本文收录在javaskill.cn中，内有完整的JAVA知识地图，欢迎访问
canal是什么
摘自项目github
基于数据库增量日志解析，提供增量数据订阅&消费，目前主要支持了mysql
原理相对比较简单：
canal模拟mysql slave的交互协议，伪装自己为mysql slave，向mysql master发送dump协议
mysql master收到dump请求，开始推送binary log给slave(也就是canal)
canal解析binary log对象(原始为byte流)
canal解决了什么问题
起源：
早期，阿里巴巴B2B公司因为存在杭州和美国双机房部署，存在跨机房同步的业务需求。不过早期的数据库同步业务，主要是基于trigger的方式获取增量变更，不过从2010年开始，阿里系公司开始逐步的尝试基于数据库的日志解析，获取增量变更进行同步，由此衍生出了增量订阅&消费的业务，从此开启了一段新纪元。
本文在探讨什么
canal的基础操作，本文不再赘述，可参考github上的quick start和client api，包含了demo
本文探讨的是，基于canal的流式api的消息分发，以及如何防止消息丢失和重复处理
什么是流式API
摘自项目github
流式api设计：
image
- 每次get操作都会在meta中产生一个mark，mark标记会递增，保证运行过程中mark的唯一性
- 每次的get操作，都会在上一次的mark操作记录的cursor继续往后取，如果mark不存在，则在last ack cursor继续往后取
- 进行ack时，需要按照mark的顺序进行数序ack，不能跳跃ack. ack会删除当前的mark标记，并将对应的mark位置更新为last ack cursor
- 一旦出现异常情况，客户端可发起rollback情况，重新置位：删除所有的mark, 清理get请求位置，下次请求会从last ack cursor继续往后取
image.png
关注点
异步的ack带来了更好的性能，也带来了一些问题
rollback后，mark会清空，回到上次ack的位置。如果get的速度比ack快，当rollback()后，就会出现重复消息
本文针对这个问题，给出一个较为简单的解决方案
思路
mark清空后，再次get，获取到的batchId会继续递增（保存在服务端），但是消息是已经处理过的，此时我们不希望消息继续被分发或者处理
如何判断消息是否消费过，或者说，该次数据库变更，是否已经解析过
1.在业务上进行判断
2.更好的方式，通过当前处理的消息在binlog中的位置进行判断
String logFileName = entry.getHeader().getLogfileName();
long offset = entry.getHeader().getLogfileOffset();
使用这两行代码，就可以方便的获取当前消息在具体哪个binlog文件的哪个位置，输出类似如下
logfileName = mysql-bin.000001,offset = 41919
代码
测试代码，请勿用于生产
public static void main(String args[]) {
new Thread(SimpleCanalClient::receiver).start();
new Thread(SimpleCanalClient::ack).start();
}
开局启动两个线程，一个接收，一个确认
private static void receiver() {
// 创建链接
CanalConnector connector = CanalConnectors.newSingleConnector(new InetSocketAddress(AddressUtils.getHostIp(),
11111), "example", "", "");
int batchSize = 1;
int count = 0;
try {
connector.connect();
connector.subscribe(".*\\..*");
connector.rollback();
int total = 20;
while (count < total) {
count++;
Message message = connector.getWithoutAck(batchSize); // 获取指定数量的数据
long batchId = message.getId();
int size = message.getEntries().size();
if (batchId != -1 && size != 0) {
printEntry(message.getEntries(),batchId);
}
try {
Thread.sleep(1000);
} catch (InterruptedException e) {
}
}
STOP.set(true);
System.out.println("receiver exit");
} finally {
connector.disconnect();
}
}
接收者，客户端连上服务端后，每秒钟获取一次数据，一次获取1条，20秒后关闭
一秒钟是模拟业务处理时间
调用了printEntry的业务处理方法
注意刚启动时的rollback()，把上次停止后未提交的数据回滚，因为不确定是否已处理
private static void printEntry(List entries,long batchId) {
for (Entry entry : entries) {
if (entry.getEntryType() == EntryType.TRANSACTIONBEGIN || entry.getEntryType() == EntryType.TRANSACTIONEND) {
continue;
}
String logFileName = entry.getHeader().getLogfileName();
long offset = entry.getHeader().getLogfileOffset();
long lastOffset = getTagFromRedis(logFileName);
if(offset<=lastOffset){
