142022.01
技术干货 | MySQL8.0 如何快速回收膨胀的UNDO表空间
2022.01.14背景介绍
项目选型中，在KVM(16c 16G ssd160G )的 Linux7.6系统上部署了MySQL MGR集群 (GreatSQL 8.0.25)。
使用 sysbench 创建了100仓数据，且针对表创建为 partition 表，进行连续12小时的稳定下压测，来评估对应的架构能支撑的业务并发数，以及最高的TPS/QPS是多少。
在使用256并发连续进行12小时压测后，发现节点的SSD磁盘空间使用率达到 95% 以上，当时第一时间查看log目录，发现log目录已经达到 100G+，以为是 binlog 设置的时间太长导致的 binlog 没有及时清理造成的，去清理 binlogbinlog 过期时间设置的 1800s，实际 binlog 和 MGR 的 relay-group 空间占用在11G左右而 du -sh * 查看到的日志文件大小时，发现其中undo大小1个是71G另一个4.1G，且MGR的3个节点的undo均是这个情况，急需释放空间。
但是MySQL8.0是否支持类似Oracle似的undo在线的替换来进行收缩呢？答案是肯定的，而且和Oracle有些类似。
Oracle/MySQL undo表空间设置自动扩展，如果业务上有跑批量或者大表的DML操作时，引起大事物或针对多张大表关联更新时间较长，可能短时间内会将undo"撑大"，Oracle可通过创建一个新的undo，通过在线替换的方式，将膨胀的undo使用drop删除以释放空间。
MySQL 8.0同样可以使用这种方式来处理，因大事物或长事物引起的undo过大占用空间较多的情况。
方法如下
1、添加新的undo文件undo003。mysql8.0中默认innodb_undo_tablespace为2个，不足2个时，不允许设置为inactive，且默认创建的undo受保护，不允许删除。
2、将膨胀的 undo 临时设置为inactive，以及 innodb_undo_log_truncate=on，自动 truncate 释放膨胀的undo空间。
3、重新将释放空间之后的undo设置为active，可重新上线使用。
具体操作如下
[greatdb@mysql ~]$ mysql -ugreatsql -pgreatsql -h172.16.130.15 -P3307mysql: [Warning] Using a password on the command line interface can be insecure.Welcome to the MySQL monitor. Commands end with ; or \g.Your MySQL connection id is 74Server version: 8.0.25-15 GreatSQL, Release 15, Revision c7feae175e0Copyright (c) 2000, 2019, Oracle and/or its affiliates. All rights reserved.Oracle is a registered trademark of Oracle Corporation and/or itsaffiliates. Other names may be trademarks of their respectiveowners.Type 'help;' or '\h' for help. Type '\c' to clear the current input statement.mysql[(none)]> show variables like '%undo%';+--------------------------+-----------------------------+| Variable_name | Value |+--------------------------+-----------------------------+| innodb_max_undo_log_size | 4294967296 || innodb_undo_directory | /app/dbdata/sqlnode3306/log || innodb_undo_log_encrypt | OFF || innodb_undo_log_truncate | ON || innodb_undo_tablespaces | 2 |+--------------------------+-----------------------------+5 rows in set (0.01 sec)
1、查看undo大小
mysql[(none)]> system du -sh /app/dbdata/datanode3307/log/undo* 4.1G /app/dbdata/datanode3307/log/undo_001 71G /app/dbdata/datanode3307/log/undo_002 -----12小时连续稳定性压测，导致节点undo过大，达到71G
2、添加新的undo表空间undo003。系统默认是2个undo，大小设置4G
mysql[(none)]> mysql[(none)]> create undo tablespace undo003 add datafile '/app/dbdata/datanode3307/log/undo003.ibu'; Query OK, 0 rows affected (0.21 sec) 注意：创建添加新的undo必须以.ibu结尾，否则触发如下错误提示 mysql[(none)]> create undo tablespace undo003 add datafile '/app/dbdata/datanode3307/log/undo_003.' ; ERROR 3121 (HY000): The ADD DATAFILE filepath must end with '.ibu'.
