写在最前

这篇是关于如何使用 Flomesh^[1] 服务网格来强化 Spring Cloud 的服务治理能力，降低 Spring Cloud 微服务架构落地服务网格的门槛，实现“自主可控”。

文档在 github^[2] 上持续更新，欢迎大家一起讨论：https://github.com/flomesh-io/flomesh-bookinfo-demo。

架构

环境搭建

搭建 Kubernetes 环境，可以选择 kubeadm 进行集群搭建。也可以选择 minikube、k3s、Kind 等，本文使用 k3s。

使用 k3d^[3] 安装 k3s^[4]。k3d 将在 Docker 容器中运行 k3s，因此需要保证已经安装了 Docker。

$ k3d cluster create spring-demo -p "81:80@loadbalancer" --k3s-server-arg '--no-deploy=traefik'

安装 Flomesh

从仓库 https://github.com/flomesh-io/flomesh-bookinfo-demo.git 克隆代码。进入到 flomesh-bookinfo-demo/kubernetes目录。

所有 Flomesh 组件以及用于 demo 的 yamls 文件都位于这个目录中。

安装 Cert Manager

$ kubectl apply -f artifacts/cert-manager-v1.3.1.yamlcustomresourcedefinition.apiextensions.k8s.io/certificaterequests.cert-manager.io createdcustomresourcedefinition.apiextensions.k8s.io/certificates.cert-manager.io createdcustomresourcedefinition.apiextensions.k8s.io/challenges.acme.cert-manager.io createdcustomresourcedefinition.apiextensions.k8s.io/clusterissuers.cert-manager.io createdcustomresourcedefinition.apiextensions.k8s.io/issuers.cert-manager.io createdcustomresourcedefinition.apiextensions.k8s.io/orders.acme.cert-manager.io creatednamespace/cert-manager createdserviceaccount/cert-manager-cainjector createdserviceaccount/cert-manager createdserviceaccount/cert-manager-webhook createdclusterrole.rbac.authorization.k8s.io/cert-manager-cainjector createdclusterrole.rbac.authorization.k8s.io/cert-manager-controller-issuers createdclusterrole.rbac.authorization.k8s.io/cert-manager-controller-clusterissuers createdclusterrole.rbac.authorization.k8s.io/cert-manager-controller-certificates createdclusterrole.rbac.authorization.k8s.io/cert-manager-controller-orders createdclusterrole.rbac.authorization.k8s.io/cert-manager-controller-challenges createdclusterrole.rbac.authorization.k8s.io/cert-manager-controller-ingress-shim createdclusterrole.rbac.authorization.k8s.io/cert-manager-view createdclusterrole.rbac.authorization.k8s.io/cert-manager-edit createdclusterrole.rbac.authorization.k8s.io/cert-manager-controller-approve:cert-manager-io createdclusterrole.rbac.authorization.k8s.io/cert-manager-webhook:subjectaccessreviews createdclusterrolebinding.rbac.authorization.k8s.io/cert-manager-cainjector createdclusterrolebinding.rbac.authorization.k8s.io/cert-manager-controller-issuers createdclusterrolebinding.rbac.authorization.k8s.io/cert-manager-controller-clusterissuers createdclusterrolebinding.rbac.authorization.k8s.io/cert-manager-controller-certificates createdclusterrolebinding.rbac.authorization.k8s.io/cert-manager-controller-orders createdclusterrolebinding.rbac.authorization.k8s.io/cert-manager-controller-challenges createdclusterrolebinding.rbac.authorization.k8s.io/cert-manager-controller-ingress-shim createdclusterrolebinding.rbac.authorization.k8s.io/cert-manager-controller-approve:cert-manager-io createdclusterrolebinding.rbac.authorization.k8s.io/cert-manager-webhook:subjectaccessreviews createdrole.rbac.authorization.k8s.io/cert-manager-cainjector:leaderelection createdrole.rbac.authorization.k8s.io/cert-manager:leaderelection createdrole.rbac.authorization.k8s.io/cert-manager-webhook:dynamic-serving createdrolebinding.rbac.authorization.k8s.io/cert-manager-cainjector:leaderelection createdrolebinding.rbac.authorization.k8s.io/cert-manager:leaderelection createdrolebinding.rbac.authorization.k8s.io/cert-manager-webhook:dynamic-serving createdservice/cert-manager createdservice/cert-manager-webhook createddeployment.apps/cert-manager-cainjector createddeployment.apps/cert-manager createddeployment.apps/cert-manager-webhook createdmutatingwebhookconfiguration.admissionregistration.k8s.io/cert-manager-webhook createdvalidatingwebhookconfiguration.admissionregistration.k8s.io/cert-manager-webhook created

