贫苦家庭用户的 Envoy xDS 控制平面
点击 "阅读原文" 可以获得更好的阅读体验。
前言
之前家里的路由器一直用的都是网件 R7000
搭载的梅林固件,虽说性能也还不错,比两百块钱的小米路由器强多了,但还是不能满足我的需求,装了某魔法软件后内存蹭蹭蹭爆满啊。终于有理由换软路由了,此时不换更待何时!
经过一番对比,最后决定在某宝上入手了一款低功耗的 J3160
,4 核 4 G,700 大洋左右,刷了个 LEDE
系统,这下绝对够用了。跑了几个魔法软件和一堆容器也没耗多少资源,还是 x86 香啊!
R7000
就老老实实通过 Access Point
模式作为二级路由提供 WiFi 吧。
到这里有人可能要问了,说了这么多跟这篇文章的主题有什么关系呢?别急,下面进入主题。
01
背景
作为顶级贫苦玩家,肯定会在家里装上各种奇奇怪怪的应用,Aria2 和 Transmission 肯定不能少。作为顶级云原生狂热信徒,监控一条龙服务肯定不能少,至少应该上一套 Grafana
和 Prometheus
。然而,这么多乱七八糟的端口,我可记不住。。。
我需要一款负载均衡器来反代所有的服务,别跟我说 Nginx
,作为云原生舔狗,用 Nginx 是不可能的,必须用我的偶像 Envoy
来做反代啊,既能反代 Web 服务,还能代替防火墙的端口映射功能(就是反代 TCP 啦),最重要的是还能暴露所有 Upstream
服务的 metrics,再结合 Prometheus 和 Grafana,不香吗?(你想想,连 samba 和 UDP 服务都能监控)
第一步当然是让路由器获取外网 IP 了,现在上海电信用的都是 SDN
网关,破解都无从破解,但很多人不知道的是,其实你可以打电信客服电话让人家在后台把 SDN
网关改成桥接模式。。。 改完桥接模式就好办了,直接路由器拨号就是外网 IP。
下面就是改 LEDE
Web 服务端口,因为 Envoy 得用 80 端口,所以把它的端口改成别的,比如 81 就不错:
$ cat /etc/config/uhttpd
config uhttpd 'main'
list listen_http '0.0.0.0:81'
list listen_http '[::]:81'
list listen_proxy '127.0.0.1:8000'
list listen_https '0.0.0.0:6443'
list listen_https '[::]:6443'
option home '/www'
...
改完之后重启 httpd
服务:
$ /etc/init.d/uhttpd restart
DDNS 和申请 https 证书什么的我就不说了,不是本文的重点。
02
基于文件的 xDS 动态更新
80
和 443
端口被腾出来之后,就可以愉快地使用反代了。可是安装 Envoy 是个头疼的问题啊,编译太复杂,GetEnvoy[1] 项目又不支持 busybox
,只能通过容器跑了。配置如图:
下面就是老老实实写配置文件,没什么可说的,但问题就出在这里,Upstream 服务不多倒好办,一旦变多,Envoy
配置文件会过于冗长,很容易看花眼。虽想到了用控制平面来动态更新配置,但我没必要单独起个控制平面服务,还有没有别的办法呢?有的,其实 Envoy
是可以将文件作为配置的订阅来源的。方法很简单,首先需要参加一个 Bootstrap
引导程序配置文件,里面定义了 node 信息和动态资源:
$ cat envoy.yaml
node:
id: node0
cluster: cluster0
dynamic_resources:
lds_config:
path: /etc/envoy/lds.yaml
cds_config:
path: /etc/envoy/cds.yaml
admin:
access_log_path: "/dev/stdout"
address:
socket_address:
address: "::"
ipv4_compat: true
port_value: 15001
Envoy 将使用 inotify[2](MacOS 用的是 kqueue
)来监视文件的更改,一旦检测到更改,就立即订阅更新。查看系统是否支持 inotify:
$ ll /proc/sys/fs/inotify/
-rw-r--r-- 1 root root 0 Dec 23 16:05 max_queued_events
-rw-r--r-- 1 root root 0 Dec 23 16:05 max_user_instances
-rw-r--r-- 1 root root 0 Dec 23 16:05 max_user_watches
lds.yaml
里是 Listener 的配置,cds.yaml
里是 Cluster 的配置,先往 lds.yaml 中加入如下的配置:
version_info: "0"
resources:
- "@type": type.googleapis.com/envoy.api.v2.Listener
name: listener_http_v4
address:
socket_address:
address: 0.0.0.0
port_value: 80
filter_chains:
- filters:
- name: envoy.http_connection_manager
typed_config:
"@type": type.googleapis.com/envoy.config.filter.network.http_connection_manager.v2.HttpConnectionManager
stat_prefix: ingress_http
codec_type: AUTO
access_log:
name: envoy.file_access_log
typed_config:
"@type": type.googleapis.com/envoy.config.accesslog.v2.FileAccessLog
path: /dev/stdout
route_config:
name: http_route_v4
virtual_hosts:
- name: backend
domains:
- "router.fuckcloudnative.io"
- "mynas.fuckcloudnative.io"
routes:
- match:
prefix: "/"
redirect:
https_redirect: true
port_redirect: 8443
response_code: "FOUND"
- name: default
domains:
- "*"
routes:
- match:
