7.9 实现服务灰度发布
灰度发布又叫金丝雀发布，通过Istio提供的服务路由功能可以轻松实现灰度发布。当我们的服务需要上线新版本时，可以先只把部分流量引入到新版本，等到确认新版本没有问题时，再把全部流量切换到新版本上。通过调节服务版本路由的权重，就可以达到上述目的。当然，为了不浪费资源，对服务的新版本我们只需要部署很少的实例，然后确认服务没有问题后，再增加新版本的服务实例，减少旧版本的服务实例。
对service-go服务实行灰度发布，使用的路由规则配置文件如下：
apiVersion: networking.istio.io/v1alpha3
kind: DestinationRule
metadata:
name: service-go
spec:
host: service-go
subsets:
- name: v1
labels:
version: v1
- name: v2
labels:
version: v2
---
apiVersion: networking.istio.io/v1alpha3
kind: VirtualService
metadata:
name: service-go
spec:
hosts:
- service-go
http:
- route:
- destination:
host: service-go
subset: v1
weight: 70
- destination:
host: service-go
subset: v2
weight: 30
【实验】
1）删除service-go服务的v2版本的Pod：
$ kubectl delete deploy service-go-v2 $ kubectl get pod -l app=service-go NAME READY STATUS RESTARTS AGE service-go-v1-7cc5c6f574-2k6wv 2/2 Running 0 4m38s
2）为了模拟真实环境，我们提前把service-go服务扩容到5个实例，也就是5个Pod，使用如下方式增加实例个数：
$ kubectl scale --replicas=5 deployment/service-go-v1 $ kubectl get pod -l app=service-go service-go-v1-7cc5c6f574-9g52l 2/2 Running 0 29s service-go-v1-7cc5c6f574-kkmmh 2/2 Running 0 29s service-go-v1-7cc5c6f574-kqwg5 2/2 Running 0 7m24s service-go-v1-7cc5c6f574-pl4dt 2/2 Running 0 29s service-go-v1-7cc5c6f574-snd7b 2/2 Running 0 29s
3）创建用于测试的Pod：
$ kubectl apply -f kubernetes/dns-test.yaml
4）模拟生产环境，流量全部转发到服务service-go的v1版本：
$ kubectl apply -f istio/route/virtual-service-go-v1.yaml
5）访问service-go服务，此时所有的请求只会落到v1版本上：
$ kubectl exec dns-test -c dns-test -- curl -s http://service-go/env
{"message":"go v1"}
6）部署新版本服务，此时只创建两个v2版本的service-go服务实例用来准备接收请求：
$ kubectl apply -f service/go/service-go-v2.yaml $ kubectl scale --replicas=2 deployment/service-go-v2 $ kubectl get pod -l app=service-go NAME READY STATUS RESTARTS AGE service-go-v1-7cc5c6f574-9g52l 2/2 Running 0 6m40s service-go-v1-7cc5c6f574-kkmmh 2/2 Running 0 6m40s service-go-v1-7cc5c6f574-kqwg5 2/2 Running 0 13m service-go-v1-7cc5c6f574-pl4dt 2/2 Running 0 6m40s service-go-v1-7cc5c6f574-snd7b 2/2 Running 0 6m40s service-go-v2-7656dcc478-7xn4h 2/2 Running 0 19s service-go-v2-7656dcc478-qmkr8 2/2 Running 0 8s
7）给service-go服务v2版本分配30%流量：
$ kubectl apply -f istio/route/virtual-service-go-canary.yaml
8）访问service-go服务，此时请求会分别落到v1、v2两个版本上：
$ kubectl exec dns-test -c dns-test -- curl -s http://service-go/env
{"message":"go v1"}
$ kubectl exec dns-test -c dns-test -- curl -s http://service-go/env
{"message":"go v2"}
9）观察服务的性能、响应时间、错误率等数据，当服务达到要求时，增加service-go服务的v2版本实例数：
