背景
在最近的 v6.5.0 发布的新特性中(详见:release-6.5.0),我对一个特性特别感兴趣: 从 v6.5.0 开始,TiCDC 支持将行变更事件保存至存储服务,如 Amazon S3、Azure Blob Storage 和 NFS。参考:使用指南 简言之,这个特性能够通过 TiCDC 这一个组件,就能保存 CDC 日志到对象存储中。如果应用这个特性,我就没有必要部署 Kafka 集群了,简化了部署要求,降低了部署成本。 本文依托部署在 K8s 中的 TiDB ,测试 CDC 的新特性,OSS 采用 MinIO。
环境准备
TiDB 在 k8s 上的部署详见:快速上手 TiDB Operator,部署个测试集群即可,官方文档已经很详细了,可以参考我的 tidb-cluster.yml 如下:
# IT IS NOT SUITABLE FOR PRODUCTION USE.
# This YAML describes a basic TiDB cluster with minimum resource requirements,
# which should be able to run in any Kubernetes cluster with storage support.
apiVersion: pingcap.com/v1alpha1
kind: TidbCluster
metadata:
name: basic
spec:
version: v6.5.0
timezone: UTC
pvReclaimPolicy: Retain
enableDynamicConfiguration: true
configUpdateStrategy: RollingUpdate
discovery: {}
helper:
image: alpine:3.16.0
pd:
baseImage: uhub.service.ucloud.cn/pingcap/pd
maxFailoverCount: 0
replicas: 3
# if storageClassName is not set, the default Storage Class of the Kubernetes cluster will be used
# storageClassName: local-storage
requests:
storage: "1Gi"
config: {}
tikv:
baseImage: uhub.service.ucloud.cn/pingcap/tikv
maxFailoverCount: 0
# If only 1 TiKV is deployed, the TiKV region leader
# cannot be transferred during upgrade, so we have
# to configure a short timeout
evictLeaderTimeout: 1m
replicas: 3
# if storageClassName is not set, the default Storage Class of the Kubernetes cluster will be used
# storageClassName: local-storage
requests:
storage: "1Gi"
config:
storage:
# In basic examples, we set this to avoid using too much storage.
reserve-space: "0MB"
rocksdb:
# In basic examples, we set this to avoid the following error in some Kubernetes clusters:
# "the maximum number of open file descriptors is too small, got 1024, expect greater or equal to 82920"
max-open-files: 256
raftdb:
max-open-files: 256
tidb:
baseImage: uhub.service.ucloud.cn/pingcap/tidb
maxFailoverCount: 0
replicas: 1
service:
type: ClusterIP
config: {}
ticdc:
baseImage: uhub.service.ucloud.cn/pingcap/ticdc
replicas: 3
config:
logLevel: info
部署后集群状态如下: 部署过程中,有点小波折,关于 6.5.0 版本中 TiDB-Dashboard 依赖包版本太高的问题,我最终采用了社区大佬的镜像,替换了官方镜像完成了部署。 替换步骤为,修改官方的 tidb-dashboard.yml:
apiVersion: pingcap.com/v1alpha1
kind: TidbDashboard
metadata:
name: basic
spec:
baseImage: sabaping/tidb-dashboard:v6.5.0
version: v6.5.0
clusters:
- name: basic
requests:
storage: 10Gi
修改完成后,执行:
kubectl apply -f tidb-dashboard.yml -n tidb-cluster
即可完成更新。
MinIO 的在 k8s 上的部署详见:kubernetes 部署 minio 对象存储
配置 CDC 任务
官方参考文档:https://docs.pingcap.com/zh/tidb/stable/ticdc-sink-to-cloud-storage 我的目前是配置 TiCDC,同步 TiDB 的数据到同一个 k8s 集群的 MinIO 中,我配置中的地址都采用了 Service 地址。 我的配置如下:
./cdc cli changefeed create \
--server=http://basic-ticdc-peer:8301 \
--sink-uri="s3://tidbbinlog/?protocol=canal-json&endpoint=http://minio-headless.minio:9000&access-key=XXX&secret-access-key=XXX&force-path-style=true" \
