在GreatDB分布式部署模式中使用Chaos Mesh做混沌测试

1. 需求背景与万里安全数据库软件GreatDB分布式部署模式介绍

1.1 需求背景

混沌测试是检测分布式系统不确定性、建立系统弹性信心的一种非常好的方式，因此我们采用开源工具Chaos Mesh来做GreatDB分布式集群的混沌测试。

1.2 万里安全数据库软件GreatDB分布式部署模式介绍

万里安全数据库软件GreatDB 是一款关系型数据库软件，同时支持集中式和分布式的部署方式，本文涉及的是分布式部署方式。

分布式部署模式采用shared-nothing架构；通过数据冗余与副本管理确保数据库无单点故障；数据sharding与分布式并行计算实现数据库系统高性能；可无限制动态扩展数据节点，满足业务需要。

整体架构如下图所示：

2. 环境准备

2.1 Chaos Mesh安装

在安装Chaos Mesh之前请确保已经预先安装了helm，docker，并准备好了一个kubernetes环境。

• 使用Helm安装

1）在 Helm 仓库中添加 Chaos Mesh 仓库：

helm repo add chaos-mesh https://charts.chaos-mesh.org

2）查看可以安装的 Chaos Mesh 版本：

helm search repo chaos-mesh

3）创建安装 Chaos Mesh 的命名空间：

kubectl create ns chaos-testing

4）在docker环境下安装Chaos Mesh：

helm install chaos-mesh chaos-mesh/chaos-mesh -n=chaos-testing

• 验证安装

执行以下命令查看Chaos Mesh的运行情况：

kubectl get pod -n chaos-testing

下面是预期输出：

NAME                                       READY   STATUS    RESTARTS   AGEchaos-controller-manager-d7bc9ccb5-dbccq   1/1     Running   0          26dchaos-daemon-pzxc7                         1/1     Running   0          26dchaos-dashboard-5887f7559b-kgz46           1/1     Running   1          26d

如果3个pod的状态都是Running，表示 Chaos Mesh 已经成功安装。

2.2 准备测试需要的镜像

2.2.1 准备mysql镜像

一般情况下，mysql使用官方5.7版本的镜像，mysql监控采集器使用的是mysqld-exporter，也可以直接从docker hub下载：

docker pull mysql:5.7docker pull prom/mysqld-exporter
2.2.2 准备zookeeper镜像

zookeeper使用的是官方3.5.5版本镜像，zookeeper组件涉及的监控有jmx-prometheus-exporter 和zookeeper-exporter，均从docker hub下载：

docker pull zookeeper:3.5.5docker pull sscaling/jmx-prometheus-exporterdocker pull josdotso/zookeeper-exporter
2.2.3 准备GreatDB镜像

选择一个GreatDB的tar包，将其解压得到一个./greatdb目录，再将greatdb-service-docker.sh文件拷贝到这个解压出来的./greatdb目录里：

cp greatdb-service-docker.sh ./greatdb/

将greatdb Dockerfile放到./greatdb文件夹的同级目录下，然后执行以下命令构建GreatDB镜像：

docker build -t greatdb/greatdb:tag2021 .
2.2.4 准备GreatDB分布式集群部署/清理的镜像

下载集群部署脚本cluster-setup,集群初始化脚本init-zk 以及集群helm charts包（可咨询4.0开发/测试组获取）

将上述材料放在同一目录下，编写如下Dockerfile:

FROM debian:buster-slim as init-zkCOPY ./init-zk /root/init-zkRUN chmod +x /root/init-zkFROM debian:buster-slim as cluster-setup\# Set aliyun repo for speedRUN sed -i 's/deb.debian.org/mirrors.aliyun.com/g' /etc/apt/sources.list && \  sed -i 's/security.debian.org/mirrors.aliyun.com/g' /etc/apt/sources.listRUN apt-get -y update && \  apt-get -y install \  curl \  wgetRUN curl -L https://storage.googleapis.com/kubernetes-release/release/v1.20.1/bin/linux/amd64/kubectl -o /usr/local/bin/kubectl && \  chmod +x /usr/local/bin/kubectl && \  mkdir /root/.kube && \  wget https://get.helm.sh/helm-v3.5.3-linux-amd64.tar.gz && \  tar -zxvf helm-v3.5.3-linux-amd64.tar.gz && \  mv linux-amd64/helm /usr/local/bin/helmCOPY ./config /root/.kube/COPY ./helm /helmCOPY ./cluster-setup /

执行以下命令构建所需镜像：

docker build --target init-zk -t greatdb/initzk:latest .docker build --target cluster-setup -t greatdb/cluster-setup:v1 .
2.2.5 准备测试用例的镜像

目前测试支持的用例有：bank,bank2,pbank,tpcc,flashback等，每个用例都是一个可执行文件。

以flashback测例为例构建测试用例的镜像，先将用例下载到本地，在用例的同一目录下编写如下内容的Dockerfile：

FROM debian:buster-slimCOPY ./flashback /RUN cd / && chmod +x ./flashback

执行以下命令构建测试用例镜像：

docker build -t greatdb/testsuite-flashback:v1 .

