HTTP长连接与短连接使⽤⽅法及测试详解
HTTP短连接（⾮持久连接）是指，客户端和服务端进⾏⼀次HTTP请求/响应之后，就关闭连接。所以，下⼀次的HTTP请求/响应操作就需要重新建⽴连接。
HTTP长连接（持久连接）是指，客户端和服务端建⽴⼀次连接之后，可以在这条连接上进⾏多次请求/响应操作。持久连接可以设置过期时间，也可以不设置。
我为什么没有说HTTP/1.0 默认短连接，HTTP/1.1起，默认长连接呢？因为我第⼀次看这个说法的时候，以为⾃⼰懂了，其实并没有懂。长短连接操作上有什么区别，有的地⽅出现的持久连接⼜是怎么回事？
使⽤设置
这⾥的设置，我们都以HTTP1.1协议为例⼦。
设置HTTP短连接
在⾸部字段中设置Connection:clo，则在⼀次请求/响应之后，就会关闭连接。
设置HTTP长连接，有过期时间
在⾸部字段中设置Connection:keep-alive 和Keep-Alive: timeout=60，表明连接建⽴之后，空闲时间超过60秒之后，就会失效。如果在空闲第58秒时，再次使⽤此连接，则连接仍然有效，使⽤完之后，重新计数，空闲60秒之后过期。
设置HTTP长连接，⽆过期时间
在⾸部字段中只设置Connection:keep-alive，表明连接永久有效。
实现原理
了解怎么设置之后，就开始⽤起来。然⽽，问题来了。在请求头中设置Connection:keep-alive，为什么连接空闲⼀段时间之后，还是断开了呢？这是因为connection字段只有服务端设置才有效。
HTTP操作是请求/响应成对出现的，即先有客户端发出请求，后有服务端处理请求。所以，⼀次HTTP操作的终点操作在服务端上，关闭也是由服务端发起的。
接下来我们做做测试，以及show code。下⾯的测试都是使⽤Spring RestTemplate，封装apache http client进⾏的。为⽅便讲解代码，先说明长连接的情况，最后再对其他形式做测试总结。
客户端连接失效时间⼤于服务端失效时间
如下，为请求⽇志。客户端设置Connection: Keep-Alive和Keep-Alive: timeout=60，服务端设置Connection: Keep-Alive和Keep-Alive: timeout=5。
## 客户端设置有效期为60s
[2017-04-26 14:08:00 DEBUG] (org.apache.http.wire:?) - http-outgoing-0 >> "POST /adx-api/api/creative/upload HTTP/1.1[\r][\n]"
[2017-04-26 14:08:00 DEBUG] (org.apache.http.wire:?) - http-outgoing-0 >> "Accept: application/json, application/*+json, text/html, application/json, text/javascript[\r][\n]"
[2017-04-26 14:08:00 DEBUG] (org.apache.http.wire:?) - http-outgoing-0 >> "Content-Type: application/json;chart=UTF-8[\r][\n]"
[2017-04-26 14:08:00 DEBUG] (org.apache.http.wire:?) - http-outgoing-0 >> "Ur-Agent: Mozilla/5.0 (Windows NT 6.1) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/31.0.1650.16 Safari/537.36[\r][\n]"
[2017-04-26 14:08:00 DEBUG] (org.apache.http.wire:?) - http-outgoing-0 >> "Accept-Encoding: gzip,deflate[\r][\n]"
[2017-04-26 14:08:00 DEBUG] (org.apache.http.wire:?) - http-outgoing-0 >> "Accept-Language: zh-CN[\r][\n]"
[2017-04-26 14:08:00 DEBUG] (org.apache.http.wire:?) - http-outgoing-0 >> "Connection: keep-alive[\r][\n]"
[2017-04-26 14:08:00 DEBUG] (org.apache.http.wire:?) - http-outgoing-0 >> "Keep-Alive: timeout=60[\r][\n]"
[2017-04-26 14:08:00 DEBUG] (org.apache.http.wire:?) - http-outgoing-0 >> "Content-Length: 396[\r][\n]"
[2017-04-26 14:08:00 DEBUG] (org.apache.http.wire:?) - http-outgoing-0 >> "Host: bizdomain[\r][\n]"
[2017-04-26 14:08:00 DEBUG] (org.apache.http.wire:?) - http-outgoing-0 >> "[\r][\n]"
[2017-04-26 14:08:00 DEBUG] (org.apache.http.wire:?) - http-outgoing-0 >> "request data"
##服务端设置有效期为5s
[2017-04-26 14:08:00 DEBUG] (org.apache.http.wire:?) - http-outgoing-0 << "HTTP/1.1 200 OK[\r][\n]"
[2017-04-26 14:08:00 DEBUG] (org.apache.http.wire:?) - http-outgoing-0 << "Date: Wed, 26 Apr 2017 06:07:58 GMT[\r][\n]"
[2017-04-26 14:08:00 DEBUG] (org.apache.http.wire:?) - http-outgoing-0 << "Server: Apache-Coyote/1.1[\r][\n]"
