KingbaseES V8R3 备份恢复系列之 -- sys_rman备份过程分析
案例说明：
本案例通过对KingbaseES sys_rman物理备份过程的详细描述，有助于在执行sys_rman过程中发生故障的分析。
适用版本：
KingbaseES V8R3
一、sys_rman执行过程简介
1. 调用select sys_start_backup()开始备份，sys_start_backup()进行创建基础备份的准备工作，包括：
1)判断WAL归档是否已开启
如果没有开启,备份依然会进行，但在备份结束后会显示提醒信息：
NOTICE: WAL archiving is not enabled; you must ensure that all required WAL segments are copied through other means to complete the backup.
2)强制进入全页写模式
判断当前配置是否为全页写模式，如果当前full_page_writes设置为off，则强制更改为on
TEST=# show full_page_writes ;
full_page_writes
------------------
on
(1 row)
3）创建一个检查点
4）查看当前对应的wal（xlog）日志文件及lsn
select lsn::text as lsn,sys_catalog.sys_xlogfile_name(lsn)::text as wal_segment_name
from sys_catalog.sys_start_backup('sys_rman backup start at ' ||current_timestamp,false,false) as lsn;
2、执行系统层面的文件拷贝
3、调用SELECT * FROM sys_stop_backup()结束备份
1）执行wal（xlog）日志文件归档
2）归档完成后，备份结束
二、sys_rman执行备份分析
Tips：
可以通过以下两个参数，查看sys_rman执行的详细过程
1、查看sys_rman全备过程
执行备份初始化：
[kingbase@node101 bin]$ ./sys_rman -U SYSTEM -W 123456 -d TEST
-B /data/kingbase/bk/v8r3 -D /home/kingbase/cluster/HAR3/db/data init
执行备份：
[kingbase@node101 bin]$ ./sys_rman -U SYSTEM -W 123456 -d TEST -B /data/kingbase/bk/v8r3 -b full -D /home/kingbase/cluster/HAR3/db/data -v --progress backup
# 获取当前数据库配置参数
LOG: (query) SELECT current_setting($1)
LOG: (param:0) = block_size
LOG: (query) SELECT current_setting($1)
LOG: (param:0) = wal_block_size
LOG: (query) SELECT sys_is_in_recovery()
# 备份开始调用sys_backup_start()
LOG: ========================================
LOG: backup start
LOG: ----------------------------------------
# configuration
BACKUP_MODE=FULL
LOG: ----------------------------------------
LOG: backup destination is initialized
LOG: database backup start
LOG: (query) SELECT proname FROM sys_proc WHERE proname='pg_ptrack_clear'
LOG: (query) SELECT sys_start_backup($1, $2)
LOG: (param:0) = 2023-03-29 13:48:16 with sys_rman
LOG: (param:1) = true
# 获取当前wal日志文件及lsn
LOG: wait_archive_lsn() wait for lsn:3959423016 /data/kingbase/bk/v8r3/wal/0000000500000000000000EB
# 创建相关目录并开始copy文件
LOG: make directory "base"
LOG: make directory "base/1"
.......
LOG: make directory "sys_twophase"
LOG: make directory "sys_xlog"
WARNING: Start thread num:2917
LOG: (1/2917) base/14946/2830_vm
LOG: copied 0
Progress 0/2887LOG: (2/2917) base/14943/3348
LOG: copied 0
Progress 1/2887LOG: (3/2917) base/14943/3348_vm
LOG: copied 0
Progress 2/2887LOG: (4/2917) base/16392/14808
LOG: copied 0
Progress 3/2887LOG: (5/2917) base/14946/2830
LOG: copied 0
Progress 4/2887LOG: (6/2917) base/14946/2613_vm
LOG: copied 0
.......
LOG: (2888/2917) global/sys_internal.init
LOG: (2889/2917) kingbase.auto.conf
......
