微信支付用的数据库开源了

2020-01-02 14:36:38

腾讯TBase是一款腾讯自研高性能HTAP数据库，提供高性能的OLTP和OLAP能力，同时保证可扩展全局一致性分布式事务（ACID），为用户提供高一致性的分布式数据库服务和高性能的数据仓库服务。一方面解决了传统数据库扩展不足、数据sharding之后数据库事务的严格一致性难题、数据安全、跨地域容灾等问题，同时具备了高性能事务处理、数据治理、混合负载支持等能力。

在OLTP方面，TBase采用MVCC+全局时钟+2PC+SSI的方式来实现全局一致性分布式事务，同时引入大量性能优化的设计来减少全局事务带来的开销。在小规模集群上，TBase能够提供超过300万TPMTotal的事务处理吞吐量（工业界标准TPCC测试集）。

交易毫秒内完成

TBase已经覆盖多个行业的标杆用户，其中对内支持了微信广告、微信支付、腾讯地图等海量数据业务，一笔交易毫秒内即可完成，支撑了微信支付50倍的交易增长。

TBase是一个提供写可靠性，多主节点数据同步的关系数据库集群平台。你可以将TBase配置一台或者多台主机上，TBase数据存储在多台物理主机上面。数据表的存储有两种方式，分别是distributed或者replicated ，当向TBase发送查询 SQL时，TBase会自动向数据节点发出查询语句并获取终结果。

TBase采用分布式集群架构（如下图），该架构分布式为无共享(share nothing)模式，节点之间相应独立，各自处理自己的数据，处理后的结果可能向上层汇总或在节点间流转，各处理单元之间通过网络协议进行通信，并行处理和扩展能力更好，这也意味着只需要简单的x86服务器就可以部署TBase数据库集群。

下面简单解读一下TBase的三大模块：

Coordinator：协调节点（简称CN）

业务访问入口，负责数据的分发和查询规划，多个节点位置对等，每个节点都提供相同的数据库视图；在功能上CN上只存储系统的全局元数据，并不存储实际的业务数据。
Datanode：数据节点（简称DN）

每个节点还存储业务数据的分片在功能上，DN节点负责完成执行协调节点分发的执行请求。
GTM:全局事务管理器(Global Transaction Manager)

负责管理集群事务信息，同时管理集群的全局对象，比如序列等。

接下来，让我们来看看如何从源码开始，完成到TBase集群环境的搭建。

TBase源码编译安装

1. 创建tbase用户

注意：所有需要安装TBase集群的机器上都需要创建

mkdir /data useradd -d /data/tbase tbase

2. 源码获取

git clone https://github.com/Tencent/TBase

3. 源码编译

cd ${SOURCECODE_PATH}
rm -rf ${INSTALL_PATH}/tbase_bin_v2.0
chmod +x configure*
./configure --prefix=${INSTALL_PATH}/tbase_bin_v2.0  --enable-user-switch --with-openssl  --with-ossp-uuid CFLAGS=-g
make clean
make -sj
make install
chmod +x contrib/pgxc_ctl/make_signature
cd contrib
make -sj
make install

本文的使用环境中，上述两个参数如下

${SOURCECODE_PATH}=/data/tbase/TBase-master

${INSTALL_PATH}=/data/tbase/install

4. 集群安装

4.1）集群规划

下面以两台服务器上搭建1GTM主，1GTM备，2CN主（CN主之间对等，因此无需备CN），2DN主，2DN备的集群，该集群为具备容灾能力的小配置

机器1：10.215.147.158 机器2：10.240.138.159

集群规划如下：

4.2）机器间的ssh互信配置

参考Linux ssh互信配置

4.3）环境变量配置

集群所有机器都需要配置

[tbase@TENCENT64 ~]$ vim ~/.bashrc
export TBASE_HOME=/data/tbase/install/tbase_bin_v2.0
export PATH=$TBASE_HOME/bin:$PATH
export LD_LIBRARY_PATH=$TBASE_HOME/lib:${LD_LIBRARY_PATH}

以上，已经配置好了所需要基础环境，可以进入到集群初始化阶段，为了方便用户，TBase提供了专用的配置和操作工具：pgxc_ctl来协助用户快速搭建并管理集群，首先需要将前文所述的节点的ip，端口，目录写入到配置文件 pgxc_ctl.conf 中。

4.4）初始化pgxc_ctl.conf文件

[tbase@TENCENT64 ~]$ mkdir /data/tbase/pgxc_ctl
[tbase@TENCENT64 ~]$ cd /data/tbase/pgxc_ctl
[tbase@TENCENT64 ~/pgxc_ctl]$ vim pgxc_ctl.conf

如下，是结合上文描述的IP，端口，数据库目录，二进制目录等规划来写的pgxc_ctl.conf文件。具体实践中只需按照自己的实际情况配置好即可.

