Flume 拓扑结构

1. Agent 实现原理

1.1 Flume 流式处理

推送事件（Put）

duPut：将此数据写入临时缓冲区putList

doCommit：检查channel 内存队列是否足够合并

doRollBack：当 channel 内存队列空间不足，回滚数据

拉取事件（Take）

doTake：将数据取到临时缓冲区 takeList

doCommit：如果数据全部发送成功，则清除临时缓冲区 takeList

doRollback：数据发送过程中如果出现异常，rollback 将临时缓冲区 takeList 中的数据归还给 channel 内存队列

1.2 Agent 重要组件

Agent 除了 Source、Channel、Sink 外，还有 Channel Selectors、Sink Processors、Interceptors。

1.2.1 Channel Selectors

Channel Selector ：选出 Event 将要被发往哪个 Channel。

其共有三种类型， 分别是 Replicating（复制）、Multiplexing（多路复用）、Custom Channel（自定义）。

• Replicating：会将同一个 Event 发往所有的 Channe，默认项
• Multiplexing：会根据相应的原则，将不同的 Event 发往不同的 Channel
• Custom Channel：需要自定义实现 ChannelSelector 接口

例子：

# Replicating Channel
a1.sources = r1
a1.channels = c1 c2 c3
a1.sources.r1.selector.type = replicating
a1.sources.r1.channels = c1 c2 c3
a1.sources.r1.selector.optional = c3

# Multiplexing Channel 
a1.sources = r1
a1.channels = c1 c2 c3 c4
a1.sources.r1.selector.type = multiplexing
a1.sources.r1.selector.header = state
a1.sources.r1.selector.mapping.CZ = c1
a1.sources.r1.selector.mapping.US = c2 c3
a1.sources.r1.selector.default = c4

# 自定义 Channel 
a1.sources = r1
a1.channels = c1
a1.sources.r1.selector.type = org.example.MyChannelSelector

1.2.2 Sink Processors

SinkProcessor 共 有 三四种 类 型 ， 分 别 是 Default Sink Processor、FailoverSinkProcessor（故障转移）、 LoadBalancingSinkProcessor（负载均衡接）、Custom Sink Processor（自定义）。

• Default Sink Processor：只接受一个 Sink
• Failover Sink Processor：维护一个按优先级排序的接收器列表，确保只要有一个可用的事件就会被处理(交付)。
• LoadBalancing Sink Processor：负载均衡接收器处理器提供了在多个接收器上负载平衡流的能力。

例子：

# Failover Sink Processor
a1.sinkgroups = g1
a1.sinkgroups.g1.sinks = k1 k2
a1.sinkgroups.g1.processor.type = failover
a1.sinkgroups.g1.processor.priority.k1 = 5
a1.sinkgroups.g1.processor.priority.k2 = 10
a1.sinkgroups.g1.processor.maxpenalty = 10000

# Load balancing Sink Processor
a1.sinkgroups = g1
a1.sinkgroups.g1.sinks = k1 k2
a1.sinkgroups.g1.processor.type = load_balance
a1.sinkgroups.g1.processor.backoff = true
a1.sinkgroups.g1.processor.selector = random

LoadBalancing Sink Processor 和 FailoverSinkProcessor 对应的是 Sink Group

1.2.3 Interceptors

Interceptors ：Flume具有在运行中修改/删除事件的能力，这就需要在拦截器的帮助下完成。通常一个拦截器返回的事件列表被传递给下一个拦截器。

常用的拦截器：

• Timestamp Interceptor：这个拦截器在事件头中插入它处理事件的时间(以毫秒为单位)。
• Host Interceptor：这个拦截器插入运行此代理的主机的主机名或IP地址。

例子：

a1.sources = r1
a1.sinks = k1
a1.channels = c1
a1.sources.r1.interceptors = i1 i2
a1.sources.r1.interceptors.i1.type = host
a1.sources.r1.interceptors.i1.preserveExisting = false
a1.sources.r1.interceptors.i1.hostHeader = hostname
a1.sources.r1.interceptors.i2.type = timestamp
a1.sinks.k1.filePrefix = FlumeData.%{CollectorHost}.%Y-%m-%d
a1.sinks.k1.channel = c1

2. 拓扑结构

2.1 串行模式

2.1.1 模型结构

该模式是将多个 Flumev串联起来，为了跨越多个代理或跃点传输数据，前一个代理的 Sink 和当前跃点的 Source 需要为avro类型，且 Sink 指向 Source 的主机名(或IP地址)和端口。