prefix: "/"
route:
cluster: lede
http_filters:
- name: envoy.router
域名改成你自己的就好,路由分为两部分,通过那两个域名访问的就会被转到 https
,其他的都转到 lede Web 服务。其实 http 转 https 的那部分路由可以删掉,因为国内的运营商基本上都把 80
端口封了,外网是无法访问的。第二部分不能删除,删除之后就不能通过内网访问 lede Web 界面了。
再加入 https 的配置:
- "@type": type.googleapis.com/envoy.api.v2.Listener
name: listener_https_v4
address:
socket_address:
address: 0.0.0.0
port_value: 8443
filter_chains:
- filter_chain_match:
server_names: "router.fuckcloudnative.io"
transport_socket:
name: envoy.transport_sockets.tls
typed_config:
"@type": type.googleapis.com/envoy.api.v2.auth.DownstreamTlsContext
common_tls_context:
tls_certificates:
- certificate_chain:
filename: "/etc/ssl/router.fuckcloudnative.io/3207748_router.fuckcloudnative.io.pem"
private_key:
filename: "/etc/ssl/router.fuckcloudnative.io/3207748_router.fuckcloudnative.io.key"
filters:
- name: envoy.http_connection_manager
typed_config:
"@type": type.googleapis.com/envoy.config.filter.network.http_connection_manager.v2.HttpConnectionManager
stat_prefix: ingress_https
codec_type: AUTO
access_log:
name: envoy.file_access_log
typed_config:
"@type": type.googleapis.com/envoy.config.accesslog.v2.FileAccessLog
path: /dev/stdout
route_config:
name: https_route_v4_default
virtual_hosts:
- name: default
domains:
- "*"
routes:
- match:
prefix: "/"
route:
cluster: lede
http_filters:
- name: envoy.router
接下来往 cds.yaml
中加入 Cluster 配置:
version_info: "0"
resources:
- "@type": type.googleapis.com/envoy.api.v2.Cluster
name: lede
connect_timeout: 1s
type: strict_dns
dns_lookup_family: V4_ONLY
lb_policy: ROUND_ROBIN
load_assignment:
cluster_name: lede
endpoints:
- lb_endpoints:
- endpoint:
address:
socket_address:
address: 127.0.0.1
port_value: 81
由于 Docker 对 inotify
的支持不太友好,有时不会检测不到文件系统的更改,所以最好的办法是强制更改,原理很简单,将文件重命名,然后再改回来。写一个脚本就好了:
$ cat apply.sh
mv cds.yaml cds.yaml.temp
mv cds.yaml.temp cds.yaml
mv lds.yaml lds.yaml.temp
mv lds.yaml.temp lds.yaml
注意:必须先更新
CDS
,后更新LDS
。
执行脚本之后,查看 Envoy 日志,发现配置已经生效:
$ docker logs -f envoy
[2019-12-23 09:22:14.644][1][info][upstream] [source/common/upstream/cds_api_impl.cc:71] cds: add 1 cluster(s), remove 0 cluster(s)
[2019-12-23 09:22:14.648][1][info][upstream] [source/common/upstream/cds_api_impl.cc:87] cds: add/update cluster 'lede'
[2019-12-23 09:22:30.186][1][info][upstream] [source/server/lds_api.cc:71] lds: add/update listener 'listener_http_v4'
[2019-12-23 09:22:45.881][1][warning][config] [source/server/listener_impl.cc:287] adding listener '0.0.0.0:8443': filter chain match rules require TLS Inspector listener filter, but it isn't configured, trying to inject it (this might fail if Envoy is compiled without it)
[2019-12-23 09:22:45.882][1][info][upstream] [source/server/lds_api.cc:71] lds: add/update listener 'listener_https_v4'
ipv6
上面只是 ipv4
的配置,如果你的宽带开启了 ipv6
,还可以开启 ipv6 端口。至于我为什么要将 ipv4 和 ipv6 分开呢,因为据我测试,电信运营商只封了 ipv4 的 80
和 443
端口,ipv6 还可以用,所以我需要为 ipv4 和 ipv6 分配不同的路由策略。在 lds.yaml
