placement-rules介绍
TiDB的placement-rules是一种用于控制数据在TiKV集群中放置位置的规则系统。它允许用户通过SQL接口、配置文件的方式，来配置数据的副本数量、存放位置、主机类型等属性，以满足不同的业务需求和优化数据的高可用性，还可以实现读写分离。
TiDB的placement-rules用于指导 PD 针对不同类型的数据生成对应的调度。通过组合不同的调度规则，用户可以精细地控制任何一段连续数据的副本数量、存放位置、主机类型、是否参与 Raft 投票、是否可以担任 Raft leader 等属性。
Placement Rules 特性在4.0版本开始引入的一套副本规则系统，Placement Rules 特性在 TiDB v5.0 及以上的版本中默认开启。
placement-rules作用
1、placement-rule允许用户精细控制任何一段连续数据的副本数量、Raft角色、放置位置等属性；
2、它由多条规则（Rule）组成，每条规则可以指定不同的副本数量、Raft角色、放置位置等属性，以及这条规则生效的key range；
3、利用placement-rules可以实现两地三中心的高可用性。
4、把重要的数据存储为5副本，把不太重要的数据存储为3副本。
如上图所示，通过placement-rules可以实现表的精细化控制。meta原数据表设置为5个voter副本；table-01设置为3个voter副本，2个learner副本；table-02设置为3个voter副本，2个follower副本；table-03设置为1个voter副本，2个follower副本，1个learner副本。
placement-rules使用方法
设置placement-rules规则常用的两种方式为：
1、通过配置文件方式来设置规则；
2、通过sql来设置规则（placement-rules in sql）。
通过配置文件来设置规则
1、查看初始化placement-rules规则
通过pd-ctl来查看初始化规则
tiup ctl:v7.5.4 pd -u 10.2.103.125:2379 config placement-rules show
2、设置5副本
tiup ctl:v7.5.4 pd -u 10.2.103.125:2379 config placement-rules save --in="rule_new_1.json"
rule_new_1.json如下：
注意： 如果需要删除某条规则，只需要将规则的 count 置为 0 即可，对应 GroupID+ID 相同的规则会被删除。 由默认变为指定的规则，必须把"id"为"default"规则的count设置为0，否则不能实现预期。 设置完新规则后，一定要进行config placement-rules show验证。
[
{
"group_id": "pd",
"id": "default",
"start_key": "",
"end_key": "",
"role": "voter",
"is_witness": false,
"count": 0,
"location_labels": [
"zone",
"host"
]
},
{
"group_id": "pd",
"id": "zone1",
"start_key": "",
"end_key": "",
"role": "voter",
"count": 1,
"label_constraints": [
{"key": "host", "op": "in", "values": ["h-125"]}
],
"location_labels": ["zone", "rack", "host"]
},
{
"group_id": "pd",
"id": "zone2",
"start_key": "",
"end_key": "",
"role": "voter",
"count": 1,
"label_constraints": [
{"key": "host", "op": "in", "values": ["h-162"]}
],
"location_labels": ["zone", "rack", "host"]
},
{
"group_id": "pd",
"id": "zone3",
"start_key": "",
"end_key": "",
"role": "voter",
"count": 1,
"label_constraints": [
{"key": "host", "op": "in", "values": ["h-163"]}
],
"location_labels": ["zone", "rack", "host"]
},
{
"group_id": "pd",
"id": "zone4",
"start_key": "",
"end_key": "",
"role": "voter",
"count": 1,
"label_constraints": [
{"key": "host", "op": "in", "values": ["h-54"]}
],
"location_labels": ["zone", "rack", "host"]
},
{
"group_id": "pd",
"id": "zone5",
"start_key": "",
"end_key": "",
"role": "voter",
"count": 1,
"label_constraints": [
{"key": "host", "op": "in", "values": ["h-78"]}
],
"location_labels": ["zone", "rack", "host"]
}
]
3、指定leader在zone1上
tiup ctl:v7.5.4 pd -u 10.2.103.125:2379 config placement-rules save --in="rule_new_2.json"
rule_new_2.json如下：
[
{
"group_id": "pd",
"id": "default",
"start_key": "",
"end_key": "",
"role": "voter",
"is_witness": false,
"count": 0,
"location_labels": [
"zone",
"host"
]
},
{
"group_id": "pd",
"id": "zone1",
"start_key": "",
"end_key": "",
"role": "voter",
"count": 1,
"label_constraints": [
