高并发高可用复杂系统中的缓存架构(十九) kafka在此项目中的应用

多级缓存的架构主要用来解决的问题是：时效高低数据的维度存储

时效性不高的数据，比如一些商品的基本信息，如果发生了变更，假设在 5 分钟之后再更新到页面中， 供用户观察到，也是 ok 的，那么我们采取的是异步更新缓存的策略

时效性要求很高的数据，如库存，采取的是数据库 + 缓存双写的技术方案，也解决了双写的一致性的问题

上面这两条可能直接看觉得好像差不多的，这里忽略了一个解释，对于页面来说，需要静态的生成页面， 这个过程可能稍微耗时一些，而对于双写来说则快太多了，它不负责页面渲染等工作，只需要把缓存数据更新即可

缓存数据生产服务，监听一个消息队列，然后数据源服务（商品信息管理服务）发生了数据变更之后， 就将数据变更的消息推送到消息队列中

缓存数据生产服务可以去消费到这个数据变更的消息，然后根据消息的指示提取一些参数， 然后调用对应的数据源服务的接口拉取数据，这个时候一般是从 mysql 库中拉去的

消息队列这里采用的是 kafka，这里选择 kafka 另外一个原因： 后面我们还要用 zookeeper 来解决缓存的分布式并发更新的问题（如分布式锁解决）

而 kafka 集群是依赖 zookeeper 集群，所以先搭建 zookeeper 集群，再搭建 kafka 集群

zookeeper + kafka 的集群，都至少是三节点

配置文件介绍

  
  [root@master config]# more server.properties 
  # Licensed to the Apache Software Foundation (ASF) under one or more
  # contributor license agreements.  See the NOTICE file distributed with
  # this work for additional information regarding copyright ownership.
  # The ASF licenses this file to You under the Apache License, Version 2.0
  # (the "License"); you may not use this file except in compliance with
  # the License.  You may obtain a copy of the License at
  #
  #    http://www.apache.org/licenses/LICENSE-2.0
  #
  # Unless required by applicable law or agreed to in writing, software
  # distributed under the License is distributed on an "AS IS" BASIS,
  # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
  # See the License for the specific language governing permissions and
  # limitations under the License.
  ​
  # see kafka.server.KafkaConfig for additional details and defaults
  ​
  ############################# Server Basics #############################
  ​
  # The id of the broker. This must be set to a unique integer for each broker.
  ##每一个broker在集群中的唯一标示，要求是正数。在改变IP地址，不改变broker.id的话不会影响consumers
  broker.id=0
  ​
  # Switch to enable topic deletion or not, default value is false
  ## 是否允许自动创建topic ，若是false，就需要通过命令创建topic
  delete.topic.enable=true
  ​
  ############################# Socket Server Settings #############################
  ​
  # The address the socket server listens on. It will get the value returned from 
  # java.net.InetAddress.getCanonicalHostName() if not configured.
  #   FORMAT:
  #     listeners = listener_name://host_name:port
  #   EXAMPLE:
  #     listeners = PLAINTEXT://your.host.name:9092
  #listeners=PLAINTEXT://:9092
  ##提供给客户端响应的端口
  port=9092
  host.name=192.168.1.128
  ​
  # Hostname and port the broker will advertise to producers and consumers. If not set, 
  # it uses the value for "listeners" if configured.  Otherwise, it will use the value
  # returned from java.net.InetAddress.getCanonicalHostName().
  #advertised.listeners=PLAINTEXT://your.host.name:9092
  ​
  # Maps listener names to security protocols, the default is for them to be the same. See the config documentation for more details
  #listener.security.protocol.map=PLAINTEXT:PLAINTEXT,SSL:SSL,SASL_PLAINTEXT:SASL_PLAINTEXT,SASL_SSL:SASL_SSL
  ​
  # The number of threads handling network requests
  ## broker 处理消息的最大线程数，一般情况下不需要去修改
  num.network.threads=3
  ​
  # The number of threads doing disk I/O
  ## broker处理磁盘IO 的线程数 ，数值应该大于你的硬盘数
  num.io.threads=8
  ​
  # The send buffer (SO_SNDBUF) used by the socket server
  ## socket的发送缓冲区，socket的调优参数SO_SNDBUFF
  socket.send.buffer.bytes=102400
  ​
  # The receive buffer (SO_RCVBUF) used by the socket server
  ## socket的接受缓冲区，socket的调优参数SO_RCVBUFF
  socket.receive.buffer.bytes=102400
  ​
  # The maximum size of a request that the socket server will accept (protection against OOM)
  ## socket请求的最大数值，防止serverOOM，message.max.bytes必然要小于socket.request.max.bytes，会被topic创建时的指定参数覆盖
  socket.request.max.bytes=104857600
  ​
  ​
  ############################# Log Basics #############################
  ​
