Flume对Nginx群集日志收集方案

Flume是一个分布式、可靠、和高可用的海量日志聚合的系统，支持在系统中定制各类数据发送方，用于收集数据；同时，Flume提供对数据进行简单处理，并写到各种数据接受方（可定制）的能力。

flume的基本概念
如下图

每个agent都具有三个元素，source、channel、sink。
source：数据流的源。产生event。
channel：可以理解成数据流的管道。传递event
sink :  数据流的终点。消耗event
注：source可以上一节点的sink，sink可以指定为下一节点的source。比较常见的场景如下图

为以收集日志，并做实时的集中存储，元素相应类型如下
1.source ： client端使用exec类型，通过tail –F 产生event。server端使用avro类型，用于接收client端发出的event
2.channel ： 使用file类型。(测试期使用了mem类似)
3. sink: client端使用avro类型，传递给server端。server端使用file_roll类型，指定相应目录储存日志。最终方案会使用hdfs

flume具体配置如下：
1、将flume 解压到 /usr/local/flume下
2、agent端配置

flume允许对环境资源使用做设置，需要修改配置，/PREFIX/conf/flume-env.sh  可以通过实际情况进行调整
JAVA_OPTS="-Xms100m -Xmx200m -Dcom.sun.management.jmxremote"
#此处PREFIX 代表 /usr/local/flume

配置（/PREFIX/conf/flume-client.properties）
# Name the components on this agent 
a1.sources = r1 
a1.channels = c1 
a1.sinks = k1

# Describe/configure the source 
a1.sources.r1.type = exec 
a1.sources.r1.command = tail -n 0 -F /space/nginxlogs/access.web.log 
a1.sources.r1.channels = c1

# Describe/configure the channels (后面有memory channel配置方案)
a1.channels.c1.type = file
a1.channels.c1.keep-alive = 10
a1.channels.c1.write-timeout = 10
a1.channels.c1.checkpointDir = /space/guang_mobileapi_flume/checkpoint
a1.channels.c1.useDualCheckpoints = true
a1.channels.c1.backupCheckpointDir =  /space/guang_mobileapi_flume/backupcheckpoint
a1.channels.c1.dataDirs = /space/guang_mobileapi_flume/data

# Describe/configure the sink 
a1.sinks.k1.type = avro 
a1.sinks.k1.hostname = 192.168.10.35 
a1.sinks.k1.port = 44444 
a1.sinks.k1.channel = c1

（# Describe/configure the channels  （次方案可以替换前面的file channel配置方案） 
a1.channels.c1.type = memory 
a1.channels.c1.capacity = 1000 
a1.channels.c1.transactionCapacity = 100
）

server配置：

# Name the components on this agent
a1.sources = r1
a1.sinks = k1
a1.channels = c1

# Describe/configure the source
a1.sources.r1.type = avro
a1.sources.r1.channels = c1
a1.sources.r1.bind = 0.0.0.0
a1.sources.r1.port = 44444

# Describe the sink
a1.sinks.k1.type = hdfs
a1.sinks.k1.hdfs.path = /logs/web/nginx/%Y%m%d
a1.sinks.k1.hdfs.rollInterval = 3600 
a1.sinks.k1.hdfs.fileType = DataStream 
a1.sinks.k1.hdfs.rollSize = 0 
a1.sinks.k1.hdfs.rollCount = 0
a1.sinks.k1.hdfs.useLocalTimeStamp = true 
a1.sinks.k1.hdfs.filePrefix = access.guang.j.cn
a1.sinks.k1.hdfs.inUseSuffix = .tmp
a1.sinks.k1.hdfs.idleTimeout = 300

# Use a channel which buffers events in memory
a1.channels.c1.type = file
a1.channels.c1.write-timeout = 10
a1.channels.c1.keep-alive = 10
a1.channels.c1.checkpointDir = /space/flume/checkpoint
a1.channels.c1.useDualCheckpoints = true
a1.channels.c1.backupCheckpointDir =  /space/flume/backupcheckpoint
a1.channels.c1.dataDirs = /space/flume/data

# Bind the source and sink to the channel
a1.sources.r1.channels = c1
a1.sinks.k1.channel = c1

(
# Describe the sink （前面面有hdfs方式） 
a1.sinks.k1.type = file_roll 
a1.sinks.k1.sink.directory = /tmp/test 
a1.sinks.k1.sink.rollInterval = 3600

# Use a channel which buffers events in memory  （前面有file channel配置方案） 
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
)

以上配置1小时自动释放一次。可理解成每小时截断一次。因为Hadoop目录使用日志变量，在某文件空闲5分钟后自己释放。

启动(client 和server端只有配置文件不同)
cd /PREFIX
./flume-ng agent -n a1 -c ../conf -f ../conf/flume-XXXXX.properties    2>/dev/null  &