module

        make -j2
        make install

  Installation on Ubuntu 11.10
    Note that it is recommended to use LuaJIT 2.0 instead of the standard
    Lua 5.1 interpreter where possible.

    If the standard Lua 5.1 interpreter is required however, run the
    following command to install it from the Ubuntu repository:

    apt-get install -y lua5.1 liblua5.1-0 liblua5.1-0-dev

    Everything should be installed correctly, except for one small tweak.

    Library name "liblua.so" has been changed in liblua5.1 package, it only
    comes with "liblua5.1.so", which needs to be symlinked to "/usr/lib" so
    it could be found during the configuration process.

    ln -s /usr/lib/x86_64-linux-gnu/liblua5.1.so /usr/lib/liblua.so

Community
  English Mailing List
    The openresty-en (<https://groups.google.com/group/openresty-en>)
    mailing list is for English speakers.

  Chinese Mailing List
    The openresty (<https://groups.google.com/group/openresty>) mailing list
    is for Chinese speakers.

Bugs and Patches
    Please submit bug reports, wishlists, or patches by

    1.  creating a ticket on the GitHub Issue Tracker
        (<http://github.com/chaoslawful/lua-nginx-module/issues>),

    2.  or posting to the OpenResty community.

TODO
  Short Term
    *   review and apply Brian Akin's patch for the new directive
        "lua_socket_log_errors".

    *   review and apply Brian Akin's patch for the new
        "shdict:flush_expired()" API.

    *   implement the SSL cosocket API.

    *   review and apply Jader H. Silva's patch for "ngx.re.split()".

    *   review and apply vadim-pavlov's patch for ngx.location.capture's
        "extra_headers" option

    *   use "ngx_hash_t" to optimize the built-in header look-up process for
        ngx.req.set_header, ngx.header.HEADER, and etc.

    *   add configure options for different strategies of handling the
        cosocket connection exceeding in the pools.

    *   add directives to run Lua codes when nginx stops.

    *   add APIs to access cookies as key/value pairs.

    *   add "ignore_resp_headers", "ignore_resp_body", and "ignore_resp"
        options to ngx.location.capture and ngx.location.capture_multi
        methods, to allow micro performance tuning on the user side.

    *   implement new directive "lua_ignore_client_abort".

  Longer Term
    *   add lightweight thread API (i.e., the "ngx.thread" API) as
        demonstrated in this sample code
        (<http://agentzh.org/misc/nginx/lua-thread2.lua>).

    *   add automatic Lua code time slicing support by yielding and resuming
        the Lua VM actively via Lua's debug hooks.

    *   add "stat" mode similar to mod_lua
        (<http://httpd.apache.org/docs/2.3/mod/mod_lua.html>).

Changes
    The changes of every release of this module can be obtained from the
    ngx_openresty bundle's change logs:

    http://openresty.org/#Changes

Test Suite
    The following dependencies are required to run the test suite:

    *   Nginx version >= 0.8.54

    *   Perl modules:

        *   test-nginx: http://github.com/agentzh/test-nginx

    *   Nginx modules:

        *   echo-nginx-module: http://github.com/agentzh/echo-nginx-module

        *   drizzle-nginx-module:
            http://github.com/chaoslawful/drizzle-nginx-module

        *   rds-json-nginx-module:
            http://github.com/agentzh/rds-json-nginx-module

        *   set-misc-nginx-module:
            http://github.com/agentzh/set-misc-nginx-module

        *   headers-more-nginx-module:
            http://github.com/agentzh/headers-more-nginx-module

        *   memc-nginx-module: http://github.com/agentzh/memc-nginx-module

        *   srcache-nginx-module:
            http://github.com/agentzh/srcache-nginx-module

        *   ngx_auth_request:
            http://mdounin.ru/hg/ngx_http_auth_request_module/

    *   C libraries:

        *   yajl: https://github.com/lloyd/yajl

    *   Lua modules:

        *   lua-yajl: https://github.com/brimworks/lua-yajl

            *   Note: the compiled module has to be placed in
                '/usr/local/lib/lua/5.1/'

    *   Applications:

        *   mysql: create database 'ngx_test', grant all privileges to user
            'ngx_test', password is 'ngx_test'

        *   memcached

    The order in which these modules are added during configuration is
    important as the position of any filter module in the filtering chain
    determines the final output. The correct adding order is:

    1.  ngx_devel_kit

    2.  set-misc-nginx-module

    3.  ngx_http_auth_request_module

    4.  echo-nginx-module

    5.  memc-nginx-module

    6.  lua-nginx-module (i.e. this module)

    7.  headers-more-nginx-module

    8.  srcache-nginx-module

    9.  drizzle-nginx-module

    10. rds-json-nginx-module

Copyright and License
    This module is licensed under the BSD license.

    Copyright (C) 2009-2013, by Xiaozhe Wang (chaoslawful)
    <chaoslawful@gmail.com>.

    Copyright (C) 2009-2013, by Yichun "agentzh" Zhang (章亦春)
    <agentzh@gmail.com>, CloudFlare Inc.

    All rights reserved.

    Redistribution and use in source and binary forms, with or without
    modification, are permitted provided that the following conditions are
    met:

    *   Redistributions of source code must retain the above copyright
        notice, this list of conditions and the following disclaimer.

    *   Redistributions in binary form must reproduce the above copyright
        notice, this list of conditions and the following disclaimer in the
        documentation and/or other materials provided with the distribution.

    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS
    IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
    TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A
    PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
    HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
    SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED
    TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
    PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
    LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
    NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
    SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

See Also
    *   lua-resty-memcached
        (<http://github.com/agentzh/lua-resty-memcached>) library based on
        ngx_lua cosocket.

    *   lua-resty-redis (<http://github.com/agentzh/lua-resty-redis>)
        library based on ngx_lua cosocket.

    *   lua-resty-mysql (<http://github.com/agentzh/lua-resty-mysql>)
        library based on ngx_lua cosocket.

    *   lua-resty-upload (<http://github.com/agentzh/lua-resty-upload>)
        library based on ngx_lua cosocket.

    *   lua-resty-dns (<http://github.com/agentzh/lua-resty-dns>) library
        based on ngx_lua cosocket.

    *   lua-resty-string (<http://github.com/agentzh/lua-resty-string>)
        library based on LuaJIT FFI (<http://luajit.org/ext_ffi.html>).

    *   Routing requests to different MySQL queries based on URI arguments
        (<http://openresty.org/#RoutingMySQLQueriesBasedOnURIArgs>)

    *   Dynamic Routing Based on Redis and Lua
        (<http://openresty.org/#DynamicRoutingBasedOnRedis>)

    *   Using LuaRocks with ngx_lua (<http://openresty.org/#UsingLuaRocks>)

    *   Introduction to ngx_lua
        (<https://github.com/chaoslawful/lua-nginx-module/wiki/Introduction>
        )

    *   ngx_devel_kit (<http://github.com/simpl/ngx_devel_kit>)

    *   [[HttpEchoModule]]

    *   [[HttpDrizzleModule]]

    *   postgres-nginx-module (<http://github.com/FRiCKLE/ngx_postgres>)

    *   [[HttpMemcModule]]

    *   The ngx_openresty bundle (<http://openresty.org>)

Translations
    *   Chinese (still in progress)