ClickHouse集群部署

https://packagecloud.io/altinity/clickhouse

clickhouse-client-19.1.7-1.el7.x86_64.rpm
clickhouse-common-static-19.1.7-1.el7.x86_64.rpm
clickhouse-server-19.1.7-1.el7.x86_64.rpm
clickhouse-server-common-19.1.7-1.el7.x86_64.rpm
clickhouse-test-19.1.7-1.el7.x86_64.rpm

mkdir -p /data/clickhouse/tmp /data/clickhouse/logs /data/clickhouse/lock
chown clickhouse:clickhouse -R /data/clickhouse

CLICKHOUSE_USER=clickhouse
CLICKHOUSE_GROUP=${CLICKHOUSE_USER}
SHELL=/bin/bash
PROGRAM=clickhouse-server
CLICKHOUSE_GENERIC_PROGRAM=clickhouse
CLICKHOUSE_PROGRAM_ENV=""
EXTRACT_FROM_CONFIG=${CLICKHOUSE_GENERIC_PROGRAM}-extract-from-config
CLICKHOUSE_CONFDIR=/etc/$PROGRAM
CLICKHOUSE_LOGDIR=/data/clickhouse/logs
CLICKHOUSE_LOGDIR_USER=root
CLICKHOUSE_DATADIR_OLD=/data/clickhouse-old
CLICKHOUSE_DATADIR=/data/clickhouse
LOCALSTATEDIR=/var/lock
CLICKHOUSE_BINDIR=/usr/bin
CLICKHOUSE_CRONFILE=/etc/cron.d/clickhouse-server
CLICKHOUSE_CONFIG=$CLICKHOUSE_CONFDIR/config.xml
LOCKFILE=$LOCALSTATEDIR/$PROGRAM
RETVAL=0
CLICKHOUSE_PIDDIR=/var/run/$PROGRAM
CLICKHOUSE_PIDFILE="$CLICKHOUSE_PIDDIR/$PROGRAM.pid"

<?xml version="1.0"?>
<yandex>
    <logger>
        <!-- Possible levels: https://github.com/pocoproject/poco/blob/develop/Foundation/include/Poco/Logger.h#L105 -->
        <level>trace</level>
        <!-- <level>error</level> -->
        <log>/data/clickhouse/logs/clickhouse-server.log</log>
        <errorlog>/data/clickhouse/logs/clickhouse-server.err.log</errorlog>
        <size>1000M</size>
        <count>10</count>
        <!-- <console>1</console> --> <!-- Default behavior is autodetection (log to console if not daemon mode and is tty) -->
    </logger>


    <!--display_name>production</display_name--> <!-- It is the name that will be shown in the client -->
    <http_port>8123</http_port>
    <tcp_port>9000</tcp_port>
    <!-- Port for communication between replicas. Used for data exchange. -->
    <interserver_http_port>9009</interserver_http_port>
    <!-- Hostname that is used by other replicas to request this server.


         If not specified, than it is determined analoguous to 'hostname -f' command.


         This setting could be used to switch replication to another network interface.
      -->
    <!--每个节点配置本机主机名-->
    <interserver_http_host>ckprd1</interserver_http_host>
    
    <!-- Listen specified host. use :: (wildcard IPv6 address), if you want to accept connections both with IPv4 and IPv6 from everywhere. -->
    <!-- <listen_host>::</listen_host> -->
    <!-- Same for hosts with disabled ipv6: -->
    <listen_host>0.0.0.0</listen_host>
    
    <!-- Path to data directory, with trailing slash. -->
    <path>/data/clickhouse/</path>


    <!-- Path to temporary data for processing hard queries. -->
    <tmp_path>/data/clickhouse/tmp/</tmp_path>
    
    <!-- <timezone>Europe/Moscow</timezone> -->
    <timezone>Asia/Shanghai</timezone>


    <!--集群相关配置-->
    <include_from>/etc/clickhouse-server/metrika.xml</include_from>


