转自：https://my.oschina.net/Kenyon/blog/908370
一、环境
CentOS Linux release 7.3.1611 (Core)
172.26.11.91 pd & tidb
172.26.11.92 tikv
172.26.11.93 tikv
172.26.11.94 tikv
二、安装
分别在4台服务器上上传安装包
wget http://download.pingcap.org/tidb-latest-linux-amd64.tar.gz tar -xzf tidb-latest-linux-amd64.tar.gz cd tidb-latest-linux-amd64 mkdir -p /data/tidb/log ln -s /root/software/tidb/tidb-latest-linux-amd64/bin/pd-tso-bench /usr/bin ln -s /root/software/tidb/tidb-latest-linux-amd64/bin/tikv-server /usr/bin/ ln -s /root/software/tidb/tidb-latest-linux-amd64/bin/tidb-server /usr/bin/ ln -s /root/software/tidb/tidb-latest-linux-amd64/bin/pd-server /usr/bin/ ln -s /root/software/tidb/tidb-latest-linux-amd64/bin/pd-ctl /usr/bin/三、配置使用
1.按照顺序启动
在91上启动pd服务 pd-server --name=pd1 --多个pd以不同名字命名 --data-dir=/data/tidb/pd --pd路径 --client-urls="http://172.26.11.91:2379" --peer-urls="http://172.26.11.91:2380" --initial-cluster="pd1=http://172.26.11.91:2380" --多个pd以逗号分隔 --log-file=/data/tidb/log/pd.log & 在92,93,94上启动tikv tikv-server --pd="172.26.11.91:2379" \ --addr="172.26.11.92:20160" \ --data-dir=/data/tidb/tikv \ --log-file=/data/tidb/log/tikv.log tikv-server --pd="172.26.11.91:2379" \ --addr="172.26.11.93:20160" \ --data-dir=/data/tidb/tikv \ --log-file=/data/tidb/log/tikv.log tikv-server --pd="172.26.11.91:2379" \ --addr="172.26.11.94:20160" \ --data-dir=/data/tidb/tikv \ --log-file=/data/tidb/log/tikv.log & 在91上启动tipd服务 tidb-server --store=tikv \ --tikv引擎允许分布式存储，其他如LevelDB等是本地存储 --path="172.26.11.91:2379" \ --log-file=/data/tidb/log/tidb.log &2.登陆使用
[root@test05 ~]# mysql -h 172.26.11.91 -P 4000 -u root -D test Welcome to the MySQL monitor. Commands end with ; or \g. Your MySQL connection id is 7 Server version: 5.7.1-TiDB-1.0 MySQL Community Server (GPL) Copyright (c) 2000, 2017, Oracle and/or its affiliates. All rights reserved. Oracle is a registered trademark of Oracle Corporation and/or its affiliates. Other names may be trademarks of their respective owners. Type 'help;' or '\h' for help. Type '\c' to clear the current input statement. mysql> create database db_kenyon; Query OK, 0 rows affected (2.02 sec) mysql> use db_kenyon; Database changed mysql> create table tbl_kenyon(user_code varchar(64) primary key,user_name varchar(32),ctime timestamp); Query OK, 0 rows affected (2.03 sec) mysql> insert into tbl_kenyon values('01','qiaofeng',now()),('02','murong',now()); Query OK, 2 rows affected (0.01 sec) mysql>三、高可用
tidb的数据都是保存在tikv节点上面，比如上面配置了3套tikv，每套tikv都是独立的，数据保存的方式和传统关系型不一样的是，在tidb里面或者说tikv里面是映射成kv模式存储的
把92的tikv人为挂掉，此时数据库的使用会受影响，简单的一个查询就会被挂起，直到切换成功
--切换过程
mysql> select * from tbl_kenyon;
+----+-----------+---------------------+
