1. 准备
#安装Java运行环境JRE，下载Server Jre上传到tmp目录mkdir /usr/java;\ tar xf /tmp/server-jre-8u181-linux-x64.tar.gz -C /usr/java;\ ln -s /usr/java/jdk1.8.0_181 /usr/java/default;\ tee -a /etc/profile << 'EOF'export JAVA_HOME=/usr/java/defaultexport PATH=$JAVA_HOME/bin:$PATHEOF#生效环境变量source /etc/profile#下载部署包wget -c -P /tmp/ https://mirrors.tuna.tsinghua.edu.cn/apache/kafka/1.1.1/kafka_2.11-1.1.1.tgz;\ mkdir /opt/kafka;\ tar xf /tmp/kafka_2.11-1.1.1.tgz -C /opt/kafka#添加用户及目录useradd -s /sbin/nologin kafka;\ mkdir -p /home/kafka/data/kafka /home/kafka/data/zookeeper
2. Zookeeper
2.1 配置zookeeper.properties
mv /opt/kafka/kafka_2.11-1.1.1/config/zookeeper.properties /opt/kafka/kafka_2.11-1.1.1/config/zookeeper.properties_bak;\ tee /opt/kafka/kafka_2.11-1.1.1/config/zookeeper.properties << 'EOF'#数据目录，不建议放到程序内，因为更新到时候容易丢，应该放在统一到数据目录dataDir=/home/kafka/data/zookeeper#zookeeper中使用的基本时间单位, 毫秒值tickTime=2000#这个配置项是用来配置 Zookeeper 接受客户端（这里所说的客户端不是用户连接 Zookeeper 服务器的客户端，而是 Zookeeper 服务器集群中连接到 Leader 的 Follower 服务器）初始化连接时最长能忍受多少个 tickTime 时间间隔数。这里设置为5表名最长容忍时间为 5 * 2000 = 10 秒initLimit=10#这个配置标识 Leader 与 Follower 之间发送消息，请求和应答时间长度，最长不能超过多少个 tickTime 的时间长度，总的时间长度就是 5 * 2000 = 10秒syncLimit=5#客户端连接的端口clientPort=2181#每个IP最大客户端连接数，0表示关闭限制maxClientCnxns=1000#节点信息server.41=192.168.1.41:2888:3888 server.42=192.168.1.42:2888:3888 server.43=192.168.1.43:2888:3888 EOF
2.2 配置myid
#zookeeper node-1echo "41" > /home/kafka/data/zookeeper/myid#zookeper node-2echo "42" > /home/kafka/data/zookeeper/myid#zookeeper node-3echo "43" > /home/kafka/data/zookeeper/myid
2.3 添加Systemd服务
tee /etc/systemd/system/zookeeper.service << 'EOF'[Unit] Description=Apache Zookeeper server (Kafka) Documentation=http://zookeeper.apache.org Requires=network.target remote-fs.target After=network.target remote-fs.target [Service] Type=simple User=kafka Group=kafka Environment=JAVA_HOME=/usr/java/default ExecStart=/opt/kafka/kafka_2.11-1.1.1/bin/zookeeper-server-start.sh /opt/kafka/kafka_2.11-1.1.1/config/zookeeper.properties ExecStop=/opt/kafka/kafka_2.11-1.1.1/bin/zookeeper-server-stop.sh [Install] WantedBy=multi-user.target EOF
2.4 启动
chown -R kafka. /home/kafka/data /opt/kafka;\ systemctl daemon-reload;\ systemctl enable zookeeper;\ systemctl start zookeeper;\ systemctl status zookeeper
2.5 打开防火墙
firewall-cmd --add-port=2181/tcp --permanent;\ firewall-cmd --add-port=2888/tcp --permanent;\ firewall-cmd --add-port=3888/tcp --permanent;\ firewall-cmd --reload
2.6 查看
/opt/kafka/kafka_2.11-1.1.1/bin/zookeeper-shell.sh 192.168.1.41:2181
3. Kafka
3.1 配置server.properties
mv /opt/kafka/kafka_2.11-1.1.1/config/server.properties /opt/kafka/kafka_2.11-1.1.1/config/server.properties_bak;\ tee /opt/kafka/kafka_2.11-1.1.1/config/server.properties << 'EOF'#非负整数，用于唯一标识broker，不同节点该值必须不一样，比如 1,2,3broker.id=41# broker 发布给生产者消费者的hostname，会存储在zookeeper。配置好这个host可以实现内网外网同时访问。