問題描述
nginx reload是有一定損耗的，如果你使用的是長連接的話，那么當reload nginx時長連接所有的worker進程會進行優雅退出，並當該worker進程上的所有連接都釋放時，進程才真正退出。
解決辦法
對於社區版nginx目前有三個選擇方式：
- Tengine 的Dyups模塊。
- 微博的Upsync+Consul 實現動態負載均衡。
- OpenResty的balancer_by_lua(又拍雲使用其開源的slardar(Consul balancer_by_lua))。
本文使用upsync模塊來解決配置文件修改后，reload nginx進程造成性能下降的問題。
它的功能是拉取 consul 的后端 server 的列表，並更新 Nginx 的路由信息。此模塊不依賴於任何第三方模塊。consul 作為 Nginx 的 db，利用 consul 的 KV 服務，每個 Nginx work 進程獨立的去拉取各個 upstream 的配置，並更新各自的路由。
實戰
給nginx打補丁包
這步可以不做，如果不做，編譯的時候刪除這個模塊
git clone https://github.com/xiaokai-wang/nginx_upstream_check_module
## 打補丁包
patch -p0 < /usr/local/src/nginx_upstream_check_module-master/check_1.9.2+.patch
下載nginx-upsync-module源碼
git clone https://github.com/weibocom/nginx-upsync-module.git
下載nginx源碼
wget 'http://nginx.org/download/nginx-1.10.1.tar.gz'
tar -xzvf nginx-1.10.1.tar.gz
cd nginx-1.10.1/
開始編譯
./configure --prefix=/data/app/nginx-1.10.1 --user=nginx --group=nginx --with-http_ssl_module --with-http_stub_status_module --add-module=/usr/local/src/nginx-upsync-module-master/ --add-module=/usr/local/src/nginx_upstream_check_module-master/
make
make install
啟動consul
wget https://releases.hashicorp.com/consul/0.6.4/consul_0.6.4_linux_amd64.zip
unzip consul_0.6.4_linux_amd64.zip
./consul agent -advertise=x.x.x.x -client=0.0.0.0 -dev
創建nginx配置文件
mkdir -p /usr/local/nginx/conf/servers
/usr/local/nginx/conf/nginx.conf
events {
worker_connections 4096; ## Default: 1024
}
http {
upstream test {
# fake server otherwise ngx_http_upstream will report error when startup
server 127.0.0.1:11111;
# all backend server will pull from consul when startup and will delete fake server
upsync 127.0.0.1:8500/v1/kv/upstreams/test upsync_timeout=6m upsync_interval=500ms upsync_type=consul strong_dependency=off;
upsync_dump_path /usr/local/nginx/conf/servers/servers_test.conf;
}
upstream bar {
server 127.0.0.1:8090 weight=1 fail_timeout=10 max_fails=3;
}
server {
listen 8080;
location = /proxy_test {
proxy_pass http://test;
}
location = /bar {
proxy_pass http://bar;
}
location = /upstream_show {
upstream_show;
}
}
}
測試
for i in `seq 3`;do mkdir html$i/test -p && echo $i >html$i/test/test.html; done;
docker run -d -p 8001:80 -v /root/html1/:/usr/share/nginx/html nginx
docker run -d -p 8002:80 -v /root/html2/:/usr/share/nginx/html nginx
docker run -d -p 8003:80 -v /root/html3/:/usr/share/nginx/html nginx
添加服務
curl -X PUT -d '{"weight":1, "max_fails":2, "fail_timeout":10}' http://127.0.0.1:8500/v1/kv/upstreams/test/192.168.56.12:8001
curl -X PUT -d '{"weight":1, "max_fails":2, "fail_timeout":10}' http://127.0.0.1:8500/v1/kv/upstreams/test/192.168.56.12:8002
curl -X PUT -d '{"weight":1, "max_fails":2, "fail_timeout":10}' http://127.0.0.1:8500/v1/kv/upstreams/test/192.168.56.12:8003