中加入 ipv6 的配置:
- "@type": type.googleapis.com/envoy.api.v2.Listener
name: listener_http_v6
address:
socket_address:
address: "::"
port_value: 80
filter_chains:
- filters:
- name: envoy.http_connection_manager
typed_config:
"@type": type.googleapis.com/envoy.config.filter.network.http_connection_manager.v2.HttpConnectionManager
stat_prefix: ingress_http
codec_type: AUTO
access_log:
name: envoy.file_access_log
typed_config:
"@type": type.googleapis.com/envoy.config.accesslog.v2.FileAccessLog
path: /dev/stdout
route_config:
name: http_route_v6
virtual_hosts:
- name: backend
domains:
- "router.fuckcloudnative.io"
- "mynas.fuckcloudnative.io"
routes:
- match:
prefix: "/"
redirect:
https_redirect: true
port_redirect: 443
response_code: "FOUND"
- name: default
domains:
- "*"
routes:
- match:
prefix: "/"
route:
cluster: lede
http_filters:
- name: envoy.router
- "@type": type.googleapis.com/envoy.api.v2.Listener
name: listener_https_v6
address:
socket_address:
address: "::"
port_value: 443
filter_chains:
- filter_chain_match:
server_names: "router.fuckcloudnative.io"
transport_socket:
name: envoy.transport_sockets.tls
typed_config:
"@type": type.googleapis.com/envoy.api.v2.auth.DownstreamTlsContext
common_tls_context:
tls_certificates:
- certificate_chain:
filename: "/etc/ssl/router.fuckcloudnative.io/3207748_router.fuckcloudnative.io.pem"
private_key:
filename: "/etc/ssl/router.fuckcloudnative.io/3207748_router.fuckcloudnative.io.key"
filters:
- name: envoy.http_connection_manager
typed_config:
"@type": type.googleapis.com/envoy.config.filter.network.http_connection_manager.v2.HttpConnectionManager
stat_prefix: ingress_https
codec_type: AUTO
access_log:
name: envoy.file_access_log
typed_config:
"@type": type.googleapis.com/envoy.config.accesslog.v2.FileAccessLog
path: /dev/stdout
route_config:
name: https_route_v6_default
virtual_hosts:
- name: default
domains:
- "*"
routes:
- match:
prefix: "/"
route:
cluster: lede
http_filters:
- name: envoy.router
执行 apply.sh
使配置生效,查看日志:
$ docker logs -f envoy
[2019-12-23 09:43:44.431][1][info][upstream] [source/server/lds_api.cc:71] lds: add/update listener 'listener_http_v6'
[2019-12-23 09:43:44.441][1][warning][config] [source/server/listener_impl.cc:287] adding listener '[::]:443': filter chain match rules require TLS Inspector listener filter, but it isn't configured, trying to inject it (this might fail if Envoy is compiled without it)
[2019-12-23 09:43:44.441][1][info][upstream] [source/server/lds_api.cc:71] lds: add/update listener 'listener_https_v6'
Grafana
Grafana
的安装我就不多说了,直接容器跑,配置如下:
为了能够通过反向代理正确访问 Grafana
,需要对 Grafana
的配置做一些调整,修改 grafana.ini
中的以下几个字段:
[server]
domain = foo.bar
root_url = %(protocol)s://%(domain)s/grafana/
将 domain
的值换成你自己的域名。
修改 Listener listener_https_v4
的路由:
route_config:
name: https_route_v4_default
virtual_hosts:
- name: default
domains:
- "*"
routes:
- match:
prefix: "/grafana/"
route:
