DAG dag = new DAG();
dag.addVertex("A");
dag.addVertex("B");
dag.addVertex("C");
dag.addVertex("D");
dag.addEdge("A", "B");
dag.addEdge("A", "C");
System.out.println(dag);
2 拓扑排序检测图中是否有环
public boolean isCircularity() {
Set
3 stage优化
eg
如果任务存在如下的关系 , task1 执行完后执行 task2 ,task2 执行完后执行task3 ...
Task1 -> Task2 -> Task3 -> Task4
这些task 本来就要串行执行的 可以把这些task 打包在一块 减少线程上下文的切换
eg : 复杂一点的DAG:
/**
* H
* \
* G
* \
* A -> B
* \
* C- D -E - F-> J
*
*
*
* 优化后得 ==>
*
* (H,G)
* \
* A -> B
* \
* (C,D,E) - (F,J)
*
*/
详见chain方法: 关键代码如下
private void chain_(Set sources, final LinkedHashSetMultimap foutChain, final LinkedHashSetMultimap finChain) {
sources.forEach(sourceNode -> {
ArrayList maxStage = Lists.newArrayList();
findMaxStage(sourceNode, maxStage);
if (maxStage.size() > 1) { //存在需要合并的stage
addVertex(foutChain, finChain, maxStage);//添加一个新节点
Object o = maxStage.get(maxStage.size() - 1); //最后一个节点
reChain_(foutChain, finChain, maxStage, o);
}
if (maxStage.size() == 1) {
//不存在需要合并的stage
addVertex(foutChain, finChain, sourceNode);//添加一个新节点
Set subNodes = outDegree.get(sourceNode);
addSubNodeage(foutChain, finChain, sourceNode, subNodes);
}
});
}
4 测试DAG 执行
测试程序: 详见 DAGExecTest
1 新建一个task 只打印一句话
public static class Task implements Runnable {
private String taskName;
public Task(final String taskName) {
this.taskName = taskName;
}
@Override public void run() {
try {
Thread.sleep(2000);
} catch (InterruptedException e) {
e.printStackTrace();
}
System.out.println("i am running my name is " + taskName + " finish ThreadID: " + Thread.currentThread().getId());
}
public String getTaskName() {
return taskName;
}
@Override public String toString() {
return taskName;
}
}
2 构建DAG
DAG dag = DAG.create();
Task a = new Task("a");
Task b = new Task("b");
Task c = new Task("c");
Task d = new Task("d");
Task e = new Task("e");
Task f = new Task("f");
Task g = new Task("g");
Task h = new Task("h");
Task j = new Task("j");
dag.addVertex(a);
dag.addVertex(b);
dag.addVertex(c);
dag.addVertex(d);
dag.addVertex(e);
dag.addVertex(f);
dag.addVertex(g);
dag.addVertex(h);
dag.addVertex(j);
dag.addEdge(h, g);
dag.addEdge(g, b);
dag.addEdge(a, b);
dag.addEdge(b, f);
dag.addEdge(c, d);
dag.addEdge(d, e);
dag.addEdge(e, f);
dag.addEdge(f, j);
构建完成后如图
* H
* \
* G
* \
* A -> B
* \
* C- D -E - F-> J
3 stage 切分
DAG chain = dag.chain();
执行完图入下:
* (H,G)
* \
* A -> B
* \
* (C,D,E) - (F,J)
4 执行 DAG DAGExecTest 最终结果打印如下如下:
可以发现有3个Stage stage1 包含3个task task分别在不同的线程里面执行
其中c-d-e g-c f-j是经过优化的在同一个线程里面执行,减少了不必要的上下文切换
i am running my name is a finish ThreadID: 10
i am running my name is c finish ThreadID: 11
i am running my name is h finish ThreadID: 12
i am running my name is d finish ThreadID: 11
i am running my name is g finish ThreadID: 12
i am running my name is e finish ThreadID: 11
stage 结束 : task detached:a, task chain c-d-e task chain h-g
-----------------------------------------------
i am running my name is b finish ThreadID: 14
stage 结束 : task detached:b,
-----------------------------------------------
i am running my name is f finish ThreadID: 11
i am running my name is j finish ThreadID: 11
stage 结束 : task chain f-j
测试执行关键代码如下:
chain.execute(col -> {
Set set = (Set) col;
List completableFutures = Lists.newArrayList();
StringBuilder sb = new StringBuilder();
set.stream().forEach(x -> {
if (x instanceof Task) {
CompletableFuture future = CompletableFuture.runAsync((Task) x, executorService);
completableFutures.add(future);
sb.append(" task detached:" + ((Task) x).getTaskName()).append(",");
}
if (x instanceof List) {
List taskList = (List) x;
CompletableFuture future = CompletableFuture.runAsync(()->
taskList.forEach(Task::run));
completableFutures.add(future);
sb.append(
" task chain " + Joiner.on("-").join(taskList.stream().map(Task::getTaskName).collect(Collectors.toList())));
}
});
CompletableFuture.allOf(completableFutures.toArray(new CompletableFuture[completableFutures.size()])).join();
System.out.println("stage 结束 : " + sb.toString());
System.out.println("-----------------------------------------------");
});