LOG: (2907/2917) sys_multixact/offsets/0000
# 备份结束
LOG: backup RS9PGG is valid
2、查看sys_rman增量备份过程
1、执行事务操作（产生增量数据）
PROD=# select count(*) from t1;
COUNT
----------
10010000
(1 row)
PROD=# create table t2 as select * from t1;
SELECT 10010000
PROD=# drop table t1;
NOTICE: view V_T1 depends on table T1
DROP TABLE
2、查看增量备份过程
如下图所示：最近全量备份及结束对应的LSN：
每次块的变化都包含了LSN的修改，那么也即是说，我们可以通过第一次备份开始时的全局LSN，以及当前需要备份的数据的page LSN来判断此页是否发生过修改。
[kingbase@node101 bin]$ ./sys_rman -U SYSTEM -W 123456 -d TEST -B /data/kingbase/bk/v8r3 -b page -D /home/kingbase/cluster/HAR3/db/data -v --progress backup
# 获取当前数据库配置参数
LOG: (query) SELECT current_setting($1)
LOG: (param:0) = block_size
LOG: (query) SELECT current_setting($1)
LOG: (param:0) = wal_block_size
LOG: (query) SELECT sys_is_in_recovery()
# 备份开始调用sys_backup_start()
LOG: ========================================
LOG: backup start
LOG: ----------------------------------------
# configuration
BACKUP_MODE=PAGE
LOG: ----------------------------------------
LOG: backup destination is initialized
LOG: database backup start
LOG: (query) SELECT proname FROM sys_proc WHERE proname='pg_ptrack_clear'
LOG: (query) SELECT sys_start_backup($1, $2)
LOG: (param:0) = 2023-03-29 14:07:05 with sys_rman
LOG: (param:1) = true
# 获取当前wal日志文件及lsn
LOG: wait_archive_lsn() wait for lsn:4731347992 /data/kingbase/bk/v8r3/wal/000000050000000100000019
LOG: backup only the page that there was of the update from LSN(0/EC000028)
LOG: (query) SET client_min_messages = warning;
LOG: (query) SELECT * FROM sys_switch_xlog()
LOG: wait_for_archive(): tli=5 lsn=1/1A0D34D0
LOG: wait_for_archive() wait for /home/kingbase/cluster/HAR3/db/data/sys_xlog/archive_status/00000005000000010000001A.ready
LOG: (query) SELECT txid_current();
LOG: get_xid():2232
LOG: wait_for_archive() .ready deleted in 1 try
LOG: extractPageMap
# 通过读取wal日志获取变化的数据块，执行增量备份
LOG: current_tli:5
LOG: prev_backup->start_lsn: 0/EC000028
LOG: current.start_lsn: 1/1A02A418
LOG: opening WAL segment "/data/kingbase/bk/v8r3/wal/0000000500000000000000EC"
LOG: opening WAL segment "/data/kingbase/bk/v8r3/wal/0000000500000000000000ED"
LOG: opening WAL segment "/data/kingbase/bk/v8r3/wal/0000000500000000000000EE"
LOG: opening WAL segment "/data/kingbase/bk/v8r3/wal/0000000500000000000000EF"
.......
LOG: opening WAL segment "/data/kingbase/bk/v8r3/wal/000000050000000100000018"
LOG: opening WAL segment "/data/kingbase/bk/v8r3/wal/000000050000000100000019"
LOG: opening WAL segment "/data/kingbase/bk/v8r3/wal/00000005000000010000001A"
LOG: make directory "base"
.......
# 创建相关目录并开始copy文件（只拷贝有变化块的数据文件）
LOG: make directory "sys_xlog"
WARNING: Start thread num:2916
LOG: (1/2916) base/14946/2830_vm
LOG: skip
LOG: (2/2916) base/14943/3348
LOG: skip
.....
LOG: (2615/2916) base/16392/2703
LOG: copied 8196
Progress 4/2886LOG: (2616/2916) base/16392/2696
LOG: copied 8196
Progress 5/2886LOG:
LOG: (2681/2916) base/16392/2662
LOG: copied 8196
LOG: (2686/2916) base/16392/3455
LOG: copied 8196
LOG: (2681/2916) base/16392/2662
LOG: copied 8196
LOG: (2686/2916) base/16392/3455
LOG: copied 8196
.......
