Introduction to long connection

A long connection

A long connection 、 Short connection 、 heartbeat

HTTP Deal with the TCP/IP Relationship of agreement ：

• HTTP The long connection and the short connection of are essentially TCP Long and short connections .HTTP Belongs to the application layer protocol , Use... In the transport layer TCP agreement , Use... In the network layer IP agreement . IP The protocol mainly solves the problem of network routing and addressing ,TCP The agreement mainly deals with how to IP Deliver packets reliably above the layer , Make the receiving end on the network receive all the packets sent by the sender , And the order is the same as the sending order .TCP The agreement is reliable 、 Connection oriented .

How to understand HTTP Protocol is stateless ：

• HTTP Protocol is stateless , It means that the protocol has no memory for transaction processing , The server does not know what the client status is . in other words , There is no connection between opening a web page on a server and the last web page opened on this server .HTTP Is a stateless connection oriented protocol , Stateless does not mean HTTP Can not keep TCP Connect , Not to mention HTTP It uses UDP agreement （ There is no connection ）.

What is long connection 、 Short connection ？

• stay HTTP/1.0 Short connection is used by default in . in other words , Client and server every time HTTP operation , Just one connection , Disconnect at end of task . When a client browser accesses HTML Or other types Web The page contains other Web resources （ Such as JavaScript file 、 image file 、CSS Documents, etc. ）, Every time I meet such a Web resources , The browser will recreate a HTTP conversation .
  And from HTTP/1.1 rise , Use long connection by default , To maintain connection characteristics . Using long connected HTTP agreement , This line will be added to the response header ：

Connection:keep-alive

With long connections , When a web page is opened , Transport between client and server HTTP Data TCP Connection will not close , When the client accesses this server again , Will continue to use this established connection .Keep-Alive Not permanently connected , It has a hold time , Different server software is available （ Such as Apache） Set this time in . To realize long connection, both client and server need to support long connection .

HTTP Long connection and short connection of protocol , Is essentially TCP Long connection and short connection of protocol .

TCP Connect ：

• When network communication is used TCP When the agreement , Before the real read-write operation , A connection must be established between the client and the server , When the read-write operation is completed , Both parties can release the connection when they no longer need it . The establishment of the connection depends on “ Three handshakes ”, And release requires “ Four handshakes ”, Therefore, the establishment of each connection requires resource consumption and time consumption .

Schematic diagram of establishing connection with classic triple handshake ：

The classic four times handshake closes the connection diagram ：

Long connection short connection operation process ：
The operation steps of short connection are ：

• Establishing a connection —— The data transfer —— Close the connection … Establishing a connection —— The data transfer —— Close the connection

The procedure of long connection is ：

• Establishing a connection —— The data transfer …（ Keep connected ）… The data transfer —— Close the connection

Short connection ：

Concept

client And server Establish a connection with three handshakes ,client Send request message ,server Return response , One connection is complete .

At this time, both sides can initiate close operation , But it's usually client First initiate close operation . It can be seen from the above that , The short connection is usually only in client/server Pass a request operation between .

Advantages and disadvantages of short connection ：

• It's easier to manage , The existing connections are all useful connections , There is no need for additional controls .

Use scenarios ：

• Usually, when the browser accesses the server, it is a short connection .

• For the server , Long connections will consume the resources of the server , Moreover, users use the browser to access the server relatively infrequently

• If there are hundreds of thousands , Millions of connections , The pressure on the server side will be very great , Even collapse .

• So for a large amount of concurrency , Low request frequency , Short connections are recommended .

A long connection ：
What is long connection ？

• client towards server A connection ,server Accept client Connect , The two sides establish a connection .

• Client And server After one read and write , The connection between them is not actively closed , Subsequent read and write operations will continue to use this connection .

Long connection life cycle ：

• Under normal circumstances , One TCP After a long connection is established , As long as the double does not make a shutdown request and there are no exceptions , This connection has always existed .