#!/bin/bash

pgxcInstallDir=/data/tbase/install/tbase_bin_v2.0
pgxcOwner=tbase
defaultDatabase=postgres
pgxcUser=$pgxcOwner
tmpDir=/tmp
localTmpDir=$tmpDir
configBackup=n
configBackupHost=pgxc-linker
configBackupDir=$HOME/pgxc
configBackupFile=pgxc_ctl.bak

#---- GTM ----------
gtmName=gtm
gtmMasterServer=10.215.147.158
gtmMasterPort=50001
gtmMasterDir=/data/tbase/data/gtm
gtmExtraConfig=none
gtmMasterSpecificExtraConfig=none
gtmSlave=y
gtmSlaveServer=10.240.138.159
gtmSlavePort=50001
gtmSlaveDir=/data/tbase/data/gtm
gtmSlaveSpecificExtraConfig=none

#---- Coordinators -------
coordMasterDir=/data/tbase/data/coord
coordMasterDir=/data/tbase/data/coord
coordArchLogDir=/data/tbase/data/coord_archlog

coordNames=(cn001 cn002 )
coordPorts=(30004 30004 )
poolerPorts=(31110 31110 )
coordPgHbaEntries=(0.0.0.0/0)
coordMasterServers=(10.215.147.158 10.240.138.159)
coordMasterDirs=($coordMasterDir $coordMasterDir)
coordMaxWALsernder=2
coordMaxWALSenders=($coordMaxWALsernder $coordMaxWALsernder )
coordSlave=n
coordSlaveSync=n
coordArchLogDirs=($coordArchLogDir $coordArchLogDir)

coordExtraConfig=coordExtraConfig
cat > $coordExtraConfig <<EOF
#================================================
# Added to all the coordinator postgresql.conf
# Original: $coordExtraConfig

include_if_exists = '/data/tbase/global/global_tbase.conf'

wal_level = replication
wal_keep_segments = 256
max_wal_senders = 4
archive_mode = on
archive_timeout = 1800
archive_command = 'echo 0'
log_truncate_on_rotation = on
log_filename = 'postgresql-%M.log'
log_rotation_age = 4h
log_rotation_size = 100MB
hot_standby = on
wal_sender_timeout = 30min
wal_receiver_timeout = 30min
shared_buffers = 1024MB
max_pool_size = 2000
log_statement = 'ddl'
log_destination = 'csvlog'
logging_collector = on
log_directory = 'pg_log'
listen_addresses = '*'
max_connections = 2000

EOF

coordSpecificExtraConfig=(none none)
coordExtraPgHba=coordExtraPgHba
cat > $coordExtraPgHba <<EOF

local   all             all                                     trust
host    all             all             0.0.0.0/0               trust
host    replication     all             0.0.0.0/0               trust
host    all             all             ::1/128                 trust
host    replication     all             ::1/128                 trust

EOF

coordSpecificExtraPgHba=(none none)
coordAdditionalSlaves=n
cad1_Sync=n

#---- Datanodes ---------------------
dn1MstrDir=/data/tbase/data/dn001
dn2MstrDir=/data/tbase/data/dn002
dn1SlvDir=/data/tbase/data/dn001
dn2SlvDir=/data/tbase/data/dn002
dn1ALDir=/data/tbase/data/datanode_archlog
dn2ALDir=/data/tbase/data/datanode_archlog

primaryDatanode=dn001
datanodeNames=(dn001 dn002)
datanodePorts=(40004 40004)
datanodePoolerPorts=(41110 41110)
datanodePgHbaEntries=(0.0.0.0/0)
datanodeMasterServers=(10.215.147.158 10.240.138.159)
datanodeMasterDirs=($dn1MstrDir $dn2MstrDir)
dnWALSndr=4
datanodeMaxWALSenders=($dnWALSndr $dnWALSndr)

