一、背景
基于K8s集群的私有化交付方案中，日志收集采用了ilogtail+logstash+kafka+es方案，其中ilogtail负责日志收集，logstash负责对数据转换，kafka负责对日志传递中的消峰进而减少es的写入压力，es用来保存日志数据。在私有化交付中本方案中涉及的中间件一般需要单独部署，但是在京东内网环境的部署考虑到kafka和es的高可用，则不推荐采用单独部署的方案。
二、新方案实践
1.新方案简介
在京东内网环境部署K8S并收集日志， kafka+es的替代方案考虑使用JMQ+JES，由于JMQ的底层是基于kafaka、JES的底层基于ES，所以该替换方案理论上是可行的
2.主要架构
数据流向大致如下
应用日志 -> ilogtail -> JMQ -> logstash -> JES
3.如何使用
核心改造点汇总
-
ilogtail nameservers配置
增加解析JMQ域名的nameserver（京东云主机上无法直接解析.local域名）
spec:spec:dnsPolicy: "None"dnsConfig:nameservers:- x.x.x.x # 可以解析jmq域名的nameserver
-
ilogtail flushers配置
调整发送到JMQ到配置
apiVersion: v1
kind: ConfigMap
metadata:name: ilogtail-user-cmnamespace: elastic-system
data:app_stdout.yaml: |flushers:- Type: flusher_stdoutOnlyStdout: true- Type: flusher_kafka_v2Brokers:- nameserver.jmq.jd.local:80 # jmq元数据地址Topic: ai-middle-k8s-log-prod # jmq topic ClientID: ai4middle4log # Kafka的用户ID（识别客户端并设置其唯一性），对应jmq的Group名称，重要‼️ （https://ilogtail.gitbook.io/ilogtail-docs/plugins/input/service-kafka#cai-ji-pei-zhi-v2）
-
logstash kafka&es配置
apiVersion: v1
kind: ConfigMap
metadata:name: logstash-confignamespace: elastic-systemlabels:elastic-app: logstash
data:logstash.conf: |-input {kafka {bootstrap_servers => ["nameserver.jmq.jd.local:80"] #jmq的元数据地址group_id => "ai4middle4log" # jmq的Group的名称client_id => "ai4middle4log" # jmq的Group的名称，即jmq的省略了kafka中的client_id概念，用Group名称代替consumer_threads => 2decorate_events => truetopics => ["ai-middle-k8s-log-prod"] # jmp的topicauto_offset_reset => "latest"codec => json { charset => "UTF-8" }}}output {elasticsearch {hosts => ["http://x.x.x.x:40000","http://x.x.x.x:40000","http://x.x.x.x:40000"] # es地址index => "%{[@metadata][kafka][topic]}-%{+YYYY-MM-dd}" # 索引规则user => "XXXXXX" #jes的用户名password => "xxxxx" #jes的密码ssl => "false"ssl_certificate_verification => "false"}}ilogtail 的配置如下
# ilogtail-daemonset.yaml
apiVersion: apps/v1
kind: DaemonSet
metadata:name: ilogtail-dsnamespace: elastic-systemlabels:k8s-app: logtail-ds
spec:selector:matchLabels:k8s-app: logtail-dstemplate:metadata:labels:k8s-app: logtail-dsspec:dnsPolicy: "None"dnsConfig:nameservers:- x.x.x.x # （京东云主机上）可以解析jmq域名的nameservertolerations:- operator: Exists # deploy on all nodescontainers:- name: logtailenv:- name: ALIYUN_LOG_ENV_TAGS # add log tags from envvalue: _node_name_|_node_ip_- name: _node_name_valueFrom:fieldRef:apiVersion: v1fieldPath: spec.nodeName- name: _node_ip_valueFrom:fieldRef:apiVersion: v1fieldPath: status.hostIP- name: cpu_usage_limit # iLogtail's self monitor cpu limitvalue: "1"- name: mem_usage_limit # iLogtail's self monitor mem limitvalue: "512"image: dockerhub.ai.jd.local/ai-middleware/ilogtail-community-edition/ilogtail:1.3.1imagePullPolicy: IfNotPresentresources:limits:cpu: 1000mmemory: 1Girequests:cpu: 400mmemory: 384MivolumeMounts:- mountPath: /var/run # for container runtime socketname: run- mountPath: /logtail_host # for log access on the nodemountPropagation: HostToContainername: rootreadOnly: true- mountPath: /usr/local/ilogtail/checkpoint # for checkpoint between container restartname: checkpoint- mountPath: /usr/local/ilogtail/user_yaml_config.d # mount config dirname: user-configreadOnly: true- mountPath: /usr/local/ilogtail/apsara_log_conf.jsonname: apsara-log-configreadOnly: truesubPath: apsara_log_conf.jsondnsPolicy: ClusterFirsthostNetwork: truevolumes:- hostPath:path: /var/runtype: Directoryname: run- hostPath:path: /type: Directoryname: root- hostPath:path: /etc/ilogtail-ilogtail-ds/checkpointtype: DirectoryOrCreatename: checkpoint- configMap:defaultMode: 420name: ilogtail-user-cmname: user-config- configMap:defaultMode: 420name: ilogtail-apsara-log-config-cmname: apsara-log-config
# ilogtail-user-configmap.yaml
apiVersion: v1