3、查看系统中的undo表空间信息，如下：
mysql[(none)]> select * from information_schema.INNODB_TABLESPACES where name like '%undo%'; +------------+-----------------+------+------------+-----------+---------------+------------+---------------+-------------+----------------+-----------------+----------------+---------------+------------+--------+ | SPACE | NAME | FLAG | ROW_FORMAT | PAGE_SIZE | ZIP_PAGE_SIZE | SPACE_TYPE | FS_BLOCK_SIZE | FILE_SIZE | ALLOCATED_SIZE | AUTOEXTEND_SIZE | SERVER_VERSION | SPACE_VERSION | ENCRYPTION | STATE | +------------+-----------------+------+------------+-----------+---------------+------------+---------------+-------------+----------------+-----------------+----------------+---------------+------------+--------+ | 4294967279 | innodb_undo_001 | 0 | Undo | 16384 | 0 | Undo | 4096 | 4311744512 | 4311764992 | 0 | 8.0.25 | 1 | N | active | | 4294967278 | innodb_undo_002 | 0 | Undo | 16384 | 0 | Undo | 4096 | 76067897344 | 76068229120 | 0 | 8.0.25 | 1 | N | active | | 4294967277 | undo003 | 0 | Undo | 16384 | 0 | Undo | 4096 | 16777216 | 16777216 | 0 | 8.0.25 | 1 | N | active | +------------+-----------------+------+------------+-----------+---------------+------------+---------------+-------------+----------------+-----------------+----------------+---------------+------------+--------+ 3 rows in set (0.03 sec)
4、查看到上述视图中 innodb_undo_002 大小达到76067897344 （约71G）其状态state为active。手动将其设置为 inactive，使其自动触发 innodb_undo_log_truncate 回收。
mysql[(none)]> alter undo tablespace innodb_undo_002 set inactive; Query OK, 0 rows affected (0.00 sec)
5、查看对应视图如下
mysql[(none)]> select * from information_schema.INNODB_TABLESPACES where name like '%undo%'; +------------+-----------------+------+------------+-----------+---------------+------------+---------------+------------+----------------+-----------------+----------------+---------------+------------+--------+ | SPACE | NAME | FLAG | ROW_FORMAT | PAGE_SIZE | ZIP_PAGE_SIZE | SPACE_TYPE | FS_BLOCK_SIZE | FILE_SIZE | ALLOCATED_SIZE | AUTOEXTEND_SIZE | SERVER_VERSION | SPACE_VERSION | ENCRYPTION | STATE | +------------+-----------------+------+------------+-----------+---------------+------------+---------------+------------+----------------+-----------------+----------------+---------------+------------+--------+ | 4294967279 | innodb_undo_001 | 0 | Undo | 16384 | 0 | Undo | 4096 | 4311744512 | 4311764992 | 0 | 8.0.25 | 1 | N | active | | 4294967151 | innodb_undo_002 | 0 | Undo | 16384 | 0 | Undo | 4096 | 16777216 | 2179072 | 0 | 8.0.25 | 1 | N | empty | | 4294967277 | undo003 | 0 | Undo | 16384 | 0 | Undo | 4096 | 16777216 | 16777216 | 0 | 8.0.25 | 1 | N | active | +------------+-----------------+------+------------+-----------+---------------+------------+---------------+------------+----------------+-----------------+----------------+---------------+------------+--------+ 3 rows in set (0.01 sec)
此时可以查看对应操作系统目录中的 undo_002大小，innodb_undo_002 FILE_SIZE 16777216 默认大小 2179072 ，STATE 为 empty
mysql[(none)]> system du -sh /app/dbdata/datanode3307/log/undo* 4.1G /app/dbdata/datanode3307/log/undo_001 2.1M /app/dbdata/datanode3307/log/undo_002 16M /app/dbdata/datanode3307/log/undo003.ibu
6、重新将其设置为active状态
mysql[(none)]> alter undo tablespace innodb_undo_002 set active; Query OK, 0 rows affected (0.01 sec)
