ERserver
iSeries
Commitment control
Version 5 Release 3


ERserver
iSeries
Commitment control
Version 5 Release 3
 Note
Before using this information and the product it supports, be sure to read the information in
“Notices,” on page 105.

Third Edition (August 2005)

This edition applies to version 5, release 3, modification 0 of IBM Operating System/400 (5722-SS1) and to all
subsequent releases and modifications until otherwise indicated in new editions. This version does not run on all
reduced instruction set computer (RISC) models nor does it run on CISC models.
© Copyright International Business Machines Corporation 2003, 2005. All rights reserved.
US Government Users Restricted Rights – Use, duplication or disclosure restricted by GSA ADP Schedule Contract
with IBM Corp.
Contents
Commitment control . . . . . . . . . 1 Updates to the notify object . . . . . . . . 60
What’s new for V5R3 . . . . . . . . . . . 2 Manage transactions and commitment control . . . 61
Print this topic . . . . . . . . . . . . . . 2 Display commitment control information . . . 61
Commitment control concepts . . . . . . . . 2 Optimize performance for commitment control 63
How commitment control works . . . . . . . 3 Scenarios and examples: Commitment control . . . 68
How commit and rollback operations work . . . 4 Scenario: Commitment control . . . . . . . 68
Commitment definition . . . . . . . . . . 6 Practice problem for commitment control . . . 71
How commitment control works with objects . . 12 Example: Use a transaction logging file to start
Commitment control and independent disk pools 21 an application . . . . . . . . . . . . 77
Considerations and restrictions for commitment Example: Use a notify object to start an
control . . . . . . . . . . . . . . . 22 application . . . . . . . . . . . . . 81
Commitment control for batch applications . . . 24 Example: Use a standard processing program to
Two-phase commitment control. . . . . . . 24 start an application . . . . . . . . . . . 87
XA transaction support for commitment control 41 Troubleshoot transactions and commitment control 92
SQL server mode and thread-scoped transactions Commitment control errors . . . . . . . . 93
for commitment control . . . . . . . . . 45 Detect deadlocks . . . . . . . . . . . 100
Start commitment control . . . . . . . . . . 46 Recover transactions after communications
Commit notify object . . . . . . . . . . 47 failure . . . . . . . . . . . . . . . 101
Commit lock level . . . . . . . . . . . 48 When to force commits and rollbacks and when
End commitment control . . . . . . . . . . 51 to cancel resynchronization . . . . . . . . 101
System-initiated end of commitment control . . . 52 End a long-running rollback . . . . . . . 102
Commitment control during activation group end 52 Related information for commitment control . . . 103
Implicit commit and rollback operations . . . . 53
Commitment control during normal routing step Appendix. Notices . . . . . . . . . 105
end . . . . . . . . . . . . . . . . 57 Programming Interface Information . . . . . . 107
Commitment control during abnormal system or Trademarks . . . . . . . . . . . . . . 107
job end . . . . . . . . . . . . . . . 57 Terms and conditions for downloading and
Commitment control recovery during initial printing publications . . . . . . . . . . . 107
program load after abnormal end . . . . . . 58 Code example disclaimer . . . . . . . . . 108

© Copyright IBM Corp. 2003, 2005 iii

iv iSeries: Commitment control
Commitment control
Commitment control is a function that ensures data integrity. It allows you to define and process a group
of changes to resources, such as database files or tables, as a transaction. Commitment control ensures
that either the entire group of individual changes occur on all systems that participate or that none of the
changes occur. DB2 Universal Database(TM) for iSeriesTM uses the commitment control function to commit
and rollback database transactions that are running with an isolation level other than *NONE (No
Commit).

You can use commitment control to design an application so the system can restart the application if a
job, an activation group within a job, or the system ends abnormally. With commitment control, you can
have assurance that when the application starts again, no partial updates are in the database due to
incomplete transactions from a prior failure.

See the following information to get commitment control up and running on your iSeries(TM) server.

“What’s new for V5R3” on page 2

This topic lists new information for commitment control.

“Print this topic” on page 2

Print this information in its entirety.

