1. 解析参数工具类(ParameterTool)
该类提供了从不同数据源读取和解析程序参数的简单实用方法,其解析args时,只能支持单只参数。
-
用来解析main方法传入参数的工具类
public class ParseArgsKit { public static void main(String[] args) { ParameterTool parameters = ParameterTool.fromArgs(args); String host = parameters.getRequired("redis.host"); String port = parameters.getRequired("redis.port"); System.out.println(host); System.out.println(port); } }
参数的输入格式如下:
这种解析程序参数的的优点是参数不需要按照顺序指定,但若是参数过多的话,写起来不方便,这时我们可以选择使用解析配置文件的工具类
- 用来解析配置文件的工具类,该配置文件的路径自己指定
public class ParseArgsKit { public static void main(String[] args) throws IOException { ParameterTool parameters = ParameterTool.fromPropertiesFile("E:\flink\conf.properties"); String host = parameters.getRequired("redis.host"); String port = parameters.getRequired("redis.port"); System.out.println(host); System.out.println(port); } }
配置文件conf.properties
redis.host=feng05 redis.port=4444
2. Flink工具类封装(创建KafkaSource)
RealtimeETL
package cn._51doit.flink.day06; import cn._51doit.flink.Utils.FlinkUtils; import org.apache.flink.api.common.serialization.SimpleStringSchema; import org.apache.flink.api.java.utils.ParameterTool; import org.apache.flink.streaming.api.datastream.DataStream; public class RealtimeETL { public static void main(String[] args) throws Exception { ParameterTool parameters = ParameterTool.fromPropertiesFile("E:\flink\conf.properties"); //使用Flink拉取Kafka中的数据,对数据进行清洗、过滤整理 DataStream<String> lines = FlinkUtils.createKafkaStream(parameters, SimpleStringSchema.class); lines.print(); FlinkUtils.env.execute(); } }
FlinkUtils
package cn._51doit.flink.Utils; import org.apache.flink.api.common.restartstrategy.RestartStrategies; import org.apache.flink.api.common.serialization.DeserializationSchema; import org.apache.flink.api.java.utils.ParameterTool; import org.apache.flink.runtime.state.filesystem.FsStateBackend; import org.apache.flink.streaming.api.CheckpointingMode; import org.apache.flink.streaming.api.datastream.DataStream; import org.apache.flink.streaming.api.environment.CheckpointConfig; import org.apache.flink.streaming.api.environment.StreamExecutionEnvironment; import org.apache.flink.streaming.connectors.kafka.FlinkKafkaConsumer; import java.io.IOException; import java.util.Arrays; import java.util.List; import java.util.Properties; public class FlinkUtils { public static final StreamExecutionEnvironment env = StreamExecutionEnvironment.getExecutionEnvironment(); public static <T> DataStream<T> createKafkaStream(ParameterTool parameters, Class<? extends DeserializationSchema<T>> clazz) throws IOException, IllegalAccessException, InstantiationException { // 设置checkpoint的间隔时间 env.enableCheckpointing(parameters.getLong("checkpoint.interval",300000)); env.getCheckpointConfig().setCheckpointingMode(CheckpointingMode.AT_LEAST_ONCE); //就是将job cancel后,依然保存对应的checkpoint数据 env.getCheckpointConfig().enableExternalizedCheckpoints(CheckpointConfig.ExternalizedCheckpointCleanup.RETAIN_ON_CANCELLATION); String checkPointPath = parameters.get("checkpoint.path"); if(checkPointPath != null){ env.setStateBackend(new FsStateBackend(checkPointPath)); } int restartAttempts = parameters.getInt("restart.attempts", 30); int delayBetweenAttempts = parameters.getInt("delay.between.attempts", 