Introduction to zero based im development (4): what is message timing consistency in IM systems?

Want to make APP Same thing as WeChat , Can run small programs smoothly ？ | Experience will send you to Xinjiang 、 Huawei 、 Cherry keyboard ！>>>

This article quotes Shen Jian 《 How to ensure IM Real time message “ Timing ” And “ Uniformity ”？》 The picture and content of the article （ Because I'm too lazy , The picture is not redrawn ）, The original text is linked at the end of the article .

1、 introduction

This article continues with the previous article 《 Zero basis IM Introduction to development ( 3、 ... and )： What is? IM The reliability of the system ？》, Explain IM The consistency of message timing in the system .

So called consistency , stay IM In general, it refers to the consistency of message timing , That's it ：

• 1） Context continuity of chat messages ;
• 2） Absolute chronology of chat messages .

More specifically ,IM The consistency of the message is reflected in ：

• 1） When talking alone ： Make sure that the order in which the sender sends out the chat message is consistent with the order in which the receiver sees it ;
• 2） Group chat ： Make sure that all group members see the chat message , It is consistent with the absolute time order when the sender sends the message .

IM The consistency of message timing in the system is a seemingly simple problem , One of the most difficult hot topics in technology , This article tries to explain to you in plain and simple words IM The product significance of message timing consistency problem 、 Cause of occurrence 、 Solutions, etc .

2、 Series articles

《 Zero basis IM Introduction to development ( One )： What is? IM System ？》

《 Zero basis IM Introduction to development ( Two )： What is? IM Real time performance of the system ？》

《 Zero basis IM Introduction to development ( 3、 ... and )： What is? IM The reliability of the system ？》

《 Zero basis IM Introduction to development ( Four )： What is? IM Message timing consistency of the system ？》（* this paper ）

《 Zero basis IM Introduction to development ( 5、 ... and )： What is? IM Security of the system ？ ( Release later )》

《 Zero basis IM Introduction to development ( 6、 ... and )： What is? IM The heartbeat mechanism of the system ？ ( Release later )》

《 Zero basis IM Introduction to development ( 7、 ... and )： How to understand and implement IM System message not read ？ ( Release later )》

《 Zero basis IM Introduction to development ( 8、 ... and )： How to understand and implement IM Multi terminal message roaming of the system ？ ( Release later )》

3、 Consistency of message timing , about IM The meaning of

Today, , Because of the popularity of mobile Internet , The actual social relationships of modern people , Almost entirely by IM This instant messaging social tool is organized ,IM The importance of this tool is self-evident .

IM In modern life , More and more important , But it's more and more common . Now I want to contact someone , The first thing I thought about was not to make a phone call , It's a “ WeChat ” or “QQ”. Yes ,IM This thing carries more and more meaning .

The consistency of message timing , about IM The meaning of , There is no doubt that it brings more than the so-called user experience problems . Let's take a look at an example .

hypothesis ： Your confession to the goddess is entering a critical stage , Just because it sucks IM, This leads to the incoherence of the chat message , At the moment 1000 Kilometer away, your goddess is staring at your mobile phone with a mask “ Drunken talk ”, How serious the consequences are —— Over the past six months “ Flatterer ” life 、 swallow humiliation and bear a heavy load , It's all for nothing by programmers “ Plaid Shirt ”、“ mediterranean sea ” We have been harmed .

To put it more seriously , Because it sucks IM, Let you lose this rare good gene with the help of goddess , It's a great opportunity to transform the appearance of the offspring of the family , It's very painful ...

The above example , It's just a chat , If it's a group chat , Then the problem may be magnified infinitely ： Think of a technology exchange group , It's a heated argument “php It's the best language in the world ” When it comes to this kind of topic , Suddenly I want to smash my cell phone , It's not because it's too noisy , And because the order of the messages is completely out of order , It has seriously affected the keyboard heroes to express their personal opinions .

