赞
踩
Kafka metrics 所有的 metric 都可以通过 JMX 获取,暴露kafka metrics 支持两种方式
1.在 Kafka Broker 外部, 作为一个独立进程, 通过 JMX 的 RMI 接口读取数据.
这种方式的好处是有任何调整不需要重启 Kafka Broker 进程,
缺点是多维护了一个独立的进程。
2.在 Kafka Broker 进程内部读取 JMX 数据, 这样解析数据的逻辑就在 Kafka Broker
进程内部, 如果有任何调整, 需要重启 Broker。
选择暴露 kafka-metric 方式
第一种需要外部多维护一个程序,而且还要考虑之后各种版本升级,实现起来比较繁琐,还好的是github上有许多优秀的开源kafka_exporter
下载过来直接启动就好了。简单介绍下
git项目地址:https://github.com/danielqsj/kafka_exporter
wget
https://github.com/danielqsj/kafka_exporter/releases/download/v1.2.0/kafka_exporter-1.2.0.linux-amd64.tar.gz
解压: tar -zxvf kafka_exporter-1.2.0.linux-amd64.tar.gz
切到相应目录: cd kafka_exporter-1.2.0.linux-amd64
./kafka_exporter --kafka.server=kafkaIP或者域名:9092 &
(只需填写kafka集群的一个ip即可)
[root@prometheus prometheus]# yum install screen -y
[root@prometheus prometheus]# screen
[root@prometheus prometheus]# ./kafka_exporter --kafka.server=kafka03:9092
#按Ctrl+a+d后进入后台运行模式
对应的服务端口为9308
2.下载prometheus
wget
https://github.com/prometheus/prometheus/releases/download/v2.15.1/prometheus-2.15.1.linux-amd64.tar.gz
解压
tar -zxvf prometheus-2.15.1.linux-amd64.tar.gz
prometheus.yml为promethues配置文件,可以先启动验证服务可用性
cd ./prometheus-2.15.1.linux-amd64
prometheus.yml
这个文件是对应的配置文件,在未添加kafka_exporter之前可以先启动查看下服务是否正常
添加监控配置
重启prometheus
./prometheus --config.file=prometheus.yml --web.listen-address=:9090
–storage.tsdb.path=/data/prometheus &
wget https://dl.grafana.com/oss/release/grafana-7.1.5.linux-amd64.tar.gz
tar -zxvf grafana-7.1.5.linux-amd64.tar.gz
该grafana-server二进制需要工作目录是根安装目录,其中二进制文件以及public文件夹的位置。
通过运行以下命令启动Grafana:
./bin/grafana-server web
打开grafana的web页面 ip:3000,添加promethues数据源
导入监控图标,对于grafana的监控,官方有监控图标,不需要自己搞
官方的监控界面
生产环境的监控环境配置及对应查询语句
Grafana画图也有许多优秀的开源dashboard
第二种是读取 JMX 的数据. Prometheus 官方的组件 jmx_exporter 把两种实现都提供了:
jmx_prometheus_httpserver 通过独立进程读取 JMX 的数据
jmx_prometheus_javaagent 使用 Java Agent 方式, 尽量无侵入(仅需在 java
命令行中使用 -javaagent 参数)的启动 in-process library, 读取 JMX 数据.
Prometheus 采用了 PULL 方式, Prometheus 主动抓取 metrics 数据,
而不是靠客户端主动 PUSH 数据, 因此 jmx_prometheus 都是通过暴露 HTTP
端口的方式暴露 metrics 数据, 方便 Prometheus 抓取数据.
