Environment information: Linux 6.1 + oracle11.2.0.3 RAC
problem Phenomenon:School Night 6 o'clock elective course, the number of people about 3000, 7 o'clock, the database error is as follows (database to 6 points or can be connected), the database hung live.
Tue Dec 18:00:33
Dumping diagnostic data in directory=[cdmp_20141216180033], requested by (instance=2, osid=24917 (M001)), summary=[ INCIDENT=45211].
Tue Dec 16 18:02:23 2014
Pmon failed to acquire latch, see Pmon dump
Tue Dec 16 18:03:23 2014
Pmon failed to acquire latch, see Pmon dump
Tue Dec 16 18:04:54 2014
Pmon failed to acquire latch, see Pmon dump
Tue Dec 16 18:05:55 2014
Pmon failed to acquire latch, see Pmon dump
Tue Dec 16 18:07:26 2014
Pmon failed to acquire latch, see Pmon dump
Tue Dec 16 18:08:26 2014
Pmon failed to acquire latch, see Pmon dump ..................................................................................Tue Dec 19:31:32
Exception [TYPE:SIGSEGV, Address not mapped to object] [addr:0x18] [Pc:0x9379d18, Kglic0 () +1086] [flags:0x0, Count: 1]
Errors in File/u01/app/oracle/diag/rdbms/ywkdb/ywkdb2/trace/ywkdb2_m003_28984.trc (incident=44523):
ora-07445:exception Encountered:core Dump [Kglic0 () +1086] [SIGSEGV] [addr:0x18] [pc:0x9379d18] [Address not mapped t o Object] []
Incident details in:/U01/APP/ORACLE/DIAG/RDBMS/YWKDB/YWKDB2/INCIDENT/INCDIR_44523/YWKDB2_M003_28984_I44523.TRC
Use Adrci or support Workbench to the incident.
See Note 411.1 at My Oracle support for error and packaging details.
Tue Dec 20:00:23
Exception [TYPE:SIGSEGV, Address not mapped to object] [addr:0x18] [Pc:0x9379d18, Kglic0 () +1086] [flags:0x0, Count: 1]
Errors in File/u01/app/oracle/diag/rdbms/ywkdb/ywkdb2/trace/ywkdb2_m000_30227.trc (incident=44811):
ora-07445:exception Encountered:core Dump [Kglic0 () +1086] [SIGSEGV] [addr:0x18] [pc:0x9379d18] [Address not mapped t o Object] []
Incident details in:/u01/app/oracle/diag/rdbms/ywkdb/ywkdb2/incident/incdir_44811/ywkdb2_m000_30227_ I44811.TRC
Use Adrci or support Workbench to the incident.
See Note 411.1 at My Oracle support for error and packaging details.
The final troubleshooting is a bug in this version of the database. See:ORA-7445 Core Dump [Kglic0 () +712] (document ID 1325437.1) Mmon Slave Process Reports ORA-7445 [Kglic0]/[kksitercursorstat] Error , Plus Database hangs (document ID 1487720.1)
Troubleshooting process:According to the error of the ORA-07445, troubleshoot the official Oracle document and discover that this is an Oracle bug in the official documentation, as described in the documentation, which can trigger the bug under the following conditions:
These points, the current database state are consistent with:
1). The current database version is 11.2.0.3
2). Error warning is ORA-07445: [Kglic0 ()]
3). In the TRC file, there is a kksitercursorstat hint
4). _kghdsidx_count value >1
5). Query select COUNT (*) from v$sql_bind_capture where name like ': sys% '; Total >0
According to the official document processing opinion, modify the implied parameter "_kghdsidx_count" = 1, restart the database service can resolve this problem.
In addition, according to the extension in the official documentation: BUG 12340939 ORA-7445 [kglic0] can occur capturing cursor stats for v$sqlstats may also trigger the problem, modify the implied parameters
"_cursor_stats_enabled" =false, restarting the database service can resolve this issue.
After the change, the school took the next day to choose the course no more error.
Case Summary:1. Check MoS 2 for questions. Specifically to this problem, before this problem, the choice of class came out, a point: in big Data high concurrency situation, this bug appears. Cause of the problem: During SQL Statistics analysis for query execution, we load dependency information into the Library Cache. These dependency memory pieces can be overwritten by other sessions without a error. When the original session runs into the wrong dependency information it can leads to the internal error and performance ISS UEs during subsequent queries of these Library Cache objects. ---in short: the Library cache because of the concurrency pressure, the historical information can not be found or
During SQL Statistics analysis for query execution, we load dependency information into the Library Cache. These dependency memory pieces can be overwritten by other sessions without a error. When the original session runs into the wrong dependency information it can leads to the internal error and performance ISS UEs during subsequent queries of these Library Cache objects. ---In short: there is a problem with the statistics on the table
Database bug caused by high concurrency pressure
