RocketMQ FilterServer 使用详解

1. 概述

FilterServer 是 RocketMQ 的服务端消息过滤机制，允许在 Broker 端进行消息过滤，减少网络传输和客户端处理压力。

2. Filter Server架构

3. FilterServer 使用方式

3.1 基于 Tag 的简单过滤（常用）

// Producer - 发送带 Tag 的消息
Message msg = new Message("TopicTest", "TagA", "Hello Filter".getBytes());
SendResult sendResult = producer.send(msg);

// Consumer - 订阅时指定 Tag
DefaultMQPushConsumer consumer = new DefaultMQPushConsumer("consumer_group");
// 只消费 TagA 的消息
consumer.subscribe("TopicTest", "TagA");
consumer.registerMessageListener(new MessageListenerConcurrently() {
    @Override
    public ConsumeConcurrentlyStatus consumeMessage(List<MessageExt> msgs,
            ConsumeConcurrentlyContext context) {
        // 这里只会收到 TagA 的消息
        return ConsumeConcurrentlyStatus.CONSUME_SUCCESS;
    }
});

3.2 基于 SQL92 的复杂过滤

发送时指定properties，消费时根据properties的属性进行sql过滤

注意⚠️: 设置如下broker配置，默认为false

enablePropertyFilter=true

3.2.1 Producer 发送带属性的消息

public class SQLFilterProducer {
    public static void main(String[] args) throws Exception {
        DefaultMQProducer producer = new DefaultMQProducer("please_rename_unique_group_name");
        producer.start();
        
        // 发送带属性的消息
        for (int i = 0; i < 10; i++) {
            Message msg = new Message("TopicTest", 
                "TagA", 
                ("Hello RocketMQ " + i).getBytes(RemotingHelper.DEFAULT_CHARSET));
            
            // 设置消息属性
            msg.putUserProperty("a", String.valueOf(i));
            msg.putUserProperty("b", "hello");
            msg.putUserProperty("c", "world");
            msg.putUserProperty("price", String.valueOf(100 + i));
            msg.putUserProperty("region", i % 2 == 0 ? "Beijing" : "Shanghai");
            
            SendResult sendResult = producer.send(msg);
            System.out.printf("%s%n", sendResult);
        }
        
        producer.shutdown();
    }
}

3.2.2 Consumer 使用 SQL 过滤

public class SQLFilterConsumer {
    public static void main(String[] args) throws Exception {
        DefaultMQPushConsumer consumer = new DefaultMQPushConsumer("please_rename_unique_group_name");
        
        // 使用 SQL92 语法进行消息过滤
        // 只消费 a between 0 and 3 的消息
        consumer.subscribe("TopicTest", 
            MessageSelector.bySql("a between 0 and 3"));
        
        // 或者使用多个条件
        // consumer.subscribe("TopicTest", 
        //     MessageSelector.bySql("a > 2 AND b = 'hello' AND region = 'Beijing'"));
        
        consumer.registerMessageListener(new MessageListenerConcurrently() {
            @Override
            public ConsumeConcurrentlyStatus consumeMessage(List<MessageExt> msgs,
                    ConsumeConcurrentlyContext context) {
                for (MessageExt msg : msgs) {
                    System.out.printf("Receive message: %s, Properties: %s%n", 
                        new String(msg.getBody()),
                        msg.getProperties());
                }
                return ConsumeConcurrentlyStatus.CONSUME_SUCCESS;
            }
        });
        
        consumer.start();
        System.out.printf("SQL Filter Consumer Started.%n");
    }
}

3.3 自定义类过滤（高级用法）

3.3.1 创建自定义过滤类

// 自定义过滤类
public class MyMessageFilter implements MessageFilter {
    @Override
    public boolean match(MessageExt msg, FilterContext context) {
        // 自定义过滤逻辑
        String body = new String(msg.getBody());
        String region = msg.getProperty("region");
        
        // 示例：只接受北京地区且包含"important"的消息
        return "Beijing".equals(region) && body.contains("important");
    }
}

3.3.2 Consumer 使用 Class Filter

// Consumer - 订阅时指定 fullClassName
DefaultMQPushConsumer consumer = new DefaultMQPushConsumer("consumer_group");

consumer.subscribe("TopicTest", "full class name", "class source code");
consumer.registerMessageListener(new MessageListenerConcurrently() {
    @Override
    public ConsumeConcurrentlyStatus consumeMessage(List<MessageExt> msgs,
            ConsumeConcurrentlyContext context) {
        // 这里只会收到 TagA 的消息
        return ConsumeConcurrentlyStatus.CONSUME_SUCCESS;
    }
});

 /**
     * Subscribe a topic to consuming subscription.
     *
     * @param topic topic to consume.
     * @param fullClassName full class name,must extend org.apache.rocketmq.common.filter.MessageFilter
     * @param filterClassSource class source code,used UTF-8 file encoding,must be responsible for your code safety
     */
    @Override
    public void subscribe(String topic, String fullClassName, String filterClassSource) throws MQClientException {
        this.defaultMQPushConsumerImpl.subscribe(withNamespace(topic), fullClassName, filterClassSource);
    }

4. SQL92 语法支持

4.1 支持的运算符

-- 数值比较
a > 10
a <= 100
a BETWEEN 0 AND 100

-- 字符串比较
b = 'hello'
b IS NOT NULL
b IN ('hello', 'world')

-- 逻辑运算符
a > 10 AND b = 'hello'
a < 5 OR b = 'world'
NOT (a > 10)

-- 空值判断
c IS NULL
c IS NOT NULL

4.2 实际使用示例

// 价格在100-200之间的北京订单
consumer.subscribe("OrderTopic", 
    MessageSelector.bySql("price BETWEEN 100 AND 200 AND region = 'Beijing'"));

// 特定用户ID的消息
consumer.subscribe("UserTopic", 
    MessageSelector.bySql("userId IN (1001, 1002, 1003)"));

// 排除测试消息
consumer.subscribe("LogTopic", 
    MessageSelector.bySql("env != 'test' AND level = 'ERROR'"));

5. 过滤方式对比

总结

FilterServer 最适合的场景：

• 需要服务端过滤减少网络传输
• 过滤条件相对固定且简单
• 不同消费者需要不同数据视图
• 消息量较大，客户端处理压力大

简单过滤用 Tag，复杂过滤用 SQL92，特殊需求用自定义过滤类。