2.3 将准备好的镜像上传到私有仓库中

创建私有仓库和上传镜像操作请参考：https://zhuanlan.zhihu.com/p/78543733

3. Chaos Mesh的使用

3.1 搭建GreatDB分布式集群

在上一章2.2.4 中cluster-setup目录下执行以下命令块去搭建测试集群：

./cluster-setup  \-clustername=c0 \-namespace=test \-enable-monitor=true \-mysql-image=mysql:5.7 \-mysql-replica=3 \-mysql-auth=1 \-mysql-normal=1 \-mysql-global=1 \-mysql-partition=1 \-zookeeper-repository=zookeeper \-zookeeper-tag=3.5.5 \-zookeeper-replica=3 \-greatdb-repository=greatdb/greatdb \-greatdb-tag=tag202110 \-greatdb-replica=3 \-greatdb-serviceHost=172.16.70.249

输出信息：

liuxinle@liuxinle-OptiPlex-5060:~/k8s/cluster-setup$ ./cluster-setup \> -clustername=c0 \> -namespace=test \> -enable-monitor=true \> -mysql-image=mysql:5.7 \> -mysql-replica=3 \> -mysql-auth=1 \> -mysql-normal=1 \> -mysql-global=1 \> -mysql-partition=1 \> -zookeeper-repository=zookeeper \> -zookeeper-tag=3.5.5 \> -zookeeper-replica=3 \> -greatdb-repository=greatdb/greatdb \> -greatdb-tag=tag202110 \> -greatdb-replica=3 \> -greatdb-serviceHost=172.16.70.249INFO[2021-10-14T10:41:52+08:00] SetUp the cluster ...                         NameSpace=testINFO[2021-10-14T10:41:52+08:00] create namespace ...                         INFO[2021-10-14T10:41:57+08:00] copy helm chart templates ...                INFO[2021-10-14T10:41:57+08:00] setup ...                                     Component=MySQLINFO[2021-10-14T10:41:57+08:00] exec helm install and update greatdb-cfg.yaml ... INFO[2021-10-14T10:42:00+08:00] waiting mysql pods running ...               INFO[2021-10-14T10:44:27+08:00] setup ...                                     Component=ZookeeperINFO[2021-10-14T10:44:28+08:00] waiting zookeeper pods running ...           INFO[2021-10-14T10:46:59+08:00] update greatdb-cfg.yaml                      INFO[2021-10-14T10:46:59+08:00] setup ...                                     Component=greatdbINFO[2021-10-14T10:47:00+08:00] waiting greatdb pods running ...             INFO[2021-10-14T10:47:21+08:00] waiting cluster running ...                  INFO[2021-10-14T10:47:27+08:00] waiting prometheus server running...         INFO[2021-10-14T10:47:27+08:00] Dump Cluster Info                            INFO[2021-10-14T10:47:27+08:00] SetUp success.                                ClusterName=c0 NameSpace=test

