Cloning Around

"We never choose which words to use, for as long as they mean what
they mean to mean, we don’t care if they make sense or nonsense." 
― Norton Juster, The Phantom Tollbooth

Sometimes it’s necessary to clone an existing ORACLE_HOME; one case for this is when the business requires a new ORACLE_HOME when CPU patchsets are applied. With some operating systems cloning a home is simiply a matter of creating a suitable archive (tar, pax, cpio) and extracting the contents to the destination directory. Some operating systems, however, aren’t as well-mannered. Linux is one such operating system and cloning an ORACLE_HOME requires the use the Universal Installer to set things ‘right’ once the copy has completed. Unfortunately this cloning process tends to have issues of its own; one such issue is the setting of the ddl_lock_timeout parameter. Let’s look at the process, the error and how to get around it.

Presuming the business does require a new home to patch it to the latest CPU so the original home can be preserved let’s proceed with the cloning process. This imaginary database will be named GLORBLE (reminding us that this is not an real database) and the original database home will be /u01/zapp/oracle/product/12.2.0/glorble. The April 2018 CPU will be applied so the cloned home will be /u01/app/oracle/product/12.2.0/glorble_apr2018, to make it obvious that it contains software patched with the April 2018 CPU.

The destination directory is created and the original database home files are copied to it from an archive — pax, tar, cpio, gzip can all be used to create such an archive. Presuming this will be a repeated task the clone command can be saved as a script (possibly named clone_home.sh):

$ORACLE_HOME/perl/bin/perl $ORACLE_HOME/clone/bin/clone.pl ORACLE_BASE="/u01/app/oracle/"
ORACLE_HOME="/u01/app/oracle/product/12.2.0/glorble_apr2018" 
OSDBA_GROUP=dba OSOPER_GROUP=oper -defaultHomeName

For readability the command has been split into three lines; it must be a single line in the script or it won’t execute.

The environment must be set to ‘point’ to the home being cloned, then the script can be executed. Prodigous output is generated that won’t be provided here; the ‘problems’ begin after the home is cloned and the root.sh script has been run. Once cloned utilities like sqlplus will perform generally as expected until it becomes desirable or necessary to set a dynamic init parameter, like ddl_lock_timeout, when an unexpected ORA-00068 error is thrown, even when the parameter value is within the accepted range:

SQL> alter session set ddl_lock_timeout = 5;
ERROR:
ORA-00068: invalid value "" for parameter ddl_lock_timeout, must be
between 0 and 1000000

The clue appears to be in the value reported by Oracle; something in the linking step of the cloning process apparently failed. What that may be is not clear and Oracle Support has no document providing any explanation other than an upgrade process was not completed. Fortunately this home will be patched in the next step, which requires the most recent release of OPatch (patch number p6880880) which simply needs to be unzipped into the cloned home. Using the current OPatch the cloned home is processed:

$ cd 
$ $ORACLE_HOME/OPatch/opatch apply
...

As if by magic the ORA-00068 errors disappear. In reality the patch process applies software updates amd re-links the software in a manner different than the Perl script used to clone the home, which executes the setup command in $ORACLE_HOME/oui/bin. It isn’t clear WHAT is different between the two processes but no errors are thrown after patching is completed.

It’s truly an oddity; Oracle supplied tools and utiliities are expected to work and produce usable results. In this case something was missed in the overall cloning process and for those cloning an existing ORACLE_HOME for another database (which isn’t necessary but isn’t wrong) because they are running Oracle on an Oracle VM built from a template and have no software distribution to use this could be a major problem.

Take care when using this cloning process as things may not be as they seem on the surface. Many bug fixes have gone into the Perl script Oracle provides yet it appears there is at least one that hasn’t been addressed (possibly because it hasn’t been reported). This has been reported to Oracle support so hopefully an answer is forthcoming. In the interim if a home is cloned it should be patched to apply the latest security patches and to avoid the ORA-00068 surprises.

And that just makes sense.