Flink实时ETL练习项目
2020-12-07 本文已影响0人
羋学僧
一、需求背景
针对算法产生的日志数据进行清洗拆分
- 1、算法产生的日志数据是嵌套json格式，需要拆分打平
- 2、针对算法中的国家字段进行大区转换
- 3、把数据回写到 Kafka
二、数据格式
Kafka中的算法日志数据格式
{"dt":"2019-11-19 20:33:39","countryCode":"TW","data":[{"type":"s1","score":0.8,"level":"D"},{"type":"s2","score":0.1,"level":"B"}]}
{"dt":"2019-11-19 20:33:41","countryCode":"KW","data":[{"type":"s2","score":0.2,"level":"A"},{"type":"s1","score":0.2,"level":"D"}]}
{"dt":"2019-11-19 20:33:43","countryCode":"HK","data":[{"type":"s5","score":0.5,"level":"C"},{"type":"s2","score":0.8,"level":"B"}]}
{"dt":"2019-11-19 20:33:39","countryCode":"TW","data":[{"type":"s1","score":0.8,"level":"D"},{"type":"s2","score":0.1,"level":"B"}]}
Flink中ETL输出数据格式
{"dt":"2019-11-19 20:33:39","countryCode":"AREA_CT"，"type":"s1","score":0.8,"level":"D"}
{"dt":"2019-11-19 20:33:39","countryCode":"AREA_CT"，"type":"s2","score":0.1,"level":"B"}
国家地区信息数据
存储字Redis中
三、产生数据
国家地区信息数据生成语句
hset areas AREA_US US
hset areas AREA_CT TW,HK
hset areas AREA_AR PK,KW,SA
hset areas AREA_IN IN
算法日志数据生成
/**
* 模拟数据源
*/
public class kafkaProducer {
public static void main(String[] args) throws Exception{
Properties prop = new Properties();
//指定kafka broker地址
prop.put("bootstrap.servers", "bigdata03:9092");
//指定key value的序列化方式
prop.put("key.serializer", StringSerializer.class.getName());
prop.put("value.serializer", StringSerializer.class.getName());
//指定topic名称
String topic = "data";
//创建producer链接
KafkaProducer<String, String> producer = new KafkaProducer<String,String>(prop);
//{"dt":"2018-01-01 10:11:11","countryCode":"US","data":[{"type":"s1","score":0.3,"level":"A"},{"type":"s2","score":0.2,"level":"B"}]}
while(true){
String message = "{\"dt\":\""+getCurrentTime()+"\",\"countryCode\":\""+getCountryCode()+"\",\"data\":[{\"type\":\""+getRandomType()+"\",\"score\":"+getRandomScore()+",\"level\":\""+getRandomLevel()+"\"},{\"type\":\""+getRandomType()+"\",\"score\":"+getRandomScore()+",\"level\":\""+getRandomLevel()+"\"}]}";
System.out.println(message);
//同步的方式，往Kafka里面生产数据
producer.send(new ProducerRecord<String, String>(topic,message));
Thread.sleep(2000);
}
//关闭链接
//producer.close();
}
public static String getCurrentTime(){
SimpleDateFormat sdf = new SimpleDateFormat("YYYY-MM-dd HH:mm:ss");
return sdf.format(new Date());
}
public static String getCountryCode(){
String[] types = {"US","TW","HK","PK","KW","SA","IN"};
Random random = new Random();
int i = random.nextInt(types.length);
return types[i];
}
public static String getRandomType(){
String[] types = {"s1","s2","s3","s4","s5"};
Random random = new Random();
int i = random.nextInt(types.length);
return types[i];
}
public static double getRandomScore(){
double[] types = {0.3,0.2,0.1,0.5,0.8};
Random random = new Random();
int i = random.nextInt(types.length);
return types[i];
}
public static String getRandomLevel(){
String[] types = {"A","A+","B","C","D"};
Random random = new Random();
int i = random.nextInt(types.length);
return types[i];
}
}
四、Redis数据读取
map:
key:US value:AREA_US
key:TW value:AREA_CT
key:HK value:AREA_CT
public class RedisSource implements SourceFunction<HashMap<String,String>> {
private Logger logger=LoggerFactory.getLogger(RedisSource.class);
private Jedis jedis;
private boolean isRunning=true;
@Override
public void run(SourceContext<HashMap<String, String>> sourceContext) throws Exception {
this.jedis = new Jedis("bigdata02",6379);
HashMap<String, String> map = new HashMap<>();
while(isRunning){
try{
map.clear();
Map<String, String> areas = jedis.hgetAll("areas");
for(Map.Entry<String,String> entry:areas.entrySet()){
String area = entry.getKey();
String value = entry.getValue();
String[] fields = value.split(",");
for (String country:fields){
map.put(country,area);
}
}
if(map.size() > 0){
sourceContext.collect(map);
}
}catch (JedisConnectionException e){