执行如下命令，查看集群pod状态：

kubectl get pod -n test -o wide

输出信息：

NAME                                    READY   STATUS      RESTARTS   AGE     IP             NODE                     NOMINATED NODE   READINESS GATESc0-auth0-mysql-0                        2/2     Running     0          10m     10.244.87.18   liuxinle-optiplex-5060              c0-auth0-mysql-1                        2/2     Running     0          9m23s   10.244.87.54   liuxinle-optiplex-5060              c0-auth0-mysql-2                        2/2     Running     0          8m39s   10.244.87.57   liuxinle-optiplex-5060              c0-greatdb-0                            2/2     Running     1          5m3s    10.244.87.58   liuxinle-optiplex-5060              c0-greatdb-1                            2/2     Running     0          4m57s   10.244.87.20   liuxinle-optiplex-5060              c0-greatdb-2                            2/2     Running     0          4m50s   10.244.87.47   liuxinle-optiplex-5060              c0-glob0-mysql-0                        2/2     Running     0          10m     10.244.87.51   liuxinle-optiplex-5060              c0-glob0-mysql-1                        2/2     Running     0          9m23s   10.244.87.41   liuxinle-optiplex-5060              c0-glob0-mysql-2                        2/2     Running     0          8m38s   10.244.87.60   liuxinle-optiplex-5060              c0-nor0-mysql-0                         2/2     Running     0          10m     10.244.87.29   liuxinle-optiplex-5060              c0-nor0-mysql-1                         2/2     Running     0          9m29s   10.244.87.4    liuxinle-optiplex-5060              c0-nor0-mysql-2                         2/2     Running     0          8m45s   10.244.87.25   liuxinle-optiplex-5060              c0-par0-mysql-0                         2/2     Running     0          10m     10.244.87.55   liuxinle-optiplex-5060              c0-par0-mysql-1                         2/2     Running     0          9m26s   10.244.87.13   liuxinle-optiplex-5060              c0-par0-mysql-2                         2/2     Running     0          8m42s   10.244.87.21   liuxinle-optiplex-5060              c0-prometheus-server-6697649b76-fkvh9   2/2     Running     0          4m36s   10.244.87.37   liuxinle-optiplex-5060              c0-zookeeper-0                          1/1     Running     1          7m35s   10.244.87.44   liuxinle-optiplex-5060              c0-zookeeper-1                          1/1     Running     0          6m41s   10.244.87.30   liuxinle-optiplex-5060              c0-zookeeper-2                          1/1     Running     0          6m10s   10.244.87.49   liuxinle-optiplex-5060              c0-zookeeper-initzk-7hbfs               0/1     Completed   0          7m35s   10.244.87.17   liuxinle-optiplex-5060              

看到c0-zookeeper-initzk-7hbfs的状态是Completed，其他pod的状态为Running，表示集群搭建成功。

3.2 在GreatDB分布式集群中使用Chaos Mesh做混沌测试

Chaos Mesh在kubernetes环境支持注入的故障类型包括：模拟Pod故障、模拟网络故障、模拟压力场景等，这里我们以模拟Pod故障中的pod-kill为例。

将实验配置写入到文件中 pod-kill.yaml，内容示例如下：

apiVersion: chaos-mesh.org/v1alpha1kind: PodChaos   # 要注入的故障类型metadata:  name: pod-failure-example  namespace: test   # 测试集群pod所在的namespacespec:  action: pod-kill   # 要注入的具体故障类型  mode: all    # 指定实验的运行方式，all（表示选出所有符合条件的 Pod）  duration: '30s'    # 指定实验的持续时间   selector:     labelSelectors:      "app.kubernetes.io/component": "greatdb"    # 指定注入故障目标pod的标签，通过kubectl describe pod c0-greatdb-1 -n test 命令返回结果中Labels后的内容得到

创建故障实验，命令如下：

kubectl create -n test -f pod-kill.yaml

创建完故障实验之后，执行命令 kubectl get pod -n test -o wide 结果如下：

NAME                                    READY   STATUS              RESTARTS   AGE     IP             NODE                     NOMINATED NODE   READINESS GATESc0-auth0-mysql-0                        2/2     Running             0          14m     10.244.87.18   liuxinle-optiplex-5060              c0-auth0-mysql-1                        2/2     Running             0          14m     10.244.87.54   liuxinle-optiplex-5060              c0-auth0-mysql-2                        2/2     Running             0          13m     10.244.87.57   liuxinle-optiplex-5060              c0-greatdb-0                            0/2     ContainerCreating   0          2s               liuxinle-optiplex-5060              c0-greatdb-1                            0/2     ContainerCreating   0          2s               liuxinle-optiplex-5060              c0-glob0-mysql-0                        2/2     Running             0          14m     10.244.87.51   liuxinle-optiplex-5060              c0-glob0-mysql-1                        2/2     Running             0          14m     10.244.87.41   liuxinle-optiplex-5060              c0-glob0-mysql-2                        2/2     Running             0          13m     10.244.87.60   liuxinle-optiplex-5060              c0-nor0-mysql-0                         2/2     Running             0          14m     10.244.87.29   liuxinle-optiplex-5060              c0-nor0-mysql-1                         2/2     Running             0          14m     10.244.87.4    liuxinle-optiplex-5060              c0-nor0-mysql-2                         2/2     Running             0          13m     10.244.87.25   liuxinle-optiplex-5060              c0-par0-mysql-0                         2/2     Running             0          14m     10.244.87.55   liuxinle-optiplex-5060              c0-par0-mysql-1                         2/2     Running             0          14m     10.244.87.13   liuxinle-optiplex-5060              c0-par0-mysql-2                         2/2     Running             0          13m     10.244.87.21   liuxinle-optiplex-5060              c0-prometheus-server-6697649b76-fkvh9   2/2     Running             0          9m24s   10.244.87.37   liuxinle-optiplex-5060              c0-zookeeper-0                          1/1     Running             1          12m     10.244.87.44   liuxinle-optiplex-5060              c0-zookeeper-1                          1/1     Running             0          11m     10.244.87.30   liuxinle-optiplex-5060              c0-zookeeper-2                          1/1     Running             0          10m     10.244.87.49   liuxinle-optiplex-5060              c0-zookeeper-initzk-7hbfs               0/1     Completed           0          12m     10.244.87.17   liuxinle-optiplex-5060              