    <remote_servers incl="clickhouse_remote_servers" />
    <zookeeper incl="zookeeper-servers" optional="true" />


    <macros incl="macros" optional="true" />
</yandex>

<?xml version="1.0"?>
<yandex>
    <!-- cluster setting -->
    <clickhouse_remote_servers>
        <mcd_prod_cluster_1st>
            <!-- shard01 -->
            <shard>
                <!-- Optional. Shard weight when writing data. Default: 1. -->
                <weight>1</weight>
                <!-- Optional. Whether to write data to just one of the replicas. Default: false (write data to all replicas). 
                1. internal_replication为false时候，（前提是往分布表写入数据）会自动往同一shard下所有备份表写入相同数据，不需要任何其他外力，单独设置这个参数即可；但是会出现各备份之间数据不同步的情况，因为此种情况下往分布式表里面写数据，后台算法会先按照weight将数据分成shard数量堆，然后将对应堆的数据分别写入该shard下面的所有备份表中，有可能存在同样的数据写入A备份成功但是写入B备份失败的情况，这里是没有校验的；
                2. internal_replication为true时，一定要配合zookeeper和ReplicatedMergeTree引擎表使用，如果不配合这些，经本人测试查询数据时会出现严重错误，请切记
                -->
                <internal_replication>false</internal_replication>
                <replica>
                    <host>ckprd1</host>
                    <port>9000</port>
                </replica>


                <replica>
                    <host>ckprd2</host>
                    <port>9000</port>
                </replica>
            </shard>
            
            <!-- shard02 -->
            <shard>
                <weight>1</weight>
                <internal_replication>false</internal_replication>
                <replica>
                    <host>ckprd3</host>
                    <port>9000</port>
                </replica>
                <replica>
                    <host>ckprd4</host>
                    <port>9000</port>
                </replica>
            </shard>
        </mcd_prod_cluster_1st>
    </clickhouse_remote_servers>


    <!-- ReplicatedMergeTree('/clickhouse/tables/{layer}-{shard}/hits', '{replica}') 
    1. /clickhouse/tables/{layer}-{shard}/hits作为整体可以理解为表在zookeeper中的定位和识别符
    2. 同一个{layer}-{shard}下面的表互为备份，会自动同步
    3. layer我将之理解为集群识别符，虽然其可以与上面clickhouse_remote_servers配置中的cluster名称不一样；同一个配置文件是可以配置多个集群的，因此有layer这一说法
    4. hits可以与表名相同，但是表名改变时这里不会变，因为这里的含义是，表在zookeeper中的定位和识别符
    5. replica是备份序号的识别符，只要不同即可，可以设置为与该节点hostname相同（注意replica一定要不同）
    6. macros中的值设置会对应到ReplicatedMergeTree中的变量（就是大括号的部分），这样在每个节点建立local表时，可以使用完全相同的SQL语句，因为变量由这里的配置控制
    -->
    
    <macros>
        <layer>mcd_prod_cluster_1st</layer>
        <!--根据前面shard的配置，例子中总共2个shard-->
        <shard>shard01</shard>
        <!--每个节点配置本地主机名即可，或者唯一的数字id-->
        <replica>ckprd1</replica>
    </macros>


    <!-- 监听网络（貌似重复） -->
    <networks>
       <ip>::/0</ip>
    </networks>
    
    <!-- zookeeper  -->
    <zookeeper-servers>
        <node index="1">
            <host>ckprd1</host>
            <port>2181</port>
        </node>
        <node index="2">
            <host>ckprd2</host>
            <port>2181</port>
        </node>
        <node index="3">
            <host>ckprd3</host>
            <port>2181</port>
        </node>
    </zookeeper-servers>


    <!-- 数据压缩算法  -->
    <clickhouse_compression>
        <case>
            <min_part_size>10000000000</min_part_size>
            <min_part_size_ratio>0.01</min_part_size_ratio>
            <method>lz4</method>
        </case>
    </clickhouse_compression>
</yandex>