Got a mail from my grid-control monitoring: database unavailable. After a very short look: database crashed. No changes made the last month. Version 10.2.0.4 on Suse Linux 10.1, 64-bit, on a VM (ESX).
Alert-log: ORA-07445: exception encountered: core dump [_IO_vfprintf_internal()+1] [SIGSEGV] [Address not mapped to object] [0x000000000]   Mon Dec 29 10:04:23 2008
Pmon-trace file: ” ORA-00822: MMAN process terminated with error “.
Solution on internet: warm start the instance. Yeah, that’s right … up again.
Cause ( Metalink): looks like a BUG !
Note 749524.1 looks very much alike, only the ” [_IO_vfprintf_internal()+1″ – part of the call stack doesn’t fit:
” The issue has been identified in unpublished Bug 7039896
Processes may wait for the “shared pool” latch for a long time when sga_target or memory_target is being used.
The stack trace for the process holding the latch has the function “kghquiesce_regular_extent” in it and appears to be spinning (using CPU). Typically the latch holder will be the MMAN process.
Unpublished Bug 7039896 is fixed in 10.2.0.5, 22.214.171.124 and 11.2 rel.
At the moment of writing this note 10.2.0.5 patchset is not available. ”
So, to fix the issue:
2. or apply 126.96.36.199 patchset where is included the fix;
3. or apply one of the following workaround:
- do not use SGA_TARGET / MEMORY_TARGET
- or set the parameter _enable_shared_pool_durations=false in the init.ora or spfile
Not completely sure if this is the bug, but the particular database is for development AND installed on a VM (ESX). Since the installation 6 months ago, nothing like this happened, so I’ll take my chances and wait for 10.2.0.5 or higher….