MQ interview questions sorting

Why do you use MQ

advantage

decoupling 
The relevant capabilities of the same business are required by many projects , It is also very likely to be needed by other projects in the future . Decouple this part of the capability , adopt MQ Unified message processing . If there are other projects or businesses that need relevant functions , Subscribe to this , Handle related business . Reduce direct project dependencies .
asynchronous 
Large scale project related collaborators , More and more teams are involved , The relationship between projects is getting deeper and deeper . The whole link is very long , And troubleshooting is not easy .
Peak shaving 
For some businesses , At some point in time , It may be because the instantaneous request is too large , This causes all requests to access the database directly and puts pressure on the database , Thus affecting the original business of the project . commonly MySQL The processing capacity of the database is 2k Time /s,MQ Peak processing capacity 1w+. Can effectively deal with

shortcoming

Reduced system availability

Because of the new addition MQ, because MQ The usability of 、 Reliability imposes an additional burden on system availability .

How to solve the usability problem 
Different MQ The solution to this is different .
ActiveMQ
Zookeeper And multiple activeMQ, Change persistence to better performance LevelDB Replace the default KahaDB. 
advantage ： Simple implementation of availability . 
shortcoming ： Each time actually available MQ only one , Once there is one MQ Problems arise , Relevant messages will be lost .

RocketMQ

single master Pattern , When something goes wrong , The entire service will be unavailable . Generally, it will not be directly produced and used , Suitable for personal study ;
many master Pattern , Multiple master Nodes form a cluster . Single master Problems do not affect . advantage ： The highest performance of all modes . shortcoming ： Single master During node downtime , Messages that are not consumed are not available until the node is restored , The real-time performance of messages is affected . Be careful ： Using synchronous disk brushing can ensure that messages are not lost , meanwhile topic Corresponding queue It should be distributed among the centralized powers master node , Not just in one place master Node ;
many master many slave Asynchronous replication mode , In a multiple master In mode , Every master Each node has at least one corresponding slave; 
master Nodes are readable and writable ,slave Read only Don't write . advantage ： stay master outage , Consumers can directly from slave Read message , The real-time performance of messages will not be affected , Performance is almost as good as many master equally . shortcoming ： Using asynchronous replication may cause message loss problems .
many master many slave Synchronous double write , Similar to the previous mode , The difference lies in master And slave Data synchronization between . advantage ： Synchronous double write synchronization mode can ensure that data is not lost . shortcoming ： It takes a long time to send a single message , The performance will be poor ;
Synchronous brush set , The difference between asynchronous disk brushing ： Lie Duo master Write data to disk mode synchronously .

	Synchronous replication	Asynchronous replication
Concept	Immediate wait Master and Slave Only after all of them are successfully written, can they be fed back to the client	as long as Master Write a successful , Feedback the client to write the success status
reliability	High reliability , if Master Something goes wrong ,Slave There's all the backup data on the computer , Easy to recover	if Master Something goes wrong , There may be some data that hasn't been written yet Slave, May lose
efficiency	Because it is synchronous replication , It will increase the data write delay , Reduce system throughput	Because just write Master that will do , Therefore, the data writing delay is low , High throughput

Recommendations in practical applications Master and Slave Set to ASYNC_FLUSH Asynchronous disk brushing mode , The master and slave are configured as SYNC_MASTER Synchronous replication method , So even if one of the machines fails , It can still guarantee that the data will not be lost .

How to solve the reliability problem 
introduce MQ Resulting in data reliability problems , Or how to deal with message loss ？ Whole MQ In the link of , There are only three possible locations where messages may be lost , One is the production side , One is MQ Oneself , The last one is downstream consumers . 
Different MQ This solution is different , It needs to be analyzed separately .

ActiveMQ
Production end
Every time the program is sent to MQ After the news , Can receive MQ Of its Ack Respond to ,

The complexity of the system increases

Added MQ, It introduces whether the message will be consumed repeatedly , Message loss issues .

Data consistency

introduce MQ after , A single transaction can be split into multiple transactions . The problem of distributed transaction consistency is introduced .

MQ How to deal with the problem of repeated consumption

Idempotent check

The upstream cannot issue repeatedly 
The program guarantees not to create messages repeatedly , Use business transactions for control .
Downstream cannot repeat consumption
Put the message information into storage , Using a database 、Redis Store messages . A message is stored using a unique primary key , Check whether there exists each time , There is no continue processing , Otherwise, ignore or update according to the business ;
Put the message information into storage , The unique primary key constraint is performed according to the business , If there are duplicate data , A unique primary key conflict will be reported , Can't insert ;
If only redis Of set operation , There is no need to deal with . Even repetition does not affect ;