cluster: grafana
- match:
prefix: "/"
route:
cluster: lede
修改 Listener listener_https_v6
的路由:
route_config:
name: https_route_v6_default
virtual_hosts:
- name: default
domains:
- "*"
routes:
- match:
prefix: "/grafana/"
route:
cluster: grafana
- match:
prefix: "/"
route:
cluster: lede
向 cds.yaml
中添加 Cluster:
- "@type": type.googleapis.com/envoy.api.v2.Cluster
name: grafana
connect_timeout: 1s
type: strict_dns
dns_lookup_family: V4_ONLY
lb_policy: ROUND_ROBIN
load_assignment:
cluster_name: grafana
endpoints:
- lb_endpoints:
- endpoint:
address:
socket_address:
address: 127.0.0.1
port_value: 3000
应用更新:
$ ./apply.sh
然后就可以通过 subpath
访问 Grafana 了。
TCP
一般情况下,路由器的端口映射都是通过 iptables
来做的,但我既然用了 Envoy,端口映射肯定还是要用 Envoy 来实现,毕竟 Grafana
真香。
Envoy 通过 TCP 代理即可实现端口映射功能,比如我想将 samba
服务暴露到公网,只需向 lds.yaml
中加入配置:
- "@type": type.googleapis.com/envoy.api.v2.Listener
name: listener_smb_local
address:
socket_address:
address: "::"
ipv4_compat: true
port_value: 139
filter_chains:
- filters:
- name: envoy.tcp_proxy
typed_config:
"@type": type.googleapis.com/envoy.config.filter.network.tcp_proxy.v2.TcpProxy
stat_prefix: smb_local
cluster: smb_local
access_log:
name: envoy.file_access_log
typed_config:
"@type": type.googleapis.com/envoy.config.accesslog.v2.FileAccessLog
path: /dev/stdout
- "@type": type.googleapis.com/envoy.api.v2.Listener
name: listener_smb_internet
address:
socket_address:
address: "::"
ipv4_compat: true
port_value: 4450
filter_chains:
- filters:
- name: envoy.tcp_proxy
typed_config:
"@type": type.googleapis.com/envoy.config.filter.network.tcp_proxy.v2.TcpProxy
stat_prefix: smb_internet
cluster: smb_internet
access_log:
name: envoy.file_access_log
typed_config:
"@type": type.googleapis.com/envoy.config.accesslog.v2.FileAccessLog
path: /dev/stdout
其中 ipv4_compat: true
表示同时监听 ipv4 和 ipv6。445
端口也被运营商封了,所以可以使用 4450 端口。
再向 cds.yaml
中加入如下的配置:
- "@type": type.googleapis.com/envoy.api.v2.Cluster
name: smb_local
connect_timeout: 1s
type: strict_dns
dns_lookup_family: V4_ONLY
lb_policy: ROUND_ROBIN
load_assignment:
cluster_name: smb_local
endpoints:
- lb_endpoints:
- endpoint:
address:
socket_address:
address: 192.168.100.20
port_value: 139
- "@type": type.googleapis.com/envoy.api.v2.Cluster
name: smb_internet
connect_timeout: 1s
type: strict_dns
dns_lookup_family: V4_ONLY
lb_policy: ROUND_ROBIN
load_assignment:
cluster_name: smb_internet
endpoints:
- lb_endpoints:
- endpoint:
address:
socket_address:
address: 192.168.100.20
port_value: 445
将其中的地址改成你的 samba
服务内网地址。
配置生效后,就可以通过外网连接你的 samba 服务了。
当然了,我自己的 Upstream 服务远远不止这些,我只是针对每一种类型举一个示例,大家可以举一反三。看看我的:
$ docker logs -f envoy
[2019-12-23 10:16:58.199][1][info][upstream] [source/common/upstream/cds_api_impl.cc:87] cds: add/update cluster 'prometheus'
[2019-12-23 10:16:58.201][1][info][upstream] [source/common/upstream/cds_api_impl.cc:87] cds: add/update cluster 'transmission'