“Commitment control concepts” on page 2

Read this information to know how commitment control works.

“Start commitment control” on page 46

Read this information to get information about starting commitment control

“End commitment control” on page 51

Read this information to know what prerequisites are necessary to end commitment control and
how to end commitment control.

“System-initiated end of commitment control” on page 52

Read tasks you need to perform when the system initiates the end of commitment control.

“Manage transactions and commitment control” on page 61

Read the tasks you need to perform to manage a system with commitment control.

“Scenarios and examples: Commitment control” on page 68

Read these scenarios and examples to see how one company sets up commitment control. Read
code examples for programs that use commitment control.

“Troubleshoot transactions and commitment control” on page 92

Read this information when you need to troubleshoot commitment control.

“Related information for commitment control” on page 103

See the topics, manuals, IBM(R) Redbooks(TM), and external Web sites that relate to commitment
control.

Note: Read the Code example disclaimer for important legal information.

© Copyright IBM Corp. 2003, 2005 1

What’s new for V5R3
For V5R3, there is an improvement and addition to commitment control. The following item contains a
summary of the improvement and addition.
v “End a long-running rollback” on page 102
End long-running rollbacks that consume critical processor time, lock resources, or take up storage
space.

How to see what is new or changed

To help you see where technical changes have been made, this information uses:
v The

image mark where new or changed information begins.

v The

image mark where new or changed information ends.

To find other information about what’s new or changed this release, see the Memo to Users.

Print this topic

To view or download the PDF version, select Commitment control (889 KB).

Saving PDF files

To save a PDF on your workstation for viewing or printing:

1. Right-click the PDF in your browser (right-click the link above).
2. Click Save Target As...
3. Navigate to the directory in which you would like to save the PDF.
4. Click Save.

Downloading Adobe Acrobat Reader

If you need Adobe Acrobat Reader to view or print these PDFs, you can download a copy from the
Adobe Web site (www.adobe.com/products/acrobat/readstep.html)

Commitment control concepts

These pages provide information to help you know how commitment control works, how it interacts
with your system, and how it interacts with other systems in your network:
v “How commitment control works” on page 3
v “How commit and rollback operations work” on page 4
v “Commitment definition” on page 6

2 iSeries: Commitment control

v “How commitment control works with objects” on page 12
v “Commitment control and independent disk pools” on page 21
v “Considerations and restrictions for commitment control” on page 22
v “Commitment control for batch applications” on page 24
v “Two-phase commitment control” on page 24
v “XA transaction support for commitment control” on page 41
v “SQL server mode and thread-scoped transactions for commitment control” on page 45

How commitment control works

Commitment control is a function that allows you to define and process a group of changes to resources,
such as database files or tables, as a transaction. Commitment control ensures that either the entire group
of individual changes occur on all systems that participate or that none of the changes occur. For
example, when you transfer funds from a savings to a checking account, more than one change occurs as
a group. To you, this transfer seems like a single change. However, more than one change occurs to the
database because both savings and checking accounts are updated. To keep both accounts accurate, either
all the changes or none of the changes must occur to the checking and savings account.

Commitment control allows you to:

v Ensure that all changes within a transaction are completed for all resources that are affected.
v Ensure that all changes within a transaction are removed if processing is interrupted.
v Remove changes that are made during a transaction when the application determines that a transaction
is in error.

You can also design an application so that commitment control can restart the application if a job, an
activation group within a job, or the system ends abnormally. With commitment control, you can have
assurance that when the application starts again, no partial updates are in the database due to incomplete
transactions from a prior failure.

Transaction

A transaction is a group of individual changes to objects on the system that appears as a single atomic
change to the user.

Note: iSeries(TM) navigator uses the term transaction, while the

character-based interface uses the term Logical Unit of
Work (LUW). The two terms are interchangeable. This
topic, unless specifically referring to the character-based
interface, uses the term transaction.

A transaction can be any of the following:

v Inquiries in which no database file changes occur.
v Simple transactions that change one database file.
v Complex transactions that changes one or more database files.
v Complex transactions that change one or more database files, but these changes represent only a part
of a logical group of transactions.
v Simple or complex transactions that involve database files at more than one location. The database files
can be:
– On a single remote system.
– On the local system and one or more remote systems.
– Assigned to more than one journal on the local system. Each journal can be thought of as a local
location.