| id | cname | ctime | +----+-----------+---------------------+ | 1 | qiaofeng | 2017-05-23 10:50:43 | | 2 | murong | 2017-05-23 10:50:43 | | 3 | saodiseng | 2017-05-23 10:50:43 | +----+-----------+---------------------+ 3 rows in set (10.24 sec) --切换以后 mysql> select * from tbl_kenyon; +----+-----------+---------------------+ | id | cname | ctime | +----+-----------+---------------------+ | 1 | qiaofeng | 2017-05-23 10:50:43 | | 2 | murong | 2017-05-23 10:50:43 | | 3 | saodiseng | 2017-05-23 10:50:43 | +----+-----------+---------------------+ 3 rows in set (0.00 sec)可以发现经过投票，PD已经连到94上去了（93,92上都能看到），此时读取表数据很快
2017/05/23 17:59:12.151 server.rs:153: [INFO] TiKV is ready to serve 2017/05/23 17:59:12.517 raft.rs:846: [INFO] [region 2] 3 [term: 1487] received a MsgHeartbeat message with higher term from 7 [term: 1488] 2017/05/23 17:59:12.517 raft.rs:681: [INFO] [region 2] 3 became follower at term 1488 2017/05/23 17:59:12.525 server.rs:460: [INFO] resolve store 6 address ok, addr 172.26.11.94:20160 2017/05/23 17:59:13.517 apply.rs:621: [INFO] [region 2] 3 execute admin command cmd_type: CompactLog compact_log {compact_index: 6437 compact_term: 1488} at [term: 1488, index: 6439] 2017/05/23 17:59:13.644 raftlog_gc.rs:117: [INFO] [region 2] collected 225 log entries 2017/05/23 17:59:43.517 apply.rs:621: [INFO] [region 2] 3 execute admin command cmd_type: CompactLog compact_log {compact_index: 6498 compact_term: 1488} at [term: 1488, index: 6500] 2017/05/23 17:59:43.643 raftlog_gc.rs:117: [INFO] [region 2] collected 61 log entries这是因为92是Region中的leader，假如不是leader的tikv服务器受影响如93,94，数据因为默认做了三个副本（也可以配置5个或者7个副本），服务并不会受影响，但是在日志中会不停地告警
四、水平扩展
其实主要是以上组件模块的扩展，对于tidb来说，本身是无状态的，比较容易扩展，pd也可以部署成集群的模式，通过Haproxy、F5或者其他第三方软件来实现，比较难的Tikv的水平扩展。在每个TiKV的节点里，逻辑上划分了一个或多个store，每个store里又划了一个或多个Region，数据就是存放在Region里面，每个Region的默认值是64M，扩容的过程类似以下细胞分裂的过程，比传统的RDBMS采用的Sharding方式以及中间件模式要透明很多，也许以后市面上的诸多中间件日子要难过了。
1.添加pd
动态添加pd
pd-server --name=pd2 \
--client-urls="http://172.26.11.95:2379" \ --peer-urls="http://172.26.11.95:2380" \ --join="http://172.26.11.91:2379" --要加入的原pd集群 动态删除pd pd-ctl -u http://172.26.11.91:2379 >> member delete pd22.添加tikv
添加tikv，比较简单，直接注册一个新的tikv，剩下的数据迁移工作就交给pd，以下在91上新注册一个tikv tikv-server --pd="172.26.11.91:2379" \ --addr="172.26.11.91:20160" \ --data-dir=/data/tidb/tikv \ --log-file=/data/tidb/log/tikv.log & 查看store，新增了一个1001的store，另外也能看出当前的leader在93上面 [root@test05 ~]# pd-ctl -u http://172.26.11.91:2379 » store { "count": 4, "stores": [ { "store": { "id": 6, "address": "172.26.11.94:20160", "state": 0, "state_name": "Up" }, "status": { "store_id": 6, "capacity": "21 GB", "available": "21 GB", "leader_count": 0, "region_count": 1, "sending_snap_count": 0, "receiving_snap_count": 0, "applying_snap_count": 0, "is_busy": false, "start_ts": "2017-05-23T18:05:07+08:00", "last_heartbeat_ts": "2017-05-24T17:52:03.842239159+08:00", "uptime": "23h46m56.842239159s" } }, { "store": { "id": 1001, "address": "172.26.11.91:20160", "state": 