前言
从上周一直在调研找一款好用的kafka监控，我测试使用过的KafkaOffsetMonitor、Burrow、kafka-monitor、Kafka-Manager，他们各有优缺点，具体情况我这里就不展开描述了，大家可以到它们的git上去查看， 并且它们基本上都是监控topic的写入和读取等等，没有提供对于整体集群的监控信息，比如集群的分片、延时、内存使用情况等等，无意中发现了jmxtrans，jmxtrans它是一个通过jmx采集java应用的数据采集器，他的输出可以是Graphite、StatsD、Ganglia、InfluxDb等等，刚好我们现有的监控是通过InfluxDb做数据存储的，通过Grafana做展示，下面就给大家介绍一下jmxtrans+InfluxDb+Grafana监控kafka的整体解决方案，并且不需要任何额外的开发工作，完全使用原生的。
二、环境介绍
1、角色
a、10.10.10.10 InfluxDb
b、10.10.10.100 Grafana
c、10.10.30.69 jmxtrans
d、kafka集群
10.10.20.14 node1
10.10.20.15 node2
10.10.20.16 node3
10.10.20.17 node4
2、软件版本
influxdb-1.2.4-1.x86_64
grafana-4.1.1-1484211277.x86_64
jmxtrans-266.rpm
kafka_2.10-0.9.0.0.jar.asc
3、架构图
三、配置规划
1、jmxtrans我们可以分别在每台kafka节点上部署，也可以部署到一台机器上，我这里是选择了后者，因为我的集群小，这样配置文件可以集中管理，如果集群比较大，可以考虑分散部署。
2、关于jmxtrans的配置文件，分全局指标（每个kafka节点）和topic指标，全局指标每个节点一个配置文件，命名规则：base_10.10.20.14.json，topic指标是每个topic一个配置文件，命名规则：falcon_monitor_us_17.json
四、监控指标
1、全局指标
每秒输入的流量
"obj" : "kafka.server:type=BrokerTopicMetrics,name=BytesInPerSec"
"attr" : [ "Count" ]
"resultAlias":"BytesInPerSec"
"tags" : {"application" : "BytesInPerSec"}
每秒输入的流量
"obj" : "kafka.server:type=BrokerTopicMetrics,name=BytesOutPerSec"
"attr" : [ "Count" ]
"resultAlias":"BytesOutPerSec"
"tags" : {"application" : "BytesOutPerSec"}
每秒输入的流量
"obj" : "kafka.server:type=BrokerTopicMetrics,name=BytesRejectedPerSec"
"attr" : [ "Count" ]
"resultAlias":"BytesRejectedPerSec"
"tags" : {"application" : "BytesRejectedPerSec"}
每秒的消息写入总量
"obj" : "kafka.server:type=BrokerTopicMetrics,name=MessagesInPerSec"
"attr" : [ "Count" ]
"resultAlias":"MessagesInPerSec"
"tags" : {"application" : "MessagesInPerSec"}
每秒FetchFollower的请求次数
"obj" : "kafka.network:type=RequestMetrics,name=RequestsPerSec,request=FetchFollower"
"attr" : [ "Count" ]
"resultAlias":"RequestsPerSec"
"tags" : {"request" : "FetchFollower"}
每秒FetchConsumer的请求次数
"obj" : "kafka.network:type=RequestMetrics,name=RequestsPerSec,request=FetchConsumer"
"attr" : [ "Count" ]
"resultAlias":"RequestsPerSec"
"tags" : {"request" : "FetchConsumer"}
每秒Produce的请求次数
"obj" : "kafka.network:type=RequestMetrics,name=RequestsPerSec,request=Produce"
"attr" : [ "Count" ]
"resultAlias":"RequestsPerSec"
"tags" : {"request" : "Produce"}
内存使用的使用情况
"obj" : "java.lang:type=Memory"
"attr" : [ "HeapMemoryUsage", "NonHeapMemoryUsage" ]
"resultAlias":"MemoryUsage"
"tags" : {"application" : "MemoryUsage"}
GC的耗时和次数
"obj" : "java.lang:type=GarbageCollector,name=*"
"attr" : [ "CollectionCount","CollectionTime" ]
"resultAlias":"GC"
"tags" : {"application" : "GC"}
线程的使用情况
"obj" : "java.lang:type=Threading"
"attr" : [ "PeakThreadCount","ThreadCount" ]
"resultAlias":"Thread"
"tags" : {"application" : "Thread"}
副本落后主分片的最大消息数量
"obj" : "kafka.server:type=ReplicaFetcherManager,name=MaxLag,clientId=Replica"
"attr" : [ "Value" ]
"resultAlias":"ReplicaFetcherManager"
"tags" : {"application" : "MaxLag"}