  # A comma seperated list of directories under which to store log files
  ##kafka数据的存放地址，多个地址的话用逗号分割/data/kafka-logs-1，/data/kafka-logs-2
  log.dirs=/tmp/kafka-logs
  ​
  # The default number of log partitions per topic. More partitions allow greater
  # parallelism for consumption, but this will also result in more files across
  # the brokers.
  ##每个topic的分区个数，若是在topic创建时候没有指定的话会被topic创建时的指定参数覆盖
  num.partitions=1
  ​
  # The number of threads per data directory to be used for log recovery at startup and flushing at shutdown.
  # This value is recommended to be increased for installations with data dirs located in RAID array.
  ##我们知道segment文件默认会被保留7天的时间，超时的话就
  ##会被清理，那么清理这件事情就需要有一些线程来做。这里就是
  ##用来设置恢复和清理data下数据的线程数量
  num.recovery.threads.per.data.dir=1
  ​
  ############################# Log Flush Policy #############################
  ​
  # Messages are immediately written to the filesystem but by default we only fsync() to sync
  # the OS cache lazily. The following configurations control the flush of data to disk.
  # There are a few important trade-offs here:
  #    1. Durability: Unflushed data may be lost if you are not using replication.
  #    2. Latency: Very large flush intervals may lead to latency spikes when the flush does occur as there will be a lot of data to flush.
  #    3. Throughput: The flush is generally the most expensive operation, and a small flush interval may lead to exceessive seeks.
  # The settings below allow one to configure the flush policy to flush data after a period of time or
  # every N messages (or both). This can be done globally and overridden on a per-topic basis.
  ​
  # The number of messages to accept before forcing a flush of data to disk
  #log.flush.interval.messages=10000
  ​
  # The maximum amount of time a message can sit in a log before we force a flush
  #log.flush.interval.ms=1000
  ​
  ############################# Log Retention Policy #############################
  ​
  # The following configurations control the disposal of log segments. The policy can
  # be set to delete segments after a period of time, or after a given size has accumulated.
  # A segment will be deleted whenever *either* of these criteria are met. Deletion always happens
  # from the end of the log.
  ​
  # The minimum age of a log file to be eligible for deletion due to age
  ##segment文件保留的最长时间，默认保留7天（168小时），
  ##超时将被删除，也就是说7天之前的数据将被清理掉。
  log.retention.hours=168
  ​
  # A size-based retention policy for logs. Segments are pruned from the log as long as the remaining
  # segments don't drop below log.retention.bytes. Functions independently of log.retention.hours.
  #log.retention.bytes=1073741824
  ​
  # The maximum size of a log segment file. When this size is reached a new log segment will be created.
  ###日志文件中每个segment的大小，默认为1G
  log.segment.bytes=1073741824
  ​
  # The interval at which log segments are checked to see if they can be deleted according
  # to the retention policies
  ##上面的参数设置了每一个segment文件的大小是1G，那么
  ##就需要有一个东西去定期检查segment文件有没有达到1G，
  ##多长时间去检查一次，就需要设置一个周期性检查文件大小
  ##的时间（单位是毫秒）。
  log.retention.check.interval.ms=300000
  ​
  ############################# Zookeeper #############################
  ​
  # Zookeeper connection string (see zookeeper docs for details).
  # This is a comma separated host:port pairs, each corresponding to a zk
  # server. e.g. "127.0.0.1:3000,127.0.0.1:3001,127.0.0.1:3002".
  # You can also append an optional chroot string to the urls to specify the
  # root directory for all kafka znodes.
  #zookeeper.connect=localhost:2181
  ##消费者集群通过连接Zookeeper来找到broker。
  ##zookeeper连接服务器地址
  zookeeper.connect=master:2181,worker1:2181,worker2:2181
  ​
  # Timeout in ms for connecting to zookeeper
  zookeeper.connection.timeout.ms=6000
  
