Flume 是 Cloudera 提供的一个高可用的,高可靠的,分布式的海量日志采集、聚合和传输的系统。Flume 基于流式架构,灵活简单。

Flume最主要的作用就是,实时读取服务器本地磁盘的数据,将数据写入到HDFS。
Flume 组成架构如下图所示。

Agent 是一个 JVM 进程,它以事件的形式将数据从源头送至目的。Agent 主要有 3 个部分组成,Source、Channel、Sink。
Source 是负责接收数据到 Flume Agent 的组件。Source 组件可以处理各种类型、各种格式的日志数据,包括 avro、thrift、exec、jms、spooling directory、netcat、taildir、sequence generator、syslog、http、legacy。
Sink 不断地轮询 Channel 中的事件且批量地移除它们,并将这些事件批量写入到存储或索引系统、或者被发送到另一个 Flume Agent。Sink 组件目的地包括 hdfs、logger、avro、thrift、ipc、file、HBase、solr、自定义。
Channel 是位于 Source 和 Sink 之间的缓冲区。因此,Channel 允许 Source 和 Sink 运作在不同的速率上。Channel 是线程安全的,可以同时处理几个 Source 的写入操作和几个Sink 的读取操作。
Flume 自带两种 Channel:Memory Channel 和 File Channel。Memory Channel 是内存中的队列。Memory Channel 在不需要关心数据丢失的情景下适用。如果需要关心数据丢失,那么 Memory Channel 就不应该使用,因为程序死亡、机器宕机或者重启都会导致数据丢失。
File Channel 将所有事件写到磁盘。因此在程序关闭或机器宕机的情况下不会丢失数据。
传输单元,Flume 数据传输的基本单元,以 Event 的形式将数据从源头送至目的地。Event 由 Header 和 Body 两部分组成,Header 用来存放该 event 的一些属性,为 K-V 结构,Body 用来存放该条数据,形式为字节数组。

http://flume.apache.org/http://flume.apache.org/FlumeUserGuide.htmlhttp://archive.apache.org/dist/flume/将 apache-flume-1.9.0-bin.tar.gz 上传到 linux 的/opt/software 目录下

解压 apache-flume-1.9.0-bin.tar.gz 到/opt/model/目录下
[song@hadoop102 software]$ tar -zxf /opt/software/apache-flume-1.9.0-bin.tar.gz -C /opt/model/
[song@hadoop102 module]$ mv /opt/model/apache-flume-1.9.0-bin /opt/model/flume
[song@hadoop102 lib]$ rm /opt/model/flume/lib/guava-11.0.2.jar

[song@hadoop102 software]$ sudo yum install -y nc
[song@hadoop102 flume-telnet]$ sudo netstat -nlp | grep 44444
创建 Flume Agent 配置文件 flume-netcat-logger.conf

在 flume 目录下创建 job 文件夹并进入 job 文件夹。
[song@hadoop102 flume]$ mkdir job
[song@hadoop102 flume]$ cd job/
[song@hadoop102 job]$ vim flume-netcat-logger.conf
# Name the components on this agent
a1.sources = r1
a1.sinks = k1
a1.channels = c1
# Describe/configure the source
a1.sources.r1.type = netcat
a1.sources.r1.bind = localhost
a1.sources.r1.port = 44444
# Describe the sink
a1.sinks.k1.type = logger
# Use a channel which buffers events in memory
a1.channels.c1.type = memory
a1.channels.c1.capacity = 1000
a1.channels.c1.transactionCapacity = 100
# Bind the source and sink to the channel
a1.sources.r1.channels = c1
a1.sinks.k1.channel = c1
脚本解析:

第一种写法:
[song@hadoop102 flume]$ bin/flume-ng agent --conf conf/ --name a1 --conf-file job/flume-netcat-logger.conf
-Dflume.root.logger=INFO,console
第二种写法:
[song@hadoop102 flume]$ bin/flume-ng agent -c conf/ -n a1 -f job/flume-netcat-logger.conf -Dflume.root.logger=INFO,console
参数说明:
–conf/-c:表示配置文件存储在 conf/目录
–name/-n:表示给 agent 起名为 a1
–conf-file/-f:flume 本次启动读取的配置文件是在 job 文件夹下的 flume-telnet.conf文件。
-Dflume.root.logger=INFO,console :-D 表示 flume 运行时动态修改 flume.root.logger 参数属性值,并将控制台日志打印级别设置为 INFO 级别。日志级别包括:log、info、warn、error。
[song@hadoop102 ~]$ nc localhost 44444
hello



