文章目录
一、需求说明
1、以案例驱动理解
二、技术点
三、代码实现（一）
1、调用底层的Process（可做类似map的操作），将Json字符串解析成MyBehavior对象
2、提取EventTime,转换成Timestamp格式,生成WaterMark
3、按照指定事件分组
4、把分好组的数据，划分窗口：假设窗口总长10分钟， 步长1分钟滑动一次
5、窗口内的数据进行聚合，拿出窗口Star时间和窗口End时间
四、定义的单独类MyBehavior 和 ItemViewCount
1、MyBehavior
2、ItemViewCount
五、最终结果
六、代码实现（二） 更高级
1、单独类 MyWindowAggFunction
2、单独类 MyWindowFunction
七、对聚合好的窗口内数据排序
1、分组
2、排序
一、需求说明
统计一定时间段内的，热门商品/品牌TopN
1、以案例驱动理解
-
数据：
-
{"userId": "u001", "itemId": "p1001", "categoryId": "c11", type: "pv", "timestamp": "2020-03-08 11:11:11"} {"userId": "u002", "itemId": "p1001", "categoryId": "c11", type: "pv", "timestamp": "2020-03-08 11:11:11"} {"userId": "u003", "itemId": "p1001", "categoryId": "c11", type: "pv", "timestamp": "2020-03-08 11:11:11"} {"userId": "u003", "itemId": "p1001", "categoryId": "c11", type: "cart", "timestamp": "2020-03-08 11:11:11"} {"userId": "u011", "itemId": "p2222", "categoryId": "c22", type: "pv", "timestamp": "2020-03-08 11:11:11"} {"userId": "u012", "itemId": "p2222", "categoryId": "c22", type: "pv", "timestamp": "2020-03-08 11:11:11"} {"userId": "u012", "itemId": "p2222", "categoryId": "c22", type: "pv", "timestamp": "2020-03-08 11:12:01"} {"userId": "u001", "itemId": "p1001", "categoryId": "c11", type: "pv", "timestamp": "2020-03-08 11:12:01"} {"userId": "u002", "itemId": "p1001", "categoryId": "c11", type: "pv", "timestamp": "2020-03-08 11:12:01"} {"userId": "u003", "itemId": "p1001", "categoryId": "c11", type: "pv", "timestamp": "2020-03-08 11:12:01"} {"userId": "u003", "itemId": "p1001", "categoryId": "c11", type: "cart", "timestamp": "2020-03-08 11:12:01"} {"userId": "u011", "itemId": "p2222", "categoryId": "c22", type: "pv", "timestamp": "2020-03-08 11:12:01"} {"userId": "u012", "itemId": "p2222", "categoryId": "c22", type: "pv", "timestamp": "2020-03-08 11:12:01"} {"userId": "u011", "itemId": "p2222", "categoryId": "c22", type: "pv", "timestamp": "2020-03-08 11:13:01"}二、技术点
Flink的EventTime
Flink的滑动窗口（滚动窗口也可以完成 ，但是生成的结果太突兀，没有平滑性）
Flink的定时器
三、代码实现（一）
使用window.apply( )方法 → 见第5步
窗口触发时，会执行一次apply,相当于对窗口中的全量数据进行计算（全部拿出在计算）
窗口不触发，会把数据缓存在内存中，当窗口特别长时，那么这种apply不太好
1、调用底层的Process（可做类似map的操作），将Json字符串解析成MyBehavior对象
-
import com.alibaba.fastjson.JSON; import org.apache.flink.api.java.tuple.Tuple; import org.apache.flink.streaming.api.TimeCharacteristic; import org.apache.flink.streaming.api.datastream.DataStreamSource; import org.apache.flink.streaming.api.datastream.KeyedStream; import org.apache.flink.streaming.api.datastream.SingleOutputStreamOperator; import org.apache.flink.streaming.api.datastream.WindowedStream; import org.apache.flink.streaming.api.environment.StreamExecutionEnvironment; import org.apache.flink.streaming.api.functions.ProcessFunction; import org.apache.flink.streaming.api.functions.timestamps.BoundedOutOfOrdernessTimestampExtractor; import org.apache.flink.streaming.api.functions.windowing.WindowFunction; import org.apache.flink.streaming.api.windowing.assigners.SlidingEventTimeWindows; import