OceanBase 备份清理任务卡死之谜

正常情况下，会有一个 OceanBase 备份文件清理任务定期执行。备份文件超过了预设数量，猜测可能是文件清理任务出现问题。

> 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。 > >本文约 1500 字，预计阅读需要 5 分钟。

问题背景

客户 OceanBase 集群的备份策略是每天全备，备份文件最近保留 4 天。经过观察，实际保留的备份文件多大 13 个，导致磁盘空间占用较高。

问题排查

正常情况下，会有一个备份文件清理任务定期执行。备份文件超过了预设数量，猜测可能是文件清理任务出现问题。

1. 备份文件的清理任务

先看一下自动清理时间间隔的计算公式：

自动清理的时间间隔 = min(recovery_window/2, 1d)

故正常每天会发起一次备份文件的清理任务。

> 注意：对于 oceanbase-3.1.2-20211230114204 之前的版本（不含该版本），自动清理的时间间隔为 recovery_window/2。

备份文件的清理任务卡住了！

信息采集时间为 22 日。

发现 19 日发起备份清理任务，22 日还未完成，判断为清理任务卡住了。

MySQL [oceanbase]> select * from __all_tenant_backup_clean_info; +----------------------------+----------------------------+-----------+--------+----------------------------+----------------------------+-------------+------------------------+--------+------------------+-----------+---------+------------------+--------+---------+ | gmt_create | gmt_modified | tenant_id | job_id | start_time | end_time | incarnation | type | status | parameter | error_msg | comment | clog_gc_snapshot | result | copy_id | +----------------------------+----------------------------+-----------+--------+----------------------------+----------------------------+-------------+------------------------+--------+------------------+-----------+---------+------------------+--------+---------+ | 2021-03-19 14:18:50.841475 | 2024-10-19 00:28:50.021733 | 1 | 1187 | 2024-10-19 00:28:49.998736 | 2024-10-19 00:27:09.663132 | 1 | DELETE OBSOLETE BACKUP | DOING | 1728923329993025 | | | 1728845993342237 | 0 | 0 | | 2024-10-19 00:28:50.030195 | 2024-10-19 00:28:50.030195 | 1001 | 1187 | 2024-10-19 00:28:49.998736 | 2024-10-19 00:27:09.663132 | 1 | DELETE OBSOLETE BACKUP | DOING | 1728923329993025 | | | 1728845993342237 | 0 | 0 | +----------------------------+----------------------------+-----------+--------+----------------------------+----------------------------+-------------+------------------------+--------+------------------+-----------+---------+------------------+--------+---------+ 2 rows in set (0.01 sec) 

集群备份清理有关的历史事件信息。

MySQL [oceanbase]> select * from __all_rootservice_event_history where module='backup_clean'; +----------------------------+--------------+------------+-----------+--------+---------------+--------+---------+--------+----------+-------------------------------+-------+--------+-------+--------+------------+----------------+-------------+ | gmt_create | module | event | name1 | value1 | name2 | value2 | name3 | value3 | name4 | value4 | name5 | value5 | name6 | value6 | extra_info | rs_svr_ip | rs_svr_port | +----------------------------+--------------+------------+-----------+--------+---------------+--------+---------+--------+----------+-------------------------------+-------+--------+-------+--------+------------+----------------+-------------+ | 2024-04-16 18:55:49.799602 | backup_clean | backup_set | tenant_id | 1 | backup_set_id | 1099 | copy_id | 0 | trace_id | YB42B6B49660-0005E8F1C2B2E2CD | | | | | | 182.180.150.96 | 2882 | ----------- 输出信息过多,省略部分信息 ----------- | 2024-10-19 00:28:50.010504 | backup_clean | backup_set | tenant_id | 1 | backup_set_id | 1288 | copy_id | 0 | trace_id | YB42B6B49660-0005E8F1C2B2E845 | | | | | | 182.180.150.96 | 2882 | | 2024-10-19 00:28:50.010521 | backup_clean | backup_set | tenant_id | 1 | backup_set_id | 1289 | copy_id | 0 | trace_id | YB42B6B49660-0005E8F1C2B2E845 | | | | | | 182.180.150.96 | 2882 | +----------------------------+--------------+------------+-----------+--------+---------------+--------+---------+--------+----------+-------------------------------+-------+--------+-------+--------+------------+----------------+-------------+ 191 rows in set (29.60 sec) 

