OceanBase 单机租户允许创建的最大分区数是多少?作者通过分区超限错误排查,计算出单机允许创建的最大分区数量。
作者:何文超,爱可生南区交付服务部 DBA 团队成员,主要负责 MySQL 故障处理,MySQL 高可用架构改造,OceanBase 相关技术支持。爱好足球,羽毛球。
爱可生开源社区出品,原创内容未经授权不得随意使用,转载请联系小编并注明来源。
本文共 1200 字,预计阅读需要 3 分钟。
背景
ERROR 1499 (HY000): Too many partitions (including subpartitions) were defined
创建表报错,虽然是内部错误,但是错误信息是指:创建了太多了分区。
[root@observer04 ~]# mysql -h10.186.64.125 -P2883 -uroot@wenchao_mysql#hwc_cluster:1682755171 -p"xxxx"
MySQL [lss]> CREATE TABLE `wms_order` (
`A1` varchar(100) CHARACTER SET utf8mb4 COLLATE utf8mb4_bin DEFAULT NULL COMMENT 'A1',
`A2` varchar(100) CHARACTER SET utf8mb4 COLLATE utf8mb4_bin DEFAULT NULL COMMENT 'A2',
`A3` varchar(100) CHARACTER SET utf8mb4 COLLATE utf8mb4_bin DEFAULT NULL COMMENT 'A3',
`A4` varchar(100) CHARACTER SET utf8mb4 COLLATE utf8mb4_bin DEFAULT NULL COMMENT 'A4',
`A5` varchar(100) CHARACTER SET utf8mb4 COLLATE utf8mb4_bin DEFAULT NULL COMMENT 'A5',
`A6` varchar(100) CHARACTER SET utf8mb4 COLLATE utf8mb4_bin DEFAULT NULL COMMENT 'A6',
`A7` varchar(100) CHARACTER SET utf8mb4 COLLATE utf8mb4_bin DEFAULT NULL COMMENT 'A7',
`A8` varchar(100) CHARACTER SET utf8mb4 COLLATE utf8mb4_bin DEFAULT NULL COMMENT 'A8',
`A9` varchar(100) CHARACTER SET utf8mb4 COLLATE utf8mb4_bin DEFAULT NULL COMMENT 'A9',
`A10` varchar(100) CHARACTER SET utf8mb4 COLLATE utf8mb4_bin DEFAULT NULL COMMENT 'A10'
) DEFAULT CHARSET = utf8mb4 ROW_FORMAT = DYNAMIC COMPRESSION = 'zstd_1.0' REPLICA_NUM = 3 BLOCK_SIZE = 16384 USE_BLOOM_FILTER = FALSE TABLET_SIZE = 134217728 PCTFREE = 0 COMMENT = '物流订单表'
MySQL [lss]> ERROR 1499 (HY000): Too many partitions (including subpartitions) were defined
接下来我们分析一下问题的原因。
排查
2.1 检查参数
- 检查每个 OBServer 上可以创建最大的分区数量,当前是 500000。
[root@observer04 ~]# mysql -h10.186.64.125 -P2883 -uroot@sys#hwc_cluster:1682755171 -p"xxxx" -A oceanBase
MySQL [oceanBase]> select * from __all_virtual_sys_parameter_stat where name like '%_max_partition_%';
+-------+----------+---------------+----------+-------------------------------+-----------+--------+--------------+---------------------------------------------+-------------+----------+---------------+---------+---------+-------------------+
| zone | svr_type | svr_ip | svr_port | name | data_type | value | value_strict | info | need_reboot | section | visible_level | scope | source | edit_level |
+-------+----------+---------------+----------+-------------------------------+-----------+--------+--------------+---------------------------------------------+-------------+----------+---------------+---------+---------+-------------------+
| zone1 | observer | 10.186.64.122 | 2882 | _max_partition_cnt_per_server | NULL | 500000 | NULL | specify max partition count on one observer | NULL | OBSERVER | NULL | CLUSTER | DEFAULT | DYNAMIC_EFFECTIVE |
+-------+----------+---------------+----------+-------------------------------+-----------+--------+--------------+---------------------------------------------+-------------+----------+---------------+---------+---------+-------------------+
- 检查当前分区数量的和,目前并没有超过这个限制(500000)。
MySQL [oceanBase]> select count(*) from v$partition;
+----------+
| count(*) |
+----------+
| 421485 |
+----------+
2.2 检查回收站
- 检查回收站是否开启?
[root@observer04 ~]# mysql -h10.186.64.125 -P2883 -uroot@wenchao_mysql#hwc_cluster:1682755171 -p"xxxx"
MySQL [lss]> show variables like '%recy%';
+---------------+-------+
| Variable_name | Value |
+---------------+-------+
| recyclebin | ON |
+---------------+-------+
1 row in set (0.01 sec)
- 检查回收站中是否存在未删除的分区表?