启动kafka        
  [root@master kafka]# ./bin/kafka-server-start.sh config/server.properties &
  [2018-06-25 02:31:21,931] INFO KafkaConfig values: 
          advertised.host.name = null
          advertised.listeners = null
          advertised.port = null
          authorizer.class.name = 
          auto.create.topics.enable = true
          auto.leader.rebalance.enable = true
          background.threads = 10
          broker.id = 0
          broker.id.generation.enable = true
          broker.rack = null
          compression.type = producer
          connections.max.idle.ms = 600000
          controlled.shutdown.enable = true
          controlled.shutdown.max.retries = 3
          controlled.shutdown.retry.backoff.ms = 5000
          controller.socket.timeout.ms = 30000
创建topic

  #创建topic topic名字为gilbert
  [root@master kafka]# ./bin/kafka-topics.sh --create --zookeeper master:2181,worker1:2181,worker2:2181 --replication-factor 3 --partitions 3 --topic gilbert
  Created topic "gilbert".
查看topic

  
  [root@master kafka]# ./bin/kafka-topics.sh --describe --zookeeper master:2181,worker1:2181,worker2:2181 --topic gilbert
  Topic:gilbert   PartitionCount:3        ReplicationFactor:3     Configs:
          Topic: gilbert  Partition: 0    Leader: 2       Replicas: 2,0,1 Isr: 2,0,1
          Topic: gilbert  Partition: 1    Leader: 0       Replicas: 0,1,2 Isr: 0,1,2
          Topic: gilbert  Partition: 2    Leader: 1       Replicas: 1,2,0 Isr: 1,2,0
            
  [root@master kafka]# ./bin/kafka-topics.sh --list --zookeeper master:2181,worker1:2181,worker2:2181
  gilbert
  test
创建producer

  
  ./bin/kafka-console-producer.sh --broker-list master:9092 -topic gilbert
创建consumer，分别在3台服务器上执行创建消费者

  #192.168.1.128服务器 
  [root@master kafka]# ./bin/kafka-console-consumer.sh --zookeeper master:2181,worker1:2181,worker2:2181 -topic gilbert --from-beginning
  Using the ConsoleConsumer with old consumer is deprecated and will be removed in a future major release. Consider using the new consumer by passing [bootstrap-server] instead of [zookeeper].
    
  #192.168.1.129服务器 
  [root@worker1 kafka_2.10-0.10.2.0]# ./bin/kafka-console-consumer.sh --zookeeper master:2181,worker1:2181,worker2:2181 -topic gilbert --from-beginning
  Using the ConsoleConsumer with old consumer is deprecated and will be removed in a future major release. Consider using the new consumer by passing [bootstrap-server] instead of [zookeeper].
  ​
  #192.168.1.130服务器 
  [root@worker2 kafka_2.10-0.10.2.0]#  ./bin/kafka-console-consumer.sh --zookeeper master:2181,worker1:2181,worker2:2181 -topic gilbert --from-beginning
  Using the ConsoleConsumer with old consumer is deprecated and will be removed in a future major release. Consider using the new consumer by passing [bootstrap-server] instead of [zookeeper].
  ​
  ​
在#192.168.1.128服务器上生产者控制台输入：hello kafka进行测试