该broker上的partition的数量
"obj" : "kafka.server:type=ReplicaManager,name=PartitionCount"
"attr" : [ "Value" ]
"resultAlias":"ReplicaManager"
"tags" : {"application" : "PartitionCount"}
正在做复制的partition的数量
"obj" : "kafka.server:type=ReplicaManager,name=UnderReplicatedPartitions"
"attr" : [ "Value" ]
"resultAlias":"ReplicaManager"
"tags" : {"application" : "UnderReplicatedPartitions"}
Leader的replica的数量
"obj" : "kafka.server:type=ReplicaManager,name=LeaderCount"
"attr" : [ "Value" ]
"resultAlias":"ReplicaManager"
"tags" : {"application" : "LeaderCount"}
一个请求FetchConsumer耗费的所有时间
"obj" : "kafka.network:type=RequestMetrics,name=TotalTimeMs,request=FetchConsumer"
"attr" : [ "Count","Max" ]
"resultAlias":"TotalTimeMs"
"tags" : {"application" : "FetchConsumer"}
一个请求FetchFollower耗费的所有时间
"obj" : "kafka.network:type=RequestMetrics,name=TotalTimeMs,request=FetchFollower"
"attr" : [ "Count","Max" ]
"resultAlias":"TotalTimeMs"
"tags" : {"application" : "FetchFollower"}
一个请求Produce耗费的所有时间
"obj" : "kafka.network:type=RequestMetrics,name=TotalTimeMs,request=Produce"
"attr" : [ "Count","Max" ]
"resultAlias":"TotalTimeMs"
"tags" : {"application" : "Produce"}
2、topic的监控指标
falcon_monitor_us每秒的写入流量
"kafka.server:type=BrokerTopicMetrics,name=BytesInPerSec,topic=falcon_monitor_us"
"attr" : [ "Count" ]
"resultAlias":"falcon_monitor_us"
"tags" : {"application" : "BytesInPerSec"}
falcon_monitor_us每秒的输出流量
"kafka.server:type=BrokerTopicMetrics,name=BytesOutPerSec,topic=falcon_monitor_us"
"attr" : [ "Count" ]
"resultAlias":"falcon_monitor_us"
"tags" : {"application" : "BytesOutPerSec"}
falcon_monitor_us每秒写入消息的数量
"obj" : "kafka.server:type=BrokerTopicMetrics,name=MessagesInPerSec,topic=falcon_monitor_us"
"attr" : [ "Count" ]
"resultAlias":"falcon_monitor_us"
"tags" : {"application" : "MessagesInPerSec"}
falcon_monitor_us在每个分区最后的Offset
"obj" : "kafka.log:type=Log,name=LogEndOffset,topic=falcon_monitor_us,partition=*"
"attr" : [ "Value" ]
"resultAlias":"falcon_monitor_us"
"tags" : {"application" : "LogEndOffset"}
PS：
1、参数说明
"obj"对应jmx的ObjectName，就是我们要监控的指标
"attr"对应ObjectName的属性，可以理解为我们要监控的指标的值
"resultAlias"对应metric 的名称，在InfluxDb里面就是MEASUREMENTS名
"tags" 对应InfluxDb的tag功能，对与存储在同一个MEASUREMENTS里面的不同监控指标可以做区分，我们在用Grafana绘图的时候会用到，建议对每个监控指标都打上tags
2、对于全局监控，每一个监控指标对应一个MEASUREMENTS，所有的kafka节点同一个监控指标数据写同一个MEASUREMENTS ，对于topc监控的监控指标，同一个topic所有kafka节点写到同一个MEASUREMENTS，并且以topic名称命名
五、安装
1、kafka
这里不详细介绍kafka集群的安装，主要说一下kafka的启动方式，因为我们需要通过jmx采集kafka的监控数据，所以在kafka的启动时候需要启动jmx端口，启动方式如下：
cd /data/kafka/bin/
JMX_PORT=9999 nohup ./kafka-server-start.sh ../config/server.properties >/dev/null 2>&1 &
2、InfluxDb
yum -y install influxdb ##安装
/etc/init.d/influxdb start ##启动服务
[root@ip-10-10-10-10 jmxtrans]# influx
Connected to http://localhost:8086 version 1.3.2
InfluxDB shell version: 1.3.2
> CREATE USER "root" WITH PASSWORD '123456' WITH ALL PRIVILEGES ##添加一个账号
