本文想记录和表达的东西挺多的,一时想不到什么好的标题,所以就用上面的关键字作为标题了。
在实时流式计算中,最重要的是在任何情况下,消息不重复、不丢失,即Exactly-once。本文以Kafka–>Spark Streaming–>Redis为例,一方面说明一下如何做到Exactly-once,另一方面说明一下我是如何计算实时去重指标的。
1. 关于数据源
数据源是文本格式的日志,由Nginx产生,存放于日志服务器上。在日志服务器上部署Flume Agent,使用TAILDIR Source和Kafka Sink,将日志采集到Kafka进行临时存储。日志格式如下:
2018-02-22T00:00:00+08:00|~|200|~|/test?pcid=DEIBAH&siteid=3
2018-02-22T00:00:00+08:00|~|200|~|/test?pcid=GLLIEG&siteid=3
2018-02-22T00:00:00+08:00|~|200|~|/test?pcid=HIJMEC&siteid=8
2018-02-22T00:00:00+08:00|~|200|~|/test?pcid=HMGBDE&siteid=3
2018-02-22T00:00:00+08:00|~|200|~|/test?pcid=HIJFLA&siteid=4
2018-02-22T00:00:01+08:00|~|200|~|/test?pcid=JCEBBC&siteid=9
2018-02-22T00:00:01+08:00|~|200|~|/test?pcid=KJLAKG&siteid=8
2018-02-22T00:00:01+08:00|~|200|~|/test?pcid=FHEIKI&siteid=3
2018-02-22T00:00:01+08:00|~|200|~|/test?pcid=IGIDLB&siteid=3
2018-02-22T00:00:01+08:00|~|200|~|/test?pcid=IIIJCD&siteid=5
日志是由测试程序模拟产生的,字段之间由|~|分隔。
2. 实时计算需求
分天、分小时PV;
分天、分小时、分网站(siteid)PV;
分天 UV;
3. Spark Streaming消费Kafka数据
http://spark.apache.org/docs/latest/streaming-kafka-0-10-integration.html
在Spark Streaming中消费Kafka数据,保证Exactly-once的核心有三点:
使用Direct方式连接Kafka;自己保存和维护Offset;更新Offset和计算在同一事务中完成;
后面的Spark Streaming程序(文章结尾),主要有以下步骤:
- 启动后,先从Redis中获取上次保存的Offset,Redis中的key为”topic_partition”,即每个分区维护一个Offset;
- 使用获取到的Offset,创建DirectStream;
- 在处理每批次的消息时,利用Redis的事务机制,确保在Redis中指标的计算和Offset的更新维护,在同一事务中完成。只有这两者同步,才能真正保证消息的Exactly-once。
- ./spark-submit \
- --class com.lxw1234.spark.TestSparkStreaming \
- --master local[2] \
- --conf spark.streaming.kafka.maxRatePerPartition=20000 \
- --jars /data1/home/dmp/lxw/realtime/commons-pool2-2.3.jar,\
- /data1/home/dmp/lxw/realtime/jedis-2.9.0.jar,\
- /data1/home/dmp/lxw/realtime/kafka-clients-0.11.0.1.jar,\
- /data1/home/dmp/lxw/realtime/spark-streaming-kafka-0-10_2.11-2.2.1.jar \
- /data1/home/dmp/lxw/realtime/testsparkstreaming.jar \
- --executor-memory 4G \
- --num-executors 1
在启动Spark Streaming程序时候,有个参数最好指定:
spark.streaming.kafka.maxRatePerPartition=20000(每秒钟从topic的每个partition最多消费的消息条数)
如果程序第一次运行,或者因为某种原因暂停了很久重新启动时候,会积累很多消息,如果这些消息同时被消费,很有可能会因为内存不够而挂掉,因此,需要根据实际的数据量大小,以及批次的间隔时间来设置该参数,以限定批次的消息量。
如果该参数设置20000,而批次间隔时间未10秒,那么每个批次最多从Kafka中消费20万消息。
4. Redis中的数据模型
- 分小时、分网站PV
普通K-V结构,计算时候使用incr命令递增,
Key为 “site_pv_网站ID_小时”,
如:site_pv_9_2018-02-21-00、site_pv_10_2018-02-21-01
该数据模型用于计算分网站的按小时及按天PV。
- 分小时PV
普通K-V结构,计算时候使用incr命令递增,
Key为“pv_小时”,如:pv_2018-02-21-14、pv_2018-02-22-03
该数据模型用于计算按小时及按天总PV。
- 分天UV
Set结构,计算时候使用sadd命令添加,
Key为”uv_天”,如:uv_2018-02-21、uv_2018-02-20
该数据模型用户计算按天UV(获取时候使用SCARD命令获取Set元素个数)
注:这些Key对应的时间,均由实际消息中的第一个字段(时间)而定。
5. 故障恢复
如果Spark Streaming程序因为停电、网络等意外情况终止而需要恢复,则直接重启即可;
如果因为其他原因需要重新计算某一时间段的消息,可以先删除Redis中对应时间段内的Key,然后从原始日志中截取该时间段内的消息,当做新消息添加至Kafka,由Spark Streaming程序重新消费并进行计算;
6. 附程序
依赖jar包:
commons-pool2-2.3.jar
jedis-2.9.0.jar
kafka-clients-0.11.0.1.jar
spark-streaming-kafka-0-10_2.11-2.2.1.jar
InternalRedisClient (Redis链接池)
- package com.lxw1234.spark
- import redis.clients.jedis.JedisPool