Commitment control 3
v Simple or complex transactions on the local system that involve objects other than database files.

How commit and rollback operations work

Two operations affect changes that are made under commitment control:
v “Commit operation”
A commit operation makes permanent all changes made under commitment control since the previous
commit or rollback operation. The system also releases all locks related to the transaction.
v “Rollback operation” on page 5
A rollback operation removes all changes made since the previous commit or rollback operation. The
system also releases all locks related to the transaction.

The following programming languages and APIs support commit and rollback operations:

Language or API Commit Rollback

CL COMMIT command ROLLBACK command
(R)
ILE RPG/400 COMIT operation code ROLBK operation code
(R)
ILE COBOL/400 COMMIT verb ROLLBACK verb
(R)
ILE C/400 _Rcommit function _Rrollbck function
PL/I PLICOMMIT subroutine PLIROLLBACK subroutine
SQL COMMIT statement ROLLBACK statement
SQL Call Level Interface SQLTransact() function (Use to commit and rollback a transaction)
(CLI)
XA APIs db2xa_commit() API db2xa_rollback() API

The following links provide more information about these programming languages and APIs:
v COBOL/400 User’s Guide

(on the V5R1 Supplemental Manuals Web site)

v RPG/400 User’s Guide

(on the V5R1 Supplemental Manuals Web site)

v WebSphere(R) Development Studio: ILE C/C++ Programmer’s Guide
v CL Programming
v APIs topic
v DB2(R) UDB for iSeries(TM) SQL Call Level Interface (ODBC)
v Database programming

Commit operation
A commit operation makes permanent all changes made under commitment control since the previous
commit or rollback operation. The system also releases all locks related to the transaction.

The system performs the following steps when it receives a request to commit:
v The system saves the commit identification, if one is provided, for use at recovery time.
v The system writes records to the file before performing the commit operation if both of the following
are true:
– Records were added to a local or remote database file under commitment control.

4 iSeries: Commitment control

 – SEQONLY(*YES) was specified when the file was opened so that blocked I/O feedback is used by
the system and a partial block of records exists.
Otherwise, the I/O feedback area and I/O buffers are not changed.
v The system makes a call to the commit and rollback exit program for each API commitment resource
that is present in the “Commitment definition” on page 6. If a location has more than one exit program
registered, the system calls exit programs for that location in the order that they were registered.
v If any record changes were made to resources assigned to a journal, the system writes a C CM journal
entry to every local journal associated with the commitment definition. “Sequence of journal entries
under commitment control” on page 17 shows the entries that are typically written while a
commitment definition is active.
v The system makes permanent object-level changes that are pending.
v The system unlocks record and object locks that were acquired and kept for commitment control
purposes. Those resources are made available to other users.
v The system changes information in the commitment definition to show that the current transaction has
been ended.

The system must perform all of the previous steps correctly for the commit operation to be successful.