org.apache.flink.streaming.api.windowing.time.Time; import org.apache.flink.streaming.api.windowing.windows.TimeWindow; import org.apache.flink.util.Collector; public class HotGoodsTopN { public static void main(String[] args) throws Exception { StreamExecutionEnvironment env = StreamExecutionEnvironment.getExecutionEnvironment(); // 选择EventTime作为Flink的时间 env.setStreamTimeCharacteristic(TimeCharacteristic.EventTime); // 设置checkPoint时间 env.enableCheckpointing(60000); // 设置并行度 env.setParallelism(1); DataStreamSource<String> lines = env.socketTextStream("linux01", 8888); SingleOutputStreamOperator<MyBehavior> process = lines.process(new ProcessFunction<String, MyBehavior>() { @Override public void processElement(String input, Context ctx, Collector<MyBehavior> out) throws Exception { try { // FastJson 会自动把时间解析成long类型的TimeStamp MyBehavior behavior = JSON.parseObject(input, MyBehavior.class); out.collect(behavior); } catch (Exception e) { e.printStackTrace(); //TODO 记录出现异常的数据 } } });2、提取EventTime,转换成Timestamp格式,生成WaterMark
-
// 设定延迟时间 SingleOutputStreamOperator<MyBehavior> behaviorDSWithWaterMark = process.assignTimestampsAndWatermarks(new BoundedOutOfOrdernessTimestampExtractor<MyBehavior>(Time.seconds(0)) { @Override public long extractTimestamp(MyBehavior element) { return element.timestamp; } });3、按照指定事件分组
-
// 某个商品，在窗口时间内，被（点击、购买、添加购物车、收藏）了多少次 KeyedStream<MyBehavior, Tuple> keyed = behaviorDSWithWaterMark.keyBy("itemId", "type");4、把分好组的数据，划分窗口：假设窗口总长10分钟， 步长1分钟滑动一次
-
WindowedStream<MyBehavior, Tuple, TimeWindow> window = keyed.window(SlidingEventTimeWindows.of(Time.minutes(10), Time.minutes(1)));5、窗口内的数据进行聚合，拿出窗口Star时间和窗口End时间
//参数：输入的数据类， 输出的数据类，分组字段tuple， 窗口对象TimeWindow
SingleOutputStreamOperator<ItemViewCount> result = window.apply(new WindowFunction<MyBehavior, ItemViewCount,
Tuple, TimeWindow>() {
@Override
public void apply(Tuple tuple, TimeWindow window, Iterable<MyBehavior> input,
Collector<ItemViewCount> out) throws Exception {
//拿出分组的字段
String itemId = tuple.getField(0);
String type = tuple.getField(1);
//拿出窗口的起始和结束时间
long start = window.getStart();
long end = window.getEnd();
// 编写累加的逻辑
int count = 0;
for (MyBehavior myBehavior : input) {
count += 1;
}
//输出结果
out.collect(ItemViewCount.of(itemId, type, start, end, count));
}
});
result.print();
env.execute("HotGoodsTopN");
}
}四、定义的单独类MyBehavior 和 ItemViewCount
-
MyBehavior → 解析Json字符串后生成的JavaBean
-
ItemViewCount → 最后结果输出的格式类
1、MyBehavior
import java.sql.Timestamp;
public class MyBehavior {
public String userId; // 用户ID
public String itemId; // 商品ID
public String categoryId; // 商品类目ID
public String type; // 用户行为, 包括("pv", "buy", "cart", "fav")
public long timestamp; // 行为发生的时间戳，单位秒
public long counts = 1;
public static MyBehavior of(String userId, String itemId, String categoryId, String type, long timestamp) {
MyBehavior behavior = new MyBehavior();
behavior.userId = userId;
behavior.itemId = itemId;
behavior.categoryId = categoryId;
behavior.type = type;