/etc/profile.d/my_env.sh 文件,确认 Hadoop 和 Java 环境变量配置正确
[song@hadoop102 job]$ vim flume-file-hdfs.conf
注:要想读取 Linux 系统中的文件,就得按照 Linux 命令的规则执行命令。由于 Hive日志在 Linux 系统中所以读取文件的类型选择:exec 即 execute 执行的意思。表示执行Linux 命令来读取文件。
添加如下内容
# Name the components on this agent
a2.sources = r2
a2.sinks = k2
a2.channels = c2
# Describe/configure the source
a2.sources.r2.type = exec
a2.sources.r2.command = tail -F /opt/model/hive/logs/hive.log
# Describe the sink
a2.sinks.k2.type = hdfs
a2.sinks.k2.hdfs.path = hdfs://hadoop102:8020/flume/%Y%m%d/%H
#上传文件的前缀
a2.sinks.k2.hdfs.filePrefix = logs-
#是否按照时间滚动文件夹
a2.sinks.k2.hdfs.round = true
#多少时间单位创建一个新的文件夹
a2.sinks.k2.hdfs.roundValue = 1
#重新定义时间单位
a2.sinks.k2.hdfs.roundUnit = hour
#是否使用本地时间戳
a2.sinks.k2.hdfs.useLocalTimeStamp = true
#积攒多少个 Event 才 flush 到 HDFS 一次
a2.sinks.k2.hdfs.batchSize = 100
#设置文件类型,可支持压缩
a2.sinks.k2.hdfs.fileType = DataStream
#多久生成一个新的文件
a2.sinks.k2.hdfs.rollInterval = 60
#设置每个文件的滚动大小
a2.sinks.k2.hdfs.rollSize = 134217700
#文件的滚动与 Event 数量无关
a2.sinks.k2.hdfs.rollCount = 0
# Use a channel which buffers events in memory
a2.channels.c2.type = memory
a2.channels.c2.capacity = 1000
a2.channels.c2.transactionCapacity = 100i
# Bind the source and sink to the channel
a2.sources.r2.channels = c2
a2.sinks.k2.channel = c2
注意:对于所有与时间相关的转义序列,Event Header 中必须存在以 “timestamp”的 key(除非 hdfs.useLocalTimeStamp 设置为 true,此方法会使用 TimestampInterceptor 自动添加 timestamp)

[song@hadoop102 flume]$ bin/flume-ng agent --conf conf/ --name a2 --conf-file job/flume-file-hdfs.conf
[song@hadoop102 hadoop-2.7.2]$ sbin/start-dfs.sh
[song@hadoop103 hadoop-2.7.2]$ sbin/start-yarn.sh
[song@hadoop102 hive]$ bin/hive
hive (default)>


flume-dir-hdfs.confa3.sources = r3
a3.sinks = k3
a3.channels = c3
# Describe/configure the source
a3.sources.r3.type = spooldir
a3.sources.r3.spoolDir = /opt/model/flume/upload
a3.sources.r3.fileSuffix = .COMPLETED
a3.sources.r3.fileHeader = true
#忽略所有以.tmp 结尾的文件,不上传
a3.sources.r3.ignorePattern = ([^ ]*\.tmp)
# Describe the sink
a3.sinks.k3.type = hdfs
a3.sinks.k3.hdfs.path = hdfs://hadoop102:8020/flume/upload/%Y%m%d/%H
#上传文件的前缀
a3.sinks.k3.hdfs.filePrefix = upload-
#是否按照时间滚动文件夹
a3.sinks.k3.hdfs.round = true
#多少时间单位创建一个新的文件夹
a3.sinks.k3.hdfs.roundValue = 1
#重新定义时间单位
a3.sinks.k3.hdfs.roundUnit = hour
#是否使用本地时间戳
a3.sinks.k3.hdfs.useLocalTimeStamp = true
#积攒多少个 Event 才 flush 到 HDFS 一次
a3.sinks.k3.hdfs.batchSize = 100
#设置文件类型,可支持压缩
a3.sinks.k3.hdfs.fileType = DataStream
#多久生成一个新的文件
a3.sinks.k3.hdfs.rollInterval = 60
#设置每个文件的滚动大小大概是 128M
a3.sinks.k3.hdfs.rollSize = 134217700
#文件的滚动与 Event 数量无关
a3.sinks.k3.hdfs.rollCount = 0
# Use a channel which buffers events in memory
a3.channels.c3.type = memory
a3.channels.c3.capacity = 1000
a3.channels.c3.transactionCapacity = 100
# Bind the source and sink to the channel
a3.sources.r3.channels = c3
a3.sinks.k3.channel = c3

[song@hadoop102 flume]$ bin/flume-ng agent --conf conf/ --name a3 --conf-file job/flume-dir-hdfs.conf
说明:在使用 Spooling Directory Source 时,不要在监控目录中创建并持续修改文件;上传完成的文件会以.COMPLETED 结尾;被监控文件夹每 500 毫秒扫描一次文件变动。
/opt/module/flume 目录下创建 upload 文件夹[song@hadoop102 flume]$ mkdir upload
向 upload 文件夹中添加文件
[song@hadoop102 upload]$ touch song.txt
[song@hadoop102 upload]$ touch song.tmp
[song@hadoop102 upload]$ touch song.log


