在數據庫宕機前出現ORA-00600錯誤。 日誌內容以下:sql
ORA-01595: error freeing extent (4) of rollback segment (31)) ORA-00607: Internal error occurred while making a change to a data block ORA-00600: internal error code, arguments: [kdBlkCheckError], [3], [3], [18018], [], [], [], [], [], [], [], [] Corrupt Block Found TSN = 2, TSNAME = UNDOTBS1 RFN = 3, BLK = 3, RDBA = 12582915 OBJN = 2, OBJD = -1, OBJECT = , SUBOBJECT = SEGMENT OWNER = , SEGMENT TYPE = Wed Aug 02 10:00:05 2017 Dumping diagnostic data in directory=[cdmp_20170802100005], requested by (instance=1, osid=24055 (SMON)), summary=[incident=48868]. Errors in file /oracle/app/oracle/diag/rdbms/aiboss/aiboss/trace/aiboss_smon_24055.trc (incident=61108): ORA-00600: internal error code, arguments: [kdBlkCheckError], [3], [3], [18018], [], [], [], [], [], [], [], [] Incident details in: /oracle/app/oracle/diag/rdbms/aiboss/aiboss/incident/incdir_61108/aiboss_smon_24055_i61108.trc Wed Aug 02 10:00:11 2017 PMON (ospid: 24025): terminating the instance due to error 474 System state dump requested by (instance=1, osid=24025 (PMON)), summary=[abnormal instance termination]. System State dumped to trace file /oracle/app/oracle/diag/rdbms/aiboss/aiboss/trace/aiboss_diag_24035.trc Instance terminated by PMON, pid = 24025 Wed Aug 02 10:40:19 2017
從宕機前的日誌記錄來看,數據庫遇到的是oracle BUG 12349316. 依據爲:數據庫
freeing extent (4) of rollback segment (31)) =====> 與oracle BUG 12349316引起的條件一致,都是在清理extent時引起BUG. [kdBlkCheckError], [ 3], [ 3], [ 18018] =====> 與oracle BUG 12349316 中600錯誤返回參數一致,都帶有18018. trace 日誌(aiboss_smon_24055_i61108.trc)中發現 delete extent 函數 =====> FRAME [ 32] (ktusp_delextent()+76 -> ktsxr_delete())
主要的處理思路是先跳過有問題的undo段,而後重建undo表空間oracle
修改參數文件 數據庫啓動時,查找參數文件的順序是spfile<ORACLE_SID>.ora –> init<ORACLE_SID>.ora –> init.ora. 所以Oracle 數據庫傾向於使用spfile啓動數據庫。通常環境中也都使用spfile. 如何確認數據庫使用的是spfile 仍是pfile,使用 " show parameter spfile " 命令便可查看。app
當數據庫使用的是spfile參數文件時,因爲spfile是 二進制 文件,咱們不便於直接修改,所以須要先建立出一個pfile 文本文件。ide
create pfile='/tmp/pfile.ora' from spfile; 此命令的執行不須要啓動數據庫,進入sqlplus環境便可。
在參數文件中加入如下內容: undo_management = MANUAL # UNDO 段管理方式改成manual # 其餘可添加內容: *.fast_start_parallel_rollback=high # 以4*cpu 個數開啓回滾進程,可是實際上不會真的開始這麼多。 *._allow_resetlogs_corruption = true # 若是數據庫須要恢復,且undo與redo不一致,部分redo 沒法恢復時須要此參數,容許resetlogs
啓動數據庫函數
startup mount pfile='/tmp/pfile.ora'; alter database open;
若是數據庫須要recovery,則執行如下命令:spa
recover database until cancel; alter database open resetlogs;
offline存在活動事務的的undo塊日誌
select segment_name,status,tablespace_name from dba_rollback_segs where status not in ('ONLINE', 'OFFLINE') ; SEGMENT_NAME STATUS TABLESPACE_NAME ------------------------------ ---------------- ------------------------------ _SYSSMU3_4004931649$ NEEDS RECOVERY UNDOTBS1 _SYSSMU4_1126976075$ NEEDS RECOVERY UNDOTBS1 _SYSSMU5_4011504098$ NEEDS RECOVERY UNDOTBS1
將以上內容添加至剛建立的/tmp/pfile.ora中:code
_CORRUPTED_ROLLBACK_SEGMENTS = ('_SYSSMU3_4004931649$','_SYSSMU4_1126976075$','_SYSSMU5_4011504098$')
"_corrupted_rollback_segments" 做用是不使用這幾個回滾段。orm
重啓數據庫
startup force pfile='/tmp/pfile.ora';
從新建立undo 表空間
alter tablespace undotbs1 offline ; drop tablespace undotbs1 including contents and datafiles; create undo tablespace undotbs1 datafile '/data0/aiboss/undotbs1.dbf' size 30G autoextend off; alter system set undo_tablespace='UNDOTBS1';
重啓數據庫 重啓數據庫前,須要修改/tmp/pfile.ora 參數文件,將如下參數去除:
undo_management=manual _allow_resetlogs_corruption=true fast_start_parallel_rollback=high
重啓:
startup force pfile='/tmp/pfile.ora'; create spfile from pfile='/tmp/pfile.ora'; startup force;