Hello Everyone -
Environment:
DB Version: 12.1.0.2
Platform: Linux (x86_64) OS-Version 2.6.32-573.12.1.el6.x86_64
We had an instance crash recently with the following error:
/dwdbprod_pmon_14890_i945115.trc
*** SERVICE NAME:(SYS$BACKGROUND) 2019-03-01 14:22:55.088 PMON
ORA-00600: internal error code, arguments: [kghfrh:ds],
trace is showing no cursor in SO: or an on block object.
dwdbprod_pmon_14890_i945116.trc
ORA-07445: exception encountered: core dump [kghdsx_get_first_extent()+23]
[SIGSEGV] [ADDR:0x36] [PC:0xB7BD737] [Address not mapped to object] []
ORA-00600: internal error code, arguments: [kghfrh:ds], [0x2EFA081CD0], [], [],
[], [], [], [], [], [], [], []
happening in PMON. : waited for 'pmon timer'. May be a sync/resource issue.
There are no on block objects like tables being accessed by PMON in these
traces.
Not physical corruption. Tables are fine. this is happening in memory.
Here is the fix recommended by Support.. wanted to get some thoughts on this
change..
In order to get closer to the root cause of the next memory corruption error,
we are going to recommend setting Event 10235 at level 2.
Event 10235 : The event being set at all causes certain heap checks to be
performed.
Level Description
~~~~~ ~~~~~~~~~~~~
1 Fast check on heap free (kghfrh)
2 Do 1 AND fill memory with junk on alloc / free
Level 2 also causes kghxfr to place free chunks at the END of the freelist
which can impact latch hold times. Any performance impact would be in the 1-2%
increase range.
The idea is, Level 1 checks any time a memory heap is freed, which means we
catch corruption closer to when its created; Level 2 overwrites freed memory so
that Oracle is less likely to find old memory and think it's legitimate (and
therefore corrupt).
This Event must be set via init.ora / spfile and CANNOT / SHOULD NOT be set via
ALTER SYSTEM. The instance will need a restart to bring this Event online.
EVENT="10235 trace name context forever, level 2"
How risky is this change? We have seen this error a couple of times in the past
but had crashed the instance during the last occurrence.
Any insights are appreciated.
Thanks
-Upendra