Exec source 适用于监控一个实时追加的文件,不能实现断点续传;Spooldir Source 适合用于同步新文件,但不适合对实时追加日志的文件进行监听并同步;而 Taildir Source 适合用于监听多个实时追加的文件,并且能够实现断点续传。

flume-taildir-hdfs.conf[song@hadoop102 job]$ vim flume-taildir-hdfs.conf
a3.sources = r3
a3.sinks = k3
a3.channels = c3
# Describe/configure the source
a3.sources.r3.type = TAILDIR
a3.sources.r3.positionFile = /opt/model/flume/tail_dir.json
a3.sources.r3.filegroups = f1 f2
a3.sources.r3.filegroups.f1 = /opt/model/flume/files/.*file.*
a3.sources.r3.filegroups.f2 = /opt/model/flume/files2/.*log.*
# Describe the sink
a3.sinks.k3.type = hdfs
a3.sinks.k3.hdfs.path = hdfs://hadoop102:8020/flume/upload2/%Y%m%d/%H
#上传文件的前缀
a3.sinks.k3.hdfs.filePrefix = upload-
#是否按照时间滚动文件夹
a3.sinks.k3.hdfs.round = true
#多少时间单位创建一个新的文件夹
a3.sinks.k3.hdfs.roundValue = 1
#重新定义时间单位
a3.sinks.k3.hdfs.roundUnit = hour
#是否使用本地时间戳
a3.sinks.k3.hdfs.useLocalTimeStamp = true
#积攒多少个 Event 才 flush 到 HDFS 一次
a3.sinks.k3.hdfs.batchSize = 100
#设置文件类型,可支持压缩
a3.sinks.k3.hdfs.fileType = DataStream
#多久生成一个新的文件
a3.sinks.k3.hdfs.rollInterval = 60
#设置每个文件的滚动大小大概是 128M
a3.sinks.k3.hdfs.rollSize = 134217700
#文件的滚动与 Event 数量无关
a3.sinks.k3.hdfs.rollCount = 0
# Use a channel which buffers events in memory
a3.channels.c3.type = memory
a3.channels.c3.capacity = 1000
a3.channels.c3.transactionCapacity = 100
# Bind the source and sink to the channel
a3.sources.r3.channels = c3
a3.sinks.k3.channel = c3

[song@hadoop102 flume]$ bin/flume-ng agent --conf conf/ --name a3 --conf-file job/flume-taildir-hdfs.conf
[song@hadoop102 flume]$ mkdir files
[song@hadoop102 files]$ echo hello >> file1.txt
[song@hadoop102 files2]$ echo song>> file2.log

Taildir 说明: Taildir Source 维护了一个 json 格式的 position File,其会定期的往 position File 中更新每个文件读取到的最新的位置,因此能够实现断点续传。Position File 的格式如下:

{"inode":2496272,"pos":12,"file":"/opt/module/flume/files/file1.txt"}
{"inode":2496275,"pos":12,"file":"/opt/module/flume/files/file2.txt"}
注:Linux 中储存文件元数据的区域就叫做 inode,每个 inode 都有一个号码,操作系统用 inode 号码来识别不同的文件,Unix/Linux 系统内部不使用文件名,而使用 inode 号码来识别文件。

Put事务流程:
Take事务

重要组件:
ChannelSelector
ChannelSelector 的作用就是选出 Event 将要被发往哪个 Channel。其共有两种类型,分别是 Replicating(复制)和 Multiplexing(多路复用)。
ReplicatingSelector 会将同一个 Event 发往所有的 Channel,Multiplexing 会根据相应的原则,将不同的 Event 发往不同的 Channel。
SinkProcessor
SinkProcessor 共 有 三 种 类 型 , 分 别 是 DefaultSinkProcessor 、LoadBalancingSinkProcessor 和 FailoverSinkProcessor

这种模式是将多个 flume 顺序连接起来了,从最初的 source 开始到最终 sink 传送的目的存储系统。此模式不建议桥接过多的 flume 数量, flume 数量过多不仅会影响传输速率,而且一旦传输过程中某个节点 flume 宕机,会影响整个传输系统。

Flume 支持将事件流向一个或者多个目的地。这种模式可以将相同数据复制到多个channel 中,或者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。

Flume支持使用将多个sink逻辑上分到一个sink组,sink组配合不同的SinkProcessor可以实现负载均衡和错误恢复的功能。

这种模式是我们最常见的,也非常实用,日常 web 应用通常分布在上百个服务器,大者甚至上千个、上万个服务器。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。
案例需求
使用 Flume-1 监控文件变动,Flume-1将变动内容传递给Flume-2,Flume-2负责存储到HDFS。同时Flume-1 将变动内容传递给 Flume-3,Flume-3 负责输出到 Local FileSystem。
需求分析:

