The elimination strategy of redis

Redis The data has been set TTL, It has been deleted after expiration ？ Why is there a so-called elimination strategy ？ For this reason, we need to start from redis We're talking about .

One 、 Expiration strategy

1. Delete periodically

redis Will set the expiration time for each key Put it in a separate dictionary , In the future, the dictionary will be periodically traversed to delete the expired key.

Redis By default, there will be ten expiration scans per second （100ms once ）, Expiration scan does not traverse all of the key, It's a simple greedy strategy .

• Randomly... From out of date dictionaries 20 individual key;

• Delete this 20 individual key Which has expired key;

• If it's overdue key Ratio of more than 1/4, Then repeat the steps 1;

redis The default is every time 100ms Just randomly select some key, Check if it is overdue , Delete if expired . Note that this is a random sample . Why random ？ Think about what if redis Hundreds of thousands of... Have been saved key , every other 100ms Just go through all the key Words , Will give you CPU Bring a lot of load .

2. Lazy deletion

The so-called lazy policy is to access this on the client side key When ,redis Yes key Check the expiration date of , Delete immediately if expired , I'm not going to give you anything back .

Regular deletion can cause many expiration key The time has not been deleted . So there's lazy delete . If your expired key, By regular deletion has not been deleted , Still in memory , Unless your system checks that key, Will be redis Delete . This is called lazy delete , That is, when you take the initiative to check the expired key when , If you find that key Out of date , Delete it immediately , Don't return anything .

summary ： Periodic deletion is centralized processing , Lazy delete is a piecemeal process .

Two 、 Why the elimination strategy is needed

With the above expiration policy explanation , It's easy to understand why elimination strategies are needed , Because neither regular sampling deletion nor lazy deletion is a completely accurate deletion , There will still be key Scenes that have not been deleted , So we need to replace the memory strategy .

In the use of Redis In the process of , When Redis After the cache is full ,Redis The data will be eliminated according to the configured elimination strategy . from Redis4.0 After that, there are altogether 8 Kind of elimination strategy . Let's take a look at each of them .

3、 ... and 、 Memory retirement strategy

1. noeviction

No data obsolescence , It's also Redis Default configuration . At this time , When the cache is full , Another request to write in ,Redis No more services , Direct return error .

2.volatile-random

When the cache is full , Randomly delete key value pairs with expiration time set .

3.volatile-ttl

When the cache is full , The key value alignment with expiration time is set , Delete according to the order of expiration time , The earlier the expiration date, the earlier it is deleted .

4.volatile-lru

When the cache is full , For key value pairs with expiration time set , use LRU The algorithm is eliminated , Not familiar with LRU You can read this article .

5.volatile-lfu

When the cache is full , For key value pairs with expiration time set , use LFU Algorithm for elimination .

6.allkeys-random

When the cache is full , Randomly select and delete data from all key value pairs .

7.allkeys-lru

After the cache is full , Use LRU The algorithm filters and deletes all data .

8.allkeys-lfu

When the cache is full , Use LRU The algorithm filters and deletes all data .

In daily use , Configure corresponding strategies mainly according to your data requirements , Here I give you three suggestions .

• We give priority to allkeys-lru Strategy . such , We can use LRU Algorithm to eliminate those less commonly used data , Put the most commonly used ones in the cache , So as to improve the performance of the application . If your data has a clear distinction between hot and cold , I suggest you use allkeys-lru Strategy .
• If the access frequency of your data is not different , There is no difference between hot and cold , Use it directly allkeys-random Strategy , Just randomly select the data to be eliminated .
• If your data has top requirements , Such as top news . Then we choose volatile-lru Strategy , At the same time, no expiration time is set for the top data , thus , Top data will never be deleted , And other data with expiration time , It will be more LRU The algorithm is eliminated .