kind: ConfigMap
metadata:name: ilogtail-user-cmnamespace: elastic-system
data:app_stdout.yaml: |enable: trueinputs:- Type: service_docker_stdoutStderr: trueStdout: trueK8sNamespaceRegex: ai-trainExternalK8sLabelTag:platform/resource-name: k8s_label_resource-nameplatform/task-identify: k8s_label_task-identifytask-id: k8s_label_task-idrun-id: k8s_label_run-idrequest-id: k8s_label_request-idprocessors:- Type: processor_renameSourceKeys:- k8s_label_resource-name- k8s_label_task-identify- k8s_label_task-id- k8s_label_run-id- k8s_label_request-id- _namespace_- _image_name_- _pod_uid_- _pod_name_- _container_name_- _container_ip_- __path__- _source_DestKeys:- resource_name- task_identify- task_id- run_id- request_id- namespace- image_name- pod_uid- pod_name- container_name- container_ip- path- sourceflushers:- Type: flusher_stdoutOnlyStdout: true- Type: flusher_kafka_v2Brokers:- nameserver.jmq.jd.local:80 # jmq元数据地址Topic: ai-middle-k8s-log-prod # jmq topic ClientID: ai4middle4log # Kafka的用户ID（识别客户端并设置其唯一性），对应jmq的Group名称，重要‼️ （https://ilogtail.gitbook.io/ilogtail-docs/plugins/input/service-kafka#cai-ji-pei-zhi-v2）app_file_log.yaml: |enable: trueinputs:- Type: file_logLogPath: /export/Logs/ai-dt-algorithm-toolsFilePattern: "*.log"ContainerInfo:K8sNamespaceRegex: ai-trainExternalK8sLabelTag:platform/resource-name: k8s_label_resource-nameplatform/task-identify: k8s_label_task-identifytask-id: k8s_label_task-idrun-id: k8s_label_run-idrequest-id: k8s_label_request-idprocessors:- Type: processor_add_fieldsFields:source: file- Type: processor_renameSourceKeys:- __tag__:k8s_label_resource-name- __tag__:k8s_label_task-identify- __tag__:k8s_label_task-id- __tag__:k8s_label_run-id- __tag__:k8s_label_request-id- __tag__:_namespace_- __tag__:_image_name_- __tag__:_pod_uid_- __tag__:_pod_name_- __tag__:_container_name_- __tag__:_container_ip_- __tag__:__path__DestKeys:- resource_name- task_identify- task_id- run_id- request_id- namespace- image_name- pod_uid- pod_name- container_name- container_ip- pathflushers:- Type: flusher_stdoutOnlyStdout: true- Type: flusher_kafka_v2Brokers:- nameserver.jmq.jd.local:80Topic: ai-middle-k8s-log-prodClientID: ai4middle4log
logstash 的配置如下
# logstash-configmap.yaml
---
apiVersion: v1
kind: ConfigMap
metadata:name: logstash-confignamespace: elastic-systemlabels:elastic-app: logstash
data:logstash.conf: |-input {kafka {bootstrap_servers => ["nameserver.jmq.jd.local:80"] #jmq的元数据地址#group_id => "services"group_id => "ai4middle4log" # jmq的Group的名称client_id => "ai4middle4log" # jmq的Group的名称，即jmq的省略了kafka中的client_id概念，用Group名称代替consumer_threads => 2decorate_events => true#topics_pattern => ".*"topics => ["ai-middle-k8s-log-prod"] # jmp的topicauto_offset_reset => "latest"codec => json { charset => "UTF-8" }}}filter {ruby {code => "event.set('index_date', event.get('@timestamp').time.localtime + 8*60*60)"}ruby {code => "event.set('message',event.get('contents'))"}#ruby {# code => "event.set('@timestamp',event.get('time').time.localtime)"#}mutate {remove_field => ["contents"]convert => ["index_date", "string"]#convert => ["@timestamp", "string"]gsub => ["index_date", "T.*Z",""]#gsub => ["@timestamp", "T.*Z",""]}}output {elasticsearch {#hosts => ["https://ai-middle-cluster-es-http:9200"]hosts => ["http://x.x.x.x:40000","http://x.x.x.x:40000","http://x.x.x.x:40000"] # es地址index => "%{[@metadata][kafka][topic]}-%{+YYYY-MM-dd}" # 索引规则user => "XXXXXX" #jes的用户名password => "xxxxx" #jes的密码ssl => "false"ssl_certificate_verification => "false"#cacert => "/usr/share/logstash/cert/ca_logstash.cer"}stdout {codec => rubydebug}}4.核心价值
在私有化部署的基础上通过简单改造实现了与京东内部中间件的完美融合，使得系统在高可用性上适应性更强、可用范围更广。