实时事件统计项目：优化flume:用file channel代替mem channel

Posted 2020-08-25 Code Job

背景:利用kafka+flume+morphline+solr做实时统计。

solr从12月23号开始一直没有数据。查看日志发现,因为有一个同事加了一条格式错误的埋点数据，导致大量error。

据推断，是因为使用mem channel占满，消息来不及处理，导致新来的数据都丢失了。

修改flume使用file channel:

kafka2solr.sources = source_from_kafka
kafka2solr.channels = file_channel
kafka2solr.sinks = solrSink

# For each one of the sources, the type is defined  
kafka2solr.sources.source_from_kafka.type = org.apache.flume.source.kafka.KafkaSource
kafka2solr.sources.source_from_kafka.channels = file_channel
kafka2solr.sources.source_from_kafka.batchSize = 100
kafka2solr.sources.source_from_kafka.useFlumeEventFormat=false
kafka2solr.sources.source_from_kafka.kafka.bootstrap.servers= kafkanode0:9092,kafkanode1:9092,kafkanode2:9092
kafka2solr.sources.source_from_kafka.kafka.topics = eventCount
kafka2solr.sources.source_from_kafka.kafka.consumer.group.id = flume_solr_caller
kafka2solr.sources.source_from_kafka.kafka.consumer.auto.offset.reset=latest

# file channel  
kafka2solr.channels.file_channel.type = file
kafka2solr.channels.file_channel.checkpointDir = /var/log/flume-ng/checkpoint
kafka2solr.channels.file_channel.dataDirs = /var/log/flume-ng/data


kafka2solr.sinks.solrSink.type = org.apache.flume.sink.solr.morphline.MorphlineSolrSink
kafka2solr.sinks.solrSink.channel = file_channel
#kafka2solr.sinks.solrSink.batchSize = 1000
#kafka2solr.sinks.solrSink.batchDurationMillis = 1000
kafka2solr.sinks.solrSink.morphlineFile = morphlines.conf
kafka2solr.sinks.solrSink.morphlineId=morphline1
kafka2solr.sinks.solrSink.isIgnoringRecoverableExceptions=true

使得数据持久化到磁盘不会丢失。