Bug 18505843 - TEMP_UNDO_ENABLED to true causing rows to be not preserved after commit over database link

Problem Statement 

Recently we have noticed global temporary tables (GTT) with the option "Preserve rows after commit" not working as expected post-migration to oracle 19c with the distributed transaction which uses database links as compared to 12c where it was working fine.


Observation

It has been identified with the 19c parameter temp_undo_enabled parameter is by default set to true which was false in 12c has an impact on the above behavior. To understand the reason we would need to understand about parameter temp_undo_enabled.


temp_undo_enabled parameter means undo generated for the changes to temporary tablespace will be written to permanent undo tablespace when it is set to false, when this parameter value is set to true it means undo generated for temporary tables will be written to temporary tablespaces. This will mean lesser redo log generation and better database performance. It also reduces the storage requirement due to lesser redo log generation.


In cases where this parameter is set to false which is default values in 12c, in that case, undo generated for temporary tablespace are stored in permanent undo tablespace and hence generated the redo data as well. Oracle always mains temp tables are not supported for distributed transactions i.e. remote transactions over database link.  


Oracle has also logged one bug in this regard as "Bug 18505843 - TEMP_UNDO_ENABLED to true causing rows to be not preserved after commit over database link". Though the Oracle development team has closed this as NOT a bug and expected behavior as distributed transactions are not supported for temporary tables. Although no errors were given in the past for this, these restrictions exist. 


Solution

A strategic solution must be, not to use distributed transactions on temporary tables but it may not be easy for some to change the application logic due to various restrictions in place. In that case, set the parameter TEMP_UNDO_ENABLED as false. 


References

Doc ID 2169650.1 - Unexpected behavior of GTT when temp_undo_enabled set to True and using database links

Comments

Post a Comment

Popular posts from this blog

How to Solve - "WAIT FOR EMON PROCESS NTFNS"

Event Monitor Process in short called as EMON is generally used by oracle to send notification. Often applications uses Database Change Notification process also called as DBCN to be aware of changes in the database, based on these notifications they take further actions. Often clients subscribe to the database and notifications are send as callback to them. In some cases you will notice database sessions are showing wait event "WAIT FOR EMON PROCESS NTFNS" and it can even make your database un-responsive. One observation can be these processes are taking more CPU (may not in be in all cases). As stated EMON process deals with database notification and it's spawn and gets down itself and Administrators have no control on theses processes. Anyone can go through oracle documents to read more EMON functioning process.  EMON uses oracle advanced queues and database queue notifications into them and EMON process these notifications to clients based on their subscriptions. In s
Read more

Query Regression - "OR" Transformation Oracle 19c

Recently while upgrading Oracle Database from version 12.1 to 19.10 found there were couple of queries which was running slow post upgrade. Usually these type of behavior is seen while upgrading to newer version and oracle provide various ways to fix them. The difference between good and bad plan was an introduction of  "MERGE JOIN CARTESIAN" (aka MJC) and that's in where regressed query in 19c taking most of the time. MJC sometimes works for queries but in this case it was clearly not helping.  We had following options here:     1. Copy the profile from 12c to 19c database and it worked, but the problem was not just to specific query but with many similar query. This means profile option is not very attractive in this case.     2. We also tried running query with OFE 12.1 to use the same optimizer features as previous version and query behaves well. Though this option can't be placed as we need to use 19c OFE and can be placed only for testing purpose.           alte
Read more

ORA - 12537: TNS: connection closed

Often when active clone to be done to create a new environment, we land up into couple of issues. I will discuss about one issues during this process. Action : Create a new database on different machine from a source. Error : ORA - 12537: TNS: connection closed In order to duplicate the environment was using below rman log /tmp/duplicate.log connect target sys/<password>@<source tns alias> connect auxiliary sys/<password>@<target tns alias> run{ allocate channel ch1 device type disk; allocate channel ch2 device type disk; allocate channel ch3 device type disk; allocate channel ch4 device type disk; allocate auxiliary channel ax1 device type disk; allocate auxiliary channel ax2 device type disk; allocate auxiliary channel ax3 device type disk; allocate auxiliary channel ax4 device type disk; DUPLICATE TARGET DATABASE TO <TARGET_DBNAME> FROM ACTIVE DATABASE NOFILENAMECHECK; } While making connection to Auxiliary database received mentioned error ORA - 12537:
Read more