- import org.apache.commons.pool2.impl.GenericObjectPoolConfig
- /**
- * @author lxw1234
- */
- /**
- * Internal Redis client for managing Redis connection {@link Jedis} based on {@link RedisPool}
- */
- object InternalRedisClient extends Serializable {
- @transient private var pool: JedisPool = null
- def makePool(redisHost: String, redisPort: Int, redisTimeout: Int,
- maxTotal: Int, maxIdle: Int, minIdle: Int): Unit = {
- makePool(redisHost, redisPort, redisTimeout, maxTotal, maxIdle, minIdle, true, false, 10000)
- }
- def makePool(redisHost: String, redisPort: Int, redisTimeout: Int,
- maxTotal: Int, maxIdle: Int, minIdle: Int, testOnBorrow: Boolean,
- testOnReturn: Boolean, maxWaitMillis: Long): Unit = {
- if(pool == null) {
- val poolConfig = new GenericObjectPoolConfig()
- poolConfig.setMaxTotal(maxTotal)
- poolConfig.setMaxIdle(maxIdle)
- poolConfig.setMinIdle(minIdle)
- poolConfig.setTestOnBorrow(testOnBorrow)
- poolConfig.setTestOnReturn(testOnReturn)
- poolConfig.setMaxWaitMillis(maxWaitMillis)
- pool = new JedisPool(poolConfig, redisHost, redisPort, redisTimeout)
- val hook = new Thread{
- override def run = pool.destroy()
- }
- sys.addShutdownHook(hook.run)
- }
- }
- def getPool: JedisPool = {
- assert(pool != null)
- pool
- }
- }
TestSparkStreaming
- package com.lxw1234.spark
- import org.apache.kafka.clients.consumer.ConsumerRecord
- import org.apache.kafka.common.TopicPartition
- import org.apache.kafka.common.serialization.StringDeserializer
- import org.apache.spark.SparkConf
- import org.apache.spark.rdd.RDD
- import org.apache.spark.streaming.Seconds
- import org.apache.spark.streaming.StreamingContext
- import org.apache.spark.streaming.kafka010.ConsumerStrategies
- import org.apache.spark.streaming.kafka010.HasOffsetRanges
- import org.apache.spark.streaming.kafka010.KafkaUtils
- import org.apache.spark.streaming.kafka010.LocationStrategies
- import redis.clients.jedis.Pipeline
- /**
- * @author lxw1234
- * 获取topic最小的offset
- * ./kafka-run-class.sh kafka.tools.GetOffsetShell --broker-list datadev1:9092 --topic lxw1234 --time -2
- */
- object TestSparkStreaming {
- def main(args : Array[String]) : Unit = {
- val brokers = "datadev1:9092"
- val topic = "lxw1234"
- val partition : Int = 0 //测试topic只有一个分区
- val start_offset : Long = 0l
- //Kafka参数
- val kafkaParams = Map[String, Object](
- "bootstrap.servers" -> brokers,
- "key.deserializer" -> classOf[StringDeserializer],
- "value.deserializer" -> classOf[StringDeserializer],
- "group.id" -> "exactly-once",
- "enable.auto.commit" -> (false: java.lang.Boolean),
- "auto.offset.reset" -> "none"
- )
- // Redis configurations
- val maxTotal = 10
- val maxIdle = 10
- val minIdle = 1
- val redisHost = "172.16.213.79"
- val redisPort = 6379
- val redisTimeout = 30000
- //默认db,用户存放Offset和pv数据
- val dbDefaultIndex = 8
- InternalRedisClient.makePool(redisHost, redisPort, redisTimeout, maxTotal, maxIdle, minIdle)