注意: 要保证 cert-manager 命名空间中所有的 pod 都正常运行：

$ kubectl get pod -n cert-managerNAME                                       READY   STATUS    RESTARTS   AGEcert-manager-webhook-56fdcbb848-q7fn5      1/1     Running   0          98scert-manager-59f6c76f4b-z5lgf              1/1     Running   0          98scert-manager-cainjector-59f76f7fff-flrr7   1/1     Running   0          98s

安装 Pipy Operator

$ kubectl apply -f artifacts/pipy-operator.yaml

执行完命令后会看到类似的结果：

namespace/flomesh createdcustomresourcedefinition.apiextensions.k8s.io/proxies.flomesh.io createdcustomresourcedefinition.apiextensions.k8s.io/proxyprofiles.flomesh.io createdserviceaccount/operator-manager createdrole.rbac.authorization.k8s.io/leader-election-role createdclusterrole.rbac.authorization.k8s.io/manager-role createdclusterrole.rbac.authorization.k8s.io/metrics-reader createdclusterrole.rbac.authorization.k8s.io/proxy-role createdrolebinding.rbac.authorization.k8s.io/leader-election-rolebinding createdclusterrolebinding.rbac.authorization.k8s.io/manager-rolebinding createdclusterrolebinding.rbac.authorization.k8s.io/proxy-rolebinding createdconfigmap/manager-config createdservice/operator-manager-metrics-service createdservice/proxy-injector-svc createdservice/webhook-service createddeployment.apps/operator-manager createddeployment.apps/proxy-injector createdcertificate.cert-manager.io/serving-cert createdissuer.cert-manager.io/selfsigned-issuer createdmutatingwebhookconfiguration.admissionregistration.k8s.io/mutating-webhook-configuration createdmutatingwebhookconfiguration.admissionregistration.k8s.io/proxy-injector-webhook-cfg createdvalidatingwebhookconfiguration.admissionregistration.k8s.io/validating-webhook-configuration created

注意：要保证 flomesh 命名空间中所有的 pod 都正常运行：

$ kubectl get pod -n flomeshNAME                               READY   STATUS    RESTARTS   AGEproxy-injector-5bccc96595-spl6h    1/1     Running   0          39soperator-manager-c78bf8d5f-wqgb4   1/1     Running   0          39s

安装 Ingress 控制器：ingress-pipy

$ kubectl apply -f ingress/ingress-pipy.yamlnamespace/ingress-pipy createdcustomresourcedefinition.apiextensions.k8s.io/ingressparameters.flomesh.io createdserviceaccount/ingress-pipy createdrole.rbac.authorization.k8s.io/ingress-pipy-leader-election-role createdclusterrole.rbac.authorization.k8s.io/ingress-pipy-role createdrolebinding.rbac.authorization.k8s.io/ingress-pipy-leader-election-rolebinding createdclusterrolebinding.rbac.authorization.k8s.io/ingress-pipy-rolebinding createdconfigmap/ingress-config createdservice/ingress-pipy-cfg createdservice/ingress-pipy-controller createdservice/ingress-pipy-defaultbackend createdservice/webhook-service createddeployment.apps/ingress-pipy-cfg createddeployment.apps/ingress-pipy-controller createddeployment.apps/ingress-pipy-manager createdcertificate.cert-manager.io/serving-cert createdissuer.cert-manager.io/selfsigned-issuer createdmutatingwebhookconfiguration.admissionregistration.k8s.io/mutating-webhook-configuration configuredvalidatingwebhookconfiguration.admissionregistration.k8s.io/validating-webhook-configuration configured

