redis data removal policy and eviction policy

Delete Policy

Wherein the data Redis

• Redis is a database-level memory, all data stored in memory, the data memory may be acquired by a status command ttl
  • XX: time-sensitive data
  • -1: permanent data
  • -2: the data has expired or deleted data or undefined data

Wherein the data Redis

Data removal policies

1. Regularly delete
2. Inert delete
3. Periodically delete

Data storage structure timeliness

Target data removal policy

To find a balance between memory consumption and CPU, a loss will result in a decline in overall Redis performance, server downtime or even lead to memory leaks

---

Regularly delete

• Create a timer, when the key is provided with the expiration time, and the expiration time arrives, executed immediately delete operation on key tasks by the timer
• Advantages: save memory, due to deletion, quick release out unnecessary memory footprint

• Disadvantage: CPU pressure, no matter how high the amount of CPU load at this time, are occupied CPU, it affects the server redis instruction throughput and response time

• Summary: in exchange for storage space with a processor performance (take time for space)

Inert delete

• The data expiration time arrives, without processing. The next time you access the data, etc.
  • If not expired, return data
  • Found to have expired, deleted, there is no return
• Advantages: saving CPU performance, must be deleted when it is found that deletion
• Disadvantages: large memory pressure, long-term memory for emerging data
• Summary: with storage space for processor performance

---

Periodically delete

• Redis periodically in rotation timeliness data library using random policy, accounting data using the expired elimination frequency is controlled
• Features 1: CPU occupancy provided with a peak performance, the detection frequency customizable settings
• Feature 2: Memory pressure is not great, long-term memory for the cold data will be continued clean-up
• Summary: periodic checking of storage space (random checks, targeted surveillance)

---

Deleting a policy comparison

Types of	Feature
Regularly delete	Save memory, no footprint regardless of time CPU resources, high-frequency take time for space
Inert delete	Memory footprint serious delay execution, CPU utilization high to get space for time
Periodically delete	Regular random memory cleanup cost per CPU resources required to maintain a fixed memory random checks, targeted surveillance

---

Eviction policy

• Redis使用内存存储数据，在执行每一个命令前，会调用freeMemoryIfNeeded()检测内存是否充足。如果内存不满足新加入数据的最低存储要求，redis要临时删除一些数据为当前指令清理存储空间。清理数据的策略被称为逐出算法。
• 注意：逐出数据的过程不是100%能够清理出足够的可使用的内存空间，如果不成功则反复执行。当对所有数据尝试完毕后，如果不能达到内存清理的要求，将出现错误信息。

影响数据逐出的相关配置

• 最大可使用内存

maxmemory

占用物理内存的比例，默认值为0，表示不限制。生产环境中根据需求设定，通常设置在50%以上。

• 每次选取待删除数据的个数

maxmemory-samples

选取数据时并不会全库扫描，导致严重的性能消耗，降低读写性能。因此采用随机获取数据的方式作为待检测删除数据

• 删除策略

maxmemory-policy

达到最大内存后的，对被挑选出来的数据进行删除的策略

---

• 检测易失数据（可能会过期的数据集server.db[i].expires）
  1. volatile-lru: 挑选最近最少使用的数据淘汰
  2. volatile-lfu:挑选最近使用次数较少的数据淘汰
  3. volatile-ttl:挑选将要过期的数据淘汰
  4. volatile-random: 任意选择数据淘汰

---

• 检测全库数据（所有数据集server.db[i].dict）
  1. allkeys-lru: 挑选最近最少使用的数据淘汰
  2. allkeys-lfu: 挑选最近使用次数最少的数据淘汰
  3. allkeys-random: 任意选择数据淘汰

• 放弃数据驱逐

  1. no-enviction(驱逐):禁止驱逐数据(redis4.0默认策略)，会引发OOM(out of memory)

maxmemory-policy volatile-lru

数据逐出策略配置依据

• 使用INFO命令输出监控信息,查询缓存hit和miss次数，根据业务需求挑优Redis配置