0, "state_name": "Up" }, "status": { "store_id": 1001, "capacity": "21 GB", "available": "15 GB", "leader_count": 0, "region_count": 0, "sending_snap_count": 0, "receiving_snap_count": 0, "applying_snap_count": 0, "is_busy": false, "start_ts": "2017-05-24T17:50:12+08:00", "last_heartbeat_ts": "2017-05-24T17:52:03.290658649+08:00", "uptime": "1m51.290658649s" } }, { "store": { "id": 1, "address": "172.26.11.92:20160", "state": 0, "state_name": "Up" }, "status": { "store_id": 1, "capacity": "21 GB", "available": "21 GB", "leader_count": 0, "region_count": 1, "sending_snap_count": 0, "receiving_snap_count": 0, "applying_snap_count": 0, "is_busy": false, "start_ts": "2017-05-23T17:59:12+08:00", "last_heartbeat_ts": "2017-05-24T17:52:06.843194072+08:00", "uptime": "23h52m54.843194072s" } }, { "store": { "id": 4, "address": "172.26.11.93:20160", "state": 0, "state_name": "Up" }, "status": { "store_id": 4, "capacity": "21 GB", "available": "21 GB", "leader_count": 1, "region_count": 1, "sending_snap_count": 0, "receiving_snap_count": 0, "applying_snap_count": 0, "is_busy": false, "start_ts": "2017-05-23T17:22:48+08:00", "last_heartbeat_ts": "2017-05-24T17:52:09.766282426+08:00", "uptime": "24h29m21.766282426s" } } ] }3.删除tikv
查看1001的store状态是0，也就是up » store 1001 { "store": { "id": 1001, "address": "172.26.11.91:20160", "state": 0, "state_name": "Up" }, "status": { "store_id": 1001, "capacity": "21 GB", "available": "15 GB", "leader_count": 0, "region_count": 0, "sending_snap_count": 0, "receiving_snap_count": 0, "applying_snap_count": 0, "is_busy": false, "start_ts": "2017-05-24T17:50:12+08:00", "last_heartbeat_ts": "2017-05-24T17:58:57.490156968+08:00", "uptime": "8m45.490156968s" } } --删除过程，state=1表示正在下线 » store delete 1001 Success! » store 1001 { "store": { "id": 1001, "address": "172.26.11.91:20160", "state": 1, "state_name": "Offline" }, "status": { "store_id": 1001, "capacity": "21 GB", "available": "15 GB", "leader_count": 0, "region_count": 0, "sending_snap_count": 0, "receiving_snap_count": 0, "applying_snap_count": 0, "is_busy": false, "start_ts": "2017-05-24T17:50:12+08:00", "last_heartbeat_ts": "2017-05-24T17:59:17.690136502+08:00", "uptime": "9m5.690136502s" } } --state=2表示数据已经清理，可以关闭 » store 1001 { "store": { "id": 1001, "address": "172.26.11.91:20160", "state": 2, "state_name": "Tombstone" }, "status": { "store_id": 0, "capacity": "0 B", "available": "0 B", "leader_count": 0, "region_count": 0, "sending_snap_count": 0, "receiving_snap_count": 0, "applying_snap_count": 0, "is_busy": false, "start_ts": "1970-01-01T08:00:00+08:00", "last_heartbeat_ts": "0001-01-01T00:00:00Z", "uptime": "0s" } } » 五、总结：
1.维护相对很简单，官方文档是有中英文版本，资料更新相对及时，但是实际使用者提供的资料较少
2.Scale Out相比较传统的RDBMS方案上来看简化很多，特别是可以摒弃五花八门的中间件
3.从架构上来看，小数据量的性能应该一般，不建议使用，大数据量理论上会较好
4.期待GA版本