检查 ingress-pipy 命名空间下 pod 的状态：

$ kubectl get pod -n ingress-pipyNAME                                       READY   STATUS    RESTARTS   AGEsvclb-ingress-pipy-controller-8pk8k        1/1     Running   0          71singress-pipy-cfg-6bc649cfc7-8njk7          1/1     Running   0          71singress-pipy-controller-76cd866d78-m7gfp   1/1     Running   0          71singress-pipy-manager-5f568ff988-tw5w6      0/1     Running   0          70s

至此，你已经成功安装 Flomesh 的所有组件，包括 operator 和 ingress 控制器。

中间件

Demo 需要用到中间件完成日志和统计数据的存储，这里为了方便使用 pipy 进行 mock：直接在控制台中打印数据。

另外，服务治理相关的配置有 mock 的 pipy config 服务提供。

log & metrics

$ cat > middleware.js <<EOFpipy().listen(8123)    .link('mock').listen(9001)    .link('mock').pipeline('mock')    .decodeHttpRequest()    .replaceMessage(        req => (            console.log(req.body.toString()),            new Message('OK')        )    )    .encodeHttpResponse()EOF$ docker run --rm --name middleware --entrypoint "pipy" -v ${PWD}:/script -p 8123:8123 -p 9001:9001 flomesh/pipy-pjs:0.4.0-118 /script/middleware.js

pipy config

$ cat > mock-config.json <<EOF{  "ingress": {},  "inbound": {    "rateLimit": -1,    "dataLimit": -1,    "circuitBreak": false,    "blacklist": []  },  "outbound": {    "rateLimit": -1,    "dataLimit": -1  }}EOF$ cat > mock.js <<EOFpipy({  _CONFIG_FILENAME: 'mock-config.json',  _serveFile: (req, filename, type) => (    new Message(      {        bodiless: req.head.method === 'HEAD',        headers: {          'etag': os.stat(filename)?.mtime | 0,          'content-type': type,        },      },      req.head.method === 'HEAD' ? null : os.readFile(filename),    )  ),  _router: new algo.URLRouter({    '/config': req => _serveFile(req, _CONFIG_FILENAME, 'application/json'),    '/*': () => new Message({ status: 404 }, 'Not found'),  }),})// Config.listen(9000)  .decodeHttpRequest()  .replaceMessage(    req => (      _router.find(req.head.path)(req)    )  )  .encodeHttpResponse()EOF$ docker run --rm --name mock --entrypoint "pipy" -v ${PWD}:/script -p 9000:9000 flomesh/pipy-pjs:0.4.0-118 /script/mock.js

运行 Demo

Demo 运行在另一个独立的命名空间 flomesh-spring 中，执行命令 kubectl apply -f base/namespace.yaml 来创建该命名空间。如果你 describe 该命名空间你会发现其使用了 flomesh.io/inject=true 标签。

这个标签告知 operator 的 admission webHook 拦截标注的命名空间下 pod 的创建。

$ kubectl describe ns flomesh-springName:         flomesh-springLabels:       app.kubernetes.io/name=spring-mesh              app.kubernetes.io/version=1.19.0              flomesh.io/inject=true              kubernetes.io/metadata.name=flomesh-springAnnotations:  <none>Status:       ActiveNo resource quota.No LimitRange resource.

我们首先看下 Flomesh 提供的 CRD ProxyProfile。这个 demo 中，其定义了 sidecar 容器片段以及所使用的的脚本。检查 sidecar/proxy-profile.yaml 获取更多信息。执行下面的命令，创建 CRD 资源。

$ kubectl apply -f sidecar/proxy-profile.yaml

检查是否创建成功：

$ kubectl get pf -o wideNAME                         NAMESPACE        DISABLED   SELECTOR                                     CONFIG                                                                AGEproxy-profile-002-bookinfo   flomesh-spring   false      {"matchLabels":{"sys":"bookinfo-samples"}}   {"flomesh-spring":"proxy-profile-002-bookinfo-fsmcm-b67a9e39-0418"}   27s

As the services has startup dependencies, you need to deploy it one by one following the strict order. Before starting, check the Endpoints section of base/clickhouse.yaml.