在3台服务器上的消费者都正常接收到消息

删除topic

  
  [root@master kafka]# ./bin/kafka-topics.sh --delete --zookeeper master:2181,worker1:2181,worker2:2181 --topic test
  Topic test is marked for deletion.
  Note: This will have no impact if delete.topic.enable is not set to true

springboot集成kafka

1.生产者kafka-producer

a) pom文件

  <dependency>
    <groupId>org.springframework.boot</groupId>
    <artifactId>spring-boot-starter</artifactId>
  </dependency>
  <dependency>
    <groupId>org.springframework.kafka</groupId>
    <artifactId>spring-kafka</artifactId>
  </dependency>
  <dependency>
    <groupId>com.google.code.gson</groupId>
    <artifactId>gson</artifactId>
    <version>2.8.2</version>
  </dependency>
  <dependency>
    <groupId>org.projectlombok</groupId>
    <artifactId>lombok</artifactId>
    <optional>true</optional>
  </dependency>
  <dependency>
    <groupId>org.springframework.boot</groupId>
    <artifactId>spring-boot-starter-test</artifactId>
    <scope>test</scope>
  </dependency>
b) yml配置文件,本例为kafka3节点集群

  
  spring:
    kafka:
      bootstrap-servers: http://master:9092,http://worker1:9092,http://worker2:9092
      producer:
        retries: 0
        batch-size: 16384
        buffer-memory: 33554432
        key-serializer: org.apache.kafka.common.serialization.StringSerializer
        value-serializer: org.apache.kafka.common.serialization.StringSerializer
c) message消息实体类

  
  @Data
  public class Message {
      private Long id;    //id
      private String msg; //消息
      private Date sendTime;  //时间戳
  }
d) 生产者

  @Component
  @Slf4j
  public class KafkaProducer {
  ​
      @Autowired
      private KafkaTemplate<String, String> kafkaTemplate;
  ​
      private Gson gson = new GsonBuilder().create();
  ​
      //发送消息方法
      public void send() {
          Message message = new Message();
          message.setId(System.currentTimeMillis());
          message.setMsg(UUID.randomUUID().toString());
          message.setSendTime(new Date());
          log.info("+++++++++++++++++++++  message = {}", gson.toJson(message));
          //topic-ideal为主题
          kafkaTemplate.send("topic-ideal", gson.toJson(message));
      }
  }
e) 测试类,运行kafkaProducer方法即可

  @RunWith(SpringRunner.class)
  @SpringBootTest
  public class KafkaProducerApplicationTests {
  ​
      @Autowired
      private KafkaProducer kafkaProducer;
  ​
      @Test
      public void kafkaProducer(){
          this.kafkaProducer.send();
      }
  ​
      @Test
      public void contextLoads() {
      }
  ​
  }

2. 消费者kafka-consumer

a) pom文件

  <dependency>
    <groupId>org.springframework.boot</groupId>
    <artifactId>spring-boot-starter</artifactId>
  </dependency>
  <dependency>
    <groupId>org.springframework.kafka</groupId>
    <artifactId>spring-kafka</artifactId>
  </dependency>
  <dependency>
    <groupId>com.google.code.gson</groupId>
    <artifactId>gson</artifactId>
    <version>2.8.2</version>
  </dependency>
  <dependency>
    <groupId>org.projectlombok</groupId>
    <artifactId>lombok</artifactId>
    <optional>true</optional>
  </dependency>
b) yml配置文件

  
  server:
    port: 9999
  spring:
    kafka:
      bootstrap-servers: http://master:9092,http://worker1:9092,http://worker2:9092
      consumer:
        group-id: ideal-consumer-group
        auto-offset-reset: earliest
        enable-auto-commit: true
        auto-commit-interval: 20000
        key-deserializer: org.apache.kafka.common.serialization.StringDeserializer
        value-deserializer: org.apache.kafka.common.serialization.StringDeserializer
c) 消费者类

  
  @Component
  @Slf4j
  public class KafkaConsumer {
  ​
      @KafkaListener(topics = {"topic-ideal"})
      public void consumer(ConsumerRecord<?, ?> record){
          Optional<?> kafkaMessage = Optional.ofNullable(record.value());
          if (kafkaMessage.isPresent()) {
              Object message = kafkaMessage.get();
              log.info("----------------- record =" + record);
              log.info("------------------ message =" + message);
          }
      }
  }
运行消费者kafka-consumer，再运行kafka-producer工程测试类KafkaProducerApplicationTests中kafkaProducer()方法,可以看到消费者后台正常接收消息 