30000); env.setRestartStrategy(RestartStrategies.fixedDelayRestart(restartAttempts, delayBetweenAttempts)); Properties properties = parameters.getProperties(); String topics = parameters.getRequired("kafka.topics"); List<String> topicList = Arrays.asList(topics.split(",")); FlinkKafkaConsumer<T> flinkKafkaConsumer = new FlinkKafkaConsumer<T>(topicList, clazz.newInstance(), properties); //在Checkpoint的时候将Kafka的偏移量不保存到Kafka特殊的Topic中,默认是true flinkKafkaConsumer.setCommitOffsetsOnCheckpoints(false); return env.addSource(flinkKafkaConsumer); } }
此处的重点是FlinkKafkaConsumer这个类的使用,下图显示的是其中一种构造方法
参数一:topic名或 topic名的列表
Flink Kafka Consumer 需要知道如何将来自Kafka的二进制数据转换为Java/Scala对象。DeserializationSchema接口允许程序员指定这个序列化的实现。该接口的 T deserialize(byte[]message) 会在收到每一条Kafka的消息的时候被调用。我们通常会实现 AbstractDeserializationSchema,它可以描述被序列化的Java/Scala类型到Flink的类型(TypeInformation)的映射。如果用户的代码实现了DeserializationSchema,那么就需要自己实现getProducedType(...) 方法。
为了方便使用,Flink提供了一些已实现的schema:
(1) TypeInformationSerializationSchema (andTypeInformationKeyValueSerializationSchema) ,他们会基于Flink的TypeInformation来创建schema。这对于那些从Flink写入,又从Flink读出的数据是很有用的。这种Flink-specific的反序列化会比其他通用的序列化方式带来更高的性能。
(2)JsonDeserializationSchema (andJSONKeyValueDeserializationSchema) 可以把序列化后的Json反序列化成ObjectNode,ObjectNode可以通过objectNode.get(“field”).as(Int/String/…)() 来访问指定的字段。
(3)SimpleStringSchema可以将消息反序列化为字符串。当我们接收到消息并且反序列化失败的时候,会出现以下两种情况: 1) Flink从deserialize(..)方法中抛出异常,这会导致job的失败,然后job会重启;2) 在deserialize(..) 方法出现失败的时候返回null,这会让Flink Kafka consumer默默的忽略这条消息。请注意,如果配置了checkpoint 为enable,由于consumer的失败容忍机制,失败的消息会被继续消费,因此还会继续失败,这就会导致job被不断自动重启。
参数二:
反序列化约束,以便于Flink决定如何反序列化从Kafka获得的数据。
参数三
Kafka consumer的属性配置,下面三个属性配置是必须的:
3 日志采集架构图
(1)以前学习离线数仓时,采集数据是使用flume的agent级联的方式,中间层是为了增大吞吐(负载均衡),和容错(failOver),这两个可以同时实现(多个sink)
这种agent级联的方式是一种过时的做法了,在flume1.7前一半使用这种,flume1.7后,有kafkachannel,这种方式就被取代了,其一级agent实现不了容错。更好的方式如下
(2)直接source+kafkaChannel的形式,kafka直接解决掉高吞吐量和容错的问题,并且一级agent中还实现了容错如下图
4. 测流输出
测流输出与split+select相似。当单存的过滤出某类数据时,用filter效率会高点,但若是对某个数据进行分类时,若再使用filter的话,则要过滤多次,即运行多次任务,效率比较低。若是使用测流输出,运行一次即可
SideOutPutDemo
package cn._51doit.flink.day06; import org.apache.flink.api.common.functions.MapFunction; import org.apache.flink.api.java.tuple.Tuple3; import org.apache.flink.streaming.api.datastream.DataStream; import org.apache.flink.streaming.api.datastream.DataStreamSource; import org.apache.flink.streaming.api.datastream.SingleOutputStreamOperator; import org.apache.flink.streaming.api.environment.StreamExecutionEnvironment; import org.apache.flink.streaming.api.functions.ProcessFunction; import org.apache.flink.util.Collector; import org.apache.flink.util.OutputTag; public class SideOutPutDemo { public static void main(String[] args) throws Exception { StreamExecutionEnvironment env = StreamExecutionEnvironment.getExecutionEnvironment(); DataStreamSource<String> lines = env.socketTextStream("feng05", 8888); SingleOutputStreamOperator<Tuple3<String, String, String>> tpData = lines.map(new MapFunction<String, Tuple3<String, String, String>>() { @Override public Tuple3<String, String, String> map(String value) throws Exception { String[] fields = value.split(" "); String event = fields[0]; String guid = fields[1]; String