behavior.timestamp = timestamp;
return behavior;
}
public static MyBehavior of(String userId, String itemId, String categoryId, String type, long timestamp,
long counts) {
MyBehavior behavior = new MyBehavior();
behavior.userId = userId;
behavior.itemId = itemId;
behavior.categoryId = categoryId;
behavior.type = type;
behavior.timestamp = timestamp;
behavior.counts = counts;
return behavior;
}
@Override
public String toString() {
return "MyBehavior{" + "userId='" + userId + '\'' + ", itemId='" + itemId + '\''
+ ", categoryId='" + categoryId + '\'' + ", type='" + type + '\''
+ ", timestamp=" + timestamp + "," + new Timestamp(timestamp)
+ "counts=" + counts + '}';
}
public String getUserId() {
return userId;
}
public String getItemId() {
return itemId;
}
public String getCategoryId() {
return categoryId;
}
public String getType() {
return type;
}
public long getTimestamp() {
return timestamp;
}
public long getCounts() {
return counts;
}
}2、ItemViewCount
import java.sql.Timestamp;
public class ItemViewCount {
public String itemId; // 商品ID
public String type; // 事件类型
public long windowStart; // 窗口开始时间戳
public long windowEnd; // 窗口结束时间戳
public long viewCount; // 商品的点击量
public static ItemViewCount of(String itemId, String type, long windowStart, long windowEnd, long viewCount) {
ItemViewCount result = new ItemViewCount();
result.itemId = itemId;
result.type = type;
result.windowStart = windowStart;
result.windowEnd = windowEnd;
result.viewCount = viewCount;
return result;
}
@Override
public String toString() {
return "{" +
"itemId='" + itemId + '\'' +
"type='" + type + '\'' +
", windowStart=" + windowStart + " , " + new Timestamp(windowStart) +
", windowEnd=" + windowEnd + " , " + new Timestamp(windowEnd) +
", viewCount=" + viewCount +
'}';
}
}五、最终结果
-
1分钟窗口一滑动一统计
-
11:11:12:01统计一次之前的，11:13:01统计一次之前的
{itemId='p1001'type='pv', windowStart=1583636520000 , 2020-03-08 11:02:00.0, windowEnd=1583637120000 , 2020-03-08 11:12:00.0, viewCount=3}
{itemId='p1001'type='cart', windowStart=1583636520000 , 2020-03-08 11:02:00.0, windowEnd=1583637120000 , 2020-03-08 11:12:00.0, viewCount=1}
{itemId='p2222'type='pv', windowStart=1583636520000 , 2020-03-08 11:02:00.0, windowEnd=1583637120000 , 2020-03-08 11:12:00.0, viewCount=2}
{itemId='p1001'type='cart', windowStart=1583636580000 , 2020-03-08 11:03:00.0, windowEnd=1583637180000 , 2020-03-08 11:13:00.0, viewCount=2}
{itemId='p1001'type='pv', windowStart=1583636580000 , 2020-03-08 11:03:00.0, windowEnd=1583637180000 , 2020-03-08 11:13:00.0, viewCount=6}
{itemId='p2222'type='pv', windowStart=1583636580000 , 2020-03-08 11:03:00.0, windowEnd=1583637180000 , 2020-03-08 11:13:00.0, viewCount=5}六、代码实现（二）
优化点：在窗口内增量聚合 (来一个加一个，内存中只保存一个数字而已)
/** 使用这种aggregate聚合方法：
*
* public <ACC, V, R> SingleOutputStreamOperator<R> aggregate(
* AggregateFunction<T, ACC, V> aggFunction,
* WindowFunction<V, R, K, W> windowFunction) {}
*/
SingleOutputStreamOperator<ItemViewCount> windowAggregate = window.aggregate(new MyWindowAggFunction(),