Rollback operation
A rollback operation removes all changes made since the previous commit or rollback operation. The
system also releases all locks related to the transaction. The system performs the following steps when it
receives a request to roll back:
v The system clears records from the I/O buffer if both of the following are true:
– If records were added to a local or remote database file under commitment control.
– If SEQONLY(*YES) was specified when the file was opened so that blocked I/O is used by the
system and a partial block of records exists that has not yet been written to the database.
Otherwise, the I/O feedback area and I/O buffers remain unchanged.
v The system makes a call to the commit or rollback exit program for each API commitment resource
that is present in the commitment definition. If a location has more than one exit program registered,
the system calls the exit programs for that location in reverse order from the order in which they were
registered.
v If a record was deleted from a file, the system adds the record back to the file.
v The system removes any changes to records that have been made during this transaction, and places
the original records (the before-images) back into the file.
v If any records were added to the file during this transaction, they remain in the file as deleted records.
v If any record changes were made to resources assigned to a journal during the transaction, the system
adds a journal entry of C RB to the journal indicating that a rollback operation occurred. The journal
also contains images of the record changes that were rolled back. Before the rollback operation was
requested, the before-images and after-images of changed records were placed in the journal. The
system also writes C RB entry to the default journal if any committable resources are assigned to that
journal.
v The system positions the open files under commitment control at one of the following positions:
– The last record accessed in the previous transaction
– At the open position if no commit operation has been performed for the file using this commitment
definition
This consideration is important if you are doing sequential processing.
v The system does not roll back noncommittable changes for database files. For example, opened files are
not closed, and cleared files are not restored. The system does not reopen or reposition any files that
were closed during this transaction.
v The system unlocks record locks that were acquired for commitment control purposes and makes those
records available to other users.
Commitment control 5
v The commit identification currently saved by the system remains the same as the commit identification
provided with the last commit operation for the same commitment definition.
v The system reverses or rolls back object-level committable changes made during this transaction.
v Object locks that were acquired for commitment control purposes are unlocked and those objects are
made available to other users.
v The system establishes the previous commitment boundary as the current commitment boundary.
v The system changes information in the commitment definition to show that the current transaction has
been ended.

The system must perform all of the previous steps correctly for the rollback operation to be successful.

Commitment definition
You create a commitment definition when you use the Start Commitment Control (STRCMTCTL)
command to start commitment control on your system. Also, DB2(R) UDB for iSeries(TM) automatically
creates a commitment definition when the isolation level is other than No Commit. The commitment
definition contains information that pertains to the resources that are being changed under commitment
control within that job. The system maintains the commitment control information in the commitment
definition as the commitment resources change, until the commitment definition is ended. Each active
transaction on the system is represented by a commitment definition. A subsequent transaction can reuse
a commitment definition after each commit or rollback of an active transaction.

A commitment definition generally includes:

v The parameters on the STRCMTCTL command.
v The current status of the commitment definition.
v Information about database files and other committable resources that contain changes that are made
during the current transaction.

For commitment definitions with job-scoped locks, only the job that starts commitment control knows
that commitment definition. No other job knows that commitment definition.

Programs can start and use multiple commitment definitions. Each commitment definition for a job
identifies a separate transaction that has committable resources associated with it. These transactions can
be committed or rolled back independently from transactions that are associated with other commitment
definitions that are started for the job.

The following provide more details about commitment definitions:

v “Scope for a commitment definition”
v “Commitment definition names” on page 9
v “Example: Jobs and commitment definitions” on page 10

For rules about commitment definitions and independent disk pools see “Commitment control and
independent disk pools” on page 21.

Scope for a commitment definition

The scope of a commitment definition determines which programs will use that commitment definition,
and how locks acquired during transactions are scoped. The interface that starts the commitment
definition determines the scope of the commitment definition. There are four possible scopes for a
commitment definition, which fall under two general categories:

Commitment definitions with job-scoped locks

v Activation-group-level commitment definition
v Job-level commitment definition

6 iSeries: Commitment control

 v Explicitly named commitment definition

Commitment definitions with transaction-scoped locks.

v Transaction scoped commitment definition

Commitment definitions with job-scoped locks can be used only by programs that run in the job that
started the commitment definitions. In comparison, more than one job can use commitment definitions
with transaction-scoped locks.

Applications typically use either activation-group-level or job-level commitment definitions. These

commitment definitions are created either explicitly with the Start Commitment Control (STRCMTCTL)
command, or implicitly by the system when an SQL application runs with an isolation level other than
*NONE.

Activation-group-level commitment definition

The most common scope is to the activation group. The activation-group-level commitment definition is
the default scope when the STRCMTCTL command explicitly starts the commitment definition, or an SQL
application that runs with an isolation level other than No Commit implicitly starts the commitment
definition. Only programs that run within that activation group use that commitment definition. Many
activation-group-level commitment definitions can be active for a job at one time. However, each
activation-group-level commitment definition can be associated only with a single activation group. The
programs that run within that activation group can associate their committable changes only with that
activation-group-level commitment definition.