mysql[(none)]> select * from information_schema.INNODB_TABLESPACES where name like '%undo%'; +------------+-----------------+------+------------+-----------+---------------+------------+---------------+------------+----------------+-----------------+----------------+---------------+------------+--------+ | SPACE | NAME | FLAG | ROW_FORMAT | PAGE_SIZE | ZIP_PAGE_SIZE | SPACE_TYPE | FS_BLOCK_SIZE | FILE_SIZE | ALLOCATED_SIZE | AUTOEXTEND_SIZE | SERVER_VERSION | SPACE_VERSION | ENCRYPTION | STATE | +------------+-----------------+------+------------+-----------+---------------+------------+---------------+------------+----------------+-----------------+----------------+---------------+------------+--------+ | 4294967279 | innodb_undo_001 | 0 | Undo | 16384 | 0 | Undo | 4096 | 4311744512 | 4311764992 | 0 | 8.0.25 | 1 | N | active | | 4294967151 | innodb_undo_002 | 0 | Undo | 16384 | 0 | Undo | 4096 | 16777216 | 2195456 | 0 | 8.0.25 | 1 | N | active | | 4294967277 | undo003 | 0 | Undo | 16384 | 0 | Undo | 4096 | 16777216 | 16777216 | 0 | 8.0.25 | 1 | N | active | +------------+-----------------+------+------------+-----------+---------------+------------+---------------+------------+----------------+-----------------+----------------+---------------+------------+--------+ 3 rows in set (0.01 sec)
7、有人说，为什么不能直接针对膨胀的undo设置为inactive，系统默认创建的undo表空间默认2个，处于active小于2个时，会有如下提示：
mysql[(none)]> mysql[(none)]> show variables like 'innodb_undo_tablespaces'; +--------------------------+-----------------------------+ | Variable_name | Value | +--------------------------+-----------------------------+ | innodb_undo_tablespaces | 2 | +--------------------------+-----------------------------+ 5 rows in set (0.01 sec) mysql[(none)]> alter undo tablespace innodb_undo_002 set inactive; ERROR 3655 (HY000): Cannot set innodb_undo_002 inactive since there would be less than 2 undo tablespaces left active. mysql[(none)]>
8、新创建添加的可以正常设置为inactive之后，使用drop方式删除，如下：
mysql[(none)]> alter undo tablespace undo003 set inactive; Query OK, 0 rows affected (0.00 sec)
mysql[(none)]> drop undo tablespace undo003; Query OK, 0 rows affected (0.01 sec)
mysql[(none)]> select * from information_schema.INNODB_TABLESPACES where name like '%undo%'; +------------+-----------------+------+------------+-----------+---------------+------------+---------------+------------+----------------+-----------------+----------------+---------------+------------+--------+ | SPACE | NAME | FLAG | ROW_FORMAT | PAGE_SIZE | ZIP_PAGE_SIZE | SPACE_TYPE | FS_BLOCK_SIZE | FILE_SIZE | ALLOCATED_SIZE | AUTOEXTEND_SIZE | SERVER_VERSION | SPACE_VERSION | ENCRYPTION | STATE | +------------+-----------------+------+------------+-----------+---------------+------------+---------------+------------+----------------+-----------------+----------------+---------------+------------+--------+ | 4294967279 | innodb_undo_001 | 0 | Undo | 16384 | 0 | Undo | 4096 | 4311744512 | 4311764992 | 0 | 8.0.25 | 1 | N | active | | 4294967151 | innodb_undo_002 | 0 | Undo | 16384 | 0 | Undo | 4096 | 16777216 | 2244608 | 0 | 8.0.25 | 1 | N | active | +------------+-----------------+------+------------+-----------+---------------+------------+---------------+------------+----------------+-----------------+----------------+---------------+------------+--------+ 2 rows in set (0.01 sec)
总结
通过以上操作我们就可以针对在undo遇到大事务并持续增长的情况下，通过新增临时undo，手动释放系统默认的2个undo表空间大小。
当然，截断 UNDO 表空间文件对数据库性能有一定影响，尽量在相对空闲时间进行。
当UNDO表空间被截断时，UNDO表空间中的回滚段将被停用。其他UNDO表空间中的活动回滚段负责整个系统负载，这可能会导致性能略有下降。性能受影响的程度取决于诸多因素:
1、UNDO表空间的数量
2、UNDO记录日志的数据量
3、UNDO表空间大小
4、磁盘I/O系统的速度
5、现有长期运行的事务
因此，避免潜在性能影响的最简单方法:
1、就是通过create undo tablespace undo_XXX add datafile '/path/undo_xxx.ibu';多添加几个UNDO表空间；
2、如果条件允许，磁盘上采用高性能的SSD来存储数据，存储REDO、UNDO等。
引起UNDO过度膨胀的原因大多是因为基础数据量大，业务并发高，表关联操作较频繁，出现大且长的事物操作，导致UNDO一直处于active状态，不能及时释放回滚段等。
大事物引起的问题由来已久，即使我们能规避99%的大事物，但实际业务遇到那1%的大事物刚性需求发过来时，还要MySQL各种场景、各种架构和业务层好好磨合。
Enjoy GreatSQL :)