>
3、Grafana
yum -y install grafana ##安装
/etc/init.d/grafana-server start ##启动服务
4、jmxtrans
wget http://central.maven.org/maven2/org/jmxtrans/jmxtrans/266/jmxtrans-266.rpm
rpm -ivh jmxtrans-266.rpm ##安装
/etc/init.d/jmxtrans start ##启动
六、配置
这里主要介绍jmxtrans采集数据的配置文件撰写和Grafana绘图的配置注意事项，kafka和InfluxDb的配置这里不做描述。
1、jmxtrans
a、jmxtrans默认读取/var/lib/jmxtrans下的配置文件去采集数据的，所以我们把采集kafka监控数据的配置文件都在这个目录下，下面是我的配置文件命名规范：
[root@ip-10-10-30-69 jmxtrans]# ll
total 96
-rw-r--r-- 1 root root 1657 Aug 18 17:03 article-feedback-10min-json_14.json
-rw-r--r-- 1 root root 1657 Aug 18 17:03 article-feedback-10min-json_15.json
-rw-r--r-- 1 root root 1657 Aug 18 17:04 article-feedback-10min-json_16.json
-rw-r--r-- 1 root root 1657 Aug 18 17:04 article-feedback-10min-json_17.json
-rw-r--r-- 1 root root 8430 Aug 22 08:24 base_10.10.20.14.json
-rw-r--r-- 1 root root 8431 Aug 22 08:24 base_10.10.20.15.json
-rw-r--r-- 1 root root 8431 Aug 22 08:25 base_10.10.20.16.json
-rw-r--r-- 1 root root 8431 Aug 22 08:25 base_10.10.20.17.json
-rw-r--r-- 1 root root 2027 Aug 21 16:19 falcon_monitor_us_14.json
-rw-r--r-- 1 root root 2027 Aug 21 16:20 falcon_monitor_us_15.json
-rw-r--r-- 1 root root 2484 Aug 21 20:58 falcon_monitor_us_16.json
-rw-r--r-- 1 root root 2027 Aug 21 16:20 falcon_monitor_us_17.json
-rw-r--r-- 1 root root 2147 Aug 21 17:43 highgmp-articles-through-primary_14.json
-rw-r--r-- 1 root root 2147 Aug 21 17:46 highgmp-articles-through-primary_15.json
-rw-r--r-- 1 root root 2147 Aug 21 17:46 highgmp-articles-through-primary_16.json
-rw-r--r-- 1 root root 2147 Aug 21 17:47 highgmp-articles-through-primary_17.json
[root@ip-10-10-30-69 jmxtrans]# pwd
/var/lib/jmxtrans
b、全局监控的配置文件，以10.10.20.14为例：
[root@ip-10-10-30-69 jmxtrans]# cat base_10.10.20.14.json
{
"servers" : [ {
"port" : "9999",
"host" : "10.10.20.14",
"queries" : [ {
"obj" : "kafka.server:type=BrokerTopicMetrics,name=BytesInPerSec",
"attr" : [ "Count","OneMinuteRate" ],
"resultAlias":"BytesInPerSec",
"outputWriters" : [ {
"@class" : "com.googlecode.jmxtrans.model.output.InfluxDbWriterFactory",
"url" : "http://10.10.10.10:8086/",
"username" : "root",
"password" : "root",
"database" : "jmxDB",
"tags" : {"application" : "BytesInPerSec"}
} ]
},
{
"obj" : "kafka.server:type=BrokerTopicMetrics,name=BytesOutPerSec",
"attr" : [ "Count","OneMinuteRate" ],
"resultAlias":"BytesOutPerSec",
"outputWriters" : [ {
"@class" : "com.googlecode.jmxtrans.model.output.InfluxDbWriterFactory",
"url" : "http://10.10.10.10:8086/",
"username" : "root",
"password" : "root",
"database" : "jmxDB",
"tags" : {"application" : "BytesOutPerSec"}
} ]
},
{
"obj" : "kafka.server:type=BrokerTopicMetrics,name=BytesRejectedPerSec",
"attr" : [ "Count","OneMinuteRate" ],
"resultAlias":"BytesRejectedPerSec",
"outputWriters" : [ {
"@class" : "com.googlecode.jmxtrans.model.output.InfluxDbWriterFactory",