4、 It's hard to guarantee the consistency of message timing ？

4.1 Basic cognition

In general IM In the eyes of users , News is nothing more than passing from one mobile phone to another , What's the difficulty with timing ？

Yes , Ordinary users think so , Technically speaking , He just simply will IM The process of sending and receiving messages is understood as the working mode of single thread .

actually , stay IM In this high performance scenario , In order to pursue high throughput 、 High concurrency , Using multithreading 、 asynchronous IO And so on. .

under these circumstances ,“ High concurrency ” And “ The order ” about IM On the server side , It's a contradiction , It's a little bit fishy and bear's paws （ It's hard to have both ）.

therefore , To achieve IM Temporal consistency of messages in scenarios , A trade-off needs to be made , And there are many technical dimensions to consider . This leads to the specific technology implementation has no fixed routine , And because of the uneven technical capabilities of developers , It makes a lot of IM There will be big problems in the actual effect of the system , For users, it will also be reflected directly in the product experience .

The following is a detailed description of the technical difficulties .

4.2 There is no global clock

As shown in the figure above , A really usable production system , It shows that it is impossible for all services to run on one server , The distributed environment is for sure .

that ： In a distributed environment , client + All kinds of background services on the server side , They're all distributed on different machines , Local clocks are used between machines , There is no such thing as “ Global clock ”（ There's no real global clock ）, Then the so-called message timing has no real time sequence benchmark . So the timing of messages is obviously not “ Local time ” It can be completely decided .

4.3 Multi sender problem

When the server is distributed , Out-of-service “ Local time ” To ensure timing , So can we use the receiver's local time to represent the time sequence ？

Unfortunately , Due to the presence of multiple clients （ For example, when talking in groups ）, Even the local time of a server , It doesn't mean “ Absolute timing ”.

As shown in the figure above ： In absolute timing ,APP1 Send out first msg1,APP2 Post issue msg2, All sent to the server web1, Network transmission is not guaranteed msg1 It must precede msg2 Arrived at , So even with a server web1 According to the time of , It can't be described precisely msg1 And msg2 The absolute timing of .

4.4 Multi receiver problem

The timing of the sender cannot be guaranteed , Suppose there is only one sender , Can the local time of the sender be used to represent the time sequence ？ Unfortunately , Due to the presence of multiple receivers , Cannot use sender's local time , Express “ Absolute timing ”.

Pictured above , In absolute timing ,web1 Send out first msg1, Post issue msg2, Due to the network transmission and the existence of multiple receivers , No guarantee msg1 First received, first processed , There is no guarantee that msg1 And msg2 The processing sequence of .

4.5 Network transmission and multithreading

Since it is difficult for multi sender and multi receiver to guarantee absolute timing , So suppose there is only a single sender and a single receiver , Can the absolute timing consistency of messages be guaranteed ？

The conclusion is pessimistic , Because of network transmission and multithreading , It still doesn't work .

As shown in the figure above ,web1 Send out first msg1、 Post issue msg2, Even if msg1 First arrive （ Network transmission is not guaranteed msg1 First arrive ）, Because of multithreading , There's no guarantee msg1 Be dealt with first .

5、 How to ensure the absolute consistency of message timing ？

Through the summary of the previous chapter , We're right IM The reason for the problem of temporal consistency of messages in , Have a more profound understanding of .

From a purely technical point of view , hypothesis ：

• 1） There's only one sender ;
• 2） A receiver ;
• 3） There is only one upstream and downstream connection socket Connect ;
• 4） Communicate by blocking .

In this case , There is no guarantee that the message sent first will be processed first , And then be shown to the receiver of the message first ？

Yes , Sure ！

But this is unlikely to happen in actual production IM System , It will be sent to the delivery party 、 Single receiver 、 single socket Connect 、 Blocking mode , In this way IM Once it's done , The product manager will show you immediately ...

6、 Practical optimization ideas

6.1 One to one chat message consistency assurance ideas

Suppose two people talk one on one , The sender A In turn msg1、msg2、msg3 Three messages to the receiver B, How to ensure the consistency of display timing of these three messages （ Send and display in the same order ）？

We know , The sender A In turn msg1、msg2、msg3 Three messages , After the end of the service , When it is transferred by the server , This sequence is due to the problem of multithreaded networks , It's possible to get out of order .

So the result could be ： 

As shown in the figure above , There will be a problem that the timing of the message is inconsistent with that when it is sent out （ The order of messages received is ：msg3、msg1、msg2）.

however , Actually, two people who are talking one on one , There is no need for global message timing consistency （ Because chatting only happens in the same conversation between two people ）, Just for the same sender A, issue B It's just that the message sequence is consistent .