[2017-04-26 14:08:00 DEBUG] (org.apache.http.wire:?) - http-outgoing-0 << "Content-Type: text/html;chart=utf-8[\r][\n]"
[2017-04-26 14:08:00 DEBUG] (org.apache.http.wire:?) - http-outgoing-0 << "Keep-Alive: timeout=5, max=100[\r][\n]"
街舞教程
[2017-04-26 14:08:00 DEBUG] (org.apache.http.wire:?) - http-outgoing-0 << "Connection: Keep-Alive[\r][\n]"
[2017-04-26 14:08:00 DEBUG] (org.apache.http.wire:?) - http-outgoing-0 << "Transfer-Encoding: chunked[\r][\n]"
[2017-04-26 14:08:00 DEBUG] (org.apache.http.wire:?) - http-outgoing-0 << "[\r][\n]"
[2017-04-26 14:08:00 DEBUG] (org.apache.http.wire:?) - http-outgoing-0 << "63[\r][\n]"
[2017-04-26 14:08:00 DEBUG] (org.apache.http.wire:?) - http-outgoing-0 << "respon data"
客户端设置的有效期⼤于服务端的，那么实际连接的有效期呢？三分钟之后再次请求，从连接池中lea连接的时候，提⽰Connection expired @ Wed Apr 26 14:08:05，即在上⼀次请求之后的5s失效，说明是服务端的设置⽣效了。
[2017-04-26 14:11:00 DEBUG] (org.apache.PoolingHttpClientConnectionManager:?) - Connection request: [route: {}->bizdomain:80][total kept alive: 1; route allocated: 1 of 32; total allocated: 1 of 200] [2017-04-26 14:11:00 DEBUG] (org.apache.CPool:?) - Connection [id:2][route:{}->bizdomain:80][state:null] expired @ Wed Apr 26 14:08:05 GMT+08:00 2017
源码分析
通过源代码了解⼀下连接失效时间的设置过程。
//org.apache.hain.MainClientExec#execute
......
//从连接池中lea connection
final HttpClientConnectionmanagedConn = (timeout > 0 ? timeout : 0, TimeUnit.MILLISECONDS);
......
//将conenction封装在ConnectionHolder中
final ConnectionHolder connHolder = new ConnectionHolder(this.log, Manager, managedConn);
歌曲名......
中国政法大学录取分数线
// The connection is in or can be brought to a re-usable state.
//如果返回值消息头中connection设置为clo，则返回fal
if (reuStrategy.keepAlive(respon, context)) {
// Set the idle duration of this connection
//取出respon消息头中，keep-alive的timeout值
final long duration = KeepAliveDuration(respon, context);
if (this.log.isDebugEnabled()) {
final String s;
if (duration > 0) {
s = "for " + duration + " " + TimeUnit.MILLISECONDS;
} el {
s = "indefinitely";
}
this.log.debug("Connection can be kept alive " + s);
}
//设置失效时间
connHolder.tValidFor(duration, TimeUnit.MILLISECONDS);
connHolder.markReusable();
} el {
connHolder.markNonReusable();
}
待读取响应之后，释放连接，即：leaConnection()。调⽤org.apache.PoolingHttpClientConnectionManager#releaConnection⽅法。
@Override
public void releaConnection(final HttpClientConnection managedConn,
final Object state,final long keepalive, final TimeUnit tunit) {
小雪的诗句
synchronized (managedConn) {
final CPoolEntry entry = CPoolProxy.detach(managedConn);
if (entry == null) {
return;
}
final ManagedHttpClientConnection conn = Connection();
try {
if (conn.isOpen()) {
final TimeUnit effectiveUnit = tunit != null ? tunit : TimeUnit.MILLISECONDS;
entry.tState(state);
/
/设置失效时间
entry.updateExpiry(keepalive, effectiveUnit);
}
} finally {
。。。。。。
}
}
}
}
然后再下⼀次HTTP操作，从连接池中获取连接时