2. 查看日志

2.1 备份清理调度任务失败

[root@hwc-ocp RS_cl_bigdata]# grep "ob_backup_data_clean" rootservice.log.20241022040332 ｜ grep WARN rootservice.log.20241022040332:[2024-10-22 03:48:55.440734] INFO [RS] ob_backup_data_clean_scheduler.cpp:71 [101273][788][YB42B6B49660-0005E8CB48709532] [lt=7] [dc=0] backup data scheduler init(arg={type:12, value:1729194535439601, copy_id:0}) rootservice.log.20241022040332:[2024-10-22 03:48:55.445090] WARN [RS] schedule_sys_tenant_backup_data_clean (ob_backup_data_clean_scheduler.cpp:352) [101273][788][YB42B6B49660-0005E8CB48709532] [lt=6] [dc=0] sys tenant clean info status is not stop, can do scheduler(ret=-9044, clean_info={tenant_id:1, job_id:1187, start_time:1729268929998736, end_time:1729268829663132, incarnation:1, copy_id:0, type:1, status:2, expired_time:1728923329993025, backup_set_id:0, error_msg:"", comment:"", clog_gc_snapshot:1728845993342237, result:0, backup_piece_id:0, backup_round_id:0}) rootservice.log.20241022040332:[2024-10-22 03:48:55.445176] WARN [RS] schedule_backup_data_clean (ob_backup_data_clean_scheduler.cpp:275) [101273][788][YB42B6B49660-0005E8CB48709532] [lt=12] [dc=0] failed to schedule sys tenant backup(ret=-9044) rootservice.log.20241022040332:[2024-10-22 03:48:55.445187] WARN [RS] start_schedule_backup_data_clean (ob_backup_data_clean_scheduler.cpp:89) [101273][788][YB42B6B49660-0005E8CB48709532] [lt=8] [dc=0] failed to schedule backup data clean(ret=-9044, tenant_ids=[1, 1003, 1004]) rootservice.log.20241022040332:[2024-10-22 03:53:55.447033] INFO [RS] ob_backup_data_clean_scheduler.cpp:71 [101273][788][YB42B6B49660-0005E8CB48709532] [lt=15] [dc=0] backup data scheduler init(arg={type:12, value:1729194835445853, copy_id:0}) 

判断关键字：sys tenant clean info status is not stop, can do scheduler

查看日志可推断 sys 租户备份清理任务状态不是 stop 的状态，故备份清理调度任务失败。

与上文中 __all_tenant_backup_clean_info 视图表查询结果形成呼应，判定为备份清理任务异常。

清理任务失败原因：上一次的备份清理任务状态一直为 doing。

2.2 无法识别到路径

通过备份清理任务的 trace_id 打印日志发现：频繁报错 4018，无法识别到路径。

[root@hwc-ocp RS_cl_bigdata]# grep "YB42B6B49660-0005E8F1C2B2E845" rootservice.log.20241022001100 |grep WARN [2024-10-21 23:37:26.194250] WARN [ARCHIVE] get_file_id_range_ (ob_archive_log_file_store.cpp:501) [101266][774][YB42B6B49660-0005E8F1C2B2E845] [lt=9] [dc=0] failed get_file_range(uri=file:///obbak_bigdata/cl_bigdata_new/cl_bigdata/2/incarnation_1/1003/clog/13/index/1102810163026406/13, ret=-4018, pg_key={tid:1102810163026406, partition_id:13, part_cnt:0}) [2024-10-21 23:37:26.194257] WARN [ARCHIVE] get_index_file_id_range (ob_archive_log_file_store.cpp:281) [101266][774][YB42B6B49660-0005E8F1C2B2E845] [lt=5] [dc=0] failed to get_index_file_id_range(ret=-4018, pg_key={tid:1102810163026406, partition_id:13, part_cnt:0}) [2024-10-21 23:37:26.194264] WARN [ARCHIVE] locate_file_by_log_ts_for_clear (ob_archive_log_file_store.cpp:252) [101266][774][YB42B6B49660-0005E8F1C2B2E845] [lt=5] [dc=0] get_index_file_id_range fail(ret=-4018, pg_key={tid:1102810163026406, partition_id:13, part_cnt:0}) [2024-10-21 23:37:26.217948] WARN [ARCHIVE] get_file_id_range_ (ob_archive_log_file_store.cpp:501) [101266][774][YB42B6B49660-0005E8F1C2B2E845] [lt=8] [dc=0] failed get_file_range(uri=file:///obbak_bigdata/cl_bigdata_new/cl_bigdata/2/incarnation_1/1003/clog/13/index/1102810163026406/216, ret=-4018, pg_key={tid:1102810163026406, partition_id:216, part_cnt:0}) 