- val conf = new SparkConf().setAppName("TestSparkStreaming").setIfMissing("spark.master", "local[2]")
- val ssc = new StreamingContext(conf, Seconds(10))
- //从Redis获取上一次存的Offset
- val jedis = InternalRedisClient.getPool.getResource
- jedis.select(dbDefaultIndex)
- val topic_partition_key = topic + "_" + partition
- var lastOffset = 0l
- val lastSavedOffset = jedis.get(topic_partition_key)
- if(null != lastSavedOffset) {
- try {
- lastOffset = lastSavedOffset.toLong
- } catch {
- case ex : Exception => println(ex.getMessage)
- println("get lastSavedOffset error, lastSavedOffset from redis [" + lastSavedOffset + "] ")
- System.exit(1)
- }
- }
- InternalRedisClient.getPool.returnResource(jedis)
- println("lastOffset from redis -> " + lastOffset)
- //设置每个分区起始的Offset
- val fromOffsets = Map{new TopicPartition(topic, partition) -> lastOffset}
- //使用Direct API 创建Stream
- val stream = KafkaUtils.createDirectStream[String, String](
- ssc,
- LocationStrategies.PreferConsistent,
- ConsumerStrategies.Assign[String, String](fromOffsets.keys.toList, kafkaParams, fromOffsets)
- )
- //开始处理批次消息
- stream.foreachRDD {
- rdd =>
- val offsetRanges = rdd.asInstanceOf[HasOffsetRanges].offsetRanges
- val result = processLogs(rdd)
- println("=============== Total " + result.length + " events in this batch ..")
- val jedis = InternalRedisClient.getPool.getResource
- val p1 : Pipeline = jedis.pipelined();
- p1.select(dbDefaultIndex)
- p1.multi() //开启事务
- //逐条处理消息
- result.foreach {
- record =>
- //增加小时总pv
- val pv_by_hour_key = "pv_" + record.hour
- p1.incr(pv_by_hour_key)
- //增加网站小时pv
- val site_pv_by_hour_key = "site_pv_" + record.site_id + "_" + record.hour
- p1.incr(site_pv_by_hour_key)
- //使用set保存当天的uv
- val uv_by_day_key = "uv_" + record.hour.substring(0, 10)
- p1.sadd(uv_by_day_key, record.user_id)
- }
- //更新Offset
- offsetRanges.foreach { offsetRange =>
- println("partition : " + offsetRange.partition + " fromOffset: " + offsetRange.fromOffset + " untilOffset: " + offsetRange.untilOffset)
- val topic_partition_key = offsetRange.topic + "_" + offsetRange.partition
- p1.set(topic_partition_key, offsetRange.untilOffset + "")
- }
- p1.exec();//提交事务
- p1.sync();//关闭pipeline
- InternalRedisClient.getPool.returnResource(jedis)
- }
- case class MyRecord(hour: String, user_id: String, site_id: String)
- def processLogs(messages: RDD[ConsumerRecord[String, String]]) : Array[MyRecord] = {
- messages.map(_.value()).flatMap(parseLog).collect()
- }
- //解析每条日志,生成MyRecord
- def parseLog(line: String): Option[MyRecord] = {
- val ary : Array[String] = line.split("\\|~\\|", -1);
- try {
- val hour = ary(0).substring(0, 13).replace("T", "-")
- val uri = ary(2).split("[=|&]",-1)
- val user_id = uri(1)
- val site_id = uri(3)
- return Some(MyRecord(hour,user_id,site_id))
- } catch {
- case ex : Exception => println(ex.getMessage)
- }
- return None
- }
- ssc.start()
- ssc.awaitTermination()
- }
- }
如果觉得本博客对您有帮助,请 赞助作者 。