Oracle RAC数据库环境与单实例数据库环境有很多共性，也有很多异性。对于数据库补丁的更新同样如此，都可以通过opatch来完成。但RAC环境的补丁更新有几种不同的更新方式，甚至于可以在零停机的情况下对所有节点实现滚动升级。本文主要是转述了Doc 244241.1，描述RAC环境下的patch更新方式以及在不同的情形下选择何种更新方式。

1、RAC patch的几种方式

OPatch supports 3 different patch methods on a RAC environment:

• Patching RAC as a single instance (All-Node Patch) (停机方式)

In this mode, OPatch applies the patch to the local node first, then propagates the patch to all the other nodes, and finally updates the inventory. All instances must be down during the whole patching process.

• Patching RAC using a minimum down-time strategy (Min. Downtime Patch)(最小化停机方式)

In this mode, OPatch patches the local node, asks users for a sub-set of nodes, which will be the first subset of nodes to be patched. After the initial subset of nodes are patched, Opatch propagates the patch to the other nodes and finally updates the inventory. The downtime would happen between the shutdown of the second subset of nodes and the startup of the initial subset of nodes patched.

• Patching RAC using a rolling strategy - No down time (Rolling Patch)(滚动方式)

With this method, there is no downtime. Each node would be patched and brought up while all the other nodes are up and running, resulting in no disruption of the system.

Rolling patching strategy incur no downtime, however, some rolling patches may incur downtime due to post-installation steps, i.e. running sql scripts to patch the actual database. Please refer to patch readme to find out whether post-installation steps requires downtime or not.
注意这句话，Rolling Patch不会停机，但是有些脚本可能会引发宕机。

2、不同方式Patch的步骤
All-Node Patch
. Shutdown all Oracle instances on all nodes
. Apply the patch to the RAC home on all nodes
. Bring all instances up

Minimum downtime
. Shutdown the Oracle instance on node 1 
. Apply the patch to the RAC home on node 1 
. Shutdown the Oracle instance on node 2 
. Apply the patch to the RAC home on node 2 
. Shutdown the Oracle instance on node 3 
. At this point, instances on nodes 1 and 2 can be brought up
. Apply the patch to the RAC home on node 3 
. Startup the Oracle instance on node 3

Rolling patch (no downtime)
. Shutdown the Oracle instance on node 1 
. Apply the patch to the RAC home on node 1 
. Start the Oracle instance on node 1 
. Shutdown the Oracle instance on node 2 
. Apply the patch to the RAC home on node 2 
. Start the Oracle instance on node 2 
. Shutdown the Oracle instance on node 3 
. Apply the patch to the RAC home on node 3 
. Start the Oracle instance on node 3

3、选用何种方式patch
To be eligible as a rolling patch, the patch needs to meet certain criterias, which are determined by Oracle developers. In order to be applied in a "rolling fashion", the patch must be designated as a "rolling updatable patch" or simply "rolling patch".

The algorithm used to decide which method is going to be used is the following:

If (users specify minimize_downtime)
              patching mechanism = Min. Downtime
       else if (patch is a rolling patch)
              patching mechanism = Rolling
            else
                  patching mechanism = All-Node

#从上面的算法来看最小化停机时间是首选。

4、滚动patch的可用性
When patches are released, they have a tag as "rolling" or "not rolling" patch. While most patches can be applied in a rolling fashion, some patches can not be applied in this fashion. Patches that could potentially be installed on rolling fashion include:
 
   . Patches that do not affect the contents of the database. 
   . Patches that are not related to the RAC internode communication infrastructure. 
   . Patches that change procedural logic and do not modify common header definitions of kernel modules. This includes client side patches that only affect utilities like export, import, sql*plus, sql*loader, etc.

Only individual patches -- not patch sets -- will be “rollable”. It should also be noted that a merge patch of a “rolling patch” and an ordinary patch will not be a “rolling patch”.

From 9.2.0.4 onwards, all patches released will be marked as a "rolling" or "not rolling patch", based on defined set of rules. Patches previously released are packaged as "not rolling".

Because the set of rules currently defined are very conservative, patches released as "not rolling patches", either before and after 9.2.0.4, may be eligible to be re-released as "rolling patches", after analysis from Oracle Development.
  
If you plan to apply a patch that is marked as "not rolling" and want to check if is possible to take advantage of the rolling patch strategy, please contact Oracle Support.

5、如何确认patch是否可滚动
#可使用下面的方式来查询当前的patch是否为可滚动
As database user execute the following:

- 9i or 10gR1: opatch query -is_rolling

- 10gR2: opatch query -all  [unzipped patch location] | grep rolling

- 10gR2 on Windows: opatch query -all [unzipped patch location] | findstr rolling

- Later 10gR2 or 11g: opatch query -is_rolling_patch [unzipped patch location]

The command may not work if unzipped patch location has more than one patch sub-directory, example output while checking CPU patches:
#对于解压路径下有多个子patch的情形，校验是否为滚动patch可能会失败，如下：
Failed to load the patch object.  Possible causes are:
  The specified path is not an interim Patch shiphome
  Meta-data files are missing from the patch area
  Patch location = /home/oracle/stage/8836308
  Details = Input metadata files are missing.

Patch Location "/home/oracle/stage/8836308" doesn't point to a valid patch area.
# Author : Leshami
# Blog   : http://blog.csdn.net/leshmai

OPatch failed with error code 75

6、当前的一些限制

Patching with Shared File System
Currently OPatch treats Shared File System, like CFS, as a single-instance patch.  It means that OPatch will blindly patch files under a given ORACLE_HOME knowing that other nodes will pick up the changes via the Shared File System. Unfortunately, this means that OPatch cannot take advantage of a rolling patch on a Shared File System environment; all nodes must be down throughout the patching process.

Patching one node at time

The Opatch strategies discussed above (All-Node, Min. Down-Time, and Rolling) presumes that all nodes will be patched at the same time. Additionally, each node can be patched individually, at different times, using the "-local" key word, which will patch only the local node.

转：http://blog.csdn.net/leshami/article/details/38727151#comments