实现步骤:
[song@hadoop102 job]$ cd group1/
[song@hadoop102 datas]$ mkdir flume3
flume-flume-hdfs 和 flume-flume-dir。编辑配置文件[song@hadoop102 group1]$ vim flume-file-flume.conf
添加如下内容
# Name the components on this agent
a1.sources = r1
a1.sinks = k1 k2
a1.channels = c1 c2
# 将数据流复制给所有 channel
a1.sources.r1.selector.type = replicating
# Describe/configure the source
a1.sources.r1.type = exec
a1.sources.r1.command = tail -F /opt/model/hive/logs/hive.log
a1.sources.r1.shell = /bin/bash -c
# Describe the sink
# sink 端的 avro 是一个数据发送者
a1.sinks.k1.type = avro
a1.sinks.k1.hostname = hadoop102
a1.sinks.k1.port = 4141
a1.sinks.k2.type = avro
a1.sinks.k2.hostname = hadoop102
a1.sinks.k2.port = 4142
# Describe the channel
a1.channels.c1.type = memory
a1.channels.c1.capacity = 1000
a1.channels.c1.transactionCapacity = 100
a1.channels.c2.type = memory
a1.channels.c2.capacity = 1000
a1.channels.c2.transactionCapacity = 100
# Bind the source and sink to the channel
a1.sources.r1.channels = c1 c2
a1.sinks.k1.channel = c1
a1.sinks.k2.channel = c2
[song@hadoop102 group1]$ vim flume-flume-hdfs.conf
添加如下内容
# Name the components on this agent
a2.sources = r1
a2.sinks = k1
a2.channels = c1
# Describe/configure the source
# source 端的 avro 是一个数据接收服务
a2.sources.r1.type = avro
a2.sources.r1.bind = hadoop102
a2.sources.r1.port = 4141
# Describe the sink
a2.sinks.k1.type = hdfs
a2.sinks.k1.hdfs.path = hdfs://hadoop102:8020/flume2/%Y%m%d/%H
#上传文件的前缀
a2.sinks.k1.hdfs.filePrefix = flume2-
#是否按照时间滚动文件夹
a2.sinks.k1.hdfs.round = true
#多少时间单位创建一个新的文件夹
a2.sinks.k1.hdfs.roundValue = 1
#重新定义时间单位
a2.sinks.k1.hdfs.roundUnit = hour
#是否使用本地时间戳
a2.sinks.k1.hdfs.useLocalTimeStamp = true
#积攒多少个 Event 才 flush 到 HDFS 一次
a2.sinks.k1.hdfs.batchSize = 100
#设置文件类型,可支持压缩
a2.sinks.k1.hdfs.fileType = DataStream
#多久生成一个新的文件
a2.sinks.k1.hdfs.rollInterval = 30
#设置每个文件的滚动大小大概是 128M
a2.sinks.k1.hdfs.rollSize = 134217700
#文件的滚动与 Event 数量无关
a2.sinks.k1.hdfs.rollCount = 0
# Describe the channel
a2.channels.c1.type = memory
a2.channels.c1.capacity = 1000
a2.channels.c1.transactionCapacity = 100
# Bind the source and sink to the channel
a2.sources.r1.channels = c1
a2.sinks.k1.channel = c1
[song@hadoop102 group1]$ vim flume-flume-dir.conf
添加如下内容
# Name the components on this agent
a3.sources = r1
a3.sinks = k1
a3.channels = c2
# Describe/configure the source
a3.sources.r1.type = avro
a3.sources.r1.bind = hadoop102
a3.sources.r1.port = 4142
# Describe the sink
a3.sinks.k1.type = file_roll
a3.sinks.k1.sink.directory = /opt/model/flume/datas/flume3
# Describe the channel
a3.channels.c2.type = memory
a3.channels.c2.capacity = 1000
a3.channels.c2.transactionCapacity = 100
# Bind the source and sink to the channel
a3.sources.r1.channels = c2
a3.sinks.k1.channel = c2
提示:输出的本地目录必须是已经存在的目录,如果该目录不存在,并不会创建新的目录。
flume-flume-dir,flume-flume-hdfs,flume-file-flume。[song@hadoop102 flume]$ bin/flume-ng agent --conf conf/ --name a3 --conf-file job/group1/flume-flume-dir.conf
[song@hadoop102 flume]$ bin/flume-ng agent --conf conf/ --name a2 --conf-file job/group1/flume-flume-hdfs.conf
[song@hadoop102 flume]$ bin/flume-ng agent --conf conf/ --name a1 --conf-file job/group1/flume-file-flume.conf
[song@hadoop102 hadoop-2.7.2]$ sbin/start-dfs.sh
[song@hadoop103 hadoop-2.7.2]$ sbin/start-yarn.sh
[song@hadoop102 hive]$ bin/hive
hive (default)>
检查 HDFS 上数据

检查/opt/model/flume3/datas/flume3 目录中数据


[song@hadoop102 job]$ cd group2/
flume-flume-console1 和 flume-flume-console2,编辑配置文件[song@hadoop102 group2]$ vim flume-netcat-flume.conf
# Name the components on this agent
a1.sources = r1
a1.channels = c1
a1.sinkgroups = g1
a1.sinks = k1 k2
# Describe/configure the source
a1.sources.r1.type = netcat
a1.sources.r1.bind = localhost
a1.sources.r1.port = 44444
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
# Describe the sink
a1.sinks.k1.type = avro
a1.sinks.k1.hostname = hadoop102
a1.sinks.k1.port = 4141
a1.sinks.k2.type = avro
a1.sinks.k2.hostname = hadoop102
a1.sinks.k2.port = 4142
# Describe the channel
a1.channels.c1.type = memory
a1.channels.c1.capacity = 1000
a1.channels.c1.transactionCapacity = 100
# Bind the source and sink to the channel
a1.sources.r1.channels = c1
a1.sinkgroups.g1.sinks = k1 k2
a1.sinks.k1.channel = c1
a1.sinks.k2.channel = c1
flume-flume-console1.conf,配置上级 Flume 输出的 Source,输出是到本地控制台,编辑配置文件[song@hadoop102 group2]$ vim flume-flume-console1.conf
# Name the components on this agent
a2.sources = r1
a2.sinks = k1
a2.channels = c1
# Describe/configure the source
a2.sources.r1.type = avro
a2.sources.r1.bind = hadoop102
a2.sources.r1.port = 4141
# Describe the sink
a2.sinks.k1.type = logger
# Describe the channel
a2.channels.c1.type = memory
a2.channels.c1.capacity = 1000
a2.channels.c1.transactionCapacity = 100
# Bind the source and sink to the channel
a2.sources.r1.channels = c1
a2.sinks.k1.channel = c1
flume-flume-console2.conf,配置上级 Flume 输出的 Source,输出是到本地控制台,编辑配置文件[song@hadoop102 group2]$ vim flume-flume-console2.conf
# Name the components on this agent
a3.sources = r1
a3.sinks = k1
a3.channels = c2
# Describe/configure the source
a3.sources.r1.type = avro
a3.sources.r1.bind = hadoop102
a3.sources.r1.port = 4142
# Describe the sink
a3.sinks.k1.type = logger
# Describe the channel
a3.channels.c2.type = memory
a3.channels.c2.capacity = 1000
a3.channels.c2.transactionCapacity = 100
# Bind the source and sink to the channel
a3.sources.r1.channels = c2
a3.sinks.k1.channel = c2
flume-flume-console2,flume-flume-console1,flume-netcat-flume。[song@hadoop102 flume]$ bin/flume-ng agent --conf conf/ --name a3 --conf-file job/group2/flume-flume-console2.conf -Dflume.root.logger=INFO,console
[song@hadoop102 flume]$ bin/flume-ng agent --conf conf/ --name a2 --conf-file job/group2/flume-flume-console1.conf -Dflume.root.logger=INFO,console
[song@hadoop102 flume]$ bin/flume-ng agent --conf conf/ --name a1 --conf-file job/group2/flume-netcat-flume.conf
$ nc localhost 44444
注:使用 jps -ml 查看 Flume 进程。

