作者： Liuhaoao 原文来源： https://tidb.net/blog/170d6d47
近期有个生产系统，计划做一套dr-autosync的集群，但是之前并没有这种类型系统的生产实施经验，就一点点的摸索，好在最后是顺利搭建成功了，把搭建过程分享出来给大家参考下。
1、集群架构
2、规划拓扑
根据集群架构规划拓扑文件
global:
user: "tidb"
ssh_port: 22
deploy_dir: "/tidb/tidb-deploy"
data_dir: "/tidb/tidb-data"
arch: "arm64"
monitored:
node_exporter_port: 19100
blackbox_exporter_port: 19115
server_configs:
tidb:
tikv:
pd:
dashboard.enable-telemetry: false
log.file.max-backups: 100
log.file.max-days: 90
replication.isolation-level: logic
replication.location-labels:
- dc
- logic
- rack
- host
replication.max-replicas: 5
schedule.max-store-down-time: 30m
pd_servers:
- host: 10.3.65.1
client_port: 12379
peer_port: 12380
- host: 10.3.65.2
client_port: 12379
peer_port: 12380
- host: 10.3.65.3
client_port: 12379
peer_port: 12380
- host: 10.3.65.1
client_port: 12379
peer_port: 12380
- host: 10.3.65.2
client_port: 12379
peer_port: 12380
tidb_servers:
- host: 10.3.65.1
port: 24000
status_port: 20080
- host: 10.3.65.2
port: 24000
status_port: 20080
- host: 10.3.65.3
port: 24000
status_port: 20080
- host: 10.3.65.1
port: 24000
status_port: 20080
- host: 10.3.65.2
port: 24000
status_port: 20080
- host: 10.3.65.3
port: 24000
status_port: 20080
tikv_servers:
- host: 10.3.65.1
port: 20160
status_port: 20180
config:
server.labels:
dc: dc1
logic: logic1
rack: rack1
host: host1
- host: 10.3.65.2
port: 20160
status_port: 20180
config:
server.labels:
dc: dc1
logic: logic2
rack: rack1
host: host1
- host: 10.3.65.3
port: 20160
status_port: 20180
config:
server.labels:
dc: dc1
logic: logic3
rack: rack1
host: host1
- host: 10.3.65.1
port: 20160
status_port: 20180
config:
server.labels:
dc: dc2
logic: logic4
rack: rack1
host: host1
- host: 10.3.65.2
port: 20160
status_port: 20180
config:
server.labels:
dc: dc2
logic: logic5
rack: rack1
host: host1
- host: 10.3.65.3
port: 20160
status_port: 20180
config:
server.labels:
dc: dc2
logic: logic6
rack: rack1
host: host1
monitoring_servers:
- host: 10.3.65.3
port: 29090
ng_port: 22020
grafana_servers:
- host: 10.3.65.3
port: 23000
alertmanager_servers:
- host: 10.3.65.3
web_port: 29093
cluster_port: 29094
3、集群部署
1、部署集群
tiup cluster deploy dr-auto-sync v6.5.4 dr-auto-sync.yaml --user tidb -p
2、编写dr-auto-sync集群的json文件
vim rule.json
[
{
"group_id": "pd",
"group_index": 0,
"group_override": false,
"rules": [
{
"group_id": "pd",
"id": "dc1",
"start_key": "",
"end_key": "",
"role": "voter",
"count": 3,
"location_labels": ["dc", "logic", "rack", "host"],
"label_constraints": [{"key": "dc", "op": "in", "values": ["dc1"]}]
},
{
"group_id": "pd",
"id": "dc2",
"start_key": "",
"end_key": "",
"role": "follower",
"count": 2,
"location_labels": ["dc", "logic", "rack", "host"],
"label_constraints": [{"key": "dc", "op": "in", "values": ["dc2"]}]
},
{
"group_id": "pd",
"id": "dc2-1",
"start_key": "",
"end_key": "",
"role": "learner",
"count": 1,
"location_labels": ["dc", "logic", "rack", "host"],
"label_constraints": [{"key": "dc", "op": "in", "values": ["dc2"]}]
}
]
}
]
3、配置placement rule json文件，使其生效
[tidb@tidb141 ~]$ tiup ctl:v6.5.4 pd -u 10.3.65.141:22379 -i
» config placement-rules rule-bundle save --in="/home/tidb/rule.json"
检查配置是否生效
» config placement-rules show
4、修改dr-auto-sync 模式
config set replication-mode dr-auto-sync
5、配置dr-auto-sync 的机房标签
config set replication-mode dr-auto-sync label-key dc
6、配置主机房
config set replication-mode dr-auto-sync primary dc1
7、配置从机房
config set replication-mode dr-auto-sync dr dc2
8、配置主机房副本数量
config set replication-mode dr-auto-sync primary-replicas 3
9、配置从机房副本数量
config set replication-mode dr-auto-sync dr-replicas 2
10、如果集群为跨机房部署的dr-auto-sync 架构，需要确保pd leader 始终位于主机房，可以配置主机房pd权重高于备机房，数值越大权重越高，越优先考虑成为pd leader
tiup ctl:v6.5.3 pd –u 192.168.113.1:12379 -i
member leader_prioritypd-192.168.113.1-12379 100
member leader_prioritypd-192.168.113.2-12379 100
member leader_prioritypd-192.168.113.3-12379 100
member leader_prioritypd-192.168.113.4-12379 50
member leader_prioritypd-192.168.113.5-12379 50
11、检查集群同步状态
[tidb@tidb141 ~]$ curl http://10.3.65.141:22379/pd/api/v1/replication_mode/status
4、测试
1、手动关停备机房tikv节点，等待约一分钟左右，检查同步级别是否自动降级为async
[tidb@tidb141 ~]$ tiup cluster stop dr-auto-sync -N 10.3.65.142:10160,10.3.65.142:40160,10.3.65.142:50160
同步级别自动降级为async（异步）
2、启动关停的tikv节点，等待约一分钟左右，检查同步级别是否自动升级为sync
[tidb@tidb141 ~]$ tiup cluster start dr-auto-sync -N 10.3.65.142:10160,10.3.65.142:40160,10.3.65.142:50160
同步级别自动升级为sync，符合预期
至此，dr-auto-sync集群部署成功
5、总结
dr-auto-sync集群，较普通集群其实区别不大，只要按需规划好集群拓扑及、abels、json文件，基本上不会有什么问题，把它当作普通集群部署就可以，但有几点需要注意：
1、跨机房的话，需要配置pd的权重，防止pd leader跑到备机房，影响整体性能
2、我使用的6.5.4版本，dr-auto-sync有个bug，配置完成后，需要reload一下tikv节点，触发region leader重新选举，同步链路才会升级为sync状态，否则会一直卡在sync_recover阶段