The most comprehensive redis transaction control in 2022 (with illustration)

Redis Transaction control

Concept of transactions

redis A transaction is a queue of command execution , Wrapping a set of predefined commands as a whole （ A line ）. When performing , One time, in order of addition , There will be no interruption or interference in the middle .

In a queue , Disposable 、 Sequence 、 To execute a series of orders exclusively

1、 The basic operation of a transaction

Open transaction

multi

effect ： Set the start position of the transaction , After this instruction is executed , All subsequent instructions are added to the transaction

Perform transactions

exec

effect ： Set the end position of the transaction , Execute transactions at the same time . And multi Pairs appear , Use in pairs

Be careful ： The command to join the transaction temporarily enters the task queue , Not immediately , Only execute exec It's the order that begins

Cancel the business

discard

effect ： Terminate the definition of the current transaction , It happened in multi after ,exec Before

If you cancel the transaction and then execute exec Will prompt (error) ERR EXEC without MULTI error message

2、 Transaction considerations

1. Syntax input error

   • It means that the command writing grammar is incorrect
   • If there is a syntax error in the command contained in the defined transaction , All commands in the overall transaction will not be executed . Including commands with correct syntax .

   

2. Running error

   • Indicates that the command format is correct , But it can't be executed correctly . For example list Conduct incr operation
   • Commands that run correctly execute , Running the wrong command will not be executed

   

   Be careful ： The data corresponding to the executed command will not be rolled back automatically , The programmer needs to implement rollback in the code

3. Manual transaction rollback

   1. Record the status before the data affected during the operation
      • Single data ：string
      • More data ：hash、list、set、zset
   2. Set the command to restore all modified items
      • Single data ： direct set（ Pay attention to the surrounding properties , For example, timeliness ）
      • More data ： Modify the corresponding value or clone and copy as a whole

3、 Transaction execution based on specific conditions 【 Transaction lock 】

Business scenario

Tmall double 11 During the hot sale , Replenish the sold out goods ,4 Every salesman has the right to replenish . Replenishment may be a series of operations , Multiple continuous operations are involved , How to ensure that the operation will not be repeated ？

Business analysis

1. It is possible for multiple clients to operate on the same set of data at the same time , And once the data is modified by the operation , Will not apply to continue operation
2. Lock the data to be manipulated before the operation , As soon as it changes , Terminate the current operation

Solution

1. Yes key Add monitoring lock , In execution exec If key There is a change , Terminate transaction execution

   watch key1 [key2……]
   

   

   obtain name by nil Why ：

   1. Modified in other clients name value
   2. watch A change in value was detected and the transaction was terminated
   3. The value obtained after exiting the transaction is the value modified by the second client

2. Cancel for all key Surveillance

   In the monitored key If it is modified, the transaction will be finalized , If you cancel monitoring, modifying the original monitored value will not affect the transaction execution

   unwatch
   

4、 Transaction execution based on specific conditions 【 Distributed lock 】

Business scenario

Tmall double 11 During the hot sale , Replenish the sold out goods , And the replenishment is completed . Customers are enthusiastic about buying ,3 Buy all goods in seconds . The inventory has been completely emptied in this replenishment , How to prevent the last item from being purchased by more than one person at the same time ？【 Oversold problem 】

Business analysis

1. Use watch Monitor one key Is there any change that can't solve the problem , The specific data to be monitored here
2. although redis It's single threaded , But when multiple clients operate on the same data at the same time , How to avoid not being modified at the same time ？

Solution

1. Use setnx Set a public lock

   setnx lock-key value
   

   utilize setnx The return value characteristic of the command , If there is a value, the setting fails , If there is no value, the setting is successful

   • For the return setting successful , Have control , Carry out specific business operations in the next step
   • For return settings that failed , No control , To line up or wait for

2. After the operation, pass del Operate the release lock

Be careful ： The above solution is a design concept , Rely on norms to ensure , Risky

5、 Transaction execution based on specific conditions 【 Distributed locks improve time limits 】

Business scenario

Mechanisms that rely on distributed locks , When a user operates, the corresponding client goes down , And the lock has been obtained at this time . How to solve ？

Business analysis

1. Since the lock operation is controlled by the user, lock and unlock , There must be a risk of not unlocking after locking
2. The unlocking operation that needs to be performed cannot rely solely on user control , At the system level, the corresponding minimum treatment scheme shall be given

Solution

Use expire For lock key Add time limit , Don't release , Abandon lock

//  Add a time limit to the lock （ second ）
expire lock-key second
//  Add a time limit to the lock （ millisecond ）
pexpire lock-key milliseconds

Because operations are usually microseconds or milliseconds , Therefore, the locking time should not be set too large . The specific time needs to be confirmed after business test

1. for example ： The longest execution time for holding a lock 127ms, The shortest execution time 7ms.
2. Testing the maximum execution time of a million times corresponds to the maximum time consumption of a command , Testing millions of network delays takes an average of
3. Lock time setting is recommended ： Most time consuming 120%+ Average network latency 110%
4. If the business takes the most time << Average network latency , Usually it is 2 An order of magnitude , It takes a long time to take one of them