Flume 的数量会影响传输速率，如果其中某个节点宕机，会影响整个传输系统。

2.1.2 串行实现

1️⃣ Step1：node1主机，添加配置文件 tailDir2avro.conf

# Name the components on this agent
taildir2avro.sources = r1
taildir2avro.sinks = k1
taildir2avro.channels = c1

# Describe/configure the source
taildir2avro.sources.r1.type = TAILDIR
taildir2avro.sources.r1.filegroups = g1 g2
taildir2avro.sources.r1.filegroups.g1 = /root/log.txt
taildir2avro.sources.r1.filegroups.g2 = /root/log/.*.txt

# Describe the sink
taildir2avro.sinks.k1.type = avro
taildir2avro.sinks.k1.hostname = node2
taildir2avro.sinks.k1.port = 12345

# Use a channel which buffers events in memory
taildir2avro.channels.c1.type = memory
taildir2avro.channels.c1.capacity = 30000
taildir2avro.channels.c1.transactionCapacity = 3000

# Bind the source and sink to the channel
taildir2avro.sources.r1.channels = c1
taildir2avro.sinks.k1.channel = c1

2️⃣ Step2：node2 主机，添加配置文件 avro2hdfs.conf

# Name the components on this agent
avro2hdfs.sources = r1
avro2hdfs.sinks = k1
avro2hdfs.channels = c1

# Describe/configure the source
avro2hdfs.sources.r1.type = avro
avro2hdfs.sources.r1.bind = node2
avro2hdfs.sources.r1.port = 12345


# Describe the sink
avro2hdfs.sinks.k1.type = hdfs
avro2hdfs.sinks.k1.hdfs.path = hdfs://node1:8020/flume/%y-%m-%d-%H-%M
avro2hdfs.sinks.k1.hdfs.fileType = DataStream
avro2hdfs.sinks.k1.hdfs.writeFormat = Text
avro2hdfs.sinks.k1.hdfs.useLocalTimeStamp = true
avro2hdfs.sinks.k1.hdfs.rollInterval = 300
avro2hdfs.sinks.k1.hdfs.rollSize = 130000000
avro2hdfs.sinks.k1.hdfs.rollCount = 0


# Use a channel which buffers events in memory
avro2hdfs.channels.c1.type = memory
avro2hdfs.channels.c1.capacity = 30000
avro2hdfs.channels.c1.transactionCapacity = 3000

# Bind the source and sink to the channel
avro2hdfs.sources.r1.channels = c1
avro2hdfs.sinks.k1.channel = c1

3️⃣ Step3：启动 node2 上的 Flume

flume-ng agent -c /opt/flume-1.9.0/conf/ -f /opt/flume-1.9.0/agents/avro2hdfs.conf -n avro2hdfs -Dflume.root.logger=INFO,console

4️⃣ Step4：查看node2 的 12345 端口是否被监听，确定已在监听状态

5️⃣ Step5：运行一会 node1 的 printer.sh 脚本，启动 node1 的 Flume

flume-ng agent -c /opt/flume-1.9.0/conf/ -f /opt/flume-1.9.0/agents/taildir2avro.conf -n taildir2avro -Dflume.root.logger=INFO,console

6️⃣ Step6：查看 node2 控制台输出，node1 以成功向 node2 的12345端口发送数据了

7️⃣ Step7：查看web端是否有数据提交

2.2 复制和多路复用

2.2.1 模型结构

Flume支持将事件流向一个或多个目的地，这种模式可以将相同数据源复制到多个channel中，或将不同数据分发到不同的channel中，sink可以选择传送到不同的目的地。

2.2.2 模型实现

实现： node1 上实现两种 channel、sink，一种将数据上传至 hdfs，另一种发送给 node2，node2再上传一份。

1️⃣ Step1：在 node1 上添加配置文件 taildir2hdfs_avro.conf

# Name the components on this agent
taildir2hdfs_avro.sources = r1
taildir2hdfs_avro.sinks = k1 k2
taildir2hdfs_avro.channels = c1 c2

# Describe/configure the source
taildir2hdfs_avro.sources.r1.type = TAILDIR
taildir2hdfs_avro.sources.r1.filegroups = g1 g2
taildir2hdfs_avro.sources.r1.filegroups.g1 = /root/log.txt
taildir2hdfs_avro.sources.r1.filegroups.g2 = /root/log/.*.txt