When iSeries(TM) Navigator, the Work with Commitment Definitions (WRKCMTDFN) command, the
Display Job (DSPJOB) command, or the Work with Job (WRKJOB) command display an
activation-group-level commitment definition, these fields display the following:
v The commitment definition field displays the name of the activation group. It shows the special value
*DFTACTGRP to indicate the default activation group.
v The activation group field displays the activation group number.
v The job field displays the job that started the commitment definition.
v The thread field displays *NONE.

Job-level commitment definition

A commitment definition can be scoped to the job only by issuing STRCMTCTL CMTSCOPE(*JOB). Any
program running in an activation group that does not have an activation-group-level commitment
definition started uses the job-level commitment definition, if it has already been started by another
program for the job. You can only start a single job-level commitment definition for a job.

When iSeries Navigator, the Work with Commitment Definitions (WRKCMTDFN) command, the Display
Job (DSPJOB) command, or the Work with Job (WRKJOB) command display a job-level commitment
definition, these fields display the following:
v The commitment definition field displays the special value *JOB.
v The activation group field displays a blank.
v The job field displays the job that started the commitment definition.
v The thread field displays *NONE.

For a given activation group, the programs that run within that activation group can use only a single
commitment definition. Therefore programs that run within an activation group can either use the
job-level or the activation-group-level commitment definition, but not both at the same time. In a
multi-threaded job that does not use SQL server mode, transactional work for a program will be scoped
to the appropriate commitment definition with respect to the activation group of the program, regardless

Commitment control 7
of which thread performs it. If multiple threads use the same activation group, they must cooperate to
perform the transactional work and ensure that commits and rollbacks occur at the correct time.

Even when the job-level commitment definition is active for the job, a program can still start the
activation-group-level commitment definition if no program running within that activation group has
performed any commitment control requests or operations for the job-level commitment definition.
Otherwise, you must first end the job-level commitment definition before you can start the
activation-group-level commitment definition. Commitment control requests or operations for the
job-level commitment definition that can prevent the activation-group-level commitment definition from
being started include:
v Opening (full or shared) a database file under commitment control.
v Using the Add Commitment Resource (QTNADDCR) API to add an API commitment resource.
v Committing a transaction.
v Rolling back a transaction.
v Adding a remote resource under commitment control.
v Using the Change Commitment Options (QTNCHGCO) API to changing commitment options.
v Bringing the commitment definition to a rollback required state using the Rollback Required
(QTNRBRQD) API.
v Sending a user journal entry that includes the current commit cycle identifier by using the Send
Journal Entry (QJOSJRNE) API with the Include Commit Cycle Identifier parameter.

Likewise, if the programs within an activation group are currently using the activation-group-level
commitment definition, the commitment definition must first be ended before programs running within
that same activation group can use the job-level commitment definition.

When opening a database file, the open scope for the opened file can be either to the activation group or
to the job with one restriction: if a program is opening a file under commitment control and the file is
scoped to the job, then the program making the open request must use the job-level commitment
definition.

Explicitly-named commitment definition

Explicitly-named commitment definitions are started by the system when it needs to perform its own
commitment control transactions without affecting any transactions used by an application. An example
of a function that starts these types of commitment definitions is the problem log. An application cannot
start explicitly named commitment definitions.

When iSeries Navigator, the Work with Commitment Definitions (WRKCMTDFN) command, the Display
Job (DSPJOB) command, or the Work with Job (WRKJOB) command display an explicitly-named
commitment definition, these fields display the following:
v The commitment definition field displays the name given to it by the system.
v The activation group field displays a blank.
v The job field displays the job that started the commitment definition.
v The thread field displays *NONE.

Transaction-scoped commitment definitions

Transaction-scoped commitment definitions are started with the XA APIs for Transaction Scoped Locks.

These APIs use commitment control protocols that are thread based or SQL connection based, and not
activation group based. In other words, the APIs are used to associate the commitment definition with a
particular thread or SQL connection while the transactional work is performed, and to commit or rollback

8 iSeries: Commitment control

the transactions. The system attaches these commitment definitions to the threads that perform the
transactional work, with respect to the API protocols. They can be used by threads in different jobs.