选择方案2
我们这里选择第二种jmx_prometheus_javaagent 方式收集kafka指标
部署流程:
下载jmx_prometheus_javaagent和kafka.yml
wget
https://raw.githubusercontent.com/prometheus/jmx_exporter/master/example_configs/kafka-0-8-2.yml
添加几行代码:
export JMX_PORT=“9999”
export
KAFKA_OPTS="-javaagent:/path/jmx_prometheus_javaagent-0.6.jar=9991:/path/kafka-0-8-2.yml"
然后重启kafka。
访问 http://localhost:9991/metrics 可以看到各种指标了。
部分监控指标解释,不一定准确,请参考。还有参考 monitoring
kafafka 有详细的指标信息
指标 | 解释 |
---|---|
kafka_server_replicafetchermanager_maxlag | Max |
kafka_server_replicamanager_isrexpands_total | ISR expansion rate 扩大率(ISR是in-sync replicas的简写) |
kafka_server_replicamanager_isrshrinks_total | ISR shrink rate 收缩率 |
kafka_server_replicamanager_underreplicatedpartitions | # of under replicated partitions (|ISR| < |all replicas|) |
kafka_network_requestmetrics_responsesendtimems | Time to send the response Produce |
kafka_network_socketserver_networkprocessoravgidlepercent | The average fraction of time the network processors are idle |
kafka_network_requestmetrics_responsesendtimems | |
kafka_network_requestmetrics_requestqueuetimems | Time the request waiting in the request queue Produce |
kafka_network_requestmetrics_remotetimems | Time the request waits for the follower Produce |
kafka_network_requestmetrics_localtimems | Time the request being processed at the leader Produce |
kafka_log_logflushstats_logflushrateandtimems_count | Log flush latency |
kafka_server_replicafetchermanager_minfetchrate | Max lag in messages btw follower and leader replicas > 4000 |
kafka_controller_controllerstats_uncleanleaderelectionspersec | Unclean leader election has occurred last 15m |
kafka_server_replicamanager_underreplicatedpartitions | Under replicated partitions |
kafka_controller_kafkacontroller_activecontrollercount | 活跃的 Controller 的数量 |
kafka_controller_controllerstats_uncleanleaderelectionspersec | 争议的 leader 选举次数 |
kafka_controller_controllerstats_controlledshutdownrateandtimems | 将ISR中处于关闭状态的副本从集合中去除掉,返回一个新的ISR集合,然后选取第一个副本作为leader,然后令当前AR作为接收LeaderAndIsr请求的副本。 |
kafka_controller_kafkacontroller_offlinepartitionscount | 从活着的ISR中选择一个broker作为leader,如果ISR中没有活着的副本,则从assignedReplicas中选择一个副本作为leader,leader选举成功后注册到Zookeeper中,并更新所有的缓存。 |
broker指标 | |
kafka_server_brokertopicmetrics_messagesin_total | 所有topic消息(进出)流量 消息写入总量 |
kafka_server_brokertopicmetrics_bytesrejected_total | 扔掉的流量 |
kafka_server_brokertopicmetrics_failedfetchrequests_total | 当前机器fetch请求失败的数量 |
kafka_server_brokertopicmetrics_bytesout_total | 输出的流量 |
kafka_server_brokertopicmetrics_bytesin_total | 输入的流量 |
kafka_server_brokertopicmetrics_failedproducerequests_total | 当前机器produce请求失败的数量 |
kafka_server_replicamanager_partitioncount | 该broker上的partition的数量 |
kafka_server_replicamanager_leadercount | Leader的replica的数量 |
kafka_network_requestmetrics_totaltimems{FetchConsumer\FetchFollower\Produce} | 一个请求FetchConsumer\FetchFollower\Produce耗费的所有时间 |
kafka.server:type=ReplicaManager,name=UnderReplicatedPartitions **
含义: 正在复制的 Partition 的数量.
建议报警阈值: > 0 就建议报警. 但如果 Kafka 集群正在 reassign partition 时,
这个值也会 >0
kafka.controller:type=KafkaController,name=OfflinePartitionsCount
含义: 没有 Leader 的 Partition 的数量. 处于这个状态的 Partition 是不可读也不可写
建议报警阈值: >0 一旦出现就报警.
kafka.controller:type=KafkaController,name=ActiveControllerCount
含义: 活跃的 Controller 的数量.