# Describe the sink
taildir2hdfs_avro.sinks.k1.type = hdfs
taildir2hdfs_avro.sinks.k1.hdfs.path = hdfs://node1:8020/flume/node1/%y-%m-%d-%H-%M
taildir2hdfs_avro.sinks.k1.hdfs.fileType = DataStream
taildir2hdfs_avro.sinks.k1.hdfs.writeFormat = Text
taildir2hdfs_avro.sinks.k1.hdfs.useLocalTimeStamp = true
taildir2hdfs_avro.sinks.k1.hdfs.rollInterval = 300
taildir2hdfs_avro.sinks.k1.hdfs.rollSize = 30000000
taildir2hdfs_avro.sinks.k1.hdfs.rollCount = 0

taildir2hdfs_avro.sinks.k2.type = avro
taildir2hdfs_avro.sinks.k2.hostname = node2
taildir2hdfs_avro.sinks.k2.port = 12345

# Use a channel which buffers events in memory
taildir2hdfs_avro.channels.c1.type = memory
taildir2hdfs_avro.channels.c1.capacity = 10000
taildir2hdfs_avro.channels.c1.transactionCapacity = 1000

taildir2hdfs_avro.channels.c2.type = memory
taildir2hdfs_avro.channels.c2.capacity = 10000
taildir2hdfs_avro.channels.c2.transactionCapacity = 1000

# Bind the source and sink to the channel
taildir2hdfs_avro.sources.r1.channels = c1 c2
taildir2hdfs_avro.sinks.k1.channel = c1
taildir2hdfs_avro.sinks.k2.channel = c2

2️⃣ Step2：在 node2 上添加配置文件 avro2hdfs.conf

# Name the components on this agent
avro2hdfs.sources = r1
avro2hdfs.sinks = k1
avro2hdfs.channels = c1

# Describe/configure the source
avro2hdfs.sources.r1.type = avro
avro2hdfs.sources.r1.bind = node2
avro2hdfs.sources.r1.port = 12345


# Describe the sink
avro2hdfs.sinks.k1.type = hdfs
avro2hdfs.sinks.k1.hdfs.path = hdfs://node1:8020/flume/node2/%y-%m-%d-%H-%M
avro2hdfs.sinks.k1.hdfs.fileType = DataStream
avro2hdfs.sinks.k1.hdfs.writeFormat = Text
avro2hdfs.sinks.k1.hdfs.useLocalTimeStamp = true
avro2hdfs.sinks.k1.hdfs.rollInterval = 300
avro2hdfs.sinks.k1.hdfs.rollSize = 30000000
avro2hdfs.sinks.k1.hdfs.rollCount = 0


# Use a channel which buffers events in memory
avro2hdfs.channels.c1.type = memory
avro2hdfs.channels.c1.capacity = 10000
avro2hdfs.channels.c1.transactionCapacity = 1000

# Bind the source and sink to the channel
avro2hdfs.sources.r1.channels = c1
avro2hdfs.sinks.k1.channel = c1

3️⃣ Step3：启动 node2 上的 Flume

flume-ng agent -c /opt/flume-1.9.0/conf/ -f /opt/flume-1.9.0/agents/avro2hdfs.conf -n avro2hdfs -Dflume.root.logger=INFO,console

4️⃣ Step4：启动 node1 上的 Flume

flume-ng agent -c /opt/flume-1.9.0/conf/ -f /opt/flume-1.9.0/agents/taildir2hdfs_avro.conf -n taildir2hdfs_avro -Dflume.root.logger=INFO,console

5️⃣ Step5：在web端查看是否有文件上传

2.3 负载均衡和故障转移

2.3.1 模型结构

2.3.2 模型实现

实现： node1 向 node2、node3 发送数据，node2、node3 负责上传数据

1️⃣ Step1：在 node1 上添加配置文件 taildir2avro.conf

# Name the components on this agent
taildir2avro.sources = r1
taildir2avro.sinks = k1 k2
taildir2avro.channels = c1 c2

# Describe/configure the source
taildir2avro.sources.r1.type = TAILDIR
taildir2avro.sources.r1.filegroups = g1 g2
taildir2avro.sources.r1.filegroups.g1 = /root/log.txt
taildir2avro.sources.r1.filegroups.g2 = /root/log/.*.txt

# Describe the sink
taildir2avro.sinks.k1.type = avro
taildir2avro.sinks.k1.hostname = node2
taildir2avro.sinks.k1.port = 12345

taildir2avro.sinks.k2.type = avro
taildir2avro.sinks.k2.hostname = node3
taildir2avro.sinks.k2.port = 12345