[2019-12-23 10:16:58.204][1][info][upstream] [source/common/upstream/cds_api_impl.cc:87] cds: add/update cluster 'mynas'
[2019-12-23 10:16:58.205][1][info][upstream] [source/common/upstream/cds_api_impl.cc:87] cds: add/update cluster 'aria2'
[2019-12-23 10:16:58.207][1][info][upstream] [source/common/upstream/cds_api_impl.cc:87] cds: add/update cluster 'aria2_bt'
[2019-12-23 10:16:58.209][1][info][upstream] [source/common/upstream/cds_api_impl.cc:87] cds: add/update cluster 'aria2_dht'
[2019-12-23 10:16:58.211][1][info][upstream] [source/common/upstream/cds_api_impl.cc:87] cds: add/update cluster 'smb_local'
[2019-12-23 10:16:58.213][1][info][upstream] [source/common/upstream/cds_api_impl.cc:87] cds: add/update cluster 'smb_internet'
[2019-12-23 10:16:58.215][1][info][upstream] [source/common/upstream/cds_api_impl.cc:87] cds: add/update cluster 'transmission_bt'
[2019-12-23 10:16:58.217][1][info][upstream] [source/common/upstream/cds_api_impl.cc:87] cds: add/update cluster 'time'
[2019-12-23 10:16:58.233][1][info][upstream] [source/server/lds_api.cc:71] lds: add/update listener 'listener_http_v6'
[2019-12-23 10:16:58.243][1][warning][config] [source/server/listener_impl.cc:287] adding listener '0.0.0.0:8443': filter chain match rules require TLS Inspector listener filter, but it isn't configured, trying to inject it (this might fail if Envoy is compiled without it)
[2019-12-23 10:16:58.243][1][info][upstream] [source/server/lds_api.cc:71] lds: add/update listener 'listener_https_v4'
[2019-12-23 10:16:58.254][1][warning][config] [source/server/listener_impl.cc:287] adding listener '[::]:443': filter chain match rules require TLS Inspector listener filter, but it isn't configured, trying to inject it (this might fail if Envoy is compiled without it)
[2019-12-23 10:16:58.255][1][info][upstream] [source/server/lds_api.cc:71] lds: add/update listener 'listener_https_v6'
[2019-12-23 10:16:58.255][1][info][upstream] [source/server/lds_api.cc:71] lds: add/update listener 'listener_ntp'
[2019-12-23 10:16:58.260][1][info][upstream] [source/server/lds_api.cc:71] lds: add/update listener 'listener_aria2'
[2019-12-23 10:16:58.263][1][info][upstream] [source/server/lds_api.cc:71] lds: add/update listener 'listener_aria2_bt'
[2019-12-23 10:16:58.265][1][info][upstream] [source/server/lds_api.cc:71] lds: add/update listener 'listener_aria2_dht'
[2019-12-23 10:16:58.269][1][info][upstream] [source/server/lds_api.cc:71] lds: add/update listener 'listener_smb_local'
[2019-12-23 10:16:58.272][1][info][upstream] [source/server/lds_api.cc:71] lds: add/update listener 'listener_smb_internet'
[2019-12-23 10:16:58.275][1][info][upstream] [source/server/lds_api.cc:71] lds: add/update listener 'listener_transmission_bt'
监控截图:
脚注
[1]GetEnvoy: https://www.getenvoy.io/
[2]inotify: https://www.infoq.cn/article/inotify-linux-file-system-event-monitoring
你可能还喜欢
点击下方图片即可阅读
云原生是一种信仰 🤘
扫码关注公众号
后台回复◉图谱◉领取史上最强 Kubernetes 知识图谱
点击 "阅读原文" 获取更好的阅读体验!