[song@hadoop102 module]$ xsync flume
在 hadoop102、hadoop103 以及 hadoop104 的/opt/model/flume/job 目录下创建一个,group3 文件夹。
[song@hadoop102 job]$ mkdir group3
[song@hadoop103 job]$ mkdir group3
[song@hadoop104 job]$ mkdir group3
[song@hadoop102 group3]$ vim flume1-logger-flume.conf
# Name the components on this agent
a1.sources = r1
a1.sinks = k1
a1.channels = c1
# Describe/configure the source
a1.sources.r1.type = exec
a1.sources.r1.command = tail -F /opt/model/group.log
a1.sources.r1.shell = /bin/bash -c
# Describe the sink
a1.sinks.k1.type = avro
a1.sinks.k1.hostname = hadoop104
a1.sinks.k1.port = 4141
# Describe the channel
a1.channels.c1.type = memory
a1.channels.c1.capacity = 1000
a1.channels.c1.transactionCapacity = 100
# Bind the source and sink to the channel
a1.sources.r1.channels = c1
a1.sinks.k1.channel = c1
[song@hadoop102 group3]$ vim flume2-netcat-flume.conf
# Name the components on this agent
a2.sources = r1
a2.sinks = k1
a2.channels = c1
# Describe/configure the source
a2.sources.r1.type = netcat
a2.sources.r1.bind = hadoop103
a2.sources.r1.port = 44444
# Describe the sink
a2.sinks.k1.type = avro
a2.sinks.k1.hostname = hadoop104
a2.sinks.k1.port = 4141
# Use a channel which buffers events in memory
a2.channels.c1.type = memory
a2.channels.c1.capacity = 1000
a2.channels.c1.transactionCapacity = 100
# Bind the source and sink to the channel
a2.sources.r1.channels = c1
a2.sinks.k1.channel = c1
[song@hadoop104 group3]$ touch flume3-flume-logger.conf
[song@hadoop104 group3]$ vim flume3-flume-logg
# Name the components on this agent
a3.sources = r1
a3.sinks = k1
a3.channels = c1
# Describe/configure the source
a3.sources.r1.type = avro
a3.sources.r1.bind = hadoop104
a3.sources.r1.port = 4141
# Describe the sink
# Describe the sink
a3.sinks.k1.type = logger
# Describe the channel
a3.channels.c1.type = memory
a3.channels.c1.capacity = 1000
a3.channels.c1.transactionCapacity = 100
# Bind the source and sink to the channel
a3.sources.r1.channels = c1
a3.sinks.k1.channel = c1
flume3-flume-logger.conf,flume2-netcat-flume.conf,flume1-logger-flume.conf[song@hadoop104 flume]$ bin/flume-ng agent --conf conf/ --name a3 --conf-file job/group3/flume3-flume-logger.conf -Dflume.root.logger=INFO,console
[song@hadoop102 flume]$ bin/flume-ng agent --conf conf/ --name a2 --conf-file job/group3/flume1-logger-flume.conf
[song@hadoop103 flume]$ bin/flume-ng agent --conf conf/ --name a1 --conf-file job/group3/flume2-netcat-flume.conf
[song@hadoop103 module]$ echo 'hello' > group.log
[song@hadoop102 flume]$ telnet hadoop102 44444
该案例中,我们以端口数据模拟日志,以是否包含”song”模拟不同类型的日志,我们需要自定义 interceptor 区分数据中是否包含”song”,将其分别发往不同的分析系统(Channel)。