提供中间件的访问 endpoid，将 base/clickhouse.yaml、base/metrics.yaml 和 base/config.yaml 中的 ip 地址改为本机的 ip 地址（不是 127.0.0.1）。

修改之后，执行如下命令：

$ kubectl apply -f base/clickhouse.yaml$ kubectl apply -f base/metrics.yaml$ kubectl apply -f base/config.yaml$ kubectl get endpoints samples-clickhouse samples-metrics samples-configNAME                 ENDPOINTS            AGEsamples-clickhouse   192.168.1.101:8123   3msamples-metrics      192.168.1.101:9001   3ssamples-config       192.168.1.101:9000   3m

部署注册中心

$ kubectl apply -f base/discovery-server.yaml

检查注册中心 pod 的状态，确保 3 个容器都运行正常。

$ kubectl get podNAME                                           READY   STATUS        RESTARTS   AGEsamples-discovery-server-v1-85798c47d4-dr72k   3/3     Running       0          96s

部署配置中心

$ kubectl apply -f base/config-service.yaml

部署 API 网关以及 bookinfo 相关的服务

$ kubectl apply -f base/bookinfo-v1.yaml$ kubectl apply -f base/bookinfo-v2.yaml$ kubectl apply -f base/productpage-v1.yaml$ kubectl apply -f base/productpage-v2.yaml

检查 pod 状态，可以看到所有 pod 都注入了容器。

$ kubectl get podssamples-discovery-server-v1-85798c47d4-p6zpb       3/3     Running   0          19hsamples-config-service-v1-84888bfb5b-8bcw9         1/1     Running   0          19hsamples-api-gateway-v1-75bb6456d6-nt2nl            3/3     Running   0          6h43msamples-bookinfo-ratings-v1-6d557dd894-cbrv7       3/3     Running   0          6h43msamples-bookinfo-details-v1-756bb89448-dxk66       3/3     Running   0          6h43msamples-bookinfo-reviews-v1-7778cdb45b-pbknp       3/3     Running   0          6h43msamples-api-gateway-v2-7ddb5d7fd9-8jgms            3/3     Running   0          6h37msamples-bookinfo-ratings-v2-845d95fb7-txcxs        3/3     Running   0          6h37msamples-bookinfo-reviews-v2-79b4c67b77-ddkm2       3/3     Running   0          6h37msamples-bookinfo-details-v2-7dfb4d7c-jfq4j         3/3     Running   0          6h37msamples-bookinfo-productpage-v1-854675b56-8n2xd    1/1     Running   0          7m1ssamples-bookinfo-productpage-v2-669bd8d9c7-8wxsf   1/1     Running   0          6m57s

添加 Ingress 规则

执行如下命令添加 Ingress 规则。

$ kubectl apply -f ingress/ingress.yaml

测试前的准备

访问 demo 服务都要通过 ingress 控制器。因此需要先获取 LB 的 ip 地址。

//Obtain the controller IP//Here, we append port. ingressAddr=`kubectl get svc ingress-pipy-controller -n ingress-pipy -o jsonpath='{.spec.clusterIP}'`:81

这里我们使用了是 k3d 创建的 k3s，命令中加入了 -p 81:80@loadbalancer 选项。我们可以使用 127.0.0.1:81 来访问 ingress 控制器。这里执行命令 ingressAddr=127.0.0.1:81。

Ingress 规则中，我们为每个规则指定了 host，因此每个请求中需要通过 HTTP 请求头 Host 提供对应的 host。

或者在 /etc/hosts 添加记录：

$ kubectl get ing ingress-pipy-bookinfo -n flomesh-spring -o jsonpath="{range .spec.rules[*]}{.host}{'\n'}"api-v1.flomesh.cnapi-v2.flomesh.cnfe-v1.flomesh.cnfe-v2.flomesh.cn//添加记录到 /etc/hosts127.0.0.1 api-v1.flomesh.cn api-v2.flomesh.cn fe-v1.flomesh.cn fe-v2.flomesh.cn

验证

$ curl http://127.0.0.1:81/actuator/health -H 'Host: api-v1.flomesh.cn'{"status":"UP","groups":["liveness","readiness"]}//OR$ curl http://api-v1.flomesh.cn:81/actuator/health{"status":"UP","groups":["liveness","readiness"]}

