skywalking docker镜像构建k8s部署

前言碎语

skywalking是个非常不错的apm产品，但是在使用过程中有个非常蛋疼的问题，在基于es的存储情况下，es的数据一有问题，就会导致整个skywalking web ui服务不可用，然后需要agent端一个服务一个服务的停用，然后服务重新部署后好，全部走一遍。这种问题同样也会存在skywalking的版本升级迭代中。而且apm 这种过程数据是允许丢弃的，默认skywalking中关于trace的数据记录只保存了90分钟。故博主准备将skywalking的部署容器化，一键部署升级。下文是整个skywalking 容器化部署的过程。

目标：将skywalking的docker镜像运行在k8s的集群环境中提供服务

docker镜像构建

FROM registry.cn-xx.xx.com/keking/jdk:1.8
ADD apache-skywalking-apm-incubating/  /opt/apache-skywalking-apm-incubating/
RUN ln -sf /usr/share/zoneinfo/Asia/Shanghai  /etc/localtime \
    && echo 'Asia/Shanghai' >/etc/timezone \
    && chmod +x /opt/apache-skywalking-apm-incubating/config/setApplicationEnv.sh \
    && chmod +x /opt/apache-skywalking-apm-incubating/webapp/setWebAppEnv.sh \
    && chmod +x /opt/apache-skywalking-apm-incubating/bin/startup.sh \
    && echo "tail -fn 100 /opt/apache-skywalking-apm-incubating/logs/webapp.log" >> /opt/apache-skywalking-apm-incubating/bin/startup.sh

EXPOSE 8080 10800 11800 12800
CMD /opt/apache-skywalking-apm-incubating/config/setApplicationEnv.sh \
     && sh /opt/apache-skywalking-apm-incubating/webapp/setWebAppEnv.sh \
     && /opt/apache-skywalking-apm-incubating/bin/startup.sh

在编写Dockerfile时需要考虑几个问题：skywalking中哪些配置需要动态配置（运行时设置）？怎么保证进程一直运行（skywalking 的startup.sh和tomcat中 的startup.sh类似）？

application.yml

#cluster:
#  zookeeper:
#    hostPort: localhost:2181
#    sessionTimeout: 100000
naming:
  jetty:
    #OS real network IP(binding required), for agent to find collector cluster
    host: 0.0.0.0
    port: 10800
    contextPath: /
cache:
#  guava:
  caffeine:
remote:
  gRPC:
    # OS real network IP(binding required), for collector nodes communicate with each other in cluster. collectorN --(gRPC) --> collectorM
    host: #real_host
    port: 11800
agent_gRPC:
  gRPC:
    #os real network ip(binding required), for agent to uplink data(trace/metrics) to collector. agent--(grpc)--> collector
    host: #real_host
    port: 11800
    # Set these two setting to open ssl
    #sslCertChainFile: $path
    #sslPrivateKeyFile: $path

    # Set your own token to active auth
    #authentication: xxxxxx
agent_jetty:
  jetty:
    # OS real network IP(binding required), for agent to uplink data(trace/metrics) to collector through HTTP. agent--(HTTP)--> collector
    # SkyWalking native Java/.Net/node.js agents don't use this.
    # Open this for other implementor.
    host: 0.0.0.0
    port: 12800
    contextPath: /
analysis_register:
  default:
analysis_jvm:
  default:
analysis_segment_parser:
  default:
    bufferFilePath: ../buffer/
    bufferOffsetMaxFileSize: 10M
    bufferSegmentMaxFileSize: 500M
    bufferFileCleanWhenRestart: true
ui:
  jetty:
    # Stay in `localhost` if UI starts up in default mode.
    # Change it to OS real network IP(binding required), if deploy collector in different machine.
    host: 0.0.0.0
    port: 12800
    contextPath: /
storage:
  elasticsearch:
    clusterName: #elasticsearch_clusterName
    clusterTransportSniffer: true
    clusterNodes: #elasticsearch_clusterNodes
    indexShardsNumber: 2
    indexReplicasNumber: 0
    highPerformanceMode: true
    # Batch process setting, refer to https://www.elastic.co/guide/en/elasticsearch/client/java-api/5.5/java-docs-bulk-processor.html
    bulkActions: 2000 # Execute the bulk every 2000 requests
    bulkSize: 20 # flush the bulk every 20mb
    flushInterval: 10 # flush the bulk every 10 seconds whatever the number of requests
    concurrentRequests: 2 # the number of concurrent requests
    # Set a timeout on metric data. After the timeout has expired, the metric data will automatically be deleted.
    traceDataTTL: 2880 # Unit is minute
    minuteMetricDataTTL: 90 # Unit is minute
    hourMetricDataTTL: 36 # Unit is hour
    dayMetricDataTTL: 45 # Unit is day
    monthMetricDataTTL: 18 # Unit is month
#storage:
#  h2:
#    url: jdbc:h2:~/memorydb
#    userName: sa
configuration:
  default:
    #namespace: xxxxx
    # alarm threshold
    applicationApdexThreshold: 2000
    serviceErrorRateThreshold: 10.00
    serviceAverageResponseTimeThreshold: 2000
    instanceErrorRateThreshold: 10.00
    instanceAverageResponseTimeThreshold: 2000
    applicationErrorRateThreshold: 10.00
    applicationAverageResponseTimeThreshold: 2000
    # thermodynamic
    thermodynamicResponseTimeStep: 50
    thermodynamicCountOfResponseTimeSteps: 40
    # max collection's size of worker cache collection, setting it smaller when collector OutOfMemory crashed.
    workerCacheMaxSize: 10000
#receiver_zipkin:
#  default:
#    host: localhost
#    port: 9411
#    contextPath: /

