Solving an unusual Oracle upgrade issue
I recently came across a very unusual situation. During an upgraded review, I noticed the dba_registry_sqlpatch was empty when it shouldn’t be:
SQL> select patch_id, patch_uid, version, action, action_time, status, description from dba_registry_sqlpatch; no rows selected SQL>The expected output should be (from another CDB in the same home):
PATCH_ID PATCH_UID VERSION ACTION ACTION_TIME STATUS DESCRIPTION ---------- ---------- -------------------- --------------- --------------------------------------------------------------------------- --------------- ---------------------------------------------------------------------------------------------------- 24917972 20791781 12.1.0.2 APPLY 37-APR-17 11.19.49.103261 AM SUCCESS Database PSU 12.1.0.2.170117, Oracle JavaVM Component (JAN2017) 24732082 20904347 12.1.0.2 APPLY 17-APR-17 11.19.49.322985 AM SUCCESS DATABASE PATCH SET UPDATE 12.1.0.2.170117 24917972 20791781 12.1.0.2 ROLLBACK 29-NOV-17 08.35.57.888426 PM SUCCESS Database PSU 12.1.0.2.170117, Oracle JavaVM Component (JAN2017) 26635845 21564421 12.1.0.2 APPLY 29-NOV-17 08.35.57.890421 PM SUCCESS Database PSU 12.1.0.2.171017, Oracle JavaVM Component (OCT2017) 26713565 21602269 12.1.0.2 APPLY 29-NOV-17 08.35.57.956378 PM SUCCESS DATABASE PATCH SET UPDATE 12.1.0.2.171017 27338041 22036385 12.1.0.2 APPLY 12-JUN-18 01.45.24.163558 PM SUCCESS DATABASE PATCH SET UPDATE 12.1.0.2.180417The result is basically the same if querying cdb_registry_sqlpatch. First, find the MOS dba_registry_sqlpatch or registry$sqlpatch View Is Not Reflecting the Complete Updated Information after Patching (Doc ID 2039738.1). The problem is that it applies to 12.1 and it is caused by a bug in OPatch version 12.1.0.1.6, but the OPatch version is 12.2.0.1.8.
$ORACLE_HOME/OPatch/opatch version OPatch Version: 12.2.0.1.8If this is a match for you, the proposed solution to note is:
1. Download and use latest opatch version 12.1.0.1.8. (Patch 6880880) 2. Take the backup & delete the contents of dba_registry_sqlpatch to remove the invalid entries: SQL>delete ... 3. Re-run the datapatchBut what was my problem then? Well, after a while, I noticed the MOS Note Datapatch may skip the application of SQL payload for certain patches included in a given bundle in a RAC environment. (Doc ID 2069046.1). It includes a PLSQL validation script, by the way. Have a look if you think it is suspect. And it was a match for me. It seems the client used to have problems with opatchauto in the past and had to run the ‘ datapatch -verbose’ manually. The solution? Run this for every CDB contained in the cluster. The registry$sqlpatch table is now reporting the correct patch history for all CDBs. Hope this helps!
