Oracle SRVCTL Issue in

Thursday May 12th 2016 by David Fitzjarrell

When using Oracle in a Grid configuration the srvctl utility can change the group on the Oracle executables, creating problems with locating the spfile and with disk access.

Every once in a while Oracle produces interesting behavior, and has done so with release using the Grid infrastructure. On AIX an issue was reported regarding an unexplained group ownership change. Let's look at the situation and see what Oracle did.

The reported behavior involves either the srvctl or crsctl utility and creates problems when either a database (using srvctl) or CRS (using crsctl) is started. Error messages that can be received include the following:

ORA-15025: could not open disk "/dev/griddisk/griddisk12"
ORA-27041: unable to open file
IBM AIX RISC System/6000 Error: 13: Permission denied

Both utilities will change the group ownership of the 'oracle' executable from its original setting to 'oinstall', or some variant of it, making any disk unreadable to the Oracle process. Oracle Support has filed this as Bug 19436183. Interestingly it doesn't affect the setUID and setGID bits on that file. Doing a bit of research uncovers a similar issue with Linux, filed as Bug 14197338. The latter bug shows no workaround or resolution, and it has been reported to change the group ownership to the entire group list that the 'oracle' user is assigned; Bug 19436183 provides the workaround of manually changing the group ownership back to its original value. That bug also reports that using sqlplus to start the database does not affect the group ownership, although that will require starting each node individually rather than having the convenience of using srvctl to start the entire RAC cluster.

Remember that Bug 19436183 also reports that using crsctl to start CRS results in the same issue; it would seem that starting the Cluster services in will need to be completed, then the group ownership of the Oracle executable changed so that sqlplus can be used to finally start the databases. For a non-RAC implementation (using ASM) this isn't as inconvenient as it becomes for a RAC implementation, since RAC clusters can span from 2 to 255 nodes starting each instance with sqlplus could be a tedious task, indeed. As stated previously no fix is reported for either bug, however apparently there have been no reports to Oracle Support that Version 12.1.0.x is affected. Since is the latest release of 11.2 it's possible that this is the terminal release of that version, and that users might seriously consider upgrading to to rid themselves of this issue.

See all articles by David Fitzjarrell

Mobile Site | Full Site