Skip to content

Latest commit

 

History

History
121 lines (91 loc) · 4.53 KB

Caching DTO beans with Spring cache and Redis.md

File metadata and controls

121 lines (91 loc) · 4.53 KB

#Caching DTO beans with Spring cache and Redis

Chinese version 微信公众号文章 大规模分布式电商系统系列文章之三：缓存系统

#Motivation Hybris creates a lot of new DTO beans object every time when you access the storefront. This impedes the responsive time and creates a lot of small short-live objects. This also gives the JVM high pressure to do the garbage collection. You can consult the topic: converters and populators to get more info.

⚠️⚠️⚠️ In-memory cache usually has better performance than network distributed cache, nevertheless distributed cache comes with better scalability. You can switch back and forth different cache providers easily based on spring cache abstration. ⚠️⚠️⚠️

#Environment ##Software:

  • Redis 3.0.7
  • Spring framework 4.1.7 embedded with hybris
  • Spring data redis 1.7.1
  • hybris 6.0
  • MariaDB 10.1.13

#Design

  • Caching the DTO beans in the Redis server.

  • Using Spring AOP to add cache apsect directly. no need to change any code. (some converters have complex data structure, maybe needs to change code then.)

  • Setting different TTL for different data type using cacheManager.

#Implementation ##Install & start Redis

brew install redis
redis-server /usr/local/etc/redis.conf

##Config spring cache

<beans>
...

    <cache:annotation-driven />

    <bean id="jedisConnFactory" class="org.springframework.data.redis.connection.jedis.JedisConnectionFactory"
          p:usePool="true"/>

    <bean id="stringRedisSerializer" class="org.springframework.data.redis.serializer.StringRedisSerializer"/>
    <bean id="int2StringRedisSerializer" class="org.springframework.data.redis.serializer.GenericToStringSerializer"
          c:type="java.lang.Integer"/>
    <bean id="genericJackson2JsonRedisSerializer" class="org.springframework.data.redis.serializer.GenericJackson2JsonRedisSerializer"/>


    <bean id="redisTemplate" class="org.springframework.data.redis.core.RedisTemplate"
          p:connectionFactory-ref="jedisConnFactory"
          p:keySerializer-ref="stringRedisSerializer"
          p:hashKeySerializer-ref="stringRedisSerializer"
          p:valueSerializer-ref="genericJackson2JsonRedisSerializer"
          p:hashValueSerializer-ref="genericJackson2JsonRedisSerializer"/>


    <bean id="cacheManager" class="org.springframework.data.redis.cache.RedisCacheManager"
          c:redisOperations-ref="redisTemplate"
          p:defaultExpiration="600"
          p:usePrefix="true"/>


    <cache:advice id="cacheAdvice">
        <cache:caching cache="beans:cache">
            <cache:cacheable method="convert" key="#source.class.getName().concat(':').concat(#source.toString())"/>
            <cache:cacheable method="convertAll*" key="#sources.toString()" unless="#result == null"/>
        </cache:caching>
    </cache:advice>

    <aop:config>
        <aop:advisor advice-ref="cacheAdvice" pointcut="bean(customerConverter) || bean(productConverter) ||
            bean(categoryUrlConverter) || bean(categoryConverter) ||
            bean(imageConverter) || bean(productUrlConverter) || bean(productReferenceConverter) ||
            bean(promotionsConverter) || bean(languageConverter) || bean(currencyConverter) ||
            bean(countryConverter) || bean(promotionResultConverter) || bean(customerReviewConverter) ||
            bean(principalConverter) || bean(classificationConverter) || bean(featureConverter)"/>
    </aop:config>
    
</beans>

#Verify ##Debug verify Debug to check whether the cached method works or not.

##Cache verify DTO beans cached in Redis

#Performance evaluation I did a 10 user load testing for 1 min accessing digital camera product listing page.

No. Guava-CPU Guava-Memory Guava-TPS Redis-CPU Redis-Memory Redis-TPS hybris-CPU hybris-Memory hybris-TPS
1 43 23 34
2 32 35 41
3 40 43 42
Average 38 34 39

#Conclusion

  • Cache DTO only if you need it
  • Cache DTO only when it becomes bottleneck
  • Choose the cache mechanism and provider wisely
  • Cache really can bypass some logic execution
  • Consider serialization overhead when distribute cache
  • Monitor and manage it

#Open issues

  • Data contention
  • Key distribution

#Next step

Winston Zhang

April 18, 2016