Common optimization schemes , stay A Go to B In the message sent out , Plus the sender A A local absolute time sequence （ For example, the local timestamp ）, To show the receiver B The timing of the presentation of .

So when the receiver B After receiving the message , Even in extreme cases, messages may arrive out of order , But because of this disorder time difference, for ordinary users, the sense of body is very short , stay UI The presentation layer is sorted according to the absolute sequence in the message and then displayed , In fact, there are not too many users' perception .

6.2 How to ensure the message consistency of many to many group chats

hypothesis N A group of friends in a IM Chat in the group , How to ensure the consistency of display timing of messages received by all group members ？

First ： You can't use the absolute timing of the sender to ensure the message order as a pair of chats , Because the sender of group chat is not only , Time is not the same .

Maybe ： We can use the server's single point to do serialization .

As shown in the figure above , here IM The sending process of group chat is ：

• 1）sender1 issue msg1,sender2 issue msg2;
• 2）msg1 and msg2 Through the access cluster , Service cluster ;
• 3）service Layer to floor take a unique seq, To determine the timing of the receiver's presentation ;
• 4）service Get msg2 Of seq yes 20,msg1 Of seq yes 30;
• 5） Send messages to multiple group friends through the delivery service , Even if you receive msg1 and msg2 The time is different , But it can be unified according to seq To show .

This method ：

• 1） Advantage is ： It can realize that the message display time sequence of all group friends is the same ;
• 2） The disadvantage is that ： This service, which generates a global increment sequence number, can easily become a system bottleneck .

Is there any further optimization method ？

From a technical point of view ： In fact, group messages do not need to ensure that the global message sequence is orderly , As long as the messages in a group are in order , In this case ,“ news id serialize ” It's a good idea .

Above, in this scenario ,service Layers no longer need to go to a unified back end to get the overall situation seq（ Serial number ）, But in service The connection pool level has been slightly modified , Make sure that messages from a group fall on the same service On , This service You can use local seq To serialize all messages of the same group , Ensure that all group members see the message in the same time sequence .

About IM How to realize message serialization under the system architecture of , Or global news ID The generation scheme of , This is another very popular technology topic .

have interest in , Read the following series in depth ：

《IM news ID Technical topics ( One )： Massive wechat IM Practice of generating chat message serial number （ Principle of algorithm ）》

《IM news ID Technical topics ( Two )： Massive wechat IM Practice of generating chat message serial number （ Disaster recovery plan ）》

《IM news ID Technical topics ( 3、 ... and )： Decrypt rongyun IM Product chat messages ID Generation strategy 》

《IM news ID Technical topics ( Four )： Deeply decrypt the distribution of meituan ID generating algorithm 》

《IM news ID Technical topics ( 5、 ... and )： Open source distributed ID generator UidGenerator Technology realization of 》

《IM news ID Technical topics ( 6、 ... and )： Deep decryption of Didi's high performance ID generator (Tinyid)》

In this series , In particular, the trend of wechat is increasing ID Generate ideas （ Be careful ： Increasing trend is not strictly increasing , Increasing trend means that there is ID It's OK to be skipped ）, For distributed IM The news of ID It's very practical .

Yes , about IM system , The absolute sense of timing is difficult to guarantee , But through the monotonous incremental message generated by the server ID The way , Using incremental ID To ensure timing , It's also a very feasible plan .

7、 To sum up

IM System architecture , The absolute timing of messages is difficult , There are many reasons , such as ： There is no global clock 、 Multiple senders 、 Multiple recipients 、 Multithreading 、 Network transmission uncertainty, etc .

When talking one-on-one , In fact, we only need to ensure that the timing of sending is consistent with that of receiving , It can basically make the user feel out of order .

In the case of many to many group chats , Ensure that all receivers in the same group have the same timing , It can also make users feel out of order , There are two ways , A single point absolute timing , Another implementation message id Serialization （ That is to implement a global incremental message ID）.

8、 Reference material

[1]  How to ensure IM Real time message “ Timing ” And “ Uniformity ”？, author ： Shen Jian

[2]  A low cost guarantee IM Discussion on the method of message timing , author ： Feng Yu

This article has been published on “ Instant messaging technology circle ” official account ：

Links are ：http://www.52im.net/thread-3189-1-1.html