测试

灰度

在 v1 版本的服务中，我们为 book 添加 rating 和 review。

# rate a book$ curl -X POST http://$ingressAddr/bookinfo-ratings/ratings \    -H "Content-Type: application/json" \    -H "Host: api-v1.flomesh.cn" \    -d '{"reviewerId":"9bc908be-0717-4eab-bb51-ea14f669ef20","productId":"2099a055-1e21-46ef-825e-9e0de93554ea","rating":3}' $ curl http://$ingressAddr/bookinfo-ratings/ratings/2099a055-1e21-46ef-825e-9e0de93554ea -H "Host: api-v1.flomesh.cn"# review a book$ curl -X POST http://$ingressAddr/bookinfo-reviews/reviews \    -H "Content-Type: application/json" \    -H "Host: api-v1.flomesh.cn" \    -d '{"reviewerId":"9bc908be-0717-4eab-bb51-ea14f669ef20","productId":"2099a055-1e21-46ef-825e-9e0de93554ea","review":"This was OK.","rating":3}'$ curl http://$ingressAddr/bookinfo-reviews/reviews/2099a055-1e21-46ef-825e-9e0de93554ea -H "Host: api-v1.flomesh.cn"

执行上面的命令之后，我们可以在浏览器中访问前端服务（http://fe-v1.flomesh.cn:81/productpage?u=normal、 http://fe-v2.flomesh.cn:81/productpage?u=normal），只有 v1 版本的前端中才能看到刚才添加的记录。

熔断

这里熔断我们通过修改 mock-config.json 中的 inbound.circuitBreak 为 true，来将服务强制开启熔断：

{  "ingress": {},  "inbound": {    "rateLimit": -1,    "dataLimit": -1,    "circuitBreak": true, //here    "blacklist": []  },  "outbound": {    "rateLimit": -1,    "dataLimit": -1  }}

$ curl http://$ingressAddr/actuator/health -H 'Host: api-v1.flomesh.cn'HTTP/1.1 503 Service UnavailableConnection: keep-aliveContent-Length: 27Service Circuit Break Open

限流

修改 pipy config 的配置，将 inbound.rateLimit 设置为 1。

{  "ingress": {},  "inbound": {    "rateLimit": 1, //here    "dataLimit": -1,    "circuitBreak": false,    "blacklist": []  },  "outbound": {    "rateLimit": -1,    "dataLimit": -1  }}

我们使用 wrk 模拟发送请求，20 个连接、20 个请求、持续 30s：

$ wrk -t20 -c20 -d30s --latency http://$ingressAddr/actuator/health -H 'Host: api-v1.flomesh.cn'Running 30s test @ http://127.0.0.1:81/actuator/health  20 threads and 20 connections  Thread Stats   Avg      Stdev     Max   +/- Stdev    Latency   951.51ms  206.23ms   1.04s    93.55%    Req/Sec     0.61      1.71    10.00     93.55%  Latency Distribution     50%    1.00s     75%    1.01s     90%    1.02s     99%    1.03s  620 requests in 30.10s, 141.07KB readRequests/sec:     20.60Transfer/sec:      4.69KB

从结果来看 20.60 req/s，即每个连接 1 req/s。

黑白名单

将 pipy config 的 mock-config.json 做如下修改：ip 地址使用的是 ingress controller 的 pod ip。

$ kgpo -n ingress-pipy ingress-pipy-controller-76cd866d78-4cqqn -o jsonpath='{.status.podIP}'10.42.0.78

{  "ingress": {},  "inbound": {    "rateLimit": -1,    "dataLimit": -1,    "circuitBreak": false,    "blacklist": ["10.42.0.78"] //here  },  "outbound": {    "rateLimit": -1,    "dataLimit": -1  }}

还是访问网关的接口

curl http://$ingressAddr/actuator/health -H 'Host: api-v1.flomesh.cn'HTTP/1.1 503 Service Unavailablecontent-type: text/plainConnection: keep-aliveContent-Length: 20Service Unavailable

引用链接

[1] Flomesh: https://flomesh.cn/
[2] github: https://github.com/flomesh-io/flomesh-bookinfo-demo
[3] k3d: https://k3d.io/
[4] k3s: https://github.com/k3s-io/k3s

ITPub博客

使用 Flomesh 强化 Spring Cloud 服务治理