"url" : "http://10.10.10.10:8086/",
"username" : "root",
"password" : "root",
"database" : "jmxDB",
"tags" : {"application" : "BytesRejectedPerSec"}
} ]
},
{
"obj" : "kafka.server:type=BrokerTopicMetrics,name=MessagesInPerSec",
"attr" : [ "Count","OneMinuteRate" ],
"resultAlias":"MessagesInPerSec",
"outputWriters" : [ {
"@class" : "com.googlecode.jmxtrans.model.output.InfluxDbWriterFactory",
"url" : "http://10.10.10.10:8086/",
"username" : "root",
"password" : "root",
"database" : "jmxDB",
"tags" : {"application" : "MessagesInPerSec"}
} ]
},
{
"obj" : "kafka.network:type=RequestMetrics,name=RequestsPerSec,request=FetchConsumer",
"attr" : [ "Count" ],
"resultAlias":"RequestsPerSec",
"outputWriters" : [ {
"@class" : "com.googlecode.jmxtrans.model.output.InfluxDbWriterFactory",
"url" : "http://10.10.10.10:8086/",
"username" : "root",
"password" : "root",
"database" : "jmxDB",
"tags" : {"request" : "FetchConsumer"}
} ]
},
{
"obj" : "kafka.network:type=RequestMetrics,name=RequestsPerSec,request=FetchFollower",
"attr" : [ "Count" ],
"resultAlias":"RequestsPerSec",
"outputWriters" : [ {
"@class" : "com.googlecode.jmxtrans.model.output.InfluxDbWriterFactory",
"url" : "http://10.10.10.10:8086/",
"username" : "root",
"password" : "root",
"database" : "jmxDB",
"tags" : {"request" : "FetchFollower"}
} ]
},
{
"obj" : "kafka.network:type=RequestMetrics,name=RequestsPerSec,request=Produce",
"attr" : [ "Count" ],
"resultAlias":"RequestsPerSec",
"outputWriters" : [ {
"@class" : "com.googlecode.jmxtrans.model.output.InfluxDbWriterFactory",
"url" : "http://10.10.10.10:8086/",
"username" : "root",
"password" : "root",
"database" : "jmxDB",
"tags" : {"request" : "Produce"}
} ]
},
{
"obj" : "java.lang:type=Memory",
"attr" : [ "HeapMemoryUsage", "NonHeapMemoryUsage" ],
"resultAlias":"MemoryUsage",
"outputWriters" : [ {
"@class" : "com.googlecode.jmxtrans.model.output.InfluxDbWriterFactory",
"url" : "http://10.10.10.10:8086/",
"username" : "root",
"password" : "root",
"database" : "jmxDB",
"tags" : {"application" : "MemoryUsage"}
} ]
},
{
"obj" : "java.lang:type=GarbageCollector,name=*",
"attr" : [ "CollectionCount","CollectionTime" ],
"resultAlias":"GC",
"outputWriters" : [ {
"@class" : "com.googlecode.jmxtrans.model.output.InfluxDbWriterFactory",
"url" : "http://10.10.10.10:8086/",
"username" : "root",
"password" : "root",
"database" : "jmxDB",
"tags" : {"application" : "GC"}
} ]
},
{
"obj" : "java.lang:type=Threading",
"attr" : [ "PeakThreadCount","ThreadCount" ],
"resultAlias":"Thread",
"outputWriters" : [ {
"@class" : "com.googlecode.jmxtrans.model.output.InfluxDbWriterFactory",
"url" : "http://10.10.10.10:8086/",
"username" : "root",
"password" : "root",
"database" : "jmxDB",
"tags" : {"application" : "Thread"}
} ]
},
{
"obj" : "kafka.server:type=ReplicaFetcherManager,name=MaxLag,clientId=Replica",
"attr" : [ "Value" ],
