Canal一般用于实时同步数据场景,那么对于实时场景HA显得尤为重要,Canal支持HA搭建,canal的HA分为两部分,canal server和canal client分别有对应的HA实现。大数据中使用Canal同步数据一般同步到Kafka中,这里Kafka相当于是Canal Client,Kafka集群自带HA属性,所以这里我们只关注Canal Server HA。Canal Server HA主要是为了减少对mysql dump的请求,不同server上的instance(不同server上的相同instance)要求同一时间只能有一个处于running,其他的处于standby状态(standby是instance的状态),Canal Server HA原理如下:
Canal HA 保证步骤如下:
运行Canal的机器:node3,node4
zookeeper地址:node3:2181,node4:2181,node5:2181
mysql地址:node2:3306
将Canal安装包上传到node3,node4,并解压到“/software/canal”目录下,修改“/software/canal/conf”下的canal.properties文件,加上zookeeper配置
#指定zookeeper集群地址
canal.zkServers = node3:2181,node4:2181,node5:2181
#配置spring的xml配置文件
canal.instance.global.spring.xml = classpath:spring/default-instance.xml
#canal将数据写入Kafka,可配:tcp, kafka, RocketMQ,tcp就是使用canal代码接收
canal.serverMode = kafka
#配置canal写入Kafka地址
canal.mq.servers = node1:9092,node2:9092,node3:9092
进入“/software/canal/conf/example”目录,修改“instance.properties”文件:
#另外一台机器改成123457,保证slaveId不重复即可
canal.instance.mysql.slaveId=123456
#配置mysql master 节点及端口
canal.instance.master.address=node2:3306
#配置连接mysql的用户名和密码,就是前面复制权限的用户名和密码
canal.instance.dbUsername=canal
canal.instance.dbPassword=canal
#配置Canal将数据导入到Kafka topic
canal.mq.topic=canal_topic
注意:两台机器上的instance目录的名字需要保证完全一致,HA模式是依赖于instance name进行管理,同时必须都选择default-instance.xml配置,此配置中才有关于zookeeper的设置信息。
#在node3上启动Canal
[root@node3 ~]# cd /software/canal/bin
[root@node3 bin]# ./startup.sh
#在node4上启动Canal
[root@node4 ~]# cd /software/canal/bin
[root@node4 bin]# ./startup.sh
启动完成后,可以查看zookeeper中对应的路径信息:
默认搭建好的Canal HA 后可以通过查看Zookeeper中的“/otter/canal/destinations/examples/running”来查看Active的Canal节点:
测试Canal HA 如下:
mysql> insert into person values (4,"s1",21),(5,"s2",22),(6,"s3",23);
可以观察到Kafka canal_topic中有监控到的数据如下:
{"data":[{"id":"4","name":"s1","age":"21"},{"id":"5","name":"s2","age":"22"},{"id":"6","name":"s3","age":"23"}],"database":"testdb","es":1618849974000,"id":2,"isDdl":false,"mysqlType":{"id":"int","name":"varchar(255)","age":"int"},"old":null,"pkNames":null,"sql":"","sqlType":{"id":4,"name":12,"age":4},"table":"person","ts":1618849975203,"type":"INSERT"}
关闭node3 Canal Server:
[root@node3 ~]# cd /software/canal/bin
[root@node3 bin]# ./stop.sh
查看zookeeper “/otter/canal/destinations/examples/running”路径Active的Canal节点:
继续向MySQL中“testdb.person”表中写入数据:
mysql> insert into person values (7,"x1",24),(8,"x2",25),(9,"x3",26);
可以观察写入到Kafka “canal_topic”中数据如下:
{"data":[{"id":"7","name":"x1","age":"24"},{"id":"8","name":"x2","age":"25"},{"id":"9","name":"x3","age":"26"}],"database":"testdb","es":1618850233000,"id":2,"isDdl":false,"mysqlType":{"id":"int","name":"varchar(255)","age":"int"},"old":null,"pkNames":null,"sql":"","sqlType":{"id":4,"name":12,"age":4},"table":"person","ts":1618850234136,"type":"INSERT"}
经过以上测试,Canal HA 生效。
注意:经过测试Canal HA 在使用zookeeper存储binlog position时,当有一个Canal Server重新启动并切换成Active节点时,每次都会重复读取最后一条数据。使用非HA 本地存储binlog position时,没有此问题。