3. 实现步骤
<dependency>
<groupId>org.apache.flumegroupId>
<artifactId>flume-ng-coreartifactId>
<version>1.9.0version>
dependency>
package com.song.interceptor;
import org.apache.flume.Context;
import org.apache.flume.Event;
import org.apache.flume.interceptor.Interceptor;
import java.util.ArrayList;
import java.util.List;
import java.util.Map;
public class TypeInterceptor implements Interceptor {
//声明一个存放事件的集合
private List<Event> addHeaderEvents;
@Override
public void initialize() {
//初始化存放事件的集合
addHeaderEvents = new ArrayList<>();
}
//单个事件拦截
@Override
public Event intercept(Event event) {
//1.获取事件中的头信息
Map<String, String> headers = event.getHeaders();
//2.获取事件中的 body 信息
String body = new String(event.getBody());
//3.根据 body 中是否有"song"来决定添加怎样的头信息
if (body.contains("song")) {
//4.添加头信息
headers.put("type", "first");
} else {
//4.添加头信息
headers.put("type", "second");
}
return event;
}
//批量事件拦截
@Override
public List<Event> intercept(List<Event> events) {
//1.清空集合
addHeaderEvents.clear();
//2.遍历 events
for (Event event : events) {
//3.给每一个事件添加头信息
addHeaderEvents.add(intercept(event));
}
//4.返回结果
return addHeaderEvents;
}
@Override
public void close() {
}
public static class Builder implements Interceptor.Builder {
@Override
public Interceptor build() {
return new TypeInterceptor();
}
@Override
public void configure(Context context) {
}
}
}
# Name the components on this agent
a1.sources = r1
a1.sinks = k1 k2
a1.channels = c1 c2
# Describe/configure the source
a1.sources.r1.type = netcat
a1.sources.r1.bind = localhost
a1.sources.r1.port = 44444
a1.sources.r1.interceptors = i1
a1.sources.r1.interceptors.i1.type = com.song.flume.interceptor.CustomInterceptor$Builder
a1.sources.r1.selector.type = multiplexing
a1.sources.r1.selector.header = type
a1.sources.r1.selector.mapping.first = c1
a1.sources.r1.selector.mapping.second = c2
# Describe the sink
a1.sinks.k1.type = avro
a1.sinks.k1.hostname = hadoop103
a1.sinks.k1.port = 4141
a1.sinks.k2.type=avro
a1.sinks.k2.hostname = hadoop104
a1.sinks.k2.port = 4242
# Use a channel which buffers events in memory
a1.channels.c1.type = memory
a1.channels.c1.capacity = 1000
a1.channels.c1.transactionCapacity = 100
# Use a channel which buffers events in memory
a1.channels.c2.type = memory
a1.channels.c2.capacity = 1000
a1.channels.c2.transactionCapacity = 100
# Bind the source and sink to the channel
a1.sources.r1.channels = c1 c2
a1.sinks.k1.channel = c1
a1.sinks.k2.channel = c2
为 hadoop103 上的 Flume4 配置一个avro source 和一个 logger sink。
a1.sources = r1
a1.sinks = k1
a1.channels = c1
a1.sources.r1.type = avro
a1.sources.r1.bind = hadoop103
a1.sources.r1.port = 4141
a1.sinks.k1.type = logger
a1.channels.c1.type = memory
a1.channels.c1.capacity = 1000
a1.channels.c1.transactionCapacity = 100
a1.sinks.k1.channel = c1
a1.sources.r1.channels = c1
为 hadoop104 上的 Flume3 配置一个 avro source 和一个 logger sink。
a1.sources = r1
a1.sinks = k1
a1.channels = c1
a1.sources.r1.type = avro
a1.sources.r1.bind = hadoop104
a1.sources.r1.port = 4242
a1.sinks.k1.type = logger
a1.channels.c1.type = memory
a1.channels.c1.capacity = 1000
a1.channels.c1.transactionCapacity = 100
a1.sinks.k1.channel = c1
a1.sources.r1.channels = c1
官方也提供了自定义 source 的接口:https://flume.apache.org/FlumeDeveloperGuide.html#source 根据官方说明自定义 MySource 需要继承 AbstractSource 类并实现 Configurable 和 PollableSource 接口。
实现相应方法:
使用场景:读取 MySQL 数据或者其他文件系统。
2. 需求
使用 flume 接收数据,并给每条数据添加前缀,输出到控制台。前缀可从 flume 配置文件中配置。