"resultAlias":"ReplicaFetcherManager",
"outputWriters" : [ {
"@class" : "com.googlecode.jmxtrans.model.output.InfluxDbWriterFactory",
"url" : "http://10.10.10.10:8086/",
"username" : "root",
"password" : "root",
"database" : "jmxDB",
"tags" : {"application" : "MaxLag"}
} ]
},
{
"obj" : "kafka.server:type=ReplicaManager,name=PartitionCount",
"attr" : [ "Value" ],
"resultAlias":"ReplicaManager",
"outputWriters" : [ {
"@class" : "com.googlecode.jmxtrans.model.output.InfluxDbWriterFactory",
"url" : "http://10.10.10.10:8086/",
"username" : "root",
"password" : "root",
"database" : "jmxDB",
"tags" : {"application" : "PartitionCount"}
} ]
},
{
"obj" : "kafka.server:type=ReplicaManager,name=UnderReplicatedPartitions",
"attr" : [ "Value" ],
"resultAlias":"ReplicaManager",
"outputWriters" : [ {
"@class" : "com.googlecode.jmxtrans.model.output.InfluxDbWriterFactory",
"url" : "http://10.10.10.10:8086/",
"username" : "root",
"password" : "root",
"database" : "jmxDB",
"tags" : {"application" : "UnderReplicatedPartitions"}
} ]
},
{
"obj" : "kafka.server:type=ReplicaManager,name=LeaderCount",
"attr" : [ "Value" ],
"resultAlias":"ReplicaManager",
"outputWriters" : [ {
"@class" : "com.googlecode.jmxtrans.model.output.InfluxDbWriterFactory",
"url" : "http://10.10.10.10:8086/",
"username" : "root",
"password" : "root",
"database" : "jmxDB",
"tags" : {"application" : "LeaderCount"}
} ]
},
{
"obj" : "kafka.network:type=RequestMetrics,name=TotalTimeMs,request=FetchConsumer",
"attr" : [ "Count","Max" ],
"resultAlias":"TotalTimeMs",
"outputWriters" : [ {
"@class" : "com.googlecode.jmxtrans.model.output.InfluxDbWriterFactory",
"url" : "http://10.10.10.10:8086/",
"username" : "root",
"password" : "root",
"database" : "jmxDB",
"tags" : {"application" : "FetchConsumer"}
} ]
},
{
"obj" : "kafka.network:type=RequestMetrics,name=TotalTimeMs,request=FetchFollower",
"attr" : [ "Count","Max" ],
"resultAlias":"TotalTimeMs",
"outputWriters" : [ {
"@class" : "com.googlecode.jmxtrans.model.output.InfluxDbWriterFactory",
"url" : "http://10.10.10.10:8086/",
"username" : "root",
"password" : "root",
"database" : "jmxDB",
"tags" : {"application" : "FetchConsumer"}
} ]
},
{
"obj" : "kafka.network:type=RequestMetrics,name=TotalTimeMs,request=Produce",
"attr" : [ "Count","Max" ],
"resultAlias":"TotalTimeMs",
"outputWriters" : [ {
"@class" : "com.googlecode.jmxtrans.model.output.InfluxDbWriterFactory",
"url" : "http://10.10.10.10:8086/",
"username" : "root",
"password" : "root",
"database" : "jmxDB",
"tags" : {"application" : "Produce"}
} ]
},
{
"obj" : "kafka.server:type=ReplicaManager,name=IsrShrinksPerSec",
"attr" : [ "Count" ],
"resultAlias":"ReplicaManager",
"outputWriters" : [ {
"@class" : "com.googlecode.jmxtrans.model.output.InfluxDbWriterFactory",
"url" : "http://10.10.10.10:8086/",
"username" : "root",
"password" : "root",