new MyWindowFunction());1、单独类 MyWindowAggFunction
-
拿到聚合字段（MyBehavior中counts）
三个泛型：
-
第一个输入的类型
-
第二个计数/累加器的类型
-
第三个输出的数据类型
//
public static class MyWindowAggFunction implements AggregateFunction<MyBehavior, Long, Long> {
//初始化一个计数器
@Override
public Long createAccumulator() {
return 0L;
}
//每输入一条数据就调用一次add方法
@Override
public Long add(MyBehavior input, Long accumulator) {
return accumulator + input.counts;
}
@Override
public Long getResult(Long accumulator) {
return accumulator;
}
//只针对SessionWindow有效，对应滚动窗口、滑动窗口不会调用此方法
@Override
public Long merge(Long a, Long b) {
return null;
}
}2、单独类 MyWindowFunction
拿到窗口的开始时间和结束时间，拿出分组字段
传入4个泛型：
-
第一个：输入的数据类型（Long类型的次数），也就是 MyWindowAggFunction中聚合后的结果值
-
第二个：输出的数据类型（ItemViewCount）
-
第三个：分组的key(分组的字段)
-
第四个：窗口对象（起始时间、结束时间）
public static class MyWindowFunction implements WindowFunction<Long, ItemViewCount, Tuple, TimeWindow> {
@Override
public void apply(Tuple tuple, TimeWindow window, Iterable<Long> input, Collector<ItemViewCount> out) throws Exception {
String itemId = tuple.getField(0);
String type = tuple.getField(1);
long windowStart = window.getStart();
long windowEnd = window.getEnd();
//窗口集合的结果
Long aLong = input.iterator().next();
//输出数据
out.collect(ItemViewCount.of(itemId, type, windowStart, windowEnd, aLong));
}七、对聚合好的窗口内数据排序
-
按照窗口的start、end进行分组，将窗口相同的数据进行排序
-
必须是在同一时间段的窗口
1、分组
KeyedStream<ItemViewCount, Tuple> soredKeyed = windowAggregate.keyBy("type", "windowStart",
"windowEnd");2、排序
SingleOutputStreamOperator<List<ItemViewCount>> sored = soredKeyed.process(new KeyedProcessFunction<Tuple, ItemViewCount, List<ItemViewCount>>() {
private transient ValueState<List<ItemViewCount>> valueState;
// 要把这个时间段的所有的ItemViewCount作为中间结果聚合在一块，引入ValueState
@Override
public void open(Configuration parameters) throws Exception {
ValueStateDescriptor<List<ItemViewCount>> VSDescriptor =
new ValueStateDescriptor<>("list-state",
TypeInformation.of(new TypeHint<List<ItemViewCount>>() {
})
);
valueState = getRuntimeContext().getState(VSDescriptor);
}
//更新valueState 并注册定时器
@Override
public void processElement(ItemViewCount input, Context ctx, Collector<List<ItemViewCount>> out) throws Exception {
List<ItemViewCount> buffer = valueState.value();
if (buffer == null) {
buffer = new ArrayList<>();
}
buffer.add(input);
valueState.update(buffer);
//注册定时器,当为窗口最后的时间时，通过加1触发定时器
ctx.timerService().registerEventTimeTimer(input.windowEnd + 1);
}
// 做排序操作
@Override
public void onTimer(long timestamp, OnTimerContext ctx, Collector<List<ItemViewCount>> out) throws Exception {
//将ValueState中的数据取出来
List<ItemViewCount> buffer = valueState.value();
buffer.sort(new Comparator<ItemViewCount>() {
@Override
public int compare(ItemViewCount o1, ItemViewCount o2) {
//按照倒序，转成int类型
return -(int) (o1.viewCount - o2.viewCount);
}
});
valueState.update(null);
out.collect(buffer);
}
});
sored.print();
env.execute("HotGoodsTopNAdv");
}
}
585
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