 

编写业务逻辑

1. 两种服务会发送来数据变更消息：商品信息服务和商品店铺信息服务，每个消息都包含服务名以及商品 id

2. 接收到消息之后，根据商品 id 到对应的服务拉取数据

   这一步，我们采取简化的模拟方式，就是在代码里面写死，会获取到什么数据，不去实际再写其他的服务去调用了

3. 商品信息：id、名称、价格、图片列表、商品规格、售后信息、颜色、尺寸

4. 商品店铺信息：其他维度

   用这个维度模拟出来缓存数据维度化拆分：id、店铺名称、店铺等级、店铺好评率

5. 分别拉取到了数据之后，将数据组织成 json 串，然后分别存储到 ehcache 中和 redis 缓存中

这里的业务逻辑代码，因为是模拟这个场景，所以事先比较简单，重要的类如下

接收到事件之后，分别处理事件

public class KafkaMessageProcessor implements Runnable {
​
    private KafkaStream kafkaStream;
    private CacheService cacheService;
    private Logger log = LoggerFactory.getLogger(getClass());
​
    public KafkaMessageProcessor(KafkaStream kafkaStream, CacheService cacheService) {
        this.kafkaStream = kafkaStream;
        this.cacheService = cacheService;
    }
​
    public void run() {
        ConsumerIterator<byte[], byte[]> it = kafkaStream.iterator();
        while (it.hasNext()) {
            String message = new String(it.next().message());
​
            // 首先将message转换成json对象
            JSONObject messageJSONObject = JSONObject.parseObject(message);
​
            // 从这里提取出消息对应的服务的标识
            String serviceId = messageJSONObject.getString("serviceId");
​
            // 如果是商品信息服务
            if ("productInfoService".equals(serviceId)) {
                processProductInfoChangeMessage(messageJSONObject);
            } else if ("shopInfoService".equals(serviceId)) {
                processShopInfoChangeMessage(messageJSONObject);
            }
        }
    }
​
    /**
     * 处理商品信息变更的消息
     */
    private void processProductInfoChangeMessage(JSONObject messageJSONObject) {
        // 提取出商品id
        Long productId = messageJSONObject.getLong("productId");
​
        // 调用商品信息服务的接口
        // 直接用注释模拟：getProductInfo?productId=1，传递过去
        // 商品信息服务，一般来说就会去查询数据库，去获取productId=1的商品信息，然后返回回来
​
        String productInfoJSON = "{\"id\": 1, \"name\": \"iphone7手机\", \"price\": 5599, \"pictureList\":\"a.jpg,b.jpg\", \"specification\": \"iphone7的规格\", \"service\": \"iphone7的售后服务\", \"color\": \"红色,白色,黑色\", \"size\": \"5.5\", \"shopId\": 1}";
        ProductInfo productInfo = JSONObject.parseObject(productInfoJSON, ProductInfo.class);
        cacheService.saveProductInfo2LocalCache(productInfo);
        log.info("获取刚保存到本地缓存的商品信息：" + cacheService.getProductInfoFromLocalCache(productId));
        cacheService.saveProductInfo2ReidsCache(productInfo);
    }
​
    /**
     * 处理店铺信息变更的消息
     */
    private void processShopInfoChangeMessage(JSONObject messageJSONObject) {
        // 提取出商品id
        Long productId = messageJSONObject.getLong("productId");
        Long shopId = messageJSONObject.getLong("shopId");
        // 这里也是模拟去数据库获取到了信息
​
        String shopInfoJSON = "{\"id\": 1, \"name\": \"小王的手机店\", \"level\": 5, \"goodCommentRate\":0.99}";
        ShopInfo shopInfo = JSONObject.parseObject(shopInfoJSON, ShopInfo.class);
        cacheService.saveShopInfo2LocalCache(shopInfo);
        log.info("获取刚保存到本地缓存的店铺信息：" + cacheService.getShopInfoFromLocalCache(shopId));
        cacheService.saveShopInfo2ReidsCache(shopInfo);
    }
}