# Use a channel which buffers events in memory
taildir2avro.channels.c1.type = memory
taildir2avro.channels.c1.capacity = 10000
taildir2avro.channels.c1.transactionCapacity = 1000

taildir2avro.channels.c2.type = memory
taildir2avro.channels.c2.capacity = 10000
taildir2avro.channels.c2.transactionCapacity = 1000

# Bind the source and sink to the channel
taildir2avro.sources.r1.channels = c1 c2
taildir2avro.sinks.k1.channel = c1
taildir2avro.sinks.k2.channel = c2

2️⃣ Step2：在 node2 上添加配置文件 avro2hdfs.conf

# Name the components on this agent
avro2hdfs.sources = r1
avro2hdfs.sinks = k1
avro2hdfs.channels = c1

# Describe/configure the source
avro2hdfs.sources.r1.type = avro
avro2hdfs.sources.r1.bind = node2
avro2hdfs.sources.r1.port = 12345


# Describe the sink
avro2hdfs.sinks.k1.type = hdfs
avro2hdfs.sinks.k1.hdfs.path = hdfs://node1:8020/flume/node2/%y-%m-%d-%H-%M
avro2hdfs.sinks.k1.hdfs.fileType = DataStream
avro2hdfs.sinks.k1.hdfs.writeFormat = Text
avro2hdfs.sinks.k1.hdfs.useLocalTimeStamp = true
avro2hdfs.sinks.k1.hdfs.rollInterval = 300
avro2hdfs.sinks.k1.hdfs.rollSize = 30000000
avro2hdfs.sinks.k1.hdfs.rollCount = 0


# Use a channel which buffers events in memory
avro2hdfs.channels.c1.type = memory
avro2hdfs.channels.c1.capacity = 10000
avro2hdfs.channels.c1.transactionCapacity = 1000

# Bind the source and sink to the channel
avro2hdfs.sources.r1.channels = c1
avro2hdfs.sinks.k1.channel = c1

3️⃣ Step3：在 node3 上添加配置文件 avro2hdfs.conf

# Name the components on this agent
avro2hdfs.sources = r1
avro2hdfs.sinks = k1
avro2hdfs.channels = c1

# Describe/configure the source
avro2hdfs.sources.r1.type = avro
avro2hdfs.sources.r1.bind = node3
avro2hdfs.sources.r1.port = 12345


# Describe the sink
avro2hdfs.sinks.k1.type = hdfs
avro2hdfs.sinks.k1.hdfs.path = hdfs://node1:8020/flume/node3/%y-%m-%d-%H-%M
avro2hdfs.sinks.k1.hdfs.fileType = DataStream
avro2hdfs.sinks.k1.hdfs.writeFormat = Text
avro2hdfs.sinks.k1.hdfs.useLocalTimeStamp = true
avro2hdfs.sinks.k1.hdfs.rollInterval = 300
avro2hdfs.sinks.k1.hdfs.rollSize = 30000000
avro2hdfs.sinks.k1.hdfs.rollCount = 0


# Use a channel which buffers events in memory
avro2hdfs.channels.c1.type = memory
avro2hdfs.channels.c1.capacity = 10000
avro2hdfs.channels.c1.transactionCapacity = 1000

# Bind the source and sink to the channel
avro2hdfs.sources.r1.channels = c1
avro2hdfs.sinks.k1.channel = c1

4️⃣ Step4：启动 node2、node3 上的 Flume

flume-ng agent -c /opt/flume-1.9.0/conf/ -f /opt/flume-1.9.0/agents/avro2hdfs.conf -n avro2hdfs -Dflume.root.logger=INFO,console

5️⃣ Step5：启动 node1 上的 Flume

flume-ng agent -c /opt/flume-1.9.0/conf/ -f /opt/flume-1.9.0/agents/taildir2avro.conf -n taildir2avro -Dflume.root.logger=INFO,console

6️⃣ Step6：控制台查看日志输出

7️⃣ Step7：在web端查看是否有文件上传

2.4 聚合模式

2.4.1 模型结构

这是比较实用的结构，在Flume中通过配置许多带有avro接收器的第一层代理来实现，这些代理都指向单个代理的avro源(同样，您可以在这种场景中使用节约源/接收器/客户端)。第二层代理上的这个源将接收到的事件合并为单个通道，该通道由接收器消费到其最终目的地。

web应用通常分布在上百个服务器。产生的日志，处理起来也非常麻烦。用flume的这种组合方式能很好的解决这一问题，每台服务器部署一个flume采集日志，传送到一个集中收集日志的flume，再由此flume上传到hdfs、hive、hbase等，进行日志分析。