建议报警阈值: != 0 就赶紧报警
kafka.server:type=ReplicaManager,name=PartitionCount
含义: 集群中 Partition 的总数
建议报警阈值: 感觉这个报警不可控.
kafka_controller_controllerstats_leaderelectionrateandtimems
含义: Leader election rate 领导人选举率
UncleanLeaderElectionsPerSec
含义: Unclean leader election rate 争议的 leader 选举次数
描述:所有的topic的消息速率(消息数/秒)
Mbean名:“kafka.server”:name=“AllTopicsMessagesInPerSec”,type=“BrokerTopicMetrics”
正常的值:
描述:所有的topic的流入数据速率(字节/秒)
Mbean名:“kafka.server”:name=“AllTopicsBytesInPerSec”,type=“BrokerTopicMetrics”
正常的值:
描述:producer或Fetch-consumer或Fetch-follower的请求速率(请求次数/秒)
Mbean名:“kafka.network”:name="{Produce|Fetch-consumer|Fetch-follower}-RequestsPerSec",type=“RequestMetrics”
正常的值:
描述:所有的topic的流出数据速率(字节/秒)
Mbean名:
“kafka.server”:name=“AllTopicsBytesOutPerSec”,type=“BrokerTopicMetrics”
正常的值:
描述:刷日志的速率和耗时
Mbean名: “kafka.log”:name=“LogFlushRateAndTimeMs”,type=“LogFlushStats”
正常的值:
描述:正在做复制的partition的数量(|ISR| < |all replicas|)
Mbean名:“kafka.server”:name=“UnderReplicatedPartitions”,type=“ReplicaManager”
正常的值:0
描述:当前的broker是否为controller
Mbean名:“kafka.controller”:name=“ActiveControllerCount”,type=“KafkaController”
正常的值:在集群中只有一个broker的这个值为1
描述:选举leader的速率
Mbean名:“kafka.controller”:name=“LeaderElectionRateAndTimeMs”,type=“ControllerStats”
正常的值:如果有broker挂了,此值非0
描述:Unclean的leader选举速率
Mbean名:“kafka.controller”:name=“UncleanLeaderElectionsPerSec”,type=“ControllerStats”
正常的值:0
描述:该broker上的partition的数量
Mbean名: “kafka.server”:name=“PartitionCount”,type=“ReplicaManager”
正常的值:应在各个broker中平均分布
描述:Leader的replica的数量
Mbean名: “kafka.server”:name=“LeaderCount”,type=“ReplicaManager”
正常的值:应在各个broker中平均分布
描述:ISR的收缩(shrink)速率
Mbean名:“kafka.server”:name=“ISRShrinksPerSec”,type=“ReplicaManager”
正常的值:如果一个broker挂掉了,一些partition的ISR会收缩。当那个broker重新起来时,一旦它的replica完全跟上,ISR会扩大(expand)。除此之外,正常情况下,此值和下面的扩大速率都是0。
描述:ISR的扩大(expansion)速率
Mbean名: “kafka.server”:name=“ISRExpandsPerSec”,type=“ReplicaManager”
正常的值:参见ISR的收缩(shrink)速率
描述:follower落后leader replica的最大的消息数量
Mbean名:“kafka.server”:name="([-.\w]+)-MaxLag",type=“ReplicaFetcherManager”
正常的值:小于replica.lag.max.messages
描述:每个follower replica落后的消息速率
Mbean名:“kafka.server”:name="([-.\w]+)-ConsumerLag",type=“FetcherLagMetrics”
正常的值:小于replica.lag.max.messages
描述:等待producer purgatory的请求数
Mbean名:“kafka.server”:name=“PurgatorySize”,type=“ProducerRequestPurgatory”
正常的值:如果ack=-1,应为非0值
描述:等待fetch purgatory的请求数
Mbean名:“kafka.server”:name=“PurgatorySize”,type=“FetchRequestPurgatory”
正常的值:依赖于consumer的fetch.wait.max.ms的设置
描述:一个请求(producer,Fetch-Consumer,Fetch-Follower)耗费的所有时间
Mbean名:“kafka.network”:name="{Produce|Fetch-Consumer|Fetch-Follower}-TotalTimeMs",type=“RequestMetrics”
正常的值:包括了queue, local, remote和response send time
描述:请求(producer,Fetch-Consumer,Fetch-Follower)在请求队列中的等待时间