查看conf/servers/servers_test.conf 文件中是否有內容
cat conf/servers/servers_test.conf
server 192.168.56.12:8003 weight=1 max_fails=2 fail_timeout=10s;
server 192.168.56.12:8002 weight=1 max_fails=2 fail_timeout=10s;
server 192.168.56.12:8001 weight=1 max_fails=2 fail_timeout=10s;
或者瀏覽器打開http://192.168.56.11:8080/upstream_show?test
顯示內容如下：
Upstream name: test; Backend server count: 3
server 192.168.56.12:8003 weight=1 max_fails=2 fail_timeout=10s;
server 192.168.56.12:8002 weight=1 max_fails=2 fail_timeout=10s;
server 192.168.56.12:8001 weight=1 max_fails=2 fail_timeout=10s;
總結
此模塊只修改upstream 中的緩存信息，不能修改或添加其他配置
測試中遇到的問題
在添加服務時出現如下錯誤，導致服務添加不能實時進行，大約需要3分鍾左右時間。
consul日志：
2016/03/22 05:34:42 [DEBUG] http: Request GET /v1/kv/upstreams/test?recurse&index=169 (149.023µs) from=127.0.0.1:38853
2016/03/22 05:34:43 [DEBUG] http: Request GET /v1/kv/upstreams/test?recurse&index=169 (146.759µs) from=127.0.0.1:38854
2016/03/22 05:34:45 [DEBUG] http: Request GET /v1/kv/upstreams/test?recurse&index=169 (149.853µs) from=127.0.0.1:38855
2016/03/22 05:34:46 [DEBUG] http: Request GET /v1/kv/upstreams/test?recurse&index=169 (111.46µs) from=127.0.0.1:38856
2016/03/22 05:34:48 [DEBUG] http: Request GET /v1/kv/upstreams/test?recurse&index=169 (142.696µs) from=127.0.0.1:38857
2016/03/22 05:34:48 [DEBUG] http: Request GET /v1/kv/upstreams/test?recurse&index=169 (112.089µs) from=127.0.0.1:38858
2016/03/22 05:34:49 [DEBUG] http: Request GET /v1/kv/upstreams/test?recurse&index=169 (114.29µs) from=127.0.0.1:38859
2016/03/22 05:34:50 [DEBUG] http: Request GET /v1/kv/upstreams/test?recurse&index=169 (148.245µs) from=127.0.0.1:38860
nginx日志
2016/03/22 05:35:09 [error] 18879#0: recv() failed (104: Connection reset by peer)
2016/03/22 05:35:09 [error] 18879#0: upsync_recv: recv error with upsync_server: 127.0.0.1:8500
2016/03/22 05:35:10 [error] 18879#0: recv() failed (104: Connection reset by peer)
2016/03/22 05:35:10 [error] 18879#0: upsync_recv: recv error with upsync_server: 127.0.0.1:8500
2016/03/22 05:35:11 [error] 18879#0: recv() failed (104: Connection reset by peer)
2016/03/22 05:35:11 [error] 18879#0: upsync_recv: recv error with upsync_server: 127.0.0.1:8500
2016/03/22 05:35:13 [error] 18879#0: recv() failed (104: Connection reset by peer)
2016/03/22 05:35:13 [error] 18879#0: upsync_recv: recv error with upsync_server: 127.0.0.1:8500
2016/03/22 05:35:13 [error] 18879#0: recv() failed (104: Connection reset by peer)
2016/03/22 05:35:13 [error] 18879#0: upsync_recv: recv error with upsync_server: 127.0.0.1:8500
2016/03/22 05:35:14 [error] 18879#0: recv() failed (104: Connection reset by peer)
問題現象
當添加一個服務時，出現此問題，新增服務不能及時添加到負載中，不影響運行正常的服務。 此時再往consul繼續添加一個服務時，可能會導致此錯誤終止，並能成功添加當前兩條服務記錄。