This module can be used to update your upstream-list without reloadding Nginx.
file: conf/nginx.conf
ATTENTION: You MUST use nginx variable to do proxy_pass
daemon off;
error_log logs/error.log debug;
events {
}
http {
include conf/upstream.conf;
server {
listen 8080;
location / {
# The upstream here must be a nginx variable
proxy_pass http://$dyups_host;
}
}
server {
listen 8088;
location / {
return 200 "8088";
}
}
server {
listen 8089;
location / {
return 200 "8089";
}
}
server {
listen 8081;
location / {
dyups_interface;
}
}
}If your original config looks like this:
proxy_pass http://upstream_name;please replace it with:
set $ups upstream_name;
proxy_pass http://$ups;$ups can be any valid nginx variable.
file: conf/upstream.conf
upstream host1 {
server 127.0.0.1:8088;
}
upstream host2 {
server 127.0.0.1:8089;
}- Only install dyups module
git clone git://github.com/yzprofile/ngx_http_dyups_module.git
# to compile as a static module
./configure --add-module=./ngx_http_dyups_module
# to compile as a dynamic module
./configure --add-dynamic-module=./ngx_http_dyups_module- Install dyups module with lua-nginx-module and upstream check module
- upstream check module: To make upstream check module work well with dyups module, you should use patched upstream check module.
- lua-nginx-module: To enable dyups LUA API, you MUST put
--add-module=./lua-nginx-modulein front of--add-module=./ngx_http_dyups_modulein the./configurecommand.
git clone git://github.com/yzprofile/ngx_http_dyups_module.git
git clone git@github.com:yzprofile/nginx_upstream_check_module.git
git clone git@github.com:openresty/lua-nginx-module.git
# to compile as a static module
./configure \
--add-module=./nginx_upstream_check_module \
--add-module=./lua-nginx-module \
--add-module=./ngx_http_dyups_module- Syntax: dyups_interface
- Default:
none - Context:
loc
This directive set the interface location where you can add or delete the upstream list. See the section of Interface for detail.
- Syntax: dyups_read_msg_timeout
time - Default:
1s - Context:
main
This directive set the interval of workers readding the commands from share memory.
- Syntax: dyups_shm_zone_size
size - Default:
2MB - Context:
main
This directive set the size of share memory which used to store the commands.
- Syntax: dyups_upstream_conf
path - Default:
none - Context:
main
This directive has been deprecated.
- Syntax: dyups_trylock
on | off - Default:
off - Context:
main
You will get a better prefomance but it maybe not stable, and you will get a '409' when the update request conflicts with others.
- Syntax: dyups_read_msg_log
on | off - Default:
off - Context:
main
You can enable / disable log of workers readding the commands from share memory. The log looks like:
2017/02/28 15:37:53 [info] 56806#0: [dyups] has 0 upstreams, 1 static, 0 deleted, all 1
/detailget all upstreams and their servers/listget the list of upstreams/upstream/namefind the upstream by it's name
/upstream/nameupdate one upstreambodycommands;bodyserver ip:port;
/upstream/namedelete one upstream
Call the interface, when you get the return code is HTTP_INTERNAL_SERVER_ERROR 500, you need to reload nginx to make the Nginx work at a good state.
If you got HTTP_CONFLICT 409, you need resend the same commands again latter.
The /list and /detail interface will return HTTP_NO_CONTENT 204 when there is no upstream.
Other code means you should modify your commands and call the interface again.
ATTENTION: You also need a third-party to generate the new config and dump it to Nginx'conf directory.