把缓存的读写封装了到 service 中。上面通过 service 去操作缓存

@Service
public class CacheServiceImpl implements CacheService {
    public static final String CACHE_NAME = "local";
​
    @Resource
    private JedisCluster jedisCluster;
​
    /**
     * 将商品信息保存到本地缓存中
     */
    @CachePut(value = CACHE_NAME, key = "'key_'+#productInfo.getId()")
    public ProductInfo saveLocalCache(ProductInfo productInfo) {
        return productInfo;
    }
​
    /**
     * 从本地缓存中获取商品信息
     */
    @Cacheable(value = CACHE_NAME, key = "'key_'+#id")
    public ProductInfo getLocalCache(Long id) {
        return null;
    }
​
    /**
     * 将商品信息保存到本地的ehcache缓存中
     */
    @CachePut(value = CACHE_NAME, key = "'product_info_'+#productInfo.getId()")
    public ProductInfo saveProductInfo2LocalCache(ProductInfo productInfo) {
        return productInfo;
    }
​
    /**
     * 从本地ehcache缓存中获取商品信息
     */
    @Cacheable(value = CACHE_NAME, key = "'product_info_'+#productId")
    public ProductInfo getProductInfoFromLocalCache(Long productId) {
        return null;
    }
​
    /**
     * 将店铺信息保存到本地的ehcache缓存中
     */
    @CachePut(value = CACHE_NAME, key = "'shop_info_'+#shopInfo.getId()")
    public ShopInfo saveShopInfo2LocalCache(ShopInfo shopInfo) {
        return shopInfo;
    }
​
    /**
     * 从本地ehcache缓存中获取店铺信息
     */
    @Cacheable(value = CACHE_NAME, key = "'shop_info_'+#shopId")
    public ShopInfo getShopInfoFromLocalCache(Long shopId) {
        return null;
    }
​
    /**
     * 将商品信息保存到redis中
     */
    public void saveProductInfo2ReidsCache(ProductInfo productInfo) {
        String key = "product_info_" + productInfo.getId();
        jedisCluster.set(key, JSONObject.toJSONString(productInfo));
    }
​
    /**
     * 将店铺信息保存到redis中
     */
    public void saveShopInfo2ReidsCache(ShopInfo shopInfo) {
        String key = "shop_info_" + shopInfo.getId();
        jedisCluster.set(key, JSONObject.toJSONString(shopInfo));
    }
}

测试业务逻辑

 

1. 创建一个 kafka topic

   # 创建 topic ，需要和程序中的一致 ：eshop-message
   bin/kafka-topics.sh --zookeeper 192.168.99.170:2181,192.168.99.171:2181,192.168.99.172:2181 --topic eshop-message --replication-factor 1 --partitions 1 --create

2. 在命令行启动一个 kafka producer

   # 创建一个生产者
   bin/kafka-console-producer.sh --broker-list 192.168.99.170:9092,192.168.99.171:9092,192.168.99.172:9092 --topic eshop-message

3. 启动系统，消费者开始监听 kafka topic

   注意：在 boot 2.1.x 中，连不上也没有日志打印。需要把几个虚拟机 hostname 映射到本地

   C:\Windows\System32\drivers\etc\hosts

4. 在 producer 中，分别发送两条消息，一个是商品信息服务的消息，一个是商品店铺信息服务的消息

   由于在本次模拟中，只使用 serviceId 作为了判定，其他数据是程序中写死的，所以这里推送两条携带 serviceId 的信息即可

   {"serviceId":"productInfoService","productId":"1"}
   ​
   {"serviceId":"shopInfoService","shopId":"1"}

5. 能否接收到两条消息，并模拟拉取到两条数据，同时将数据写入 ehcache 中，并写入 redis 缓存中

6. ehcache 通过打印日志方式来观察，redis 通过手工连接上去来查询