System.out.println("Already processed ! logfileName = "+logFileName+",offset = "+offset+",batchId "+batchId);
}else{
System.out.println("processing logfileName = "+logFileName+",offset = "+offset+",batchId "+batchId);
saveTagToRedis(logFileName,offset);
}
}
BATCH_ID_QUEUE.offer(batchId);
}
printEntry方法进行简单的消费，生产上应该在这里分发给具体的业务处理类
获取了最后处理的offset，如果当前offset小于最后处理的offset，则不处理
处理完后，把offset存入redis中，（saveTagToRedis)
最后把batchId放到了一个Queue中，方便顺序ack
private static void saveTagToRedis(String logFileName, long offset) {
try(Jedis jedis = jedisPool.getResource()){
jedis.set(logFileName, String.valueOf(offset));
jedis.append("processed-"+logFileName,String.valueOf(offset)+"\n");
}
}
private static long getTagFromRedis(String logFileName){
try (Jedis jedis = jedisPool.getResource()){
String position = jedis.get(logFileName);
if(StringUtils.isNotBlank(position)){
return Long.valueOf(position);
}else{
return -1L;
}
}
}
保存和获取最后处理位置的代码，简单的放到了redis中
private static void ack() {
CanalConnector connector = CanalConnectors.newSingleConnector(new InetSocketAddress(AddressUtils.getHostIp(),
11111), "example", "", "");
try {
connector.connect();
connector.subscribe(".*\\..*");
while (!STOP.get()) {
Long batchId = BATCH_ID_QUEUE.poll();
if(batchId!=null){
connector.ack(batchId);
System.out.println("batchId "+batchId+" ack");
}
try {
Thread.sleep(5000);
} catch (InterruptedException ignored) {
}
}
System.out.println("ack exit");
} finally {
connector.disconnect();
}
}
ack方法，从queue中获取最后的batchId（每次停止，未处理的都自动释放了）
模拟网络延迟，5秒钟才能确认一个batchId
测试
测试方法：
开启数据库，不断修改数据，看日志
第一次启动（前面消费都确认完），日志如下
"C:\Program Files\Java\jdk-10.0.1\bin\java.exe" "-javaagent:C:\Program Files\JetBrains\IntelliJ IDEA 2018.1.5\lib\idea_rt.jar=57322:C:\Program Files\JetBrains\IntelliJ IDEA 2018.1.5\bin" -Dfile.encoding=UTF-8 -classpath C:\Users\LiYang\IdeaProjects\canal-test\target\classes;C:\Users\LiYang\.m2\repository\com\alibaba\otter\canal.client\1.0.25\canal.client-1.0.25.jar;C:\Users\LiYang\.m2\repository\com\alibaba\otter\canal.protocol\1.0.25\canal.protocol-1.0.25.jar;C:\Users\LiYang\.m2\repository\com\alibaba\otter\canal.common\1.0.25\canal.common-1.0.25.jar;C:\Users\LiYang\.m2\repository\io\netty\netty-all\4.1.6.Final\netty-all-4.1.6.Final.jar;C:\Users\LiYang\.m2\repository\org\apache\zookeeper\zookeeper\3.4.5\zookeeper-3.4.5.jar;C:\Users\LiYang\.m2\repository\org\jboss\netty\netty\3.2.2.Final\netty-3.2.2.Final.jar;C:\Users\LiYang\.m2\repository\com\101tec\zkclient\0.10\zkclient-0.10.jar;C:\Users\LiYang\.m2\repository\commons-io\commons-io\2.4\commons-io-2.4.jar;C:\Users\LiYang\.m2\repository\commons-codec\commons-codec\1.9\commons-codec-1.9.jar;C:\Users\LiYang\.m2\repository\com\alibaba\fastjson\1.2.28\fastjson-1.2.28.jar;C:\Users\LiYang\.m2\repository\com\google\guava\guava\18.0\guava-18.0.jar;C:\Users\LiYang\