Flume - Error while trying to hflushOrSync异常处理

Eights

发布于 2020-07-10 14:32:39

1.4K0

发布于 2020-07-10 14:32:39

文章被收录于专栏：Eights做数据Eights做数据

文档编写目的

记录Flume的Error while trying to hflushOrSync异常处理流程

集群环境

CDH5.16.2

问题描述

业务场景

采用Flume消费Kafka的一个topic下沉到HDFS目录，作为一个Hive的外部表

Flume agent的配置

tier1.sources = testSource

#flume测试配置
#source
tier1.sources.testSource.channels = testChannel
tier1.sources.testSource.type=org.apache.flume.source.kafka.KafkaSource
tier1.sources.testSource.kafka.bootstrap.servers=xx1:9092,xx2:9092,xx3:9092
tier1.sources.testSource.kafka.topics=test-data
tier1.sources.testSource.kafka.consumer.group.id=flume-test-data
tier1.sources.testSource.kafka.batchSize = 50000
tier1.sources.testSource.kafka.batchDurationMillis = 2000

#channel
tier1.channels.testChannel.type=memory
tier1.channels.testChannel.capacity=150000
tier1.channels.testChannel.transactionCapacity=50000
    
#sink
tier1.sinks.testSink.type=hdfs
tier1.sinks.testSink.channel = testChannel
tier1.sinks.testSink.hdfs.path=hdfs://nameservice/test-data/%Y/%m/%d
tier1.sinks.testSink.hdfs.filePrefix = uaes
tier1.sinks.testSink.hdfs.writeFormat=Text
tier1.sinks.testSink.hdfs.fileType=DataStream
tier1.sinks.testSink.hdfs.callTimeout=60000
tier1.sinks.testSink.hdfs.batchSize = 10000
tier1.sinks.testSink.hdfs.rollSize = 536750000
tier1.sinks.testSink.hdfs.rollCount = 0
tier1.sinks.testSink.hdfs.rollInterval = 7200

报错信息

问题排查

通过查看不同flume的agent日志发现，同名的文件被不同的flume agent打开，在文件第二次打开后，先前打开的agent拥有的token就失效了，因此无法关闭它，就会不断的报错：Error while trying to hflushOrSync!

查找到的不同flume agent的日志如下：

flume-agent1:
WARN org.apache.flume.sink.hdfs.BucketWriter:
Closing file: hdfs://nameservice/test-data/2020/04/29/uaes.1588089601760.tmp failed.
Will retry again in 180 seconds.
flume-agent2:
WARN org.apache.flume.sink.hdfs.BucketWriter:
Closing file: hdfs://nameservice/test-data/2020/04/29/uaes.1588089601760.tmp failed.
Will retry again in 180 seconds.

查看之前的flume配置文件发现，每一个flume-agent配置的hdfsSink是完全一样的，每个flume-agent读取的source相同，有很大概率会出现多个flume-agent同时写同名文件，导致部分flume-agent无法继续。

#主要是这个配置导致的写同名文件
tier1.sinks.testSink.hdfs.filePrefix = uaes

解决方案

使用Flume的Host拦截器为各个agent处理的文件打上主机信息。

flume的host拦截器使用官方连接：

https://flume.apache.org/FlumeUserGuide.html#host-interceptor

样例配置：

a1.sources = r1
a1.channels = c1
#配置拦截器
a1.sources.r1.interceptors = i1
#给定拦截器类型
a1.sources.r1.interceptors.i1.type = host
#配置ip还是host
a1.sources.r1.interceptors.i1.useIp = false

修改hdfs Sink的filePrefix配置，添加host，主要是添加一个flume的interceptors去添加flume-agent机器的ip，其他配置不变

# 核心配置项
# source增加配置
tier1.sources.uaesSource.interceptors = testincp
tier1.sources.uaesSource.interceptors.testincp.type = host

# sink增加配置
tier1.sinks.uaesSink.hdfs.filePrefix = %{host}

修改flume-agent的配置后，刷新配置，运行一段时间查看日志，故障消失。