可以看到有带greatdb名字的pod正在被重启，说明注入故障成功。

4. 在argo中编排测试流程

Argo 是一个开源的容器本地工作流引擎，用于在Kubernetes上完成工作，可以将多步骤工作流建模为一系列任务，完成测试流程编排。

我们使用argo定义一个测试任务，基本的测试流程是固定的，如下所示：

测试流程的step1是部署测试集群，接着开启两个并行任务，step2跑测试用例，模拟业务场景，step3同时使用Chaos Mesh注入故障，step2的测试用例执行结束之后，step4终止故障注入，最后step5清理集群环境。

4.1 用argo编排一个混沌测试工作流（以flashback测试用例为例）

1）修改 cluster-setup.yaml 中的image信息，改成步骤2.2 准备测试需要的镜像中自己传上去的集群部署/清理镜像名和tag

2）修改 testsuite-flashback.yaml 中的image信息，改成步骤2.2 准备测试需要的镜像中自己传上去的测试用例镜像名和tag

3）将集群部署、测试用例和工具模板的yaml文件全部使用 kubectl apply -n argo -f xxx.yaml 命令创建资源 （这些文件定义了一些argo template，方便用户写workflow时候使用）

kubectl apply -n argo -f cluster-setup.yamlkubectl apply -n argo -f testsuite-flashback.yamlkubectl apply -n argo -f tools-template.yaml

4）复制一份workflow模板文件 workflow-template.yaml，将模板文件中注释提示的部分修改为自己的设置即可，然后执行以下命令创建混沌测试工作流：