• The operating system will not automatically shut it down , Even after the change of physical network topology, it can still be used .

• So keep a connection for a few days 、 Several months 、 It's possible for a few years or more , As long as there are no exceptions or by the user （ application layer ） Active shut down .

• The client and service order can always use this connection for data communication .

Advantages of long connections ：

• Long connection can save more TCP Set up and shut down operations , Reduce the impact of network congestion ,

• When something goes wrong , You can prompt without closing the connection ,

• Reduce CPU And memory usage , Because you don't need to establish and close connections often .

The disadvantages of long connections ：

• When there are too many connections , Affect the performance and concurrency of the server .

Use scenarios ：

The connection of database is to use TCP A long connection .

• RPC, Remote service invocation , On the server , One service process frequently calls another service process , Long connections can be used , Reduce connection time .

summary ：
1. The use of long connection and short connection needs to be judged according to the application scenario

2. Long connections are not everything , It also needs maintenance .

Implementation of long connections
heartbeat

• Most application layer protocols have HeartBeat Mechanism , Usually, the client sends a packet to the server every short period of time , Usually, the client sends a packet to the server every short period of time .

• And transmit some data that may be necessary . A typical protocol for using heartbeat packets is IM, such as QQ/MSN/ Feixin and other agreements .

• stay TCP In the mechanism of , There is a mechanism for heartbeat packets , That is to say TCP The option to ：SO_KEEPALIVE.

• The system is set by default 2 An hour's heart rate . But it can't check that the machine is powered off 、 Unplug the cable 、 Firewalls are broken .

• And the logic layer may not be so easy to deal with disconnection . commonly , It's ok if it's just for living .

Why do I need a heartbeat mechanism ？

• Because of the unreliability of the network , Are likely to TCP In the process of maintaining a long connection , Because of some emergency , For example, the network cable is pulled out , A sudden power failure, etc ,

• The connection between the server and the client will be interrupted . In these emergencies , If it happens that there is no interaction between the server and the client , Then they can't find out in a short time that the other party has dropped the line .

• Heartbeat mechanism can solve this kind of problem .

TCP Agreed KeepAlive Mechanism

• Default KeepAlive The state is not open .

• Need to put setsockopt take SOL_SOCKET.SO_KEEPALIVE Set to 1 It's opening KeepAlive state ,

And you can set three parameters ：

tcp_keepalive_time ,tcp_keepalive_probes , tcp_keepalive_intvl,

respectively ： How long does the connection go idle keepalive Of ack package 、 Send a few ack Only when the other party has broken the line when the package does not reply 、 Two ack The interval between packages .

A lot of network devices , In especial NAT Router , Because of the limitation of its hardware （ For example, memory 、CPU processing capacity ）, Can't keep all the connections on it , So when necessary , Some inactive connections will be selected in the connection pool to kick off .

Typical practice is LRU, Connect the connection with no data for the longest time to T fall .

By using TCP Of KeepAlive Mechanism （ Modify that time Parameters ）, You can make the connection generate some... Every other short period of time ack package , To reduce the risk of being kicked , Of course , The price is extra network and CPU burden .

How to realize heartbeat mechanism ？
Implement heartbeat mechanism :

• Use TCP Agreement level keepalive Mechanism .

• Implement custom heartbeat mechanism on application layer .

• Although in TCP On the level of agreement , Provides keepalive The survival mechanism , But there are several disadvantages to using it :

• It is not TCP Standard agreement for , And it's off by default .

• TCP keepalive The mechanism depends on the implementation of the operating system , default keepalive The heartbeat time is Two hours , And right keepalive The system call is required for the modification of ( Or modify the system configuration ), Lack of flexibility .

• TCP keepalive And TCP Protocol binding , So if you need to replace it with UDP When the agreement , keepalive The mechanism fails .

• Use TCP Level of keepalive The mechanism saves traffic than the custom application layer heartbeat mechanism ,