3. 分析

4. 编码
<dependency>
<groupId>org.apache.flumegroupId>
<artifactId>flume-ng-coreartifactId>
<version>1.9.0version>
dependency>
package com.song;
import org.apache.flume.Context;
import org.apache.flume.EventDeliveryException;
import org.apache.flume.PollableSource;
import org.apache.flume.conf.Configurable;
import org.apache.flume.event.SimpleEvent;
import org.apache.flume.source.AbstractSource;
import java.util.HashMap;
public class MySource extends AbstractSource implements Configurable, PollableSource {
//定义配置文件将来要读取的字段
private Long delay;
private String field;
//初始化配置信息
@Override
public void configure(Context context) {
delay = context.getLong("delay");
field = context.getString("field", "Hello!");
}
@Override
public Status process() throws EventDeliveryException {
try {
//创建事件头信息
HashMap<String, String> hearderMap = new HashMap<>();
//创建事件
SimpleEvent event = new SimpleEvent();
//循环封装事件
for (int i = 0; i < 5; i++) {
//给事件设置头信息
event.setHeaders(hearderMap);
//给事件设置内容
event.setBody((field + i).getBytes());
//将事件写入 channel
getChannelProcessor().processEvent(event);
Thread.sleep(delay);
}
} catch (Exception e) {
e.printStackTrace();
return Status.BACKOFF;
}
return Status.READY;
}
@Override
public long getBackOffSleepIncrement() {
return 0;
}
@Override
public long getMaxBackOffSleepInterval() {
return 0;
}
}
# Name the components on this agent
a1.sources = r1
a1.sinks = k1
a1.channels = c1
# Describe/configure the source
a1.sources.r1.type = com.song.MySource
a1.sources.r1.delay = 1000
#a1.sources.r1.field = song
# Describe the sink
a1.sinks.k1.type = logger
# Use a channel which buffers events in memory
a1.channels.c1.type = memory
a1.channels.c1.capacity = 1000
a1.channels.c1.transactionCapacity = 100
# Bind the source and sink to the channel
a1.sources.r1.channels = c1
a1.sinks.k1.channel = c1
[song@hadoop102 flume]$ pwd
/opt/model/flume
[song@hadoop102 flume]$ bin/flume-ng agent -c conf/ -f job/mysource.conf -n a1 -Dflume.root.logger=INFO,console
Sink 是完全事务性的。在从 Channel 批量删除数据之前,每个 Sink 用 Channel 启动一个事务。批量事件一旦成功写出到存储系统或下一个 Flume Agent,Sink 就利用 Channel 提交事务。事务一旦被提交,该 Channel 从自己的内部缓冲区删除事件。
Sink 组件目的地包括 hdfs、logger、avro、thrift、ipc、file、null、HBase、solr、自定义。官方提供的 Sink 类型已经很多,但是有时候并不能满足实际开发当中的需求,此时我们就需要根据实际需求自定义某些 Sink。
官方也提供了自定义 sink 的接口:https://flume.apache.org/FlumeDeveloperGuide.html#sink 根据官方说明自定义MySink 需要继承 AbstractSink 类并实现 Configurable 接口。
实现相应方法:
使用场景:读取 Channel 数据写入 MySQL 或者其他文件系统。

package com.song;
import org.apache.flume.*;
import org.apache.flume.conf.Configurable;
import org.apache.flume.sink.AbstractSink;
import org.slf4j.Logger;
import org.slf4j.LoggerFactory;
public class MySink extends AbstractSink implements Configurable
{
//创建 Logger 对象
private static final Logger LOG = LoggerFactory.getLogger(AbstractSink.class);
private String prefix;
private String suffix;
@Override
public Status process() throws EventDeliveryException {
//声明返回值状态信息
Status status;
//获取当前 Sink 绑定的 Channel
Channel ch = getChannel();
//获取事务
Transaction txn = ch.getTransaction();
//声明事件
Event event;
//开启事务
txn.begin();
//读取 Channel 中的事件,直到读取到事件结束循环
while (true) {
event = ch.take();
if (event != null) {
break;
}
}
try {
//处理事件(打印)
LOG.info(prefix + new String(event.getBody()) + suffix);
//事务提交
txn.commit();
status = Status.READY;
} catch (Exception e) {
//遇到异常,事务回滚
txn.rollback();
status = Status.BACKOFF;
} finally {
//关闭事务
txn.close();
}
return status;
}
@Override
public void configure(Context context) {
//读取配置文件内容,有默认值
prefix = context.getString("prefix", "hello:");
//读取配置文件内容,无默认值
suffix = context.getString("suffix");
}
}
# Name the components on this agent
a1.sources = r1
a1.sinks = k1
a1.channels = c1
# Describe/configure the source
a1.sources.r1.type = netcat
a1.sources.r1.bind = localhost
a1.sources.r1.port = 44444
# Describe the sink
a1.sinks.k1.type = com.song.MySink
#a1.sinks.k1.prefix = song:
a1.sinks.k1.suffix = :song
# Use a channel which buffers events in memory
a1.channels.c1.type = memory
a1.channels.c1.capacity = 1000
a1.channels.c1.transactionCapacity = 100
# Bind the source and sink to the channel
a1.sources.r1.channels = c1
a1.sinks.k1.channel = c1
[song@hadoop102 flume]$ bin/flume-ng agent -c conf/ -f job/mysink.conf -n a1 -Dflume.root.logger=INFO,console
[song@hadoop102 ~]$ nc localhost 44444
hello
OK
song
OK
Ganglia 由 gmond、gmetad 和 gweb 三部分组成。
gmond(Ganglia Monitoring Daemon)是一种轻量级服务,安装在每台需要收集指标数据的节点主机上。使用 gmond,你可以很容易收集很多系统指标数据,如 CPU、内存、磁盘、网络和活跃进程的数据等。
gmetad(Ganglia Meta Daemon)整合所有信息,并将其以 RRD 格式存储至磁盘的服务。
gweb(Ganglia Web)Ganglia 可视化工具,gweb 是一种利用浏览器显示 gmetad 所存储数据的 PHP 前端。在 Web 界面中以图表方式展现集群的运行状态下收集的多种不同指标数据。
hadoop102: web gmetad gmod
hadoop103: gmod
hadoop104: gmod
[song@hadoop102 flume]$ sudo yum -y install epel-release
[song@hadoop102 flume]$ sudo yum -y install ganglia-gmetad
[song@hadoop102 flume]$ sudo yum -y install ganglia-web
[song@hadoop102 flume]$ sudo yum -y install ganglia-gmond
[song@hadoop102 flume]$ sudo yum -y install ganglia-gmond
[song@hadoop102 flume]$ sudo vim /etc/httpd/conf.d/ganglia.conf
修改为红颜色的配置:
# Ganglia monitoring system php web frontend
#
Alias /ganglia /usr/share/ganglia
# Require local
# 通过 windows 访问 ganglia,需要配置 Linux 对应的主机(windows)ip 地址
Require ip 192.168.9.1
# Require ip 10.1.2.3
# Require host example.org
Location>
[song@hadoop102 flume]$ sudo vim /etc/ganglia/gmetad.conf
修改为:
data_source "my cluster" hadoop102
[song@hadoop102 flume]$ sudo vim /etc/ganglia/gmond.conf
修改为:
cluster {
name = "my cluster"
owner = "unspecified"
latlong = "unspecified"
url = "unspecified"
}
udp_send_channel {
#bind_hostname = yes # Highly recommended, soon to be default.
# This option tells gmond to use a source
address
# that resolves to the machine's hostname.
Without
# this, the metrics may appear to come from
any
# interface and the DNS names associated with
# those IPs will be used to create the RRDs.
# mcast_join = 239.2.11.71
# 数据发送给 hadoop102
host = hadoop102
port = 8649
ttl = 1
}
udp_recv_channel {
# mcast_join = 239.2.11.71
port = 8649
# 接收来自任意连接的数据
bind = 0.0.0.0
retry_bind = true
# Size of the UDP buffer. If you are handling lots of metrics
you really
# should bump it up to e.g. 10MB or even higher.
# buffer = 10485760
}
[song@hadoop102 flume]$ sudo vim /etc/selinux/config
修改为:
# This file controls the state of SELinux on the system.
# SELINUX= can take one of these three values:
# enforcing - SELinux security policy is enforced.
# permissive - SELinux prints warnings instead of enforcing.
# disabled - No SELinux policy is loaded.
SELINUX=disabled
# SELINUXTYPE= can take one of these two values:
# targeted - Targeted processes are protected,
# mls - Multi Level Security protection.
SELINUXTYPE=targeted
尖叫提示:selinux 生效需要重启,如果此时不想重启,可以临时生效之:
[song@hadoop102 flume]$ sudo setenforce 0
[song@hadoop102 flume]$ sudo systemctl start gmond
[song@hadoop102 flume]$ sudo systemctl start httpd
[song@hadoop102 flume]$ sudo systemctl start gmetad
http://hadoop102/ganglia
尖叫提示:如果完成以上操作依然出现权限不足错误,请修改/var/lib/ganglia 目录的权限:
[song@hadoop102 flume]$ sudo chmod -R 777 /var/lib/ganglia
[song@hadoop102 flume]$ bin/flume-ng agent -c conf/ -n a1 -f job/flume-netcat-logger.conf -Dflume.root.logger=INFO,console -Dflume.monitoring.type=ganglia -Dflume.monitoring.hosts=hadoop102:8649
[song@hadoop102 flume]$ nc localhost 44444
样式如图:

