水槽加载 csv 文件优于 hdfs 接收器

Question

我已将我的 Flume 源配置为 Spooldir 类型。我有很多 CSV 文件，.xl3 和 .xls，我希望我的 Flume 代理将所有文件从 spooldir 加载到 HDFS 接收器。然而水槽代理返回异常

这是我对水槽源的配置：

agent.sources.s1.type = spooldir
agent.sources.s1.spoolDir = /my-directory
agent.sources.s1.basenameHeader = true
agent.sources.batchSize = 10000

还有我的 HDFS 接收器：

agent.sinks.sk1.type = hdfs 
agent.sinks.sk1.hdfs.path = hdfs://...:8020/user/importflume/%Y/%m/%d/%H 
agent.sinks.sk1.hdfs.filePrefix = %{basename}
agent.sinks.sk1.hdfs.rollSize = 0
agent.sinks.sk1.hdfs.rollCount = 0
agent.sinks.sk1.hdfs.useLocalTimeStamp = true
agent.sinks.sk1.hdfs.batchsize =    10000
agent.sinks.sk1.hdfs.fileType = DataStream
agent.sinks.sk1.serializer = avro_event
agent.sinks.sk1.serializer.compressionCodec = snappy

Answer 1

您可以将以下配置用于假脱机目录。只需在以下配置中提供本地文件系统和 HDFS 位置的路径即可。

#Flume Configuration Starts
# Define a file channel called fileChannel on agent1
agent1.channels.fileChannel1_1.type = file 
# on linux FS
agent1.channels.fileChannel1_1.capacity = 200000
agent1.channels.fileChannel1_1.transactionCapacity = 1000
# Define a source for agent1
agent1.sources.source1_1.type = spooldir
# on linux FS
#Spooldir in my case is /home/hadoop/Desktop/flume_sink
agent1.sources.source1_1.spoolDir = 'path'
agent1.sources.source1_1.fileHeader = false
agent1.sources.source1_1.fileSuffix = .COMPLETED
agent1.sinks.hdfs-sink1_1.type = hdfs

#Sink is /flume_import under hdfs

agent1.sinks.hdfs-sink1_1.hdfs.path = hdfs://'path'
agent1.sinks.hdfs-sink1_1.hdfs.batchSize = 1000
agent1.sinks.hdfs-sink1_1.hdfs.rollSize = 268435456
agent1.sinks.hdfs-sink1_1.hdfs.rollInterval = 0
agent1.sinks.hdfs-sink1_1.hdfs.rollCount = 50000000
agent1.sinks.hdfs-sink1_1.hdfs.writeFormat=Text

agent1.sinks.hdfs-sink1_1.hdfs.fileType = DataStream
agent1.sources.source1_1.channels = fileChannel1_1
agent1.sinks.hdfs-sink1_1.channel = fileChannel1_1

agent1.sinks =  hdfs-sink1_1
agent1.sources = source1_1
agent1.channels = fileChannel1_1

您也可以参考 Flume spool 目录上的 this blog 了解更多信息。