当前位置: DBARoad > 技术专题, 本站推荐 > 文章正文

最近一次在AIX上布署RAC环境，前期安装还算顺利，虽然也遇到些小问题，不过最终都安装成功了。在升级到10.2.0.4，打上Bundle Patch后，发现一些CRS服务无法启动了。

查看状态：

$crs_stat -t
Name           Type           Target    State     Host
------------------------------------------------------------
ora....T1.inst application    ONLINE    UNKNOWN   pg_dev_db
ora....T2.inst application    ONLINE    UNKNOWN   pg_test_db
ora.OCSTEST.db application    ONLINE    OFFLINE
ora....DB.lsnr application    OFFLINE   UNKNOWN   pg_dev_db
ora...._db.gsd application    ONLINE    OFFLINE
ora...._db.ons application    OFFLINE   OFFLINE
ora...._db.vip application    OFFLINE   OFFLINE
ora....DB.lsnr application    OFFLINE   UNKNOWN   pg_test_db
ora...._db.gsd application    ONLINE    OFFLINE
ora...._db.ons application    OFFLINE   OFFLINE
ora...._db.vip application    OFFLINE   OFFLINE

尝试停止服务：

$crs_stop ora.pg_test_db.LISTENER_ZJOCSTEST_PG_TEST_DB.lsnr
Remote stop for `ora.pg_test_db.LISTENER_ZJOCSTEST_PG_TEST_DB.lsnr` failed on member `pg_test_db`
CRS-0216: Could not stop resource ‘ora.pg_test_db.LISTENER_ZJOCSTEST_PG_TEST_DB.lsnr’.

$crs_stop ora.OCSTEST.OCSTEST1.inst
Attempting to stop `ora.OCSTEST.OCSTEST1.inst` on member `pg_dev_db`
`ora.OCSTEST.OCSTEST1.inst` on member `pg_dev_db` has experienced an unrecoverable failure.
Human intervention required to resume its availability.
CRS-0216: Could not stop resource ‘ora.OCSTEST.OCSTEST1.inst’.

尝试了以下方法：
停CRS再启
crs_stop -f , crs_start -f
实例重建：srvctl remove instance , srvctl add instance

不过最终都没能成功：

$srvctl start instance -d OCSTEST -i OCSTEST1
PRKP-1001 : Error starting instance OCSTEST1 on node pg_dev_db
pg_dev_db:ora.OCSTEST.OCSTEST1.inst:
/app10g/oracle/product/10.2.0/db_1/bin/racgwrap[62]:
%ORACLE_HOME%/bin/racgmain: not found.
CRS-0215: Could not start resource ‘ora.OCSTEST.OCSTEST1.inst’.

$ls -l racgmain
-rwxr-x–x 1 ora10g dba 1911329 Apr 05 01:08 racgmain

回退Bundle Patch后，恢复正常。

后来在Metalink上查到这样一篇文章：

Cannot startup CRS after applying 10.2.0.4 patchset or CRS Bundle Patch# 3 and above
Doc ID: 739371.1

Cause
scenario 1)

This is due to changes we have introduced in init.cssd which now requires the oracle user to have CAP_NUMA_ATTACH, CAP_BYPASS_RAC_VMM, and CAP_PROPAGATE OS capabilities when running Oracle clusterware software.

These capabilities were added in AIX 5.3 TL5 (5300-05)

scenario 2)

Changes to mlockall() were introduced in AIX 5.3 TL5 (5300-05)

文中建议安装：AIX 5.3 TL5 (5300-05) or above
而这个环境的OS版本刚好是：5300-04

$oslevel -r
5300-04

在原因一中提到要求:

oracle user to have CAP_NUMA_ATTACH, CAP_BYPASS_RAC_VMM, and CAP_PROPAGATE OS capabilities when running Oracle clusterware software.

而在升级10.2.0.4的README中也提到：

Recommended OS Capabilities for CRS User to run Oracle Clusterware software
To add capabilities, enter a command similar to the following:
# /usr/bin/chuser capabilities=CAP_NUMA_ATTACH CAP_BYPASS_RAC_VMM CAP_PROPAGATE crs

注：README这里有点不对，
capabilities=CAP_NUMA_ATTACH CAP_BYPASS_RAC_VMM CAP_PROPAGATE
应改为
capabilities=CAP_NUMA_ATTACH，CAP_BYPASS_RAC_VMM，CAP_PROPAGATE

而这一步操作，正是由于中间的空格，执行失败，被我略过了。可惜现在已没法证实是不是由于这个原因引起了，这个环境又被取消了。不过这些要点，在今后的安装、升级中还是要引起注意的。

— The End —