LOG: (2916/2916) base/16392/41366
LOG: copied 443477364
Progress 20/2886
LOG: (query) SET client_min_messages = warning;
# 调用sys_stop_backup()结束备份
LOG: (query) SELECT * FROM sys_stop_backup()
# 执行wal日志归档
LOG: wait_for_archive(): tli=5 lsn=1/1DDE0AA0
LOG: wait_for_archive() wait for /home/kingbase/cluster/HAR3/db/data/sys_xlog/archive_status/00000005000000010000001D.ready
LOG: (query) SELECT txid_current();
LOG: get_xid():2233
LOG: wait_for_archive() .ready deleted in 0 try
# 备份结束
LOG: database backup completed(Backup: 443704210)
LOG: ========================================
LOG: all backup completed(read: 443704210 write: 443704210)
LOG: ========================================
INFO: validate: RS9QBT backup and archive log files by CRC
LOG: database files...
LOG: (2/2916) backup_label
LOG: (2427/2916) base/16392/1247
......
LOG: (2796/2916) base/16392/41366
LOG: (2885/2916) global/sys_control
LOG: (2892/2916) sys_clog/0000
LOG: (2900/2916) sys_logical/replorigin_checkpoint
LOG: backup RS9QBT is valid
3、查询增量备份copy的数据文件
1）如下图所示，被拷贝数据文件
2）查询数据文件对应的object
PROD=# select relname,relfilenode from sys_class where sys_relation_filepath(oid) like 'base/16392/41366' ;
RELNAME | RELFILENODE
---------+-------------
T2 | 41366
(1 row)
---如上所示，数据文件对应的t2表，是上次全备后新增加的object。
4、查看增量备份访问的wal日志信息
-rw------- 1 kingbase kingbase 16M Mar 29 13:48 0000000500000000000000EC
-rw------- 1 kingbase kingbase 16M Mar 29 14:06 0000000500000000000000ED
.......
-rw------- 1 kingbase kingbase 16M Mar 29 14:06 000000050000000100000017
-rw------- 1 kingbase kingbase 16M Mar 29 14:07 000000050000000100000018
-rw------- 1 kingbase kingbase 16M Mar 29 14:07 000000050000000100000019
-rw------- 1 kingbase kingbase 16M Mar 29 14:07 00000005000000010000001A
---以上wal日志，是在增量备份过程中读取的日志文件,从起始lsn所对应的wal日志开始。
三、查看和清理冗余备份
1、查看备份信息
[kingbase@node101 bin]$ ./sys_rman -U SYSTEM -W 123456 -d TEST -B /data/kingbase/bk/v8r3 show
==========================================================================================================
ID Recovery time Mode Current/Parent TLI Time Data start_lsn stop_lsn Status
==========================================================================================================
RS9PB9 2023-03-29 13:45:13 FULL 5 / 4 4s 525MB 0/EA000028 0/EA000130 OK
RS9KOV 2023-03-29 12:05:22 FULL 5 / 4 3s 525MB 0/E8000028 0/E80000F8 OK
RS9KI2 2023-03-29 12:01:19 FULL 5 / 4 5s 525MB 0/E6000028 0/E6000130 OK
2、清理冗余备份
[kingbase@node101 bin]$ ./sys_rman -U SYSTEM -W 123456 -d TEST -B /data/kingbase/bk/v8r3 delete RS9PB9
INFO: delete: RS9PB9 2023-03-29 13:45:13
[kingbase@node101 bin]$ ./sys_rman -U SYSTEM -W 123456 -d TEST -B /data/kingbase/bk/v8r3 delete RS9KOV
INFO: delete: RS9KOV 2023-03-29 12:05:22
[kingbase@node101 bin]$ ./sys_rman -U SYSTEM -W 123456 -d TEST -B /data/kingbase/bk/v8r3 delete RS9KI2
INFO: delete: RS9KI2 2023-03-29 12:01:19
[kingbase@node101 bin]$ ./sys_rman -U SYSTEM -W 123456 -d TEST -B /data/kingbase/bk/v8r3 show
==========================================================================================================
ID Recovery time Mode Current/Parent TLI Time Data start_lsn stop_lsn Status
==========================================================================================================
四、备份故障案例
如下所示：备份故障信息，是因为在第一次执行sys_rman备份，未做init的备份目录的初始化导致。
五、总结
通过对执行sys_rman物理备份的过程分析，可以看出，对于增量备份是通过分析wal日志获取到从上次全备以来发生变化的块；对于全备后，事务量比较大的数据库环境，执行增量备份可能效率会比较慢，需要读取大量的wal日志，来获取变化的块，然后拷贝对应的数据文件。
KINGBASE研究院