2.4.2 模型实现

实现： node1、node2 都向 node3 发送数据，由 node3 负责向 hdfs 上传数据

1️⃣ Step1：在 node1 上添加配置文件 taildir2avro.conf

# Name the components on this agent
taildir2avro.sources = r1
taildir2avro.sinks = k1
taildir2avro.channels = c1

# Describe/configure the source
taildir2avro.sources.r1.type = TAILDIR
taildir2avro.sources.r1.filegroups = g1 g2
taildir2avro.sources.r1.filegroups.g1 = /root/log.txt
taildir2avro.sources.r1.filegroups.g2 = /root/log/.*.txt

# Describe the sink
taildir2avro.sinks.k1.type = avro
taildir2avro.sinks.k1.hostname = node3
taildir2avro.sinks.k1.port = 12345

# Use a channel which buffers events in memory
taildir2avro.channels.c1.type = memory
taildir2avro.channels.c1.capacity = 10000
taildir2avro.channels.c1.transactionCapacity = 1000

# Bind the source and sink to the channel
taildir2avro.sources.r1.channels = c1
taildir2avro.sinks.k1.channel = c1

2️⃣ Step2：在 node2 上添加配置文件 taildir2avro.conf

# Name the components on this agent
taildir2avro.sources = r1
taildir2avro.sinks = k1
taildir2avro.channels = c1

# Describe/configure the source
taildir2avro.sources.r1.type = TAILDIR
taildir2avro.sources.r1.filegroups = g1 g2
taildir2avro.sources.r1.filegroups.g1 = /root/log.txt
taildir2avro.sources.r1.filegroups.g2 = /root/log/.*.txt

# Describe the sink
taildir2avro.sinks.k1.type = avro
taildir2avro.sinks.k1.hostname = node3
taildir2avro.sinks.k1.port = 12345

# Use a channel which buffers events in memory
taildir2avro.channels.c1.type = memory
taildir2avro.channels.c1.capacity = 10000
taildir2avro.channels.c1.transactionCapacity = 1000

# Bind the source and sink to the channel
taildir2avro.sources.r1.channels = c1
taildir2avro.sinks.k1.channel = c1

3️⃣ Step3：在 node3 上添加配置文件 avro2hdfs.conf

# Name the components on this agent
avro2hdfs.sources = r1
avro2hdfs.sinks = k1
avro2hdfs.channels = c1

# Describe/configure the source
avro2hdfs.sources.r1.type = avro
avro2hdfs.sources.r1.bind = node3
avro2hdfs.sources.r1.port = 12345


# Describe the sink
avro2hdfs.sinks.k1.type = hdfs
avro2hdfs.sinks.k1.hdfs.path = hdfs://node1:8020/flume/%y-%m-%d-%H-%M
avro2hdfs.sinks.k1.hdfs.fileType = DataStream
avro2hdfs.sinks.k1.hdfs.writeFormat = Text
avro2hdfs.sinks.k1.hdfs.useLocalTimeStamp = true
avro2hdfs.sinks.k1.hdfs.rollInterval = 300
avro2hdfs.sinks.k1.hdfs.rollSize = 30000000
avro2hdfs.sinks.k1.hdfs.rollCount = 0


# Use a channel which buffers events in memory
avro2hdfs.channels.c1.type = memory
avro2hdfs.channels.c1.capacity = 10000
avro2hdfs.channels.c1.transactionCapacity = 1000

# Bind the source and sink to the channel
avro2hdfs.sources.r1.channels = c1
avro2hdfs.sinks.k1.channel = c1

4️⃣ Step4：启动 node3 上的 Flume

flume-ng agent -c /opt/flume-1.9.0/conf/ -f /opt/flume-1.9.0/agents/avro2hdfs.conf -n avro2hdfs -Dflume.root.logger=INFO,console

5️⃣ Step5：启动 node1、node2 上的 Flume

flume-ng agent -c /opt/flume-1.9.0/conf/ -f /opt/flume-1.9.0/agents/taildir2avro.conf -n taildir2avro -Dflume.root.logger=INFO,console

6️⃣ Step6：控制台查看日志输出

7️⃣ Step7：在web端查看是否有文件上传

3. 写在最后

配置文件大致一样，每次测试时都先向监控的目录或文件下写点东西，每测试完清空hdfs上的文件。

 

---

❤️ END ❤️