.m2\repository\ch\qos\logback\logback-core\1.1.3\logback-core-1.1.3.jar;C:\Users\LiYang\.m2\repository\ch\qos\logback\logback-classic\1.1.3\logback-classic-1.1.3.jar;C:\Users\LiYang\.m2\repository\org\slf4j\jcl-over-slf4j\1.7.12\jcl-over-slf4j-1.7.12.jar;C:\Users\LiYang\.m2\repository\org\slf4j\slf4j-api\1.7.12\slf4j-api-1.7.12.jar;C:\Users\LiYang\.m2\repository\org\springframework\spring-core\3.2.9.RELEASE\spring-core-3.2.9.RELEASE.jar;C:\Users\LiYang\.m2\repository\commons-logging\commons-logging\1.1.3\commons-logging-1.1.3.jar;C:\Users\LiYang\.m2\repository\org\springframework\spring-aop\3.2.9.RELEASE\spring-aop-3.2.9.RELEASE.jar;C:\Users\LiYang\.m2\repository\aopalliance\aopalliance\1.0\aopalliance-1.0.jar;C:\Users\LiYang\.m2\repository\org\springframework\spring-beans\3.2.9.RELEASE\spring-beans-3.2.9.RELEASE.jar;C:\Users\LiYang\.m2\repository\org\springframework\spring-context\3.2.9.RELEASE\spring-context-3.2.9.RELEASE.jar;C:\Users\LiYang\.m2\repository\org\springframework\spring-expression\3.2.9.RELEASE\spring-expression-3.2.9.RELEASE.jar;C:\Users\LiYang\.m2\repository\org\springframework\spring-jdbc\3.2.9.RELEASE\spring-jdbc-3.2.9.RELEASE.jar;C:\Users\LiYang\.m2\repository\org\springframework\spring-tx\3.2.9.RELEASE\spring-tx-3.2.9.RELEASE.jar;C:\Users\LiYang\.m2\repository\org\springframework\spring-orm\3.2.9.RELEASE\spring-orm-3.2.9.RELEASE.jar;C:\Users\LiYang\.m2\repository\com\google\protobuf\protobuf-java\2.6.1\protobuf-java-2.6.1.jar;C:\Users\LiYang\.m2\repository\commons-lang\commons-lang\2.6\commons-lang-2.6.jar;C:\Users\LiYang\.m2\repository\redis\clients\jedis\2.9.0\jedis-2.9.0.jar;C:\Users\LiYang\.m2\repository\org\apache\commons\commons-pool2\2.4.2\commons-pool2-2.4.2.jar com.github.liyang0211.canal.test.SimpleCanalClient
processing logfileName = mysql-bin.000001,offset = 43107,batchId 151
processing logfileName = mysql-bin.000001,offset = 43404,batchId 152
processing logfileName = mysql-bin.000001,offset = 43701,batchId 153
processing logfileName = mysql-bin.000001,offset = 43998,batchId 154
batchId 151 ack
processing logfileName = mysql-bin.000001,offset = 44295,batchId 155
processing logfileName = mysql-bin.000001,offset = 44592,batchId 156
processing logfileName = mysql-bin.000001,offset = 44889,batchId 157
processing logfileName = mysql-bin.000001,offset = 45186,batchId 158
processing logfileName = mysql-bin.000001,offset = 45483,batchId 159
batchId 152 ack
processing logfileName = mysql-bin.000001,offset = 45780,batchId 160
processing logfileName = mysql-bin.000001,offset = 46077,batchId 161
processing logfileName = mysql-bin.000001,offset = 46374,batchId 162
processing logfileName = mysql-bin.000001,offset = 46671,batchId 162
processing logfileName = mysql-bin.000001,offset = 46968,batchId 163