{"key": "host", "op": "in", "values": ["h-125"]}
],
"location_labels": ["zone", "rack", "host"]
},
{
"group_id": "pd",
"id": "zone2",
"start_key": "",
"end_key": "",
"role": "follower",
"count": 1,
"label_constraints": [
{"key": "host", "op": "in", "values": ["h-162"]}
],
"location_labels": ["zone", "rack", "host"]
},
{
"group_id": "pd",
"id": "zone3",
"start_key": "",
"end_key": "",
"role": "follower",
"count": 1,
"label_constraints": [
{"key": "host", "op": "in", "values": ["h-163"]}
],
"location_labels": ["zone", "rack", "host"]
},
{
"group_id": "pd",
"id": "zone4",
"start_key": "",
"end_key": "",
"role": "follower",
"count": 1,
"label_constraints": [
{"key": "host", "op": "in", "values": ["h-54"]}
],
"location_labels": ["zone", "rack", "host"]
},
{
"group_id": "pd",
"id": "zone5",
"start_key": "",
"end_key": "",
"role": "follower",
"count": 1,
"label_constraints": [
{"key": "host", "op": "in", "values": ["h-78"]}
],
"location_labels": ["zone", "rack", "host"]
}
]
此时，leader都集中在125节点上。
注意：
此规则可以配合tidb_replica_read参数，实现读写分离，读在follower节点，写在leader节点。
set global tidb_replica_read = 'follower';
--设置读访问follower节点
通过placement-rules in sql来设置规则
1、在没设置placement-rules时，tpcc.oorder表有2个region，leader分别在“h-125”、“h-78”。
select distinct t2.db_name,t2.table_name,t2.region_id,t3.peer_id,t3.is_leader,t1.address,replace(replace(t1.label,', "value"', ''),'"key": ','')
from INFORMATION_SCHEMA.tikv_store_status t1,INFORMATION_SCHEMA.tikv_region_status t2,INFORMATION_SCHEMA.tikv_region_peers t3
where t2.db_name='tpcc' and t2.region_id=t3.region_id and t3.store_id=t1.store_id and t2.table_name='oorder' order by 1,2,3,4;
2、创建placement-rules，指定表tpcc.oorder的leader在“h-54”上。
创建policy_role_1规则，leader在“h-54”上，follower在“h-125”、“h-78”
CREATE PLACEMENT POLICY policy_role_1 LEADER_CONSTRAINTS="[+host=h-54]" FOLLOWER_CONSTRAINTS='{"+host=h-78": 1, "+host=h-125": 1,}';
MySQL [tpcc]> alter table oorder placement policy='policy_role_1';
MySQL [tpcc]> select distinct t2.db_name,t2.table_name,t2.region_id,t3.peer_id,t3.is_leader,t1.address,replace(replace(t1.label,', "value"', ''),'"key": ','')
from INFORMATION_SCHEMA.tikv_store_status t1,INFORMATION_SCHEMA.tikv_region_status t2,INFORMATION_SCHEMA.tikv_region_peers t3
where t2.db_name='tpcc' and t2.region_id=t3.region_id and t3.store_id=t1.store_id and t2.table_name='oorder' order by 1,2,3,4;
通过设置规则后，leader从“h-125”转移到“h-54”
3、创建placement-rules，指定表tpcc.oorder为五副本
创建policy_role_2规则，leader在“h-78”上，其余副本在另外几个tikv节点上
CREATE PLACEMENT POLICY policy_role_2 LEADER_CONSTRAINTS="[+host=h-78]" FOLLOWER_CONSTRAINTS='{"+host=h-125": 1, "+host=h-162": 1, "+host=h-163": 1, "+host=h-54": 1,}';
alter table oorder placement policy='policy_role_2';
4、查看已有的placement-rules in sql
--查看placement-rules
SHOW PLACEMENT;
--查看集群labels
SHOW PLACEMENT LABELS;
--删除表的placement-rules
alter table tpcc.oorder placement policy=default;
总结
1、利用好placement-rules特性，可以更好的帮助我们实现数据库高可用。
2、placement-rules易操作、灵活性高。