部分命令没有详细解释,不清楚的可以问AI
一、背景与目标
搭建一条从 MySQL 到 Kafka 的实时数据同步管道(CDC),为后续实时数仓(Flink + Doris + Paimon)提供数据源。
技术选型:
-
MySQL 5.7.28(开启 binlog)
-
Kafka 3.2.0(消息队列)
-
Debezium 1.9.7(CDC 工具)
-
Kafka Connect(Debezium 运行框架)
二、环境准备
| 组件 | 版本 | 安装路径 |
|---|---|---|
| CentOS | 7 | --- |
| Java | OpenJDK 1.8.0_412 | /opt/module/jdk1.8.0_212 |
| Kafka | 3.2.0 | /opt/module/kafka |
| Zookeeper | 3.5.7 | /opt/module/zookeeper-3.5.7 |
| MySQL | 5.7.28 | hadoop102:3306 |
| Debezium | 1.9.7.Final | /opt/module/kafka/plugins/debezium-connector-mysql |
三、MySQL 开启 binlog(数据源准备)
修改配置文件 /etc/my.cnf:
ini
[mysqld]
server-id=1
log-bin=mysql-bin
binlog_format=row
binlog-row-image=FULL
# 注意:如果配置了 binlog-do-db,只记录指定数据库的变更
# 如需同步所有数据库,请注释掉该配置
# binlog-do-db=retail_db
重启 MySQL:
bash
systemctl restart mysqld
验证 binlog 是否开启:
bash
mysql -u root -p -e "SHOW VARIABLES LIKE 'log_bin';"
# 预期输出: log_bin = ON
四、安装 Debezium MySQL Connector
bash
cd /opt/module/kafka
mkdir -p plugins
cd plugins
# 下载 Debezium 插件
wget https://mirrors.tuna.tsinghua.edu.cn/apache/kafka/3.2.0/connect-plugins/debezium-connector-mysql-1.9.7.Final-plugin.tar.gz
# 解压
tar -xzvf debezium-connector-mysql-1.9.7.Final-plugin.tar.gz
# 确认解压成功
ls -l debezium-connector-mysql/
五、配置 Kafka Connect
配置文件路径: /opt/module/kafka/config/connect-distributed.properties
properties
# Kafka Connect 集群标识
group.id=connect-cluster
# Kafka 集群地址(三台 broker)
bootstrap.servers=hadoop102:9092,hadoop103:9092,hadoop104:9092
# Debezium 插件路径
plugin.path=/opt/module/kafka/plugins
# 偏移量存储 Topic
offset.storage.topic=connect-offsets
offset.storage.replication.factor=1
offset.storage.partitions=25
# 配置存储 Topic
config.storage.topic=connect-configs
config.storage.replication.factor=1
# 状态存储 Topic
status.storage.topic=connect-status
status.storage.replication.factor=1
# 序列化格式
key.converter=org.apache.kafka.connect.json.JsonConverter
value.converter=org.apache.kafka.connect.json.JsonConverter
key.converter.schemas.enable=false
value.converter.schemas.enable=false
# REST API 地址
rest.advertised.host.name=hadoop102
rest.advertised.port=8083
六、踩坑记录(核心章节)
坑 1:connect-offsets Topic 的 cleanup.policy 必须是 compact
问题现象:
log
org.apache.kafka.common.config.ConfigException: Topic 'connect-offsets' supplied via the 'offset.storage.topic' property is required to have 'cleanup.policy=compact' to guarantee consistency and durability of source connector offsets, but found the topic currently has 'cleanup.policy=delete'.
根本原因:
-
Kafka Connect 要求
connect-offsetsTopic 的cleanup.policy必须是compact,否则无法持久化偏移量。 -
如果该 Topic 不存在,Kafka 自动创建时默认使用
cleanup.policy=delete。
解决方案:
bash
# 1. 删除原有 Topic
kafka-topics.sh --bootstrap-server hadoop102:9092 --delete --topic connect-offsets
# 2. 手动重建,指定 cleanup.policy=compact
kafka-topics.sh --bootstrap-server hadoop102:9092 --create \
--topic connect-offsets \
--partitions 25 \
--replication-factor 1 \
--config cleanup.policy=compact \
--config min.cleanable.dirty.ratio=0.01 \
--config segment.ms=10000
# 3. 验证配置是否生效
kafka-topics.sh --bootstrap-server hadoop102:9092 --describe --topic connect-offsets | grep cleanup
# 预期输出: cleanup.policy=compact
坑 2:8083 端口被占用(Address already in use)
问题现象:
log
ERROR Stopping due to error
org.apache.kafka.connect.errors.ConnectException: Unable to initialize REST server
Caused by: java.net.BindException: Address already in use
curl http://hadoop102:8083/connectors 返回 Connection refused。
根本原因:
-
上一个 Kafka Connect 进程未正常退出,仍占用 8083 端口。
-
新的 Connect 进程启动时无法绑定端口,导致启动失败。
解决方案:
bash
# 1. 查看谁占用了 8083 端口
lsof -i:8083
# 2. 强制杀掉占用进程
kill -9 <PID>
# 3. 重新启动 Kafka Connect
cd /opt/module/kafka
nohup bin/connect-distributed.sh config/connect-distributed.properties > /dev/null 2>&1 &
坑 3:Schema History Topic 缺失(The db history topic is missing)