--changefeed-id="oss-replication-task"
各参数说明:
- server=http://basic-ticdc-peer:8301 ticdc 的服务地址
- sink-uri=... 目标对象存储的配置
- s3://tidbbinlog/ TiCDC 支持的对象存储包含 s3 协议,MinIO 支持 s3 协议,此处采用 s3 协议,tidbbinlog 为 bucket 名称
- protocol=canal-json 存储到对象存储中内容的格式,v6.5.0 支持 CSV 和 Canal-JSON 格式,我惯用 canal-json
- endpoint=http://minio-headless.minio:9000 对象存储访问地址
- access-key=XXX 对象存储访问 access-key
- secret-access-key=XXX 对象存储访问 secret-access-key
- force-path-style=true 采用路径访问的方式访问对象存储,访问 MinIO 必须设置
- changefeed-id="oss-replication-task" CDC 的任务 ID
在命令行执行:
kubectl exec basic-ticdc-0 -it -n tidb-cluster -- /bin/bash
进入 pod 的命令行窗口,执行命令:
./cdc cli changefeed create \
--server=http://basic-ticdc-peer:8301 \
--sink-uri="s3://tidbbinlog/?protocol=canal-json&endpoint=http://minio-headless.minio:9000&access-key=XXX&secret-access-key=XXX&force-path-style=true" \
--changefeed-id="oss-replication-task"
返回信息如下:
Create changefeed successfully!
ID: oss-replication-task
Info: {"upstream_id":7184751775799892803,"namespace":"default","id":"oss-replication-task","sink_uri":...}
此时查询任务状态:
./cdc cli changefeed list --server=http://basic-ticdc-peer:8301
[
{
"id": "oss-replication-task",
"namespace": "default",
"summary": {
"state": "normal",
"tso": 438588139028348953,
"checkpoint": "2023-01-07 08:42:54.687",
"error": null
}
}
]
MinIO 上可以看到在对应的 bucket 下已经创建了一个文件: 文件内容为:
{"checkpoint-ts":438588195559702535}
438588195559702535 为同步数据的 commit-ts,438588195559702535 之前的数据都已经同步到了 MinIO。
测试数据变更
测试非分区表
在 TiDB 中建表和插入数据:
CREATE TABLE test.student (id varchar(20) NULL,
name varchar(20) NULL,
age varchar(100) NULL,
`desc` varchar(100) NULL
);
INSERT INTO test.student (id,name,age,`desc`)
VALUES ('1','zhangsan','18','好学生');
INSERT INTO test.student (id,name,age,`desc`)
VALUES ('2','lisi','19','也是个好学生');
INSERT INTO test.student (id,name,age,`desc`)
VALUES ('3','wangwu','19','更是个好学生');
检查 MinIO 并没有新的目录和文件产生,经检查日志:
2023-01-07T08:49:18.857903341Z [2023/01/07 08:49:18.857 +00:00] [WARN] [snapshot.go:849] ["this table is ineligible to replicate"] [tableName=student] [tableID=82]
查找原因是因为 test.student 表没有主键,建主键,并插入数据:
ALTER TABLE test.student ADD CONSTRAINT student_PK PRIMARY KEY (id);
INSERT INTO test.student (id,name,age,`desc`)
VALUES ('21','zhangsan','18','好学生');
INSERT INTO test.student (id,name,age,`desc`)
VALUES ('22','lisi','19','也是个好学生');
INSERT INTO test.student (id,name,age,`desc`)
VALUES ('23','wangwu','19','更是个好学生');
其中,CDC000001.json:
{"id":0,"database":"test","table":"student","pkNames":["id"],"isDdl":false,"type":"INSERT","es":1673227668037,"ts":1673227668465,"sql":"","sqlType":{"id":12,"name":12,"age":12,"desc":12},"mysqlType":{"id":"varchar","name":"varchar","age":"varchar","desc":"varchar"},"old":null,"data":[{"id":"21","name":"zhangsan","age":"18","desc":" 好学生 "}]}
