Spring Boot multi-level cache starter

Opinionated version of multi-level caching for Spring Boot with Redis as L2 (remote) cache and Caffeine as L1 (local) cache with a Circuit Breaker pattern for L2 cache calls.

This version does not allow setting most of the local cache properties in favor of managing local cache expiry by itself.

Usage

Maven

<dependency>
  <groupId>io.github.suppierk</groupId>
  <artifactId>spring-boot-multilevel-cache-starter</artifactId>
  <version>3.4.0.0</version>
</dependency>

Gradle

implementation 'io.github.suppierk:spring-boot-multilevel-cache-starter:3.4.0.0'

Use cases

Suitable for

• Microservices working with immutable cached entities under low latency requirements
  • The goal is to not only reduce the number of calls to external service but also reduce the number of calls to Redis

Not a good fit for

• Mutable cached entities
• Entities with short time to live (< 5 minutes)
• Cases when entities in local cache must outlive entities in distributed cache
  • Consider using only local cache instead
• Cases when all calls to Redis must be synchronized with distributed locks

Ideas

• Use well-known Spring primitives for implementation
• Microservices environment needs to fit the requirement of fault tolerance:
  • Redis calls covered by Resilience4j Circuit Breaker which allows falling back to use local cache at the cost of increased latency and more calls to external services.
• Redis TTL behaves similar to expireAfterWrite in Caffeine which allows us to set randomized expiry time for local cache:
  • This is useful to ensure that local cache entries will expire earlier for a higher chance to hit Redis instead of performing external call.
  • This also implicitly reduces the load on the Redis by spreading calls to it over time.
  • In the case of Redis connection errors, randomized expiry and Circuit Breaker will help to mitigate thundering herd problem.
• Expiry randomization follows the rule: (time-to-live / 2) * (1 ± ((expiry-jitter / 100) * RNG(0, 1))), for example:
  • If spring.cache.multilevel.time-to-live is 1h
  • And spring.cache.multilevel.local.expiry-jitter is 50 (percents)
  • Then entries in local cache will expire in approximately 15-45m:

(1h / 2) * (1 ± ((50 / 100) * RNG(0, 1))) ->
30m * (1 ± MAXRNG(0.5)) ->
30m * RANGE(0.5, 1.5) ->
15-45m

Default configuration

spring:
  data:
    redis:
      host: ${HOST:localhost}
      port: ${PORT:6379}
  cache:
    type: redis
    
    # These properties are custom
    multilevel:
      # Redis properties
      time-to-live: 1h
      use-key-prefix: false
      key-prefix: ""
      topic: "cache:multilevel:topic"
      # Local Caffeine cache properties
      local:
        max-size: 2000
        expiry-jitter: 50
        expiration-mode: after-create
        # other valid values for expiration-mode: after-update, after-read
      # Resilience4j Circuit Breaker properties for Redis
      circuit-breaker:
        failure-rate-threshold: 25
        slow-call-rate-threshold: 25
        slow-call-duration-threshold: 250ms
        sliding-window-type: count_based
        permitted-number-of-calls-in-half-open-state: 20
        max-wait-duration-in-half-open-state: 5s
        sliding-window-size: 40
        minimum-number-of-calls: 10
        wait-duration-in-open-state: 2500ms

Honorable mentions

• Circuit Breaker Redis Cache by gee4vee
• Multilevel cache Spring Boot starter by pig777

License