概述

kafka是一个开源社区常用的消息队列，虽然kafka官方（Confluent公司）提供插件从Kafka直接导入数据到HDFS的connector，但对阿里云对文件存储系统OSS却没有官方的支持。本文会举一个简单的例子，实现kafka的数据写入阿里云OSS。因为阿里云E-MapReduce服务集成了大量开源组件和阿里云产品的对接工具，所以本文直接在E-MapReduce集群上运行这个例子。

这个例子使用开源的Flume工具作为中转，将kafka和OSS连接起来。Flume开源组件将来也可能出现在E-MapReduce平台上。

场景举例

下面举一个最简单的例子，如果已经有一个线上的Kafka集群，则可以直接跳到第4步。

1. 在Kafka Home目录下启动Kafka服务进程，配置文件中Zookeeper的地址配置为E-MapReduce自带的服务地址 emr-header-1:2181
   bin/kafka-server-start.sh config/server.properties
2. 创建一个Kafka的topic，名字为test
   bin/kafka-topics.sh --create --zookeeper emr-header-1:2181 --replication-factor 1 --partitions 1 --topic test
3. 向Kafka test topic内写入数据，数据内容为本机的性能监控数据
   vmstat 1 | bin/kafka-console-producer.sh --broker-list emr-header-1:9092 --topic test
4. 在Flume Home目录下配置并启动Flume服务

新建一个配置文件：conf/kafka-example.conf。其中source指定为kafka的对应topic，sink使用HDFS Sinker，并且路径指定为OSS的路径。因为E-MapReduce服务为我们实现了一个高效的OSS FileSystem（兼容Hadoop FileSystem），所以可以直接指定OSS路径，HDFS Sinker自动将数据写入OSS。

# Name the components on this agent
a1.sources = source1
a1.sinks = oss1
a1.channels = c1

# Describe/configure the source
a1.sources.source1.type = org.apache.flume.source.kafka.KafkaSource
a1.sources.source1.zookeeperConnect = emr-header-1:2181
a1.sources.source1.topic = test
a1.sources.source1.groupId = flume
a1.sources.source1.channels = c1
a1.sources.source1.interceptors = i1
a1.sources.source1.interceptors.i1.type = timestamp
a1.sources.source1.kafka.consumer.timeout.ms = 100

# Describe the sink
a1.sinks.oss1.type = hdfs
a1.sinks.oss1.hdfs.path = oss://emr-examples/kafka/%{topic}/%y-%m-%d
a1.sinks.oss1.hdfs.rollInterval = 10
a1.sinks.oss1.hdfs.rollSize = 0
a1.sinks.oss1.hdfs.rollCount = 0
a1.sinks.oss1.hdfs.fileType = DataStream

# Use a channel which buffers events in memory
a1.channels.c1.type = memory
a1.channels.c1.capacity = 100000
a1.channels.c1.transactionCapacity = 10000

# Bind the source and sink to the channel
a1.sources.r1.channels = c1
a1.sinks.oss1.channel = c1

启动Flume服务：

bin/flume-ng agent --conf conf --conf-file conf/kafka-example.conf --name a1 -Dflume.root.logger=INFO,console --classpath '/usr/lib/hadoop-current/share/hadoop/tools/lib/*'

从日志中可以看到Flume HDFS sinker将数据写到了OSS，并且是每10秒钟轮转一次。

2016-12-05 18:41:04,794 (hdfs-oss1-call-runner-1) [INFO - org.apache.flume.sink.hdfs.BucketWriter$8.call(BucketWriter.java:618)] Renaming oss://emr-perform/kafka/test/16-12-05/Flume
Data.1480934454657.tmp to oss://emr-perform/kafka/test/16-12-05/FlumeData.1480934454657
2016-12-05 18:41:04,852 (hdfs-oss1-roll-timer-0) [INFO - org.apache.flume.sink.hdfs.HDFSEventSink$1.run(HDFSEventSink.java:382)] Writer callback called.

备注：如果遇到如下 Exception，是因为Flume自带的httpclient jar包和EMR冲突：

Exception in thread "SinkRunner-PollingRunner-DefaultSinkProcessor" java.lang.NoSuchFieldError: INSTANCE

通过删除Flume自带的 httpclient jar 包可以避免冲突（统一使用EMR Hadoop带的httpclient）：
rm ~/apache-flume-1.8.0-bin/lib/httpclient-4.2.1.jar

查看OSS上的结果

$ hadoop fs -ls oss://emr-examples/kafka/test/16-12-05/
Found 6 items
-rw-rw-rw-   1     162691 2016-12-05 18:40 oss://emr-examples/kafka/test/16-12-05/FlumeData.1480934394566
-rw-rw-rw-   1        925 2016-12-05 18:40 oss://emr-examples/kafka/test/16-12-05/FlumeData.1480934407580
-rw-rw-rw-   1       1170 2016-12-05 18:40 oss://emr-examples/kafka/test/16-12-05/FlumeData.1480934418597
-rw-rw-rw-   1       1092 2016-12-05 18:40 oss://emr-examples/kafka/test/16-12-05/FlumeData.1480934430613
-rw-rw-rw-   1       1254 2016-12-05 18:40 oss://emr-examples/kafka/test/16-12-05/FlumeData.1480934443638
-rw-rw-rw-   1        588 2016-12-05 18:41 oss://emr-examples/kafka/test/16-12-05/FlumeData.1480934454657

$ hadoop fs -cat oss://emr-examples/kafka/test/16-12-05/FlumeData.1480934443638 
 0  0      0 1911216  50036 1343828    0    0     0     0 1341 2396  1  1 98  0  0
 0  0      0 1896964  50052 1343824    0    0     0   112 1982 2511 15  1 84  0  0
 1  0      0 1896552  50052 1343828    0    0     0    76 2314 3329  3  4 94  0  0
procs -----------memory---------- ---swap-- -----io---- --system-- -----cpu-----
 r  b   swpd   free   buff  cache   si   so    bi    bo   in   cs us sy id wa st
 5  0      0 1903016  50052 1343828    0    0     0     0 2277 3249  2  4 94  0  0
 0  0      0 1902892  50052 1343828    0    0     0     0 1417 2366  5  0 95  0  0
 0  0      0 1902892  50052 1343828    0    0     0     0 1072 2243  0  0 99  0  0
 0  0      0 1902892  50068 1343824    0    0     0   144 1275 2283  1  0 99  0  0
 1  0      0 1903024  50068 1343828    0    0     0    24 1099 2071  1  1 99  0  0
 0  0      0 1903272  50068 1343832    0    0     0     0 1294 2238  1  1 99  0  0
 1  0      0 1903412  50068 1343832    0    0     0     0 1024 2094  1  0 99  0  0
 2  0      0 1903148  50076 1343836    0    0     0    68 1879 2766  1  1 98  0  0
 1  0      0 1903288  50092 1343840    0    0     0    92 1147 2240  1  0 99  0  0
 0  0      0 1902792  50092 1343844    0    0     0    28 1456 2388  1  1 98  0  0

参考资料

1. http://kafka.apache.org/quickstart
2. https://www.cloudera.com/documentation/kafka/latest/topics/kafka_flume.html