问题现象:
log
WARN Database history was not found but was expected
ERROR io.debezium.DebeziumException: The db history topic is missing. You may attempt to recover it by reconfiguring the connector to SCHEMA_ONLY_RECOVERY
根本原因:
-
Debezium 需要
schema-changes.retail_dbTopic 来存储表结构变更历史。 -
该 Topic 不存在时,Debezium 不会自动创建,而是直接报错退出。
解决方案:
bash
# 1. 手动创建 Schema History Topic
kafka-topics.sh --bootstrap-server hadoop102:9092 --create \
--topic schema-changes.retail_db \
--partitions 1 \
--replication-factor 1
# 2. 删除连接器后重新注册,使用 schema_only_recovery 模式
curl -X DELETE http://hadoop102:8083/connectors/mysql-connector-retail
curl -X POST -H "Content-Type: application/json" --data '
{
"name": "mysql-connector-retail",
"config": {
"connector.class": "io.debezium.connector.mysql.MySqlConnector",
"database.hostname": "hadoop102",
"database.port": "3306",
"database.user": "root",
"database.password": "123456",
"database.server.id": "1",
"database.server.name": "retail_db",
"database.include.list": "retail_db",
"database.history.kafka.bootstrap.servers": "hadoop102:9092,hadoop103:9092,hadoop104:9092",
"database.history.kafka.topic": "schema-changes.retail_db",
"include.schema.changes": "true",
"snapshot.mode": "schema_only_recovery",
"snapshot.locking.mode": "none",
"decimal.handling.mode": "double"
}
}' http://hadoop102:8083/connectors
坑 4:Decimal 类型序列化为乱码
问题现象:
Flink 或 kafka-console-consumer 消费数据时,price 字段显示为 "AV8s"、"Cq38" 等不可读字符,而不是 6999.00。
根本原因:
-
Debezium 默认将 MySQL 的
DECIMAL类型序列化为 Base64 编码的字节数组。 -
需要配置
decimal.handling.mode=double,让 Debezium 以数字形式输出。
解决方案:
在连接器配置中添加:
json
"decimal.handling.mode": "double"
完整连接器注册命令:
bash
curl -X POST -H "Content-Type: application/json" --data '
{
"name": "mysql-connector-retail",
"config": {
"connector.class": "io.debezium.connector.mysql.MySqlConnector",
"database.hostname": "hadoop102",
"database.port": "3306",
"database.user": "root",
"database.password": "123456",
"database.server.id": "1",
"database.server.name": "retail_db",
"database.include.list": "retail_db",
"database.history.kafka.bootstrap.servers": "hadoop102:9092,hadoop103:9092,hadoop104:9092",
"database.history.kafka.topic": "schema-changes.retail_db",
"include.schema.changes": "true",
"snapshot.mode": "initial",
"snapshot.locking.mode": "none",
"decimal.handling.mode": "double"
}
}' http://hadoop102:8083/connectors
坑 5:binlog-do-db 限制了可同步的数据库
问题现象:
Debezium 连接器状态显示 RUNNING,但 Kafka 中始终没有生成对应的 Topic,消费不到任何数据。
根本原因:
MySQL 配置了 binlog-do-db,只记录指定数据库的 binlog。如果目标数据库不在白名单中,Debezium 读不到任何变更。
解决方案:
修改 /etc/my.cnf:
ini
# 如需同步所有数据库,注释掉该配置
# binlog-do-db=gmall
# binlog-do-db=gmall2023_config
# 或添加目标数据库
binlog-do-db=retail_db
重启 MySQL:
bash
systemctl restart mysqld
七、最终验证
1. 查看连接器状态:
bash
curl http://hadoop102:8083/connectors/mysql-connector-retail/status
# 预期: "state":"RUNNING"
2. 查看 Kafka 中的 Topic:
bash
kafka-topics.sh --bootstrap-server hadoop102:9092 --list | grep retail
# 预期输出:
# retail_db
# retail_db.retail_db.orders
# retail_db.retail_db.products
# schema-changes.retail_db
3. 消费数据验证:
bash
kafka-console-consumer.sh --bootstrap-server hadoop102:9092 \
--topic retail_db.retail_db.products \
--from-beginning \
--max-messages 3
预期输出 (JSON 格式,price 为数字):
json
{"before":null,"after":{"product_id":1,"product_name":"iPhone 15","category":"Electronics","price":6999.0,"stock":50},"source":{...},"op":"r","ts_ms":...}
八、踩坑总结
| 坑 | 关键词 | 核心解决方案 |
|---|---|---|
| 1 | cleanup.policy=compact |
手动重建 connect-offsets,指定 compact |
| 2 | Address already in use |
lsof -i:8083 → kill -9 |
| 3 | The db history topic is missing |
手动创建 schema-changes.retail_db,使用 SCHEMA_ONLY_RECOVERY |
| 4 | Decimal 乱码 | 连接器配置添加 decimal.handling.mode=double |
| 5 | 数据同步不到 | 检查 binlog-do-db 是否包含目标数据库 |