processing logfileName = mysql-bin.000001,offset = 47265,batchId 164
batchId 153 ack
processing logfileName = mysql-bin.000001,offset = 47562,batchId 165
processing logfileName = mysql-bin.000001,offset = 47859,batchId 166
processing logfileName = mysql-bin.000001,offset = 48156,batchId 167
processing logfileName = mysql-bin.000001,offset = 48453,batchId 168
processing logfileName = mysql-bin.000001,offset = 48750,batchId 169
processing logfileName = mysql-bin.000001,offset = 49047,batchId 169
batchId 154 ack
receiver exit
ack exit
Process finished with exit code 0
可以看到，已经处理到了batchId 169，由于时间差，ack才到154就退出了，再次启动服务，日志如下：
"C:\Program Files\Java\jdk-10.0.1\bin\java.exe" "-javaagent:C:\Program Files\JetBrains\IntelliJ IDEA 2018.1.5\lib\idea_rt.jar=57362:C:\Program Files\JetBrains\IntelliJ IDEA 2018.1.5\bin" -Dfile.encoding=UTF-8 -classpath C:\Users\LiYang\IdeaProjects\canal-test\target\classes;C:\Users\LiYang\.m2\repository\com\alibaba\otter\canal.client\1.0.25\canal.client-1.0.25.jar;C:\Users\LiYang\.m2\repository\com\alibaba\otter\canal.protocol\1.0.25\canal.protocol-1.0.25.jar;C:\Users\LiYang\.m2\repository\com\alibaba\otter\canal.common\1.0.25\canal.common-1.0.25.jar;C:\Users\LiYang\.m2\repository\io\netty\netty-all\4.1.6.Final\netty-all-4.1.6.Final.jar;C:\Users\LiYang\.m2\repository\org\apache\zookeeper\zookeeper\3.4.5\zookeeper-3.4.5.jar;C:\Users\LiYang\.m2\repository\org\jboss\netty\netty\3.2.2.Final\netty-3.2.2.Final.jar;C:\Users\LiYang\.m2\repository\com\101tec\zkclient\0.10\zkclient-0.10.jar;C:\Users\LiYang\.m2\repository\commons-io\commons-io\2.4\commons-io-2.4.jar;C:\Users\LiYang\.m2\repository\commons-codec\commons-codec\1.9\commons-codec-1.9.jar;C:\Users\LiYang\.m2\repository\com\alibaba\fastjson\1.2.28\fastjson-1.2.28.jar;C:\Users\LiYang\.m2\repository\com\google\guava\guava\18.0\guava-18.0.jar;C:\Users\LiYang\.m2\repository\ch\qos\logback\logback-core\1.1.3\logback-core-1.1.3.jar;C:\Users\LiYang\.m2\repository\ch\qos\logback\logback-classic\1.1.3\logback-classic-1.1.3.jar;C:\Users\LiYang\.m2\repository\org\slf4j\jcl-over-slf4j\1.7.12\jcl-over-slf4j-1.7.12.jar;C:\Users\LiYang\.m2\repository\org\slf4j\slf4j-api\1.7.12\slf4j-api-1.7.12.jar;C:\Users\LiYang\.m2\repository\org\springframework\spring-core\3.2.9.RELEASE\spring-core-3.2.9.RELEASE.jar;C:\Users\LiYang\.m2\repository\commons-logging\commons-logging\1.1.3\commons-logging-1.1.3.jar;C:\Users\LiYang\.m2\repository\org\springframework\spring-aop\3.2.9.RELEASE\spring-aop-3.2.9.RELEASE.jar;C:\Users\LiYang\.m2\repository\aopalliance\aopalliance\1.0\aopalliance-1.0.jar;C:\Users\LiYang\.m2\repository\org\springframework\spring-beans\3.2.9.RELEASE\spring-beans-3.2.9.RELEASE.jar;C:\Users\LiYang\.m2\repository\org\springframework\spring-context\3.2.9.RELEASE\spring-context-3.2.9.RELEASE.jar;C:\Users\LiYang\.m2\repository\org\springframework\spring-expression\3.2.9.RELEASE\spring-expression-3.2.9.RELEASE.jar;C:\Users\LiYang\.m2\repository\org\springframework\spring-jdbc\3.2.9.RELEASE\spring-jdbc-3.2.9.RELEASE.jar;C:\Users\LiYang\.m2\repository\org\springframework\spring-tx\3.2.9.RELEASE\spring-tx-3.2.9.RELEASE.jar;C:\Users\LiYang\.m2\repository\org\springframework\spring-orm\3.2.9.RELEASE\spring-orm-3.2.9.RELEASE.jar;C:\Users\LiYang\.m2\repository\com\google\protobuf\protobuf-java\2.6.1\protobuf-java-2.6.1.jar;C:\Users\LiYang\.m2\repository\commons-lang\commons-lang\2.6\commons-lang-2.6.jar;C:\Users\LiYang\.m2\repository\redis\clients\jedis\2.9.0\jedis-2.9.0.jar;C:\Users\LiYang\.m2\repository\org\apache\commons\commons-pool2\2.4.2\commons-pool2-2.4.2.jar com.github.liyang0211.canal.test.SimpleCanalClient
Already processed ! logfileName = mysql-bin.000001,offset = 44295,batchId 170
Already processed ! logfileName = mysql-bin.000001,offset = 44592,batchId 170
Already processed ! logfileName = mysql-bin.000001,offset = 44889,batchId 170
Already processed ! logfileName = mysql-bin.000001,offset = 45186,batchId 170
Already processed ! logfileName = mysql-bin.000001,offset = 45483,batchId 170
Already processed ! logfileName = mysql-bin.000001,offset = 45780,batchId 170
Already processed ! logfileName = mysql-bin.000001,offset = 46077,batchId 171
Already processed ! logfileName = mysql-bin.000001,offset = 46374,batchId 171
Already processed ! logfileName = mysql-bin.000001,offset = 46671,batchId 171
Already processed ! logfileName = mysql-bin.000001,offset = 46968,batchId 171
Already processed ! logfileName = mysql-bin.000001,offset = 47265,batchId 171
Already processed ! logfileName = mysql-bin.000001,offset = 47562,batchId 171
Already processed ! logfileName = mysql-bin.000001,offset = 47859,batchId 172
Already processed ! logfileName = mysql-bin.000001,offset = 48156,batchId 172
Already processed ! logfileName = mysql-bin.000001,offset = 48453,batchId 172
Already processed ! logfileName = mysql-bin.000001,offset = 48750,batchId 172
Already processed ! logfileName = mysql-bin.000001,offset = 49047,batchId 172
processing logfileName = mysql-bin.000001,offset = 49344,batchId 172
processing logfileName = mysql-bin.000001,offset = 49641,batchId 173
processing logfileName = mysql-bin.000001,offset = 49938,batchId 173
batchId 170 ack
batchId 171 ack
batchId 172 ack
batchId 173 ack
receiver exit
可以看到，44295开始一直到49047位置的消息，又拉过来了，但是被检测到offset比最后处理的offset小，所以不处理
后面几条处理的，是服务停掉后，又修改的数据库
最终都确认成功，如果再次打开服务，会回到第一条日志的情况
redis内容如下
图1
图2
图2是记录了所有已处理过offset，可以看到，并没有重复处理
总结
原理非常简单，canal可以作为mq消息的一个补偿机制，也给以前需要定时全表扫描的任务提供了新的思路，另外可以关注阿里的otter，也是基于canal的，定位是“分布式数据库同步系统”，底层是canal
感谢阅读，如有错漏，欢迎指正