Mbean名:“kafka.network”:name="{Produce|Fetch-Consumer|Fetch-Follower}-QueueTimeMs",type=“RequestMetrics”
正常的值:
描述:请求(producer,Fetch-Consumer,Fetch-Follower)在leader处理请求花的时间
Mbean名:“kafka.network”:name="{Produce|Fetch-Consumer|Fetch-Follower}-LocalTimeMs",type=“RequestMetrics”
正常的值:
描述:请求(producer,Fetch-Consumer,Fetch-Follower)等待follower花费的时间
Mbean名:“kafka.network”:name="{Produce|Fetch-Consumer|Fetch-Follower}-RemoteTimeMs",type=“RequestMetrics”
正常的值:producer的ack=-1时,非0才正常
描述:发送响应花费的时间
Mbean名:“kafka.network”:name="{Produce|Fetch-Consumer|Fetch-Follower}-ResponseSendTimeMs",type=“RequestMetrics”
正常的值:
描述:consumer落后producer的消息数量
Mbean名:“kafka.consumer”:name="([-.\w]+)-MaxLag",type=“ConsumerFetcherManager”
正常的值:
建议对GC耗时和其他参数和诸如系统CPU,I/O时间等等进行监控。在client端,建议对"消息数量/字节数"的速率(全局的和对于每一个topic),请求的"速率/大小/耗时"进行监控。还有consumer端,所有partition的最大的落后情况和最小的fetch请求的速率。consumer为了能跟上,最大落后数量需要少于一个threshold并且最小fetch速率需要大于0.
Grafana画图 上传 下面 json文件
{
“__inputs”: [
{
“name”: “DS_PROMETHEUS”,
“label”: “Prometheus”,
“description”: “”,
“type”: “datasource”,
“pluginId”: “prometheus”,
“pluginName”: “Prometheus”
}
],
“__requires”: [
{
“type”: “grafana”,
“id”: “grafana”,
“name”: “Grafana”,
“version”: “5.0.4”
},
{
“type”: “panel”,
“id”: “graph”,
“name”: “Graph”,
“version”: “5.0.0”
},
{
“type”: “datasource”,
“id”: “prometheus”,
“name”: “Prometheus”,
“version”: “5.0.0”
},
{
“type”: “panel”,
“id”: “singlestat”,
“name”: “Singlestat”,
“version”: “5.0.0”
}
],
“annotations”: {
“list”: [
{
“builtIn”: 1,
“datasource”: “-- Grafana --”,
“enable”: true,
“hide”: true,
“iconColor”: “rgba(0, 211, 255, 1)”,
“name”: “Annotations & Alerts”,
“type”: “dashboard”
}
]
},
“description”: “Example of an Kafka Dashboard for DC/OS 1.12”,
“editable”: true,
“gnetId”: 9018,
“graphTooltip”: 0,
“id”: null,
“links”: [],
“panels”: [
{
“aliasColors”: {},
“bars”: false,
“dashLength”: 10,
“dashes”: false,
“datasource”: “${DS_PROMETHEUS}”,
“fill”: 1,
“gridPos”: {
“h”: 10,
“w”: 12,
“x”: 0,
“y”: 0
},
“id”: 114,
“legend”: {
“avg”: false,
“current”: false,
“max”: false,
“min”: false,
“show”: true,
“total”: false,
“values”: false
},
“lines”: true,
“linewidth”: 1,
“links”: [],
“nullPointMode”: “null”,
“percentage”: false,
“pointradius”: 5,
“points”: false,
“renderer”: “flot”,
“seriesOverrides”: [],
“spaceLength”: 10,
“stack”: false,
“steppedLine”: false,
“targets”: [
{
“expr”: “rate(process_cpu_seconds_total{job=\“kafka\”}[1m])”,
“format”: “time_series”,
“intervalFactor”: 1,
“legendFormat”: “{ {instance}}”,
“refId”: “A”
}
],
“thresholds”: [],
“timeFrom”: null,
“timeShift”: null,
“title”: “CPU Usage”,
“tooltip”: {
“shared”: true,
“sort”: 0,
“value_type”: “individual”
},
“type”: “graph”,
“xaxis”: {
