RocketMQ 的主备模式

按之前所说，只有 RocketMQ 的多主多从异步复制是可以生产使用的，因此只在这个场景下测试。另外，消息采用 Push 顺序模式消费。

假设集群采用2主2备的模式，需要启动4个 Broker，配置文件如下，

brokerName=broker-a
brokerId=0
listenPort=10911
storePathRootDir=/home/arnes/alibaba-rocketmq/data/store-a-async
storePathCommitLog=/home/arnes/alibaba-rocketmq/data/store-a-async/commitlog
brokerRole=ASYNC_MASTER
brokerName=broker-a
brokerId=1
listenPort=10921
storePathRootDir=/home/arnes/alibaba-rocketmq/data/store-a-async-slave
storePathCommitLog=/home/arnes/alibaba-rocketmq/data/store-a-async-slave/commitlog
brokerRole=SLAVE
brokerName=broker-b
brokerId=0
listenPort=20911
storePathRootDir=/home/arnes/alibaba-rocketmq/data/store-b-async
storePathCommitLog=/home/arnes/alibaba-rocketmq/data/store-b-async/commitlog
brokerRole=ASYNC_MASTER
brokerRole=ASYNC_MASTER
brokerName=broker-b
brokerId=1
listenPort=20921
storePathRootDir=/home/arnes/alibaba-rocketmq/data/store-b-async-slave
storePathCommitLog=/home/arnes/alibaba-rocketmq/data/store-b-async-slave/commitlog
brokerRole=SLAVE

另外，每个机构共通的配置项如下，

brokerClusterName=DefaultCluster
brokerIP1=192.168.232.23
namesrvAddr=192.168.232.23:9876
deleteWhen=04
fileReservedTime=120
flushDiskType=ASYNC_FLUSH

其他设置均采用默认。启动 NameServer 和所有 Broker，并试运行一下 Producer，然后看一下 TestTopic1 当前的情况，

$ sh mqadmin topicRoute -n 192.168.232.23:9876 -t TopicTest1
{
        "brokerDatas":[
                {
                        "brokerAddrs":{0:"192.168.232.23:20911",1:"192.168.232.23:20921"
                        },
                        "brokerName":"broker-b"
                },
                {
                        "brokerAddrs":{0:"192.168.232.23:10911",1:"192.168.232.23:10921"
                        },
                        "brokerName":"broker-a"
                }
        ],
        "filterServerTable":{},
        "queueDatas":[
                {
                        "brokerName":"broker-a",
                        "perm":6,
                        "readQueueNums":4,
                        "topicSynFlag":0,
                        "writeQueueNums":4
                },
                {
                        "brokerName":"broker-b",
                        "perm":6,
                        "readQueueNums":4,
                        "topicSynFlag":0,
                        "writeQueueNums":4
                }
        ]
}

可见，TestTopic1 在2个 Broker 上，且每个 Broker 备机也在运行。下面开始主备切换的实验，分别启动 Consumer 和 Producer 进程，消息采用 Pull 顺序模式消费。在消息发送接收过程中，使用 kill -9 停掉 broker-a 的主进程，模拟突然宕机。此时，TestTopic1 的状态如下，

$ sh mqadmin topicRoute -n 192.168.232.23:9876 -t TopicTest1
{
        "brokerDatas":[
                {
                        "brokerAddrs":{0:"192.168.232.23:20911",1:"192.168.232.23:20921"
                        },
                        "brokerName":"broker-b"
                },
                {
                        "brokerAddrs":{1:"192.168.232.23:10921"
                        },
                        "brokerName":"broker-a"
                }
        ],
        "filterServerTable":{},
        "queueDatas":[
                {
                        "brokerName":"broker-a",
                        "perm":6,
                        "readQueueNums":4,
                        "topicSynFlag":0,
                        "writeQueueNums":4
                },
                {
                        "brokerName":"broker-b",
                        "perm":6,
                        "readQueueNums":4,
                        "topicSynFlag":0,
                        "writeQueueNums":4
                }
        ]
}