{"id":0,"database":"test","table":"student","pkNames":["id"],"isDdl":false,"type":"INSERT","es":1673227669537,"ts":1673227670111,"sql":"","sqlType":{"id":12,"name":12,"age":12,"desc":12},"mysqlType":{"id":"varchar","name":"varchar","age":"varchar","desc":"varchar"},"old":null,"data":[{"id":"22","name":"lisi","age":"19","desc":" 也是个好学生 "}]}
{"id":0,"database":"test","table":"student","pkNames":["id"],"isDdl":false,"type":"INSERT","es":1673227670787,"ts":1673227671141,"sql":"","sqlType":{"id":12,"name":12,"age":12,"desc":12},"mysqlType":{"id":"varchar","name":"varchar","age":"varchar","desc":"varchar"},"old":null,"data":[{"id":"23","name":"wangwu","age":"19","desc":" 更是个好学生 "}]}
schema.json:
{
"Table": "student",
"Schema": "test",
"Version": 1,
"TableVersion": 438626581528444931,
"Query": "","Type": 0,"TableColumns": [
{
"ColumnName": "id",
"ColumnType": "VARCHAR",
"ColumnPrecision": "20",
"ColumnNullable": "false",
"ColumnIsPk": "true"
},
{
"ColumnName": "name",
"ColumnType": "VARCHAR",
"ColumnPrecision": "20"
},
{
"ColumnName": "age",
"ColumnType": "VARCHAR",
"ColumnPrecision": "100"
},
{
"ColumnName": "desc",
"ColumnType": "VARCHAR",
"ColumnPrecision": "100"
}
],
"TableColumnsTotal": 4
}
详细的内容解释,官方有文档说明,在此不再赘述。 为表增加字段:
ALTER TABLE test.student ADD graduation CHAR(1) NULL;
test/student 下面增加了一个目录,根据官方说明,这一层是表的版本,也就是说,如果对表执行 ddl,此处就会增加一个目录。
测试分区表
分区表建表并插入数据:
CREATE TABLE `student_partition` (
`id` INT NOT NULL,
`name` varchar(20) DEFAULT NULL,
`age` varchar(100) DEFAULT NULL,
`desc` varchar(100) DEFAULT NULL,
`graduation` char(1) DEFAULT NULL,
PRIMARY KEY (`id`)
)PARTITION BY HASH(id)
PARTITIONS 6;
INSERT INTO test.student_partition (id,name,age,`desc`)
VALUES ('21','zhangsan','18','好学生');
INSERT INTO test.student_partition (id,name,age,`desc`)
VALUES ('22','lisi','19','也是个好学生');
INSERT INTO test.student_partition (id,name,age,`desc`)
VALUES ('23','wangwu','19','更是个好学生');
查看对象存储: 发现在默认情况下,分区表和非分区表存储结构是一致的,但分区表通常是因为数据量巨大,才分区的,在处理 cdc 日志时,通常也希望能按照分区处理。 通过修改参数设置按照分区存储,执行:
kubectl exec basic-ticdc-0 -it -n tidb-cluster -- /bin/bash
vi config
在 config 里面写入:
[sink]
date-separator = 'day'
enable-partition-separator = true
其中: enable-partition-separator:开启按照分区分割目录 date-separator:按照分区分割目录的下层按天分割目录 保存 config 文件,执行:
./cdc cli changefeed create \
--server=http://basic-ticdc-peer:8301 \
--sink-uri="s3://tidbbinlog-separator/?protocol=canal-json&endpoint=http://minio-headless.minio:9000&access-key=XXX&secret-access-key=XXX&force-path-style=true" \
--changefeed-id="oss-replication-task-separator" \
--config ./config
实际情况如下: 目录为tidbbinlog-separator/test/student_partition/438627372744835072/86/2023-01-09/CDC000001.json 目录中,86 为分区的 ID,2023-01-09 为按时间分割的目录,这两种分区方式可以根据需要,分别使用和组合使用。
展望
我们在 Mysql 的 binlog 日志上,对于类似的处理方式已经有一些应用于生产,对于 json 格式的 cdc 日志,我们团队也有一些应用心得,有时间会写一篇文章补充介绍。