“buckets”: null,
“mode”: “time”,
“name”: null,
“show”: true,
“values”: []
},
“yaxes”: [
{
“format”: “short”,
“label”: null,
“logBase”: 1,
“max”: null,
“min”: null,
“show”: true
},
{
“format”: “short”,
“label”: null,
“logBase”: 1,
“max”: null,
“min”: null,
“show”: true
}
]
},
{
“aliasColors”: {},
“bars”: false,
“dashLength”: 10,
“dashes”: false,
“datasource”: “${DS_PROMETHEUS}”,
“fill”: 1,
“gridPos”: {
“h”: 10,
“w”: 12,
“x”: 12,
“y”: 0
},
“id”: 118,
“legend”: {
“avg”: false,
“current”: false,
“max”: false,
“min”: false,
“show”: true,
“total”: false,
“values”: false
},
“lines”: true,
“linewidth”: 1,
“links”: [],
“nullPointMode”: “null”,
“percentage”: false,
“pointradius”: 5,
“points”: false,
“renderer”: “flot”,
“seriesOverrides”: [],
“spaceLength”: 10,
“stack”: false,
“steppedLine”: false,
“targets”: [
{
“expr”: “sum
without(gc)(rate(jvm_gc_collection_seconds_sum{job=\“kafka\”}[5m]))”,
“format”: “time_series”,
“intervalFactor”: 1,
“legendFormat”: “{ {instance}}”,
“refId”: “A”
}
],
“thresholds”: [],
“timeFrom”: null,
“timeShift”: null,
“title”: “Time spent in GC”,
“tooltip”: {
“shared”: true,
“sort”: 0,
“value_type”: “individual”
},
“type”: “graph”,
“xaxis”: {
“buckets”: null,
“mode”: “time”,
“name”: null,
“show”: true,
“values”: []
},
“yaxes”: [
{
“format”: “short”,
“label”: null,
“logBase”: 1,
“max”: null,
“min”: null,
“show”: true
},
{
“format”: “short”,
“label”: null,
“logBase”: 1,
“max”: null,
“min”: null,
“show”: true
}
]
},
{
“aliasColors”: {},
“bars”: false,
“dashLength”: 10,
“dashes”: false,
“datasource”: “${DS_PROMETHEUS}”,
“fill”: 1,
“gridPos”: {
“h”: 9,
“w”: 12,
“x”: 0,
“y”: 10
},
“id”: 116,
“legend”: {
“avg”: false,
“current”: false,
“max”: false,
“min”: false,
“show”: true,
“total”: false,
“values”: false
},
“lines”: true,
“linewidth”: 1,
“links”: [],
“nullPointMode”: “null”,
“percentage”: false,
“pointradius”: 5,
“points”: false,
“renderer”: “flot”,
“seriesOverrides”: [],
“spaceLength”: 10,
“stack”: false,
“steppedLine”: false,
“targets”: [
{
“expr”: “sum without(area)(jvm_memory_bytes_used{job=\“kafka\”})”,
“format”: “time_series”,
“intervalFactor”: 1,
“legendFormat”: “{ {instance}}”,
“refId”: “A”
}
],
“thresholds”: [],
“timeFrom”: null,
“timeShift”: null,
“title”: “JVM Memory Used”,
“tooltip”: {
“shared”: true,
“sort”: 0,
“value_type”: “individual”
},
“type”: “graph”,
“xaxis”: {
“buckets”: null,
“mode”: “time”,
“name”: null,
“show”: true,
“values”: []
},
“yaxes”: [
{
“format”: “short”,
“label”: null,
“logBase”: 1,
“max”: null,
“min”: null,
“show”: true
},
{
“format”: “short”,
“label”: null,
“logBase”: 1,
“max”: null,
“min”: null,
“show”: true
}
]
},
{
“aliasColors”: {},
“bars”: false,
“dashLength”: 10,
“dashes”: false,