timestamp = fields[2]; return Tuple3.of(event, guid, timestamp); } }); OutputTag<Tuple3<String, String, String>> viewTag = new OutputTag<Tuple3<String, String, String>>("view-tag"){}; OutputTag<Tuple3<String, String, String>> activityTag = new OutputTag<Tuple3<String, String, String>>("activity-tag"){}; OutputTag<Tuple3<String, String, String>> orderTag = new OutputTag<Tuple3<String, String, String>>("order-tag"){}; SingleOutputStreamOperator<Tuple3<String, String, String>> outDataStream = tpData.process(new ProcessFunction<Tuple3<String, String, String>, Tuple3<String, String, String>>() { @Override public void processElement(Tuple3<String, String, String> input, Context ctx, Collector<Tuple3<String, String, String>> out) throws Exception { // 将数据打上标签 String type = input.f0; if (type.equals("pgview")) { ctx.output(viewTag, input); } else if (type.equals("activity")) { ctx.output(activityTag, input); } else { ctx.output(orderTag, input); } // 输出主流的数据,此处不输出主流数据的话,在外面则获取不到主流数据 out.collect(input); } }); // 输出的测流只能通过getSideOutput // DataStream<Tuple3<String, String, String>> viewDataStream = outDataStream.getSideOutput(viewTag); // viewDataStream.print(); outDataStream.print(); env.execute(); } }
改进使用processElement方法
package cn._51doit.flink.day06; import org.apache.flink.api.common.functions.MapFunction; import org.apache.flink.api.java.tuple.Tuple3; import org.apache.flink.configuration.Configuration; import org.apache.flink.streaming.api.datastream.DataStream; import org.apache.flink.streaming.api.datastream.DataStreamSource; import org.apache.flink.streaming.api.datastream.SingleOutputStreamOperator; import org.apache.flink.streaming.api.environment.StreamExecutionEnvironment; import org.apache.flink.streaming.api.functions.ProcessFunction; import org.apache.flink.util.Collector; import org.apache.flink.util.OutputTag; /** * 1.将数据整理成Tuple3 * 2.然后使用侧流输出将数据分类 */ public class SideOutputsDemo2 { public static void main(String[] args) throws Exception { StreamExecutionEnvironment env = StreamExecutionEnvironment.getExecutionEnvironment(); // view,pid,2020-03-09 11:42:30 // activity,a10,2020-03-09 11:42:38 // order,o345,2020-03-09 11:42:38 DataStreamSource<String> lines = env.socketTextStream("localhost", 8888); OutputTag<Tuple3<String, String, String>> viewTag = new OutputTag<Tuple3<String, String, String>>("view-tag") { }; OutputTag<Tuple3<String, String, String>> activityTag = new OutputTag<Tuple3<String, String, String>>("activity-tag") { }; OutputTag<Tuple3<String, String, String>> orderTag = new OutputTag<Tuple3<String, String, String>>("order-tag") { }; //直接调用process方法 SingleOutputStreamOperator<Tuple3<String, String, String>> tpDataStream = lines.process(new ProcessFunction<String, Tuple3<String, String, String>>() { @Override public void open(Configuration parameters) throws Exception { super.open(parameters); } @Override public void processElement(String input, Context ctx, Collector<Tuple3<String, String, String>> out) throws Exception { //1.将字符串转成Tuple2 String[] fields = input.split(","); String type = fields[0]; String id = fields[1]; String time = fields[2]; Tuple3<String, String, String> tp = Tuple3.of(type, id, time); //2.对数据打标签 //将数据打上标签 if (type.equals("view")) { //输出数据,将数据和标签关联 ctx.output(viewTag, tp); //ctx.output 输出侧流的 } else if (type.equals("activity")) { ctx.output(activityTag, tp); } else { ctx.output(orderTag, tp); } //输出主流的数据 out.collect(tp); } }); //输出的测流只能通过getSideOutput DataStream<Tuple3<String, String, String>> viewDataStream = tpDataStream.getSideOutput(viewTag); //分别处理各种类型的数据。 viewDataStream.print(); env.execute(); } }