MySQL [lss]> show recyclebin;
+-----------------------------------------+---------------+-------+----------------------------+
| OBJECT_NAME | ORIGINAL_NAME | TYPE | CREATETIME |
+-----------------------------------------+---------------+-------+----------------------------+
| __recycle_$_1682755171_1689139725669688 | mytable_1 | TABLE | 2023-07-12 13:28:45.687379 |
| __recycle_$_1682755171_1689139737584112 | mytable_1 | TABLE | 2023-07-12 13:28:57.584660 |
| __recycle_$_1682755171_1689139750594392 | t1 | TABLE | 2023-07-12 13:29:10.594118 |
+-----------------------------------------+---------------+-------+----------------------------+
3 rows in set (0.01 sec)
如果存在,需要和业务侧沟通是否可以清理。回收站的表清理后,发现分区表数量减少,但是创建表依旧报错。
- 查看回收站中中对象保留天数。
MySQL [lss]> SHOW PARAMETERS LIKE 'recyclebin_object_expire_time'G;
*************************** 1. row ***************************
zone: zone1
svr_type: observer
svr_ip: 10.186.64.122
svr_port: 2882
name: recyclebin_object_expire_time
data_type: NULL
value: 0s
info: recyclebin object expire time, default 0 that means auto purge recyclebin off. Range: [0s, +∞)
section: ROOT_SERVICE
scope: CLUSTER
source: DEFAULT
edit_level: DYNAMIC_EFFECTIVE
1 row in set (0.02 sec)
配置项 recyclebin_object_expire_time
的取值说明如下:
- 当其值为 0s 时,表示关闭自动 Purge 回收站功能。
- 当其值不为 0s 时,表示回收一段时间前进入回收站的 Schema 对象。
2.3 检查租户内存
- 找到分区数最多的 10 个租户。
[root@observer04 ~]# mysql -h10.186.64.125 -P2883 -uroot@sys#hwc_cluster:1682755171 -p"xxxx"
// 1. 找到分区数最多的 10 个租户
SELECT t2.tenant_name,t2.tenant_id, t1.replica_count
FROM
(SELECT tenant_id, COUNT(*) AS replica_count
FROM __all_virtual_partition_info
GROUP BY tenant_id
ORDER BY replica_count DESC
LIMIT 10) t1
JOIN
(SELECT tenant_id, tenant_name
FROM __all_tenant) t2
ON t1.tenant_id=t2.tenant_id
ORDER BY replica_count DESC;
+-------------------+-----------+---------------+
| tenant_name | tenant_id | replica_count |
+-------------------+-----------+---------------+
| wenchao_mysql | 1100 | 107853 |
| wenchao_01 | 1088 | 99846 |
| wenchao_02 | 1104 | 15873 |
| wenchao_03 | 1 | 3867 |
| wenchao_04 | 1044 | 3270 |
| wenchao_05 | 1066 | 2811 |
| wenchao_06 | 1079 | 2658 |
| wenchao_07 | 1103 | 2103 |
| wenchao_08 | 1057 | 2040 |
| wenchao_09 | 1016 | 1950 |
+-------------------+-----------+---------------+
10 rows in set (0.13 sec)
- 查找租户有多少表。
select count(*),svr_Ip from __all_virtual_meta_table where tenant_id=1100 and role=1 group by svr_ip;
+----------+-------------+
| count(*) | svr_Ip |
+----------+-------------+
| 11921 |10.186.64.103|
| 11868 |10.186.64.104|
| 12013 |10.186.64.105|
+----------+-------------+
3 rows in set (0.35 sec)
- 计算租户需要扩容内存大小。
- 租户当前分区总数 num=107853/副本数
- 租户可用内存上限=(1-memstore_limit_percentage)*租户 unit 的内存大小 =(1-0.8)*24GB=4.8GB
- 单个副本分区所需总内存 partition_mem=128k*(107853/3)+max(1000,(107853/3)/10)*400k=5.75GB
注意:单个副本分区所需总内存 > 租户可用内存上限,租户所需内存超限,需要对租户内存进行扩容。
- 根据租户内存计算最大分区数量。
- 单机租户允许创建的最大分区数量=(max_memory-memstore_limit)/partition_mem_n
- 单机租户允许创建的最大分区数量=(24-24*0.8)/(5.75/(107853/3))=4.8/(5.75/(107853/3))=30011
- partition_mem_n:指的是单个分区所需总内存
临时处理方案: 扩容租户内存。
根源治理: 不可能无限扩内存;给出业务方合理的分区数量限制,建议业务侧合理使用分区表,制定合理的定期清理策略。
总结
根据上述计算得出:单机租户允许创建的最大分区数量为 30011,建议业务侧注意控制分区数量,以免超限,对业务造成影响。
更多技术文章,请访问:https://opensource.actionsky.com/
关于 SQLE
爱可生开源社区的 SQLE 是一款面向数据库使用者和管理者,支持多场景审核,支持标准化上线流程,原生支持 MySQL 审核且数据库类型可扩展的 SQL 审核工具。
SQLE 获取
类型 | 地址 |
---|---|
版本库 | https://github.com/actiontech/sqle |
文档 | https://actiontech.github.io/sqle-docs/ |
发布信息 | https://github.com/actiontech/sqle/releases |
数据审核插件开发文档 | https://actiontech.github.io/sqle-docs/docs/dev-manual/plugins/howtouse |