"database" : "jmxDB",
"tags" : {"application" : "IsrShrinksPerSec"}
} ]
}
]
} ]
}
c、topic监控的配置文件，以falcon_monitor_us的10.10.20.14节点为例：
[root@ip-10-10-30-69 jmxtrans]# cat falcon_monitor_us_14.json
{
"servers" : [ {
"port" : "9999",
"host" : "10.10.20.14",
"queries" : [ {
"obj" : "kafka.server:type=BrokerTopicMetrics,name=BytesInPerSec,topic=falcon_monitor_us",
"attr" : [ "Count" ],
"resultAlias":"falcon_monitor_us",
"outputWriters" : [ {
"@class" : "com.googlecode.jmxtrans.model.output.InfluxDbWriterFactory",
"url" : "http://10.10.10.10:8086/",
"username" : "root",
"password" : "root",
"database" : "jmxDB",
"tags" : {"application" : "BytesInPerSec"}
} ]
},
{
"obj" : "kafka.server:type=BrokerTopicMetrics,name=BytesOutPerSec,topic=falcon_monitor_us",
"attr" : [ "Count" ],
"resultAlias":"falcon_monitor_us",
"outputWriters" : [ {
"@class" : "com.googlecode.jmxtrans.model.output.InfluxDbWriterFactory",
"url" : "http://10.10.10.10:8086/",
"username" : "root",
"password" : "root",
"database" : "jmxDB",
"tags" : {"application" : "BytesOutPerSec"}
} ]
},
{
"obj" : "kafka.server:type=BrokerTopicMetrics,name=MessagesInPerSec,topic=falcon_monitor_us",
"attr" : [ "Count" ],
"resultAlias":"falcon_monitor_us",
"outputWriters" : [ {
"@class" : "com.googlecode.jmxtrans.model.output.InfluxDbWriterFactory",
"url" : "http://10.10.10.10:8086/",
"username" : "root",
"password" : "root",
"database" : "jmxDB",
"tags" : {"application" : "MessagesInPerSec"}
} ]
},
{
"obj" : "kafka.log:type=Log,name=LogEndOffset,topic=falcon_monitor_us,partition=*",
"attr" : [ "Value" ],
"resultAlias":"falcon_monitor_us",
"outputWriters" : [ {
"@class" : "com.googlecode.jmxtrans.model.output.InfluxDbWriterFactory",
"url" : "http://10.10.10.10:8086/",
"username" : "root",
"password" : "root",
"database" : "jmxDB",
"tags" : {"application" : "LogEndOffset"}
} ]
}
]
} ]
}
2、Grafana配置
a、添加数据源
Url、Database、User、Password需要和jmxtrans采集数据配置文件里面的写一致，然后点击Save&Test，提示成功就正常了
b、创建一个dashboard，然后在这里配置每一个监控指标的图
c、要点说明
1、对于监控指标为Count的监控项，需要通过Grafana做计算得到我们想要的监控，比如BytesInPerSec这个指标，它的监控值是一个累计值，我们想要取到每秒的流量，肯定需要计算，(本次采集的值-上次采集的值)/60 ,jmxtrans是一分钟采集一次数据，具体配置参考下面截图：
因为我们是一分钟采集一次数据，所以group by 和derivative选1分钟；因为我们要每秒的流量，所以math这里除以60
2、X轴的单位选择，比如流量的单位、时间的单位、每秒消息的个数无单位等等，下面分布举一个例子介绍说明
设置流量的单位 ，点击需要设置的图，选择"Edit"进入编辑页面，切到Axes这个tab页，Unit--》data（Metric）--》bytes
设置时间的单位 ，点击需要设置的图，选择"Edit"进入编辑页面，切到Axes这个tab页，Unit--》time--》milliseconds(ms)
设置按原始值展示，无单位 ，点击需要设置的图，选择"Edit"进入编辑页面，切到Axes这个tab页，Unit--》none--》none
七、收获总结
1、关于jmx收集了kafka的那些指标，对应的值都是那些类型，对应这个问题走了很多弯路，各种谷歌百度拿到了有人整理过的，一个一个试，发现很多不能用，要不就是写的是错误的，要不就是版本不同，写法不一样，最后看到了jconsole这个工具，他可以连接到本地或者远程的jmx端口，能看到在收集的所有指标，在windows下装好jdk，在bin目录你可以找到这个工具。
2、关于consumer的延时，关官方介绍有一个type是 type=consumer-fetch-manager-metrics的指标，但是我这通过jconsole连进来死活没有找到，如果亲们有使用这套监控方案的，求帮忙解惑我的这个问题，谢了，官网监控指标如下：
http://kafka.apache.org/documentation/#monitoring