datanodeSlave=y
datanodeSlaveServers=(10.240.138.159 10.215.147.158)
datanodeSlavePorts=(50004 540004)
datanodeSlavePoolerPorts=(51110 51110)
datanodeSlaveSync=n
datanodeSlaveDirs=($dn1SlvDir $dn2SlvDir)
datanodeArchLogDirs=($dn1ALDir/dn001 $dn2ALDir/dn002)

datanodeExtraConfig=datanodeExtraConfig
cat > $datanodeExtraConfig <<EOF
#================================================
# Added to all the coordinator postgresql.conf
# Original: $datanodeExtraConfig

include_if_exists = '/data/tbase/global/global_tbase.conf'
listen_addresses = '*'
wal_level = replication
wal_keep_segments = 256
max_wal_senders = 4
archive_mode = on
archive_timeout = 1800
archive_command = 'echo 0'
log_directory = 'pg_log'
logging_collector = on
log_truncate_on_rotation = on
log_filename = 'postgresql-%M.log'
log_rotation_age = 4h
log_rotation_size = 100MB
hot_standby = on
wal_sender_timeout = 30min
wal_receiver_timeout = 30min
shared_buffers = 1024MB
max_connections = 4000
max_pool_size = 4000
log_statement = 'ddl'
log_destination = 'csvlog'
wal_buffers = 1GB

EOF

datanodeSpecificExtraConfig=(none none)
datanodeExtraPgHba=datanodeExtraPgHba
cat > $datanodeExtraPgHba <<EOF

local   all             all                                     trust
host    all             all             0.0.0.0/0               trust
host    replication     all             0.0.0.0/0               trust
host    all             all             ::1/128                 trust
host    replication     all             ::1/128                 trust

EOF

datanodeSpecificExtraPgHba=(none none)

datanodeAdditionalSlaves=n
walArchive=n

4.5）分发二进制包

在一个节点配置好配置文件后，需要预先将二进制包部署到所有节点所在的机器上，这个可以使用pgxc_ctl工具，执行deploy all命令来完成。

[tbase@TENCENT64 ~/pgxc_ctl]$ pgxc_ctl /usr/bin/bash Installing pgxc_ctl_bash script as /data/tbase/pgxc_ctl/pgxc_ctl_bash. Installing pgxc_ctl_bash script as /data/tbase/pgxc_ctl/pgxc_ctl_bash. Reading configuration using /data/tbase/pgxc_ctl/pgxc_ctl_bash --home /data/tbase/pgxc_ctl --configuration /data/tbase/pgxc_ctl/pgxc_ctl.conf Finished reading configuration. ******** PGXC_CTL START *************** Current directory: /data/tbase/pgxc_ctl PGXC deploy all Deploying Postgres-XL components to all the target servers. Prepare tarball to deploy ... Deploying to the server 10.215.147.158. Deploying to the server 10.240.138.159. Deployment done. 登录到所有节点，check二进制包是否分发OK [tbase@TENCENT64 ~/install]$ ls /data/tbase/install/tbase_bin_v2.0 bin include lib share

4.6）执行init all命令，完成集群初始化命令

[tbase@TENCENT64 ~]$ pgxc_ctl /usr/bin/bash Installing pgxc_ctl_bash script as /data/tbase/pgxc_ctl/pgxc_ctl_bash. Installing pgxc_ctl_bash script as /data/tbase/pgxc_ctl/pgxc_ctl_bash. Reading configuration using /data/tbase/pgxc_ctl/pgxc_ctl_bash --home /data/tbase/pgxc_ctl --configuration /data/tbase/pgxc_ctl/pgxc_ctl.conf Finished reading configuration. ******** PGXC_CTL START *************** Current directory: /data/tbase/pgxc_ctl PGXC init all Initialize GTM master .... .... Initialize datanode slave dn001 Initialize datanode slave dn002 mkdir: cannot create directory '/data1/tbase': Permission denied chmod: cannot access '/data1/tbase/data/dn001': No such file or directory pg_ctl: directory "/data1/tbase/data/dn001" does not exist pg_basebackup: could not create directory "/data1/tbase": Permission denied