webapp.yml

server:
  port: 8080

collector:
  path: /graphql
  ribbon:
    ReadTimeout: 10000
    listOfServers: #real_host:10800

security:
  user:
    admin:
      password: #skywalking_password

动态配置：密码，grpc等需要绑定主机的ip都需要运行时设置，这里我们在启动skywalking的startup.sh只之前，先执行了两个设置配置的脚本，通过k8s在运行时设置的环境变量来替换需要动态配置的参数

setApplicationEnv.sh

#!/usr/bin/env sh
sed -i "s/#elasticsearch_clusterNodes/${elasticsearch_clusterNodes}/g" /opt/apache-skywalking-apm-incubating/config/application.yml
sed -i "s/#elasticsearch_clusterName/${elasticsearch_clusterName}/g" /opt/apache-skywalking-apm-incubating/config/application.yml
sed -i "s/#real_host/${real_host}/g" /opt/apache-skywalking-apm-incubating/config/application.yml

setWebAppEnv.sh

#!/usr/bin/env sh
sed -i "s/#skywalking_password/${skywalking_password}/g" /opt/apache-skywalking-apm-incubating/webapp/webapp.yml
sed -i "s/#real_host/${real_host}/g" /opt/apache-skywalking-apm-incubating/webapp/webapp.yml

保持进程存在：通过在skywalking 启动脚本startup.sh末尾追加"tail -fn 100 /opt/apache-skywalking-apm-incubating/logs/webapp.log"，来让进程保持运行，并不断输出webapp.log的日志

Kubernetes中部署

apiVersion: extensions/v1beta1
kind: Deployment
metadata:
  name: skywalking
  namespace: uat
spec:
  replicas: 1
  selector:
    matchLabels:
      app: skywalking
  template:
    metadata:
      labels:
        app: skywalking
    spec:
      imagePullSecrets:
      - name: registry-pull-secret
      nodeSelector:
         apm: skywalking
      containers:
      - name: skywalking
        image: registry.cn-xx.xx.com/keking/kk-skywalking:5.2
        imagePullPolicy: Always
        env:
        - name: elasticsearch_clusterName
          value: elasticsearch
        - name: elasticsearch_clusterNodes
          value: 172.16.16.129:31300
        - name: skywalking_password
          value: xxx
        - name: real_host
          valueFrom:
            fieldRef:
              fieldPath: status.podIP
        resources:
          limits:
            cpu: 1000m
            memory: 4Gi
          requests:
            cpu: 700m
            memory: 2Gi

---
apiVersion: v1
kind: Service
metadata:
  name: skywalking
  namespace: uat
  labels:
    app: skywalking
spec:
  selector:
    app: skywalking
  ports:
  - name: web-a
    port: 8080
    targetPort: 8080
    nodePort: 31180
  - name: web-b
    port: 10800
    targetPort: 10800
    nodePort: 31181
  - name: web-c
    port: 11800
    targetPort: 11800
    nodePort: 31182
  - name: web-d
    port: 12800
    targetPort: 12800
    nodePort: 31183
  type: NodePort

Kubernetes部署脚本中唯一需要注意的就是env中关于pod ip的获取，skywalking中有几个ip必须绑定容器的真实ip，这个地方可以通过环境变量设置到容器里面去

文末结语

整个skywalking容器化部署从测试到可用大概耗时1天，其中花了个多小时整了下谭兄的skywalking-docker镜像（https://hub.docker.com/r/wutang/skywalking-docker/），发现有个脚本有权限问题（谭兄反馈已解决，还没来的及测试），以及有几个地方自己不是很好控制，便build了自己的docker镜像，其中最大的问题还是解决集群中网络通讯的问题，一开始我把skywalking中的服务ip都设置为0.0.0.0，然后通过集群的nodePort映射出来,这个时候的agent通过集群ip+31181是可以访问到naming服务的，然后通过naming服务获取到的collector gRPC服务缺变成了0.0.0.0:11800, 这个地址agent肯定访问不到collector的，后面通过绑定pod ip的方式解决了这个问题。