In my last post, I discussed a collaborative effort with functional business owners to devise and execute a proper cleanup plan.  I also discussed a functional review of the configuration to make sure that change pointers are being created only when needed.

So now that you have achieved control over your change pointer data, how do you make sure that it does not go out of control again?

[Read more...]

In my last post on this topic, I discussed several change pointer data forensic methods that I use to point me in the right cleanup direction.
The change pointer data forensics will determine how to design the cleanup process.  Armed with meaningful statistics, I usually approach the functional business owners to collaborate on a solution.  Typical of the many questions that I ask are “Why are change pointers configured for this message type if we are never distributing the IDOCs?”

Discussion points and procedures for the cleanup process.

1. Unless there is custom ABAP code depending on the continued existence of processed change pointers, they serve no further purpose and should be deleted.  ABAP Program RBDCPCLR should be scheduled on a regular basis so that recent “processed” change pointers are purged.

If there is a large quantity of very old “processed” change pointer records, then consider special executions of RBDCPCLR where the selected date range is old enough to collect and purge these records.

To cleanup processed change pointers, use the Processed Change Pointers section on the select option screen of program RBDCPCLR.  Make sure that the “Processed Change Pointers” check box is checked, and enter the proper date range.  You can further limit the data selection by entering a message type in the message type box.

Note that on the selection screen there is also a “Test Run” check box that will list which records would be selected for purging, without actually purging them.  I find this feature very useful, and I strongly recommend using it prior to actual purging.

2. “Unprocessed” change pointer records that are very old are most likely not useable.  If the IDOCs that they represent were triggered now, the data distributed might no longer be valid.  The important discussion point here is the definition of “very old”.  Is it one year, six months, two weeks?  This decision definitely needs input from the functional business owners.

Once the definition of “very old” has been established, schedule executions of RBDCPCLR with the “Obsolete Change Pointers” box checked, and the agreed-to “very old” date entered in the date box.  This execution will remove both “processed” and “unprocessed” change pointer records up to the specified date.

Note that on the selection screen there is also a “Test Run” check box that will list which records would be selected for purging, without actually purging them.  I find this feature very useful, and I strongly recommend using it prior to actual purging.

3. “Unprocessed” change pointers that are deemed recent must be analyzed to determine why these records are configured for creation if they are never actually processed.  Perhaps, they were needed at one time, and are no longer needed now.  What I need from the functional business owners here is an agreement to turn off, in configuration, the creation of these change pointers.  Not turning them off will only continue to add unnecessary records to the change pointer tables.

After the creation of these change pointers is turned off in configuration, then schedule an execution of RBDCPCLR using the obsolete section on the selection screen.  Enter the message type in the message type box and the current date.  This will purge all change pointers of the specified message type, regardless of their status, up to the specified date.  Special care must be taken here to make sure that the selection options are entered correctly.

Note that on the selection screen there is also a “Test Run” check box that will list which records would be selected for purging, without actually purging them.  I find this feature very useful, and I strongly recommend using it prior to actual purging.

4. Compare the list of message types that are configured to create change pointers (use SAP transaction BD50), to the list of message types in RBDMIDOC executions and the list of message types RBDCPCLR executions.  Make sure that all three lists agree.  Review these lists with the functional business owners.  Make sure that what is configured is meeting the needs of the current business processes.

I also ask the functional business owners to review the selected table-field values that, when changed or created new, will trigger the creation of change pointer records (use SAP transaction BD52).  The goal, again, is to make sure that what is configured is meeting the needs of the current business processes.

Now that we have cleaned up, how do we keep the size of the change pointer tables in check?
In my next post, I will discuss items that I recommend monitoring.

Your batch job analysis shows that you are currently running the change pointer processing program RBDMIDOC,
and the change pointer cleanup program RBDCPCLR; but you notice that the size of your change pointer tables
continues to increase.

Now, it is time to analyze the change pointer data directly.  Here are some targets that I pursue:

1.  Count.
How many change pointer records are in your tables – thousands, hundreds of thousands, millions, billions?
This simple statistic alone reveals an interesting story.

While I typically try to use transaction SE11 to count the number of records in table BDCP, I have found,
on occasion, that this method would time out due to an excessively large number of records in this table.
In this case, you can write a simple ABAP which can be scheduled to run in background.

2.  Age and status.
Using the view BDCPV, determine the age and the status of your change pointer records.  The age of a change
pointer record is indicated by the field BDCPV-CRETIME, and the status of a change pointer record is indicated
by the field BDCPV-PROCESS.

In general, are the change pointer records days, weeks, months, or years old?
Are different ancient records found to be in both the ‘processed’ and ‘unprocessed’ status?
Are different recent records found to be in both the ‘processed’ and ‘unprocessed’ status?

3.  Message type and status.
Using the BDCPV view, determine the message type and status of your change pointer records.
The message type of a change pointer record is indicated by the field BDCPV-MESTYPE.

Are there different change pointer records for some message types that are never found in the ‘processed’ state?
Are there different change pointer records for a particular message type that are in both the ‘processed’ and the ‘unprocessed’ status?

Custom Reporting
For a more sohpisticated data analysis, you might consider writing an ABAP program to categorize your
change pointer records into ageing buckets, message types, and by status.  The cross-tab report shown here
is one that I use to analyze change pointer records.   It contains three sections -
Unprocessed Change Pointers by Message Type Within Ageing Category,
Processed Change Pointers by Message Type Within Ageing Category,
and a Grand Totals Summary.

Example Scenario Leading to Runaway Change Pointer Records
Of course, there are many reasons why you might observe some of the scenarios described above.  Here is one example:

Many years ago, your company’s business processes required the daily distribution of material master
data additions and changes to a remote plant.  Your functional analysts correctly configured the
change pointers and the ALE process for message type MATMAS to send this data to your remote plant.
RBDMIDOC and RBDCPCLR were both running daily for the material master data message type
MATMAS.   Change pointer records for material master data were created whenever users added
or changed certain data fields, IDOCs were created and sent, the change pointer records were marked
as ‘processed’, and were then purged – all on a daily basis.

Last year, business process changed, and your remote plant no longer required the distribution of
material master data changes.  The batch jobs running RBDMIDOC and RBDCPCLR for the material
master data message type MATMAS were cancelled, so that IDOCs would no longer be sent to the plant,
and because there was no need to clean up after these IDOCs for this message type.

But…

No one remembered to turn off the creation of change pointers in configuration for the material master
data message type MATMAS.  So, every time a user added or changed a material, new change pointer
records were still being created, causing the change pointer tables to needlessly increase in size.
And the status of these change pointer records will forever remain ‘unprocessed’.

Do you have custom code that is creating change pointer records?
There is another source of runaway change pointer table size that I would like to discuss here.    In some SAP systems that I have worked on, I have found that change pointer records were being created via custom ABAP programs.  This was done to support a custom process, having nothing to do with the SAP intended use of the change pointer records – the support of ALE distribution of master data.  Unfortunately, the system designers also forgot to write a custom program to either clean up their “custom-purposed” change pointer records, or to allow these records to be recognized by the SAP cleanup program.

Next Steps
So, now that we have discovered unnecessary change pointer records cluttering up our SAP database,
what do we do next?
What is the recovery cleanup process?
How do we stop the creation of change pointer records that are no longer needed?

This will be discussed in my next post.