4.7）安装错误处理

一般init集群出错，终端会打印出错误日志，通过查看错误原因，更改配置即可，或者可以通过/data/tbase/pgxc_ctl/pgxc_log路径下的错误日志查看错误，排查配置文件的错误

[tbase@TENCENT64 ~]$ ll ~/pgxc_ctl/pgxc_log/ total 184 -rw-rw-r-- 1 tbase tbase 81123 Nov 13 17:22 14105_pgxc_ctl.log -rw-rw-r-- 1 tbase tbase 2861 Nov 13 17:58 15762_pgxc_ctl.log -rw-rw-r-- 1 tbase tbase 14823 Nov 14 07:59 16671_pgxc_ctl.log -rw-rw-r-- 1 tbase tbase 2721 Nov 13 16:52 18891_pgxc_ctl.log -rw-rw-r-- 1 tbase tbase 1409 Nov 13 16:20 22603_pgxc_ctl.log -rw-rw-r-- 1 tbase tbase 60043 Nov 13 16:33 28932_pgxc_ctl.log -rw-rw-r-- 1 tbase tbase 15671 Nov 14 07:57 6849_pgxc_ctl.log

通过运行 pgxc_ctl 工具，执行clean all命令删除已经初始化的文件，修改pgxc_ctl.conf文件，重新执行init all命令重新发起初始化。

[tbase@TENCENT64 ~]$ pgxc_ctl /usr/bin/bash Installing pgxc_ctl_bash script as /data/tbase/pgxc_ctl/pgxc_ctl_bash. Installing pgxc_ctl_bash script as /data/tbase/pgxc_ctl/pgxc_ctl_bash. Reading configuration using /data/tbase/pgxc_ctl/pgxc_ctl_bash --home /data/tbase/pgxc_ctl --configuration /data/tbase/pgxc_ctl/pgxc_ctl.conf Finished reading configuration. ******** PGXC_CTL START *************** Current directory: /data/tbase/pgxc_ctl PGXC clean all [tbase@TENCENT64 ~]$ pgxc_ctl /usr/bin/bash Installing pgxc_ctl_bash script as /data/tbase/pgxc_ctl/pgxc_ctl_bash. Installing pgxc_ctl_bash script as /data/tbase/pgxc_ctl/pgxc_ctl_bash. Reading configuration using /data/tbase/pgxc_ctl/pgxc_ctl_bash --home /data/tbase/pgxc_ctl --configuration /data/tbase/pgxc_ctl/pgxc_ctl.conf Finished reading configuration. ******** PGXC_CTL START *************** Current directory: /data/tbase/pgxc_ctl PGXC init all Initialize GTM master EXECUTE DIRECT ON (dn002) 'ALTER NODE dn002 WITH (TYPE=''datanode'', HOST=''10.240.138.159'', PORT=40004, PREFERRED)'; EXECUTE DIRECT EXECUTE DIRECT ON (dn002) 'SELECT pgxc_pool_reload()'; pgxc_pool_reload ------------------ t (1 row) Done.

5. 查看集群状态
当发现上面的输出时，集群已经OK，另外也可以通过pgxc_ctl工具的monitor all命令来查看集群状态

[tbase@TENCENT64 ~/pgxc_ctl]$ pgxc_ctl /usr/bin/bash Installing pgxc_ctl_bash script as /data/tbase/pgxc_ctl/pgxc_ctl_bash. Installing pgxc_ctl_bash script as /data/tbase/pgxc_ctl/pgxc_ctl_bash. Reading configuration using /data/tbase/pgxc_ctl/pgxc_ctl_bash --home /data/tbase/pgxc_ctl --configuration /data/tbase/pgxc_ctl/pgxc_ctl.conf Finished reading configuration. ******** PGXC_CTL START *************** Current directory: /data/tbase/pgxc_ctl PGXC monitor all Running: gtm master Not running: gtm slave Running: coordinator master cn001 Running: coordinator master cn002 Running: datanode master dn001 Running: datanode slave dn001 Running: datanode master dn002 Not running: datanode slave dn002