//org.apache.PoolingHttpClientConnectionManager#requestConnection调⽤org.apache.http.pool.AbstractConnPool#lea，
//调⽤getPoolEntryBlocking，调⽤org.apache.CPoolEntry#isExpired
@Override
物业管家工作内容public boolean isExpired(final long now) {
final boolean expired = super.isExpired(now);
if (expired && this.log.isDebugEnabled()) {
//⽇志中看到的内容
this.log.debug("Connection " + this + " expired @ " + new Date(getExpiry()));
}
return expired;
}
综上，连接的实际有效时间，是根据respon的设置来决定的。
饥的反义词是什么其他情况测试
客户端设置Connection: Clo
##connection:clo请求，kept alive的连接为0
[2017-04-26 13:57:00 DEBUG] (org.apache.PoolingHttpClientConnectionManager:?) - Connection request: [route: {}->bizdomain:80][total kept alive: 0; route allocated: 0 of 32; total allocated: 0 of 200] [2017-04-26 13:57:00 DEBUG] (org.apache.PoolingHttpClientConnectionManager:?) - Connection lead: [id: 0][route: {}->bizdomain:80][total kept alive: 0; route allocated: 1 of 32; total allocated: 1 of 200] [2017-04-26 13:57:00 DEBUG] (org.apache.hain.MainClientExec:?) - Opening connection {}->bizdomain:80
[2017-04-26 13:57:00 DEBUG] (org.apache.DefaultHttpClientConnectionOperator:?) - Connecting to bizdomain/127.0.0.195:80
## 建⽴新连接
[2017-04-26 13:57:00 DEBUG] (org.apache.DefaultHttpClientConnectionOperator:?) -
Connection established 127.0.0.191:49239<->127.0.0.195:80
## 客户端设置短连接
[2017-04-26 13:57:00 DEBUG] (org.apache.http.wire:?) - http-outgoing-0 >> "Connection: Clo[\r][\n]"
## 服务端返回的也是短连接
习语近人深海鱼头的做法[2017-04-26 13:57:00 DEBUG] (org.apache.http.wire:?) - http-outgoing-0 << "Connection: clo[\r][\n]"
##请求完之后，关闭连接
[2017-04-26 13:57:00 DEBUG] (org.apache.DefaultManagedHttpClientConnection:?) - http-outgoing-0: Clo connection
[2017-04-26 13:57:00 DEBUG] (org.apache.hain.MainClientExec:?) - Connection discarded
[2017-04-26 13:57:00 DEBUG] (org.apache.PoolingHttpClientConnectionManager:?) - Connection relead: [id: 0][route: {}->bizdomain:80][total kept alive: 0; route allocated: 0 of 32; total allocated: 0 of 200]如上，当服务端返回Connection: Clo时，客户端接收完响应，便会关闭连接。
客户端设置60s超时，服务端设置5s超时
##Keep-Alive: timeout=60 第⼀次请求，与connection:clo⽆差别
[2017-04-26 10:57:00 DEBUG] (org.apache.PoolingHttpClientConnectionManager:?) - Connection request: [route: {}->bizdomain:80][total kept alive: 0; route allocated: 0 of 32; total allocated: 0 of 200] [2017-04-26 10:57:00 DEBUG] (org.apache.PoolingHttpClientConnectionManager:?) - Connection lead: [id: 0][route: {}->bizdomain:80][total kept alive: 0; route allocated: 1 of 32; total allocated: 1 of 200] [2017-04-26 10:57:00 DEBUG] (org.apache.hain.MainClientExec:?) - Opening connection {}->bizdomain:80
## 客户端设置超时时间60s
[2017-04-26 10:57:00 DEBUG] (org.apache.http.wire:?) - http-outgoing-0 >> "Connection: keep-alive[\r][\n]"
[2017-04-26 10:57:00 DEBUG] (org.apache.http.wire:?) - http-outgoing-0 >> "Keep-Alive: timeout=60[\r][\n]"