When iSeries Navigator, the Work with Commitment Definitions (WRKCMTDFN) command, the Display
Job (DSPJOB) command, or the Work with Job (WRKJOB) command display a transaction-scoped
commitment definition, these fields display the following:
v The commitment definition field displays the special value *TNSOBJ.
v The activation group field displays a blank.
v The job field displays the job that started the commitment definition.
v The thread field displays the thread to which the commitment definition is attached (or *NONE if the
commitment definition is not currently attached to any thread).

Commitment definition names

The system gives names to all commitment definitions that are started for a job. The following table
shows various commitment definitions and their associated names for a particular job.

Activation group Commit scope Commitment definition name

Any Job *JOB
Default activation group Activation group *DFTACTGRP
User-named activation group Activation group Activation group name (for example,
PAYROLL)
System-named activation group Activation group Activation group number (for
example. 0000000145)
None Explicitly named QDIR001 (example of a
system-defined commitment
definition for system use only).
System-defined commitment
definition names begin with Q.
None Transaction *TNSOBJ

Only Integrated Language Environment(R) (ILE) compiled programs can start commitment control for
activation groups other than the default activation group. Therefore, a job can use multiple commitment
definitions only if the job is running one or more ILE compiled programs. For more information about
the Integrated Language Environment(R), see the Programming topic in the information center.

Original Program Model (OPM) programs run in the default activation group, and by default use the
*DFTACTGRP commitment definition. In a mixed OPM and ILE environment, jobs must use the job-level
commitment definition if all committable changes made by all programs are to be committed or rolled
back together.

An opened database file scoped to an activation group can be associated with either an
activation-group-level or job-level commitment definition. An opened database file scoped to the job can
be associated only with the job-level commitment definition. Therefore, any program, OPM or ILE, that
opens a database file under commitment control scoped to the job needs to use the job-level commitment
definition.

Application programs do not use the commitment definition name to identify a particular commitment
definition when making a commitment control request. Commitment definition names are primarily used
in messages to identify a particular commitment definition for a job.

Commitment control 9
For activation-group-level commitment definitions, the system determines which commitment definition
to use based on which activation group the requesting program is running in. This is possible because the
programs that run within an activation group at any point in time can only use a single commitment
definition.

For transactions with transaction-scoped locks, the XA APIs and the transaction related attributes added
to the CLI determine which commitment definition the invoking thread uses.

Example: Jobs and commitment definitions

The following figure shows an example of a job that uses multiple commitment definitions. It indicates
which file updates are committed or rolled back at each activation group level. The example assumes that
all of the updates that are made to the database files by all of the programs are made under commitment
control.

10 iSeries: Commitment control

Commitment control 11
Link to description

The following table shows how files are committed or rolled back if the scenario in the previous figure
changes:

Additional examples of multiple commitment definitions in a job

Effect on changes to these files:

Change in scenario F1 and F2 F3 and F4 F5 and F6 F7
PGMX performs a Still pending Rolled back Already committed Rolled back
rollback operation
instead of a commit
operation (3=
=COMMIT becomes
ROLLBACK).
PGMZ performs a Still pending Committed by PGMZ Already committed Committed
commit operation
before returning to
PGMX.
PGMZ attempts to Still pending Still pending Already committed Still pending
start commitment
control specifying
CMTSCOPE(*ACTGRP)
after updating file F7.
The attempt fails
because changes are
pending using the
job-level commitment
definition.
PGMX does not start Still pending Not under Already committed File F7 cannot be
commitment control commitment control opened because no
and does not open *JOB commitment
files F3 and F4 with definition would exist
COMMIT(*YES). (PGMX did not create
PGMZ attempts to it).
open file F7 with
COMMIT(*YES).

How commitment control works with objects

When you place an object under commitment control, it becomes a committable resource. It is registered
with the commitment definition. It participates in each commit operation and rollback operation that
occurs for that commitment definition.