“datasource”: “${DS_PROMETHEUS}”,
“fill”: 1,
“gridPos”: {
“h”: 9,
“w”: 12,
“x”: 12,
“y”: 10
},
“id”: 42,
“legend”: {
“avg”: false,
“current”: false,
“max”: false,
“min”: false,
“show”: true,
“total”: false,
“values”: false
},
“lines”: true,
“linewidth”: 1,
“links”: [],
“nullPointMode”: “null”,
“percentage”: false,
“pointradius”: 5,
“points”: false,
“renderer”: “flot”,
“seriesOverrides”: [],
“spaceLength”: 10,
“stack”: false,
“steppedLine”: false,
“targets”: [
{
“expr”: “go_memstats_heap_inuse_bytes”,
“format”: “time_series”,
“intervalFactor”: 1,
“legendFormat”: “{ {instance}}”,
“refId”: “A”
}
],
“thresholds”: [],
“timeFrom”: null,
“timeShift”: null,
“title”: “Heap Memory in Use”,
“tooltip”: {
“shared”: true,
“sort”: 0,
“value_type”: “individual”
},
“type”: “graph”,
“xaxis”: {
“buckets”: null,
“mode”: “time”,
“name”: null,
“show”: true,
“values”: []
},
“yaxes”: [
{
“format”: “decbytes”,
“label”: null,
“logBase”: 1,
“max”: null,
“min”: null,
“show”: true
},
{
“format”: “short”,
“label”: null,
“logBase”: 1,
“max”: null,
“min”: null,
“show”: true
}
],
“yaxis”: {
“align”: false,
“alignLevel”: null
}
},
{
“collapsed”: false,
“gridPos”: {
“h”: 1,
“w”: 24,
“x”: 0,
“y”: 19
},
“id”: 38,
“panels”: [],
“title”: “”,
“type”: “row”
},
{
“aliasColors”: {},
“bars”: false,
“dashLength”: 10,
“dashes”: false,
“datasource”: “${DS_PROMETHEUS}”,
“fill”: 1,
“gridPos”: {
“h”: 9,
“w”: 12,
“x”: 0,
“y”: 20
},
“id”: 120,
“legend”: {
“avg”: false,
“current”: false,
“max”: false,
“min”: false,
“show”: false,
“total”: false,
“values”: false
},
“lines”: true,
“linewidth”: 1,
“links”: [],
“nullPointMode”: “null”,
“percentage”: false,
“pointradius”: 5,
“points”: false,
“renderer”: “flot”,
“seriesOverrides”: [],
“spaceLength”: 10,
“stack”: false,
“steppedLine”: false,
“targets”: [
{
“expr”: “sum
without(instance)(rate(kafka_server_brokertopicmetrics_bytesin_total{job=\“kafka\”,topic!=\”\"}[5m]))",
“format”: “time_series”,
“intervalFactor”: 1,
“legendFormat”: “{ {topic}}”,
“refId”: “A”
}
],
“thresholds”: [],
“timeFrom”: null,
“timeShift”: null,
“title”: “Bytes In Per Topic”,
“tooltip”: {
“shared”: true,
“sort”: 0,
“value_type”: “individual”
},
“type”: “graph”,
“xaxis”: {
“buckets”: null,
“mode”: “time”,
“name”: null,
“show”: true,
“values”: []
},
“yaxes”: [
{
“format”: “short”,
“label”: null,
“logBase”: 1,
“max”: null,
“min”: null,
“show”: true
},
{
“format”: “short”,
“label”: null,
“logBase”: 1,
“max”: null,
“min”: null,
“show”: true
}
]
},
{
“aliasColors”: {},
“bars”: false,
“dashLength”: 10,
“dashes”: false,
“datasource”: “${DS_PROMETHEUS}”,
“fill”: 1,
“gridPos”: {
“h”: 9,
“w”: 12,
“x”: 12,
“y”: 20
},
“id”: 122,
“legend”: { </
Copyright © 2003-2013 www.wpsshop.cn 版权所有,并保留所有权利。