kubectl apply -n argo -f workflow-template.yaml

以下是一份workflow模板文件：

apiVersion: argoproj.io/v1alpha1kind: Workflowmetadata:  generateName: chaostest-c0-0-  name: chaostest-c0-0  namespace: argospec:  entrypoint: test-entry #测试入口，在这里传入测试参数，填写clustername、namespace、host、greatdb镜像名和tag名等基本信息  serviceAccountName: argo  arguments:    parameters:      - name: clustername        value: c0      - name: namespace        value: test      - name: host        value: 172.16.70.249      - name: port        value: 30901      - name: password        value: Bgview@2020      - name: user        value: root      - name: run-time        value: 10m      - name: greatdb-repository        value: greatdb/greatdb      - name: greatdb-tag        value: tag202110      - name: nemesis        value: kill_mysql_normal_master,kill_mysql_normal_slave,kill_mysql_partition_master,kill_mysql_partition_slave,kill_mysql_auth_master,kill_mysql_auth_slave,kill_mysql_global_master,kill_mysql_global_slave,kill_mysql_master,kill_mysql_slave,net_partition_mysql_normal,net_partition_mysql_partition,net_partition_mysql_auth,net_partition_mysql_global      - name: mysql-partition        value: 1      - name: mysql-global        value: 1      - name: mysql-auth        value: 1      - name: mysql-normal        value: 2  templates:    - name: test-entry      steps:        - - name: setup-greatdb-cluster  # step.1 集群部署. 请指定正确的参数，主要是mysql和zookeeper的镜像名、tag名            templateRef:              name: cluster-setup-template              template: cluster-setup            arguments:              parameters:                - name: namespace                  value: "{{workflow.parameters.namespace}}"                - name: clustername                  value: "{{workflow.parameters.clustername}}"                - name: mysql-image                  value: mysql:5.7.34                - name: mysql-replica                  value: 3                - name: mysql-auth                  value: "{{workflow.parameters.mysql-auth}}"                - name: mysql-normal                  value: "{{workflow.parameters.mysql-normal}}"                - name: mysql-partition                  value: "{{workflow.parameters.mysql-partition}}"                - name: mysql-global                  value: "{{workflow.parameters.mysql-global}}"                - name: enable-monitor                  value: false                - name: zookeeper-repository                  value: zookeeper                - name: zookeeper-tag                  value: 3.5.5                - name: zookeeper-replica                  value: 3                - name: greatdb-repository                  value: "{{workflow.parameters.greatdb-repository}}"                - name: greatdb-tag                  value: "{{workflow.parameters.greatdb-tag}}"                - name: greatdb-replica                  value: 3                - name: greatdb-serviceHost                  value: "{{workflow.parameters.host}}"                - name: greatdb-servicePort                  value: "{{workflow.parameters.port}}"        - - name: run-flashbacktest    # step.2 运行测试用例,请替换为你要运行的测试用例template并指定正确的参数，主要是测试使用的表个数和大小            templateRef:              name: flashback-test-template              template: flashback            arguments:              parameters:                - name: user                  value: "{{workflow.parameters.user}}"                - name: password                  value: "{{workflow.parameters.password}}"                - name: host                  value: "{{workflow.parameters.host}}"                - name: port                  value: "{{workflow.parameters.port}}"                - name: concurrency                  value: 16                - name: size                  value: 10000                - name: tables                  value: 10                - name: run-time                  value: "{{workflow.parameters.run-time}}"                - name: single-statement                  value: true                - name: manage-statement                  value: true          - name: invoke-chaos-for-flashabck-test    # step.3 注入故障，请指定正确的参数，这里run-time和interval分别定义了故障注入的时间和频次，因此省略掉了终止故障注入步骤            templateRef:              name: chaos-rto-template              template: chaos-rto            arguments:              parameters:                - name: user                  value: "{{workflow.parameters.user}}"                - name: host                  value: "{{workflow.parameters.host}}"                - name: password                  value: "{{workflow.parameters.password}}"                - name: port                  value: "{{workflow.parameters.port}}"                - name: k8s-config                  value: /root/.kube/config                - name: namespace                  value: "{{workflow.parameters.namespace}}"                - name: clustername                  value: "{{workflow.parameters.clustername}}"                - name: prometheus                  value: ''                - name: greatdb-job                  value: greatdb-monitor-greatdb                - name: nemesis                  value: "{{workflow.parameters.nemesis}}"                - name: nemesis-duration                  value: 1m                - name: nemesis-mode                  value: default                - name: wait-time                  value: 5m                - name: check-time                  value: 5m                - name: nemesis-scope                  value: 1                - name: nemesis-log                  value: true                - name: enable-monitor                  value: false                - name: run-time                  value: "{{workflow.parameters.run-time}}"                - name: interval                  value: 1m                - name: monitor-log                  value: false                - name: enable-rto                  value: false                - name: rto-qps                  value: 0.1                - name: rto-warm                  value: 5m                - name: rto-time                  value: 1m                - name: log-level                  value: debug        - - name: flashbacktest-output         # 输出测试用例是否通过的结果            templateRef:              name: tools-template              template: output-result            arguments:              parameters:                - name: info                  value: "flashback test pass, with nemesis: {{workflow.parameters.nemesis}}"        - - name: clean-greatdb-cluster           # step.4 清理测试集群，这里的参数和step.1的参数一致            templateRef:              name: cluster-setup-template              template: cluster-setup            arguments:              parameters:                - name: namespace                  value: "{{workflow.parameters.namespace}}"                - name: clustername                  value: "{{workflow.parameters.clustername}}"                - name: mysql-image                  value: mysql:5.7                - name: mysql-replica                  value: 3                - name: mysql-auth                  value: "{{workflow.parameters.mysql-auth}}"                - name: mysql-normal                  value: "{{workflow.parameters.mysql-normal}}"                - name: mysql-partition                  value: "{{workflow.parameters.mysql-partition}}"                - name: mysql-global                  value: "{{workflow.parameters.mysql-global}}"                - name: enable-monitor                  value: false                - name: zookeeper-repository                  value: zookeeper                - name: zookeeper-tag                  value: 3.5.5                - name: zookeeper-replica                  value: 3                - name: greatdb-repository                  value: "{{workflow.parameters.greatdb-repository}}"                - name: greatdb-tag                  value: "{{workflow.parameters.greatdb-tag}}"                - name: greatdb-replica                  value: 3                - name: greatdb-serviceHost                  value: "{{workflow.parameters.host}}"                - name: greatdb-servicePort                  value: "{{workflow.parameters.port}}"                - name: clean                  value: true        - - name: echo-result            templateRef:              name: tools-template              template: echo            arguments:              parameters:                - name: info                  value: "{{item}}"            withItems:              - "{{steps.flashbacktest-output.outputs.parameters.result}}"

Enjoy GreatSQL :)