The topics that follow describe these attributes of a committable resource:

v Resource type
v Location
v Commit protocol
v Access intent

The following links have more information about resources under commitment control:
v “Types of committable resources” on page 13
v “Local and remote committable resources” on page 14
v “Access intent of a committable resource” on page 15

12 iSeries: Commitment control

v “The commit protocol of a committable resource” on page 16
v “Journaled files and commitment control” on page 16
v “Sequence of journal entries under commitment control” on page 17
v “Commit cycle identifier” on page 19
v “Record locking” on page 19

Types of committable resources

The following table shows:
v The types of committable resources.
v How they are placed under commitment control.
v How they are removed from commitment control.
v Restrictions that apply to the resource type.

Resource type How to place it under How to remove it from What kinds of Restrictions
commitment control commitment control changes are
committable
FILE- local Opening under Closing the file, if no Record level changes No more than
database files commitment control1 changes are pending. 500 000 000 records
can be locked for a
single transaction2.
If changes are pending
when the file is closed,
after performing the next
commit or rollback
operation.
DDL- Running SQL under Performing a commit or Object level changes, Only object-level
object-level commitment control rollback operation after such as: changes made using
changes to the object-level change. v Create SQL SQL are under
local SQL Package commitment control.
tables and SQL
v Create SQL Table
collections.
v Drop SQL Table
DDM- remote Opening under Closing the file, if no Record level changes
distributed commitment control. changes are pending.
data Commitment control
management support for DDM has
(DDM) file more information about If changes are pending
commitment control and when the file is closed,
distributed data after performing the next
management. commit or rollback
operation.
LU 6.2- Starting the Ending the conversation
protected conversation3
conversation
DRDA(R)- Using SQL CONNECT Ending the connection
distributed statement
relational
database

Commitment control 13
Resource type How to place it under How to remove it from What kinds of Restrictions
commitment control commitment control changes are
committable
API- local API Add Commitment Remove Commitment The user program The application must
commitment Resource (QTNADDCR) Resource (QTNRMVCR) determines this. provide an exit
resource API API Journal entries may program to be called
be written by the during commit,
user program using rollback, or
the Send Journal resynchronization
Entry (QJOSJRNE) operations.
API to assist with
tracking these
changes.
TCP-TCP/IP Using SQL CONNECT Ending the SQL
connection statement to an RDB connection, or closing the
defined to use TCP/IP DDM file if no changes
connections, or opening are pending. If the DDM
a DDM file defined with file is closed with changes
a TCP/IP location pending, the connection is
closed after performing
the next commit or
rollback operation.
Notes:
1
For details on how to place a database file under commitment control, see the appropriate language reference
manual. “Related information for commitment control” on page 103 links to language manuals that you can
use.
2
You can use a QAQQINI file to reduce the limit of 500 000 000. See “Manage transaction size” on page 67
for instructions.
3
When a DDM connection is started, the DDM file specifies PTCCNV(*YES), and the DDM file is defined with
an SNA remote location, an LU6.2 resource is added with the DDM resource.
When a DRDA connection is started, an LU6.2 resource is added with the DRDA resource if both of the
following are true:
v The program is using the distributed unit of work connection protocols.
v The connection is to an RDB defined with an SNA remote location. For more information about starting
protected conversions, see APPC Programming.

Local and remote committable resources

A committable resource can be either a local resource or a remote resource.

Local committable resource

A local committable resource resides on the same system as the application. Each journal associated with
resources under commitment control can be thought of as a local location. All the resources that are
registered without a journal (optionally for both DDL resources and API resources) can be thought of as a
separate local location.

If a committable resource resides on an independent disk pool, the resource is not considered local if the
commitment definition resides on a different disk pool. See “Commitment control and independent disk
pools” on page 21 for more information about committable resources and independent disk pools.

Remote committable resources

A remote committable resource resides on a different system from the application. A remote location
exists for each unique conversation to a remote system. A commitment definition may have one or more
remote locations on one or more remote systems.
14 iSeries: Commitment control
When you place a local resource under commitment control for the system disk pool, or any independent
disk pool, you must use DRDA(R) to access resources under commitment control in any other
independent disk pool.