5. 将kafka中数据写入HDFS
- 方案一:使用flume,具体见下图:
- 方案二:使用StreamingFileSink,此种形式更加好,其可以按照需求滚动生成文件
6 KafkaProducer的使用
现在的需求是将kafka中的数据进行处理(分主题等),然后写回kafka中去。如下所示
这时可以使用flink的自定义sink往kafka中写数据,具体代码如下
KafkaSinkDemo(老版本1.9以前)
package cn._51doit.flink.day06; import org.apache.flink.api.common.serialization.SimpleStringSchema; import org.apache.flink.streaming.api.datastream.DataStreamSource; import org.apache.flink.streaming.api.environment.StreamExecutionEnvironment; import org.apache.flink.streaming.connectors.kafka.FlinkKafkaProducer; public class KafkaSinkDemo { public static void main(String[] args) throws Exception { StreamExecutionEnvironment env = StreamExecutionEnvironment.getExecutionEnvironment(); DataStreamSource<String> lines = env.socketTextStream("localhost", 8888); FlinkKafkaProducer<String> myProducer = new FlinkKafkaProducer<String>( "node-1.51doit.cn:9092,node-2.51doit.cn:9092,node-3.51doit.cn:9092", // broker list "etl-test", // target topic new SimpleStringSchema()); // serialization schema myProducer.setWriteTimestampToKafka(true); //将数据写入到Kafka lines.addSink(myProducer); env.execute(); } }
KafkaSinkDemo2(flink1.9以后)
package cn._51doit.flink.day06; import org.apache.flink.api.common.serialization.SimpleStringSchema; import org.apache.flink.api.java.utils.ParameterTool; import org.apache.flink.streaming.api.datastream.DataStream; import org.apache.flink.streaming.api.datastream.DataStreamSource; import org.apache.flink.streaming.api.environment.StreamExecutionEnvironment; import org.apache.flink.streaming.connectors.kafka.FlinkKafkaProducer; import java.util.Properties; /** * 使用新的Kafka Sink API */ public class KafkaSinkDemo2 { public static void main(String[] args) throws Exception { ParameterTool parameters = ParameterTool.fromPropertiesFile(args[0]); DataStream<String> lines = FlinkUtils.createKafkaStream(parameters, SimpleStringSchema.class); //写入Kafka的topic String topic = "etl-test"; //设置Kafka相关参数 Properties properties = new Properties(); properties.setProperty("transaction.timeout.ms",1000 * 60 * 5 + ""); properties.setProperty("bootstrap.servers", "node-1.51doit.cn:9092,node-2.51doit.cn:9092,node-3.51doit.cn:9092"); //创建FlinkKafkaProducer FlinkKafkaProducer<String> kafkaProducer = new FlinkKafkaProducer<String>( topic, //指定topic new KafkaStringSerializationSchema(topic), //指定写入Kafka的序列化Schema properties, //指定Kafka的相关参数 FlinkKafkaProducer.Semantic.EXACTLY_ONCE //指定写入Kafka为EXACTLY_ONCE语义 ); //添加KafkaSink lines.addSink(kafkaProducer); //执行 FlinkUtils.env.execute(); } }
这里需要注意一个点,要设置如下参数:
properties.setProperty("transaction.timeout.ms",1000 * 60 * 5 + "");
kafka brokers默认的最大事务超时时间为15min,生产者设置事务时不允许大于这个值。但是在默认的情况下,FlinkKafkaProducer设置事务超时属性为1h,超过了默认transaction.max.ms 15min。这个时候我们选择生产者的事务超时属性transaction.timeout.ms小于15min即可
7. 练习(未练)