一般的如果配置的不是强同步模式，gtm salve，dn slave的故障不会影响访问。

6. 集群访问
访问TBase集群和访问单机的PostgreSQL基本上无差别，我们可以通过任意一个CN访问数据库集群：例如通过连接CN节点select pgxc_node表即可查看集群的拓扑结构（当前的配置下备机不会展示在pgxc_node中），在Linux命令行下通过psql访问的具体示例如下

[tbase@TENCENT64 ~/pgxc_ctl]$ psql -h 10.215.147.158 -p 30004 -d postgres -U tbase
psql (PostgreSQL 10.0 TBase V2)
Type "help" for help.

postgres=# \d
Did not find any relations.
postgres=# select * from pgxc_node;
 node_name | node_type | node_port |   node_host    | nodeis_primary | nodeis_preferred |  node_id   | node_cluster_name
-----------+-----------+-----------+----------------+----------------+------------------+------------+-------------------
 gtm       | G         |     50001 | 10.215.147.158 | t              | f                |  428125959 | tbase_cluster
 cn001     | C         |     30004 | 10.215.147.158 | f              | f                | -264077367 | tbase_cluster
 cn002     | C         |     30004 | 10.240.138.159 | f              | f                | -674870440 | tbase_cluster
 dn001     | D         |     40004 | 10.215.147.158 | t              | t                | 2142761564 | tbase_cluster
 dn002     | D         |     40004 | 10.240.138.159 | f              | f                |  -17499968 | tbase_cluster
(5 rows)

6.1) 使用数据库前需要创建default group以及sharding

TBase使用datanode group来增加节点的管理灵活度，要求有一个default group才能使用，因此需要预先创建；一般情况下，会将节点的所有datanode节点加入到default group里另外一方面，TBase的数据分布为了增加灵活度，加了中间逻辑层来维护数据记录到物理节点的映射，我们叫sharding，所以需要预先创建sharding，命令如下：

postgres=# create default node group default_group  with (dn001,dn002);
CREATE NODE GROUP
postgres=# create sharding group to group default_group;
CREATE SHARDING GROUP

6.2) 创建数据库，用户，创建表，增删查改等操作

至此，就可以跟使用单机数据库一样来访问数据库集群了

postgres=# create database test;
CREATE DATABASE
postgres=# create user test with password 'test';
CREATE ROLE
postgres=# alter database test owner to test;
ALTER DATABASE
postgres=# \c test test
You are now connected to database "test" as user "test".
test=> create table foo(id bigint, str text) distribute by shard(id);
CREATE TABLE
test=> insert into foo values(1, 'tencent'), (2, 'shenzhen');
COPY 2
test=> select * from foo;
 id |   str
----+----------
  1 | tencent
  2 | shenzhen
(2 rows)

7. 停止集群
通过pgxc_ctl工具的stop all命令来停止集群，stop all 后面可以加上参数 -m fast或者是-m immediate来决定如何停止各个节点。

PGXC stop all -m fast Stopping all the coordinator masters. Stopping coordinator master cn001. Stopping coordinator master cn002. Done. Stopping all the datanode slaves. Stopping datanode slave dn001. Stopping datanode slave dn002. pg_ctl: PID file "/data/tbase/data/dn002/postmaster.pid" does not exist Is server running? Stopping all the datanode masters. Stopping datanode master dn001. Stopping datanode master dn002. Done. Stop GTM slave waiting for server to shut down..... done server stopped Stop GTM master waiting for server to shut down.... done server stopped PGXC monitor all Not running: gtm master Not running: gtm slave Not running: coordinator master cn001 Not running: coordinator master cn002 Not running: datanode master dn001 Not running: datanode slave dn001 Not running: datanode master dn002 Not running: datanode slave dn002

8. 启动集群
通过pgxc_ctl工具的start all命令来启动集群

[tbase@TENCENT64 ~]$ pgxc_ctl /usr/bin/bash Installing pgxc_ctl_bash script as /data/tbase/pgxc_ctl/pgxc_ctl_bash. Installing pgxc_ctl_bash script as /data/tbase/pgxc_ctl/pgxc_ctl_bash. Reading configuration using /data/tbase/pgxc_ctl/pgxc_ctl_bash --home /data/tbase/pgxc_ctl --configuration /data/tbase/pgxc_ctl/pgxc_ctl.conf Finished reading configuration. ******** PGXC_CTL START *************** Current directory: /data/tbase/pgxc_ctl PGXC start all