图例说明:

使用第三方框架 Ganglia 实时监控 Flume。

Channel Selectors,可以让不同的项目日志通过不同的Channel到不同的Sink中去。官方文档上Channel Selectors 有两种类型:Replicating Channel Selector (default)和Multiplexing Channel Selector
这两种Selector的区别是:Replicating 会 将source过来的events发往所有channel,而Multiplexing可以选择该发往哪些Channel。
batchSize 参数决定 Source 一次批量运输到 Channel 的 event 条数,适当调大这个参数可以提高 Source 搬运 Event 到 Channel 时的性能。
使用 file Channel 时 dataDirs 配置多个不同盘下的目录可以提高性能。
Capacity 参数决定 Channel 可容纳最大的 event 条数。transactionCapacity 参数决定每次 Source 往 channel 里面写的最大 event 条数和每次 Sink 从 channel 里面读的最大event 条数。transactionCapacity 需要大于 Source 和 Sink 的 batchSize 参数。
Flume 的事务机制(类似数据库的事务机制):Flume 使用两个独立的事务分别负责从Soucrce 到 Channel,以及从 Channel 到 Sink 的事件传递。
比如 spooling directory source 为文件的每一行创建一个事件,一旦事务中所有的事件全部传递到 Channel 且提交成功,那么 Soucrce 就将该文件标记为完成。
同理,事务以类似的方式处理从 Channel 到 Sink 的传递过程,如果因为某种原因使得事件无法记录,那么事务将会回滚。且所有的事件都会保持到 Channel 中,等待重新传递。
根据 Flume 的架构原理,Flume 是不可能丢失数据的,其内部有完善的事务机制,Source 到 Channel 是事务性的,Channel 到 Sink 是事务性的,因此这两个环节不会出现数据的丢失,唯一可能丢失数据的情况是 Channel 采用 memoryChannel,agent 宕机导致数据丢失,或者 Channel 存储数据已满,导致 Source 不再写入,未写入的数据丢失。
Flume 不会丢失数据,但是有可能造成数据的重复,例如数据已经成功由 Sink 发出,但是没有接收到响应,Sink 会再次发送数据,此时可能会导致数据的重复。