前言:流量控制,即Flow Control,主要是用来限定server所能承载的最大(高并发)流量峰值,以免在峰值是Server过载而宕机,对于WEB系统而言,通常是分布式部署,如果请求并发量很大,会导致整个集群崩溃,也就是通常所说的“雪崩效应”。所以,我们不仅在网络代理层面(比如nginx)设置流量控制以抵抗、拒止溢出流量,我们还应该在application server层面有一定的自我保护策略,确保当前JVM的负载应该在可控范围之内,对于JVM承载能力之外的请求,应该被合理管理。
本文主要通过开发一个Filter,来限定application的并发量:
1)对于过量的请求,首先将请求buffer在队列中。
2)当buffer队列满时,多余的请求将会被直接拒绝。(过载请求量)
3)那些buffer中被阻塞的请求,等待一定时间后任然无法被执行,则直接返回错误URL。(溢出请求量)
4)我们设定一个允许的并发量,通过java中Semaphore控制。只有获取“锁”的请求,才能继续执行。
一、web.xml配置
<filter>
<filter-name>flowControlFilter</filter-name>
<filter-class>com.demo.security.FlowControlFilter</filter-class>
<init-param>
<param-name>permits</param-name>
<param-value>128</param-value>
</init-param>
<init-param>
<param-name>timeout</param-name>
<param-value>15000</param-value>
</init-param>
<init-param>
<param-name>bufferSize</param-name>
<param-value>500</param-value>
</init-param>
<init-param>
<param-name>errorUrl</param-name>
<param-value>/error.html</param-value>
</init-param>
</filter>
<filter-mapping>
<filter-name>flowControlFilter</filter-name>
<url-pattern>/*</url-pattern>
</filter-mapping>
二、FlowControlFilter
import org.slf4j.Logger;
import org.slf4j.LoggerFactory;
import javax.servlet.*;
import javax.servlet.http.HttpServletResponse;
import java.io.IOException;
import java.util.LinkedList;
import java.util.Queue;
import java.util.concurrent.BlockingQueue;
import java.util.concurrent.LinkedBlockingQueue;
import java.util.concurrent.Semaphore;
import java.util.concurrent.TimeUnit;
import java.util.concurrent.locks.LockSupport;
/**
* 流量控制过滤器
* 用于保护当前JVM进程,在过载流量下,处于稳定可控状态。
*/
public class FlowControlFilter implements Filter {
//最大并发量
private int permits = Runtime.getRuntime().availableProcessors() + 1;//默认为500
//当并发量达到permits后,新的请求将会被buffer,buffer最大尺寸
//如果buffer已满,则直接拒绝
private int bufferSize = 500;//
//buffer中的请求被阻塞,此值用于控制最大阻塞时间
private long timeout = 30000;//默认阻塞时间
private String errorUrl;//跳转的错误页面
private BlockingQueue<Node> waitingQueue;
private Thread selectorThread;
private Semaphore semaphore;
private Object lock = new Object();
@Override
public void init(FilterConfig filterConfig) throws ServletException {
String p = filterConfig.getInitParameter("permits");
if(p != null) {
permits = Integer.parseInt(p);
if(permits < 0) {
throw new IllegalArgumentException("FlowControlFilter,permits parameter should be greater than 0 !");
}
}
String t = filterConfig.getInitParameter("timeout");
if(t != null) {
timeout = Long.parseLong(t);
if(timeout < 1) {
throw new IllegalArgumentException("FlowControlFilter,timeout parameter should be greater than 0 !");
}
}
String b = filterConfig.getInitParameter("bufferSize");
if(b != null) {
bufferSize = Integer.parseInt(b);
if(bufferSize < 0) {
throw new IllegalArgumentException("FlowControlFilter,bufferSize parameter should be greater than 0 !");
}
}
errorUrl = filterConfig.getInitParameter("errorUrl");
waitingQueue = new LinkedBlockingQueue<>(bufferSize);
semaphore = new Semaphore(permits);
selectorThread = new Thread(new SelectorRunner());
selectorThread.setDaemon(true);
selectorThread.start();
}
@Override
public void doFilter(ServletRequest request, ServletResponse response, FilterChain chain) throws IOException, ServletException {
checkSelector();
Thread t = Thread.currentThread();
HttpServletResponse httpServletResponse = (HttpServletResponse)response;
Node node = new Node(t,false);
boolean buffered = waitingQueue.offer(node);
//如果buffer已满
if (!buffered) {
if(errorUrl != null) {
httpServletResponse.sendRedirect(errorUrl);
}
return;
}
long deadline = System.currentTimeMillis() + timeout;
//进入等待队列后,当前线程阻塞
LockSupport.parkNanos(this, TimeUnit.MICROSECONDS.toNanos(timeout));
if (t.isInterrupted()) {
//如果线程是中断返回
t.interrupted();//clear status
}
//如果等待过期,则直接返回
if(deadline >= System.currentTimeMillis()) {
if(errorUrl != null) {
httpServletResponse.sendRedirect(errorUrl);
}
//对信号量进行补充
synchronized (lock) {
if(node.dequeued) {
semaphore.release();
} else {
node.dequeued = true;
}
}
return;
}
//继续执行
try {
chain.doFilter(request,response);
} finally {
semaphore.release();
checkSelector();
}
}
private void checkSelector() {
if(selectorThread != null && selectorThread.isAlive()) {
return;
}
synchronized (lock) {
if(selectorThread != null && selectorThread.isAlive()) {
return;
}
selectorThread = new Thread(new SelectorRunner());
selectorThread.setDaemon(true);
selectorThread.start();
}
}
private class SelectorRunner implements Runnable {
@Override
public void run() {
try {
while (true) {
Node node = waitingQueue.take();
//如果t,阻塞逃逸,只能在pack超时后退出
synchronized (lock) {
if(node.dequeued) {
//如果此线程已经park过期而退出了,则直接忽略
continue;
} else {
node.dequeued = true;
}
}
semaphore.acquire();
LockSupport.unpark(node.currentThread);
}
} catch (Exception e) {
//
} finally {
//全部释放阻塞
Queue<Node> queue = new LinkedList<>();
waitingQueue.drainTo(queue);
for(Node n : queue) {
if(!n.dequeued) {
LockSupport.unpark(n.currentThread);
}
}
}
}
}
private class Node {
Thread currentThread;
boolean dequeued;//是否已经出队
public Node(Thread t,boolean dequeued) {
this.currentThread = t;
this.dequeued = dequeued;
}
}
@Override
public void destroy() {
selectorThread.interrupt()
}
}
当然大家可以重新Filter,使用Spring的拦截器,不过整体思想是想通的。