$ kubectl scale --replicas=3 deployment/service-go-v2 $ kubectl get pod -l app=service-go NAME READY STATUS RESTARTS AGE service-go-v1-7cc5c6f574-9g52l 2/2 Running 0 15m service-go-v1-7cc5c6f574-kkmmh 2/2 Running 0 15m service-go-v1-7cc5c6f574-kqwg5 2/2 Running 0 22m service-go-v1-7cc5c6f574-pl4dt 2/2 Running 0 15m service-go-v1-7cc5c6f574-snd7b 2/2 Running 0 15m service-go-v2-7656dcc478-78jt7 2/2 Running 0 116s service-go-v2-7656dcc478-7xn4h 2/2 Running 0 9m service-go-v2-7656dcc478-qmkr8 2/2 Running 0 8m49s
10）给service-go服务v2版本分配50%流量：
$ kubectl apply -f istio/route/virtual-service-go-canary-50.yaml
11）减少v1版本的实例数，增加service-go服务的v2版本实例数：
$ kubectl scale --replicas=2 deployment/service-go-v1 $ kubectl scale --replicas=4 deployment/service-go-v2 $ kubectl get pod -l app=service-go NAME READY STATUS RESTARTS AGE service-go-v1-7cc5c6f574-kkmmh 2/2 Running 0 17m service-go-v1-7cc5c6f574-kqwg5 2/2 Running 0 24m service-go-v2-7656dcc478-78jt7 2/2 Running 0 4m26s service-go-v2-7656dcc478-7xn4h 2/2 Running 0 11m service-go-v2-7656dcc478-qmkr8 2/2 Running 0 11m service-go-v2-7656dcc478-t2t57 2/2 Running 0 60s
12）给service-go服务v2版本分配70%流量：
$ kubectl apply -f istio/route/virtual-service-go-canary-70.yaml
13）增加service-go服务的v2版本实例数：
$ kubectl scale --replicas=5 deployment/service-go-v2 $ kubectl get pod -l app=service-go NAME READY STATUS RESTARTS AGE service-go-v1-7cc5c6f574-kkmmh 2/2 Running 0 19m service-go-v1-7cc5c6f574-kqwg5 2/2 Running 0 26m service-go-v2-7656dcc478-4nn2q 2/2 Running 0 19s service-go-v2-7656dcc478-78jt7 2/2 Running 0 5m53s service-go-v2-7656dcc478-7xn4h 2/2 Running 0 12m service-go-v2-7656dcc478-qmkr8 2/2 Running 0 12m service-go-v2-7656dcc478-t2t57 2/2 Running 0 2m27s
14）把流量全部引入service-go服务的v2版本：
$ kubectl apply -f istio/route/virtual-service-go-v2.yaml
15）访问service-go服务，此时所有的请求只会落到v2版本上：
$ kubectl exec dns-test -c dns-test -- curl -s http://service-go/env
{"message":"go v2"}
16）减少或者删除service-go服务v1版本的实例，完成服务的灰度发布：
$ kubectl scale --replicas=0 deployment/service-go-v1 $ kubectl get pod -l app=service-go NAME READY STATUS RESTARTS AGE service-go-v2-7656dcc478-4nn2q 2/2 Running 0 2m35s service-go-v2-7656dcc478-78jt7 2/2 Running 0 8m9s service-go-v2-7656dcc478-7xn4h 2/2 Running 0 15m service-go-v2-7656dcc478-qmkr8 2/2 Running 0 15m service-go-v2-7656dcc478-t2t57 2/2 Running 0 4m43s $ kubectl delete deployment service-go-v1
17）清理。
恢复环境到实验的初始状态：
$ kubectl delete -f kubernetes/dns-test.yaml $ kubectl delete -f istio/route/virtual-service-go-v2.yaml $ kubectl scale --replicas=1 deployment/service-go-v2 $ kubectl apply -f service/go/service-go.yaml $ kubectl get pod -l app=service-go NAME READY STATUS RESTARTS AGE service-go-v1-7cc5c6f574-2282d 2/2 Running 0 16s service-go-v2-7656dcc478-7xn4h 2/2 Running 0 22m