判断关键字：failed get_file_range

2.3 IO 调用异常

继续排查，发现通过该 trace_id 过滤中发现以下日志，该日志证明 IO 调用异常，该情况通常出现于 NFS 侧。

[root@hwc-ocp RS_cl_bigdata]# grep "access storage" rootservice.log.20241022040332 ｜ grep "YB42B6B49660-0005E8F1C2B2E845" rootservice.log.20241022040332:[2024-10-22 04:02:23.121397] WARN [STORAGE] print_access_storage_log (ob_storage.cpp:39) [101266][774][YB42B6B49660-0005E8F1C2B2E845] [lt=10] [dc=0] access storage op=list_files uri=file:///obbak_bigdata/cl_bigdata_new/cl_bigdata/2/incarnation_1/1003/clog/13/index/1102810162740407/88 size=0 Byte cost_ts=2129326 us speed=0.00 MB/s rootservice.log.20241022040332:[2024-10-22 04:02:24.660028] WARN [STORAGE] print_access_storage_log (ob_storage.cpp:39) [101266][774][YB42B6B49660-0005E8F1C2B2E845] [lt=9] [dc=0] access storage op=list_files uri=file:///obbak_bigdata/cl_bigdata_new/cl_bigdata/2/incarnation_1/1003/clog/13/index/1102810162744871/4 size=0 Byte cost_ts=1537551 us speed=0.00 MB/s rootservice.log.20241022040332:[2024-10-22 04:02:33.558367] WARN [STORAGE] print_access_storage_log (ob_storage.cpp:39) [101266][774][YB42B6B49660-0005E8F1C2B2E845] [lt=13] [dc=0] access storage op=storage reader uri=file:///obbak_bigdata/cl_bigdata_new/cl_bigdata/2/incarnation_1/1003/clog/13/index/1102810162738281/141/2 size=144 Byte cost_ts=1243114 us speed=0.00 MB/s rootservice.log.20241022040332:[2024-10-22 04:02:38.908843] WARN [STORAGE] print_access_storage_log (ob_storage.cpp:39) [101263][768][YB42B6B49660-0005E8CB48109532] [lt=9] [dc=0] access storage op=get_file_length uri=file:///obbak_bigdata/cl_bigdata_new/cl_bigdata/2/incarnation_1/cluster_clog_backup_info size=0 Byte cost_ts=1176124 us speed=0.00 MB/s 

判断关键词：access storage，us speed=0.00 MB/s

3. 检查 NFS

[admin@x-shhp-oceanbase-db-p02:~]$ mount |grep nfs 2.4.8.124:/obbak_bigdata on /obbak_bigdata type nfs (rw,relatime,vers=3,rsize=1048576,wsize=1048576,namlen=255,hard,proto=tcp,timeo=600,retrans=2,sec=sys,mountaddr=2.4.8.124,mountvers=3,mountport=2050,mountproto=tcp,local_lock=none,addr=2.4.8.124) 

生产环境的 NFS 版本为 3.x，原来 3 版本为 OB 侧非标部署。

结论：生产未严格按官方文档的要求配置 NFS，可能存在很多异常问题，包括备份清理问题。

优化建议

严格按官方文档的要求配置 NFS，修改 NFS 版本为 4.x，备份卡住的问题得到解决。

更多技术文章，请访问：https://opensource.actionsky.com/

关于 SQLE

SQLE 是一款全方位的 SQL 质量管理平台，覆盖开发至生产环境的 SQL 审核和管理。支持主流的开源、商业、国产数据库，为开发和运维提供流程自动化能力，提升上线效率，提高数据质量。

✨ Github：https://github.com/actiontech/sqle

📚 文档：https://actiontech.github.io/sqle-docs/

💻 官网：https://opensource.actionsky.com/sqle/

👥 微信群：请添加小助手加入 ActionOpenSource

🔗 商业支持：https://www.actionsky.com/sqle