The following shows the types of committable resources and their locations:

Resource type Location

FILE Local
DDL Local
API Local
DDM Remote
LU62 Remote
DRDA Local or remote
TCP Remote

Access intent of a committable resource

When a resource is placed under commitment control, the resource manager indicates how the resource
will be accessed:
v Update
v Read-only
v Undetermined

The access intent determines how the resources participate together in a transaction. The following table
shows what access intents are possible for a particular type of resource and how the system determines
the access intent for a resource when it is registered:

Resource type Possible access intents How the access intent is determined
FILE Update, read-only Based on how the file was opened
DDL Update Always update
API Update Always update
DDM Update, read-only Based on how the file was opened
LU62 Undetermined Always undetermined
(R)
DRDA Update, read-only, undetermined For DRDA Level 1, the access intent is
update if no other remote resources are
registered. Otherwise, the access intent is
read-only. For DRDA Level 2, the access
intent is always undetermined.
TCP Undetermined Always undetermined

The access intent of resources that are already registered determines whether a new resource can be
registered. The following rules apply:
v A one-phase resource whose access intent is update cannot be registered when any of the following is
true:
– Resources whose access intent is update are already registered at other locations.
– Resources whose access intent is undetermined are already registered at other locations.
– Resources whose access intent is undetermined are already registered at the same location and the
resources have been changed during the current transaction.

Commitment control 15
v A two-phase resource whose access intent is update cannot be registered when a one-phase resource
whose access intent is update is already registered.

The commit protocol of a committable resource

Commit protocol is the capability a resource has to participate in one-phase or two-phase commit
processing. Local resources, except API committable resources, are always two-phase resources.

If a committable resource resides on an independent disk pool, the resource is not considered local a local
resource or a two-phase resource if the commitment definition resides on a different disk pool. See
“Commitment control and independent disk pools” on page 21 for more information about committable
resources and independent disk pools.

A two-phase resource is also called a protected resource. Remote resources and API committable
resources must be registered as one-phase resources or two-phase resources when they are placed under
commitment control. The following table shows what types of committable resources can coexist in a
commitment definition with a one-phase resource:

Resource type Can coexist with

One-phase API resource Other local resources. No remote resources.
One-phase remote resource Other one-phase resources at the same location. No local
resources.

Journaled files and commitment control

You must journal (log) a database file (resource type FILE or DDM) before it can be opened for output
under commitment control or referenced by an SQL application that uses an isolation level other than No
Commit. A file does not need to be journaled in order to open it for input only under commitment
control. An error occurs if:
v An attempt is made to open a database file for output under commitment control, but the file is not
currently journaled.
v No commitment definition is started that can be used by the file being opened under commitment
control.

If only the after images are being journaled for a database file when that file is opened under
commitment control, the system automatically starts journaling both the before and after images. The
before images are written only for changes to the file that occur under commitment control. If other
changes that are not under commitment control occur to the file at the same time, only after-images are
written for those changes.

The system automatically writes record-level committable changes and object-level committable changes
to a journal. For record-level changes, the system then uses the journal entries, if necessary, for recovery
purposes; the system does not use entries from object-level committable changes for recovery purposes.
Furthermore, the system does not automatically write journal entries for API commitment resources.
However, the exit program for the API resource can use the Send Journal Entry (QJOSJRNE) API to write
journal entries to provide an audit trail or to assist with recovery. The content of these entries is
controlled by the user exit program.

The system uses a technique other than a journal to perform recovery for object-level commitment
resources. Recovery for API commitment resources is accomplished by calling the commit and rollback
exit program associated with each particular API commitment resource. The exit program has the
responsibility for performing the actual recovery that is necessary for the situation.

For more information about journaling, see the Journal management topic.

16 iSeries: Commitment control

Sequence of journal entries under commitment control
The following table shows the sequence of entries that are typically written while a commitment
definition is active. You can use the Journal entry information finder to get more information about the
contents of the journal entries.

Commitment control entries are written to a local journal if at least one of the following is true:
v The journal is specified as the default journal on the Start Commitment Control (STRCMTCTL)
command.
v At least one file journaled to the journal is opened under commitment control.
v At least one API commitment resource associated with the journal is registered under