ORACLE 12c的数据库，以多租户方式运行，运行了一段时间，还比较稳定，分享一下相关参数修改。

1、一些参数

DEFERRED_SEGMENT_CREATION

默认是true，建议设置为false

_DATAFILE_WRITE_ERRORS_CRASH_INSTANCE

默认是true，所有datafile的IO写error，都会导致数据库crash。建议设置为false。12.2.0.1之后版本不用修改

JOB_QUEUE_PROCESSES

默认是1000，建议设置为你真正需要的数字。如果你不知道，建议设置为10。如

_OPTIMIZER_AGGR_GROUPBY_ELIM

默认是true，可能会导致groupby类的结果集错误。建议设置为false。

SESSION_CACHED_CURSORS

默认值是50，太小，容易导致内存碎片，建议设置为1000.

RAC环境设置ASM实例audit_sys_operations 参数设置为FALSE

RAC环境设置ASM实例PROCESSES参数2000

2、上线前打好补丁

请参考以下note查看Oracle相应的补丁（上线前安装最新补丁）：

Assistant: Download Reference for Oracle Database/GI Update, Revision, PSU, SPU(CPU), Bundle Patches, Patchsets and Base Releases ( Doc ID 2118136.2 )

https://support.oracle.com/epmos/faces/DocumentDisplay?parent=SrDetailText&sourceId=3-17677673745&id=2118136.2

3、内存过大导致实例宕机（SGA超过600G）

The following workarounds are available until a permanent fix for base bug will be released:

1/ Increase sga_target by 10% and set _shared_io_pool_size=<granule_size> so that system component can allocate 1 granule during startup instead of asking buffer cache.

-- You can determine your granule size with this SQL:

SQL> select bytes from v$sgainfo where name like 'Granule Size';

-- and set _shared_io_pool_size=<granule_size> by running:

SQL> alter system set "_shared_io_pool_size"= <granule_size>;

- or -

2/ Comment the following in parameter file:

dispatchers = "(PROTOCOL=TCP) (SERVICE=<sid>XDB)" . <sid> should be replaced with the relevant ORACLE_SID value.

Refer more on 2250078.1.