logger.error("redis连接一场："+ e.getCause());
}catch (Exception e){
logger.error("数据源发生了异常！！");
}
}
}
@Override
public void cancel() {
isRunning = false;
if(jedis != null){
jedis.close();
}
}
}
五、数据处理
添加flink run -m yarn-cluster
<dependency>
<groupId>org.apache.flink</groupId>
<artifactId>flink-shaded-hadoop-2-uber</artifactId>
<version>2.8.3-10.0</version>
</dependency>
/**
* 实时ETL
*/
public class DataClean {
public static void main(String[] args) throws Exception{
System.setProperty("HADOOP_USER_NAME", "bigdata");
StreamExecutionEnvironment env = StreamExecutionEnvironment.getExecutionEnvironment();
env.setParallelism(3);//假设Kafka的主题是3个分区
//设置checkpoint
env.enableCheckpointing(60000);
env.getCheckpointConfig().setCheckpointingMode(CheckpointingMode.EXACTLY_ONCE);
env.getCheckpointConfig().setMinPauseBetweenCheckpoints(10000);
env.getCheckpointConfig().setMaxConcurrentCheckpoints(1);
//flink停止的时候要不要清空checkpoint的数据
env.getCheckpointConfig().enableExternalizedCheckpoints(
CheckpointConfig.ExternalizedCheckpointCleanup.RETAIN_ON_CANCELLATION);
env.setStateBackend(new RocksDBStateBackend("hdfs://bigdata02:9000/FlinkETL/checkpoint"));
//Kafka数据源
String topic="data";
Properties properties = new Properties();
properties.put("bootstrap.servers","bigdata03:9092");
properties.put("group.id","dataclean_consumer");
properties.put("enable.auto.commit","false");
properties.put("auto.offset.reset","earliest");
FlinkKafkaConsumer011<String> consumer = new FlinkKafkaConsumer011<>(
topic,
new SimpleStringSchema(),
properties
);
DataStreamSource<String> allData = env.addSource(consumer);
// redis
DataStream<HashMap<String, String>> mapData = env.addSource(new RedisSource()).broadcast();
SingleOutputStreamOperator<String> etlDataStream = allData.connect(mapData).flatMap(new CoFlatMapFunction<String, HashMap<String, String>, String>() {
//其实不给也行。
HashMap<String, String> allMap = new HashMap<String, String>();
//在这儿一开始，我们还是需要给allmap一些初始的数据。
//alldata kafka
@Override
public void flatMap1(String line, Collector<String> collector) throws Exception {
//{"dt":"2019-11-19 20:33:39","countryCode":"TW","data":[{"type":"s1","score":0.8,"level":"D"},{"type":"s2","score":0.1,"level":"B"}]}
JSONObject jsonObject = JSONObject.parseObject(line);
String dt = jsonObject.getString("dt");
String countryCode = jsonObject.getString("countryCode");
//根据省份获取大区
String area = allMap.get(countryCode);
JSONArray data = jsonObject.getJSONArray("data");
for (int i = 0; i < data.size(); i++) {
//0 {"type":"s1","score":0.8,"level":"D"}
//1 {"type":"s2","score":0.1,"level":"B"}
JSONObject dataJSONObject = data.getJSONObject(i);
//添加日期
dataJSONObject.put("dt", dt);
//添加大区
dataJSONObject.put("area", area);
collector.collect(dataJSONObject.toString());
}
}
//mapdata redis
@Override
public void flatMap2(HashMap<String, String> map, Collector<String> collector) throws Exception {
allMap = map;
}
});
// etlDataStream.print().setParallelism(1);
String etltopic="etldata";
Properties sinkProperties = new Properties();
sinkProperties.put("bootstrap.servers","bigdata03:9092");
FlinkKafkaProducer011<String> kafkaSink = new FlinkKafkaProducer011<>(etltopic,
new SimpleStringSchema(),
sinkProperties);
etlDataStream.addSink(kafkaSink);
/**
*
* source： kafka
* sink: kafka
*
* 可以实现数据处理且只处理一次
*
* 1： checkpoint（offset）
* 2： 写到Kafka（结果数据）
* 这个两个步骤到事务一致性，可以实现这两个操作要么就一起成功，要么就一起失败。
*
*/
env.execute("data clean");
}
}
六、在集群上执行
flink run -m yarn-cluster(开辟资源＋提交任务)
Flink lib 文件夹下添加flink-shaded-hadoop-2-uber jar包
flink-shaded-hadoop-2-uber下载地址
flink-shaded-hadoop-2-uber编译地址
cd /home/bigdata/data/
java -cp etl-1.0-SNAPSHOT-jar-with-dependencies.jar com.nx.flink.producer.kafkaProducer
flink run -m yarn-cluster -yqu default -ynm etl -ys 2 -yjm 1024 -ytm 1024 -c com.nx.flink.core.DataClean etl-1.0-SNAPSHOT-jar-with-dependencies.jar
没执行成功，虚拟机内存不足