curl -H "host: dyhost" 127.0.0.1:8080<title>502 Bad Gateway</title>
nginx/1.3.13
curl -d "server 127.0.0.1:8089;server 127.0.0.1:8088;" 127.0.0.1:8081/upstream/dyhostsuccess
curl -H "host: dyhost" 127.0.0.1:80808089
curl -H "host: dyhost" 127.0.0.1:80808088
curl 127.0.0.1:8081/detailhost1 server 127.0.0.1:8088 weight=1 max_conns=0 max_fails=1 fail_timeout=10 backup=0 down=0
host2 server 127.0.0.1:8089 weight=1 max_conns=0 max_fails=1 fail_timeout=10 backup=0 down=0
dyhost server 127.0.0.1:8089 weight=1 max_conns=0 max_fails=1 fail_timeout=10 backup=0 down=0 server 127.0.0.1:8088 weight=1 max_conns=0 max_fails=1 fail_timeout=10 backup=0 down=0
curl -i -X DELETE 127.0.0.1:8081/upstream/dyhostsuccess
curl 127.0.0.1:8081/detailhost1 server 127.0.0.1:8088 weight=1 max_conns=0 max_fails=1 fail_timeout=10 backup=0 down=0
host2 server 127.0.0.1:8089 weight=1 max_conns=0 max_fails=1 fail_timeout=10 backup=0 down=0
extern ngx_flag_t ngx_http_dyups_api_enable;
ngx_int_t ngx_dyups_update_upstream(ngx_str_t *name, ngx_buf_t *buf, ngx_str_t *rv);
ngx_int_t ngx_dyups_delete_upstream(ngx_str_t *name, ngx_str_t *rv);
extern ngx_dyups_add_upstream_filter_pt ngx_dyups_add_upstream_top_filter;
extern ngx_dyups_del_upstream_filter_pt ngx_dyups_del_upstream_top_filter;NOTICE:
You should add the directive dyups_interface into your config file to active this feature
content_by_lua '
local dyups = require "ngx.dyups"
local status, rv = dyups.update("test", [[server 127.0.0.1:8088;]]);
ngx.print(status, rv)
if status ~= ngx.HTTP_OK then
ngx.print(status, rv)
return
end
ngx.print("update success")
status, rv = dyups.delete("test")
if status ~= ngx.HTTP_OK then
ngx.print(status, rv)
return
end
ngx.print("delete success")
';- Fix: memory leak of ssl session reuse.
- Each processes starts read_msg_timer separately at random timeout.
- Fix: unlocking behavior.
- Fix: compilation error without upstream check module.
- Fix: if a domain name contains multiple IP addresses, call them.
- When dyups and health check module together use, ngx_shmtx_lock block too long time and cpu full load, cause health check timeout, 502.
- Fix: build error when compiled with a higher version of OpenSSL.
- etc.
Feature: Added add/del upstream filter to make other modules operate upstream easily after upstream changed
Bugfixes: upstream connect failed caused coredump
Supported: C API and Lua API
Bugfixes: Supported sandbox before updatting
- Bugfixes: wrong string comparison for string "upstream", @chobits
- Bugfixes: that response of /detail uri has no Content-Length header, @chobits
- Feature: if you use this branch of tengine, update upstream rbtree, @SarahWang
- Feature: simplify upstream parsing methods via ngx_conf_parse api, @chobits
- Bugfixes: client timed out cause a coredumped while adding an exist upstream
- Bugfixes: when proxy_pass to a no-variable address dyups will coredump
Bugfixed: upstream will be deleted in the process of finding upstream.
- check every command to make sure they are all ok before update upstream.
done - support ip_hash and keepalive or other upstream module
done - support
weight,max_fails,fail_timeout,backupdone - support health check module, you should use this branch of Tengine or wait for it's release.
done - restore upstream configuration in
init processhandler.done
This module has been merged into Tengine.
- lua-upstream-nginx-module: You can use
lua-upstream-nginx-moduleto get more detail infomation of upstream. - yzprofile/upstream check module: To make upstream check module work well with dyups module, you should use patched upstream check module.
hg clone http://hg.nginx.org/nginx-tests/ ..
# Or use git as below if you prefer the GitHub mirror
git clone git@github.com:nginx/nginx-tests/ ..
TEST_NGINX_BINARY=/path/to/your/nginx/dir/sbin/nginx prove -I ../nginx-tests/lib ./t/dyups.tTo make the tests pass, you should also install lua-nginx-module and patched upstream check module.
- yzprofile (袁茁) yzprofile@gmail.com
- chobits (王笑臣) wangxiaochen0@gmail.com, Alibaba Inc.