## 服务端设置超时时间5s
[2017-04-26 10:57:00 DEBUG] (org.apache.http.wire:?) - http-outgoing-0 << "Keep-Alive: timeout=5, max=100[\r][\n]"
[2017-04-26 10:57:00 DEBUG] (org.apache.http.wire:?) - http-outgoing-0 << "Connection: Keep-Alive[\r][\n]"
## 服务端设置⽣效，连接可以保持5s
[2017-04-26 10:57:00 DEBUG] (org.apache.hain.MainClientExec:?) - Connection can be kept alive for 5000 MILLISECONDS
[2017-04-26 10:57:00 DEBUG] (org.apache.PoolingHttpClientConnectionManager:?) - Connection [id: 0][route: {}->bizdomain:80] can be kept alive for 5.0 conds
[2017-04-26 10:57:00 DEBUG] (org.apache.PoolingHttpClientConnectionManager:?) - Connection relead: [id: 0][route: {}->bizdomain:80][total kept alive: 1; route allocated: 1 of 32; total allocated: 1 of 200] ##Keep-Alive: timeout=60 ⾮第⼀次请求
[2017-04-26 14:11:00 DEBUG] (org.apache.PoolingHttpClientConnectionManager:?) - Connection request: [route: {}->bizdomain:80][total kept alive: 1; route allocated: 1 of 32; total allocated: 1 of 200]
## 连接在上⼀次请求结束后5s失效
[2017-04-26 14:11:00 DEBUG] (org.apache.CPool:?) - Connection [id:2][route:{}->bizdomain:80][state:null] expired @ Wed Apr 26 14:10:05 GMT+08:00 2017
客户端设置失效时间，服务端设置不失效
## 客户端设置30s超时
[2017-04-26 17:45:00 DEBUG] (org.apache.http.wire:?) - http-outgoing-0 >> "Connection: keep-alive[\r][\n]"
[2017-04-26 17:45:00 DEBUG] (org.apache.http.wire:?) - http-outgoing-0 >> "Keep-Alive: timeout=30[\r][\n]"
## 服务端设置永久连接
[2017-04-26 17:45:00 DEBUG] (org.apache.http.wire:?) - http-outgoing-0 << "Connection: keep-alive[\r][\n]"
## 连接将⼀直保持
[2017-04-26 17:45:00 DEBUG] (org.apache.hain.MainClientExec:?) - Connection can be kept alive indefinitely
综上，http连接保持时间是由服务端的消息头connection字段和keep-alive字段定的。
在上⾯前两种情况，请求的是同⼀个服务端，那么为什么⼀个返回的是短连接，⼀个返回的是长连接呢？这⾥转⼀下这篇⽂章的解释：
不论request还是respon的header中包含了值为clo的connection，都表明当前正在使⽤的tcp链接在请求处理完毕后会被断掉。以后client再进⾏新的请求时就必须创建新的tcp链接了。 HTTP Connect
ion的 clo设置允许客户端或服务器中任何⼀⽅关闭底层的连接，双⽅都会要求在处理请求后关闭它们的TCP连接。
补充
TCP长短连接
在⽹上搜资料的时候，看到很多“HTTP协议的长连接和短连接，实质上是TCP协议的长连接和短连接”。 HTTP和TCP是不同两层的东西，它们怎么会是⼀样的呢？HTTP是请求/响应模式的，就是说我们发⼀个请求⼀定要有⼀个回应。最直观的就是，浏览器上发请求，得不到响应就会⼀直转圈圈。⽽TCP并不是⼀定要有响应。⼤家以前使⽤socket模拟⼀个IM聊天，A 跟B打完招呼，完全可以不⽤等待B的回应，就⾃⼰关掉连接的。
TCP keep-alive
另外还有HTTP协议的keep-alive和TCP的keep-alive含义是有差别的。HTTP的keep-alive是为了维持连接，以便复⽤连接。通过使⽤keep-alive机制，可以减少tcp连接建⽴次数，也意味着可以减少TIME_WAIT状态连接，以此提⾼性能和提⾼httpd服务器的吞吐率(更少的tcp连接意味着更少的系统内核调⽤,socket的accept()和clo()调⽤)。但是，长时间的tcp连接容易导致系统资源⽆效占⽤。配置不当的keep-alive，有时⽐重复利⽤连接带来的损失还更⼤。
⽽tcp keep-alive是TCP的⼀种检测TCP连接状况的机制，涉及到三个参数tcp_keepalive_time, tcp_keepalive_intvl, tcp_keepalive_probes。
当⽹络两端建⽴了TCP连接之后，闲置（双⽅没有任何数据流往来）了tcp_keepalive_time后，服务器内核就会尝试向客户端发送侦测包，来判断TCP连接状况(有可能客户端崩溃、强制关闭了应⽤、主机不可达等等)。如果没有收到对⽅的回答(ack包)，则会在 tcp_keepalive_intvl后再次尝试发送侦测包，直到收到对⽅的ack。如果⼀直没有收到对⽅的ack,⼀共会尝试tcp_keepalive_probes次。如果尝试tcp_keepalive_probes，依然没有收到对⽅的ack包，则会丢弃该TCP连接。TCP连接默认闲置时间是2⼩时，⼀般设置为30分钟⾜够了。
更多关于HTTP长连接与短连接使⽤⽅法请查看下⾯的相关链接