listeners=PLAINTEXT://192.168.1.41:9092#borker进行网络处理的线程数num.network.threads=3#borker进行I/O处理的线程数num.io.threads=8#发送缓冲区buffer大小，数据不是一下子就发送的，先回存储到缓冲区了到达一定的大小后在发送，能提高性能socket.send.buffer.bytes=102400#kafka接收缓冲区大小，当数据到达一定大小后在序列化到磁盘socket.receive.buffer.bytes=102400#这个参数是向kafka请求消息或者向kafka发送消息的请请求的最大数，这个值不能超过java的堆栈大小socket.request.max.bytes=104857600#kafka存放消息的目录log.dirs=/home/kafka/data/kafka/kafka-logs#每个topic默认partition数量，根据消费者实际情况配置，配置过小会影响消费性能num.partitions=30# kafka启动恢复日志,关闭前日志刷盘的线程数，如果是raid的盘的话适当增加，比如3num.recovery.threads.per.data.dir=2#是否允许删除topic，开发环境应该是true，测试及生产环境应该是falsedelete.topic.enable=true#每个topic的副本数量，默认为1，生产最好3个以上offsets.topic.replication.factor=3 transaction.state.log.replication.factor=1 transaction.state.log.min.isr=1#日志刷盘机制默认，10000条刷一次，1秒刷一次#log.flush.interval.messages=10000#log.flush.interval.ms=1000#日志保留时间，默认为7天，如果量大，可以酌情改小，比如三天log.retention.hours=168#日志保留大小，默认是保留1G，生产上可以配置为下面50Glog.retention.bytes=53687091200#The interval at which log segments are checked to see if they can be deleted according# to the retention policieslog.retention.check.interval.ms=300000#Zookeeper host和port ，这个看zookeeper环境而设置zookeeper.connect=192.168.1.41:2181,192.168.1.42:2181,192.168.1.43:2181#Zokeeper 连接超时时间zookeeper.connection.timeout.ms=6000#是让coordinator推迟空消费组接收到成员加入请求后本应立即开启的rebalance。在实际使用时，假设你预估你的所有consumer组成员加入需要在10s内完成，那么你就可以设置该参数10000group.initial.rebalance.delay.ms=10000 EOF
3.2 server.properties分节点特殊配置
#kafka node-1sed -i '/broker.id=/c\broker.id=41' /opt/kafka/kafka_2.11-1.1.1/config/server.properties;\ sed -i '/listeners=/c\listeners=PLAINTEXT://192.168.1.41:9092' /opt/kafka/kafka_2.11-1.1.1/config/server.properties#kafka node-2sed -i '/broker.id=/c\broker.id=42' /opt/kafka/kafka_2.11-1.1.1/config/server.properties;\ sed -i '/listeners=/c\listeners=PLAINTEXT://192.168.1.42:9092' /opt/kafka/kafka_2.11-1.1.1/config/server.properties#kafka node-3sed -i '/broker.id=/c\broker.id=43' /opt/kafka/kafka_2.11-1.1.1/config/server.properties;\ sed -i '/listeners=/c\listeners=PLAINTEXT://192.168.1.43:9092' /opt/kafka/kafka_2.11-1.1.1/config/server.properties
3.3 添加Systemd服务
tee /etc/systemd/system/kafka.service << 'EOF'[Unit] Description=Apache Kafka server (broker) Documentation=http://kafka.apache.org/documentation.html Requires=network.target remote-fs.target After=network.target remote-fs.target zookeeper.service [Service] Type=simple User=kafka Group=kafka Environment=JAVA_HOME=/usr/java/default ExecStart=/opt/kafka/kafka_2.11-1.1.1/bin/kafka-server-start.sh /opt/kafka/kafka_2.11-1.1.1/config/server.properties ExecStop=/opt/kafka/kafka_2.11-1.1.1/bin/kafka-server-stop.sh [Install] WantedBy=multi-user.target EOF
3.4 启动
chown -R kafka. /home/kafka/data /opt/kafka;\ systemctl daemon-reload;\ systemctl enable kafka;\ systemctl start kafka;\ systemctl status kafka
3.5 防火墙
firewall-cmd --add-port=9092/tcp --permanent;\ firewall-cmd --reload
3.6 查看
#查看topic/opt/kafka/kafka_2.11-1.1.1/bin/kafka-topics.sh --zookeeper 127.0.0.1:2181 --list#创建topic/opt/kafka/kafka_2.11-1.1.1/bin/kafka-topics.sh --zookeeper 127.0.0.1:2181 --create --replication-factor 1 --partitions 1 --topic test#删除topic/opt/kafka/kafka_2.11-1.1.1/bin/kafka-topics.sh --zookeeper 127.0.0.1:2181 --delete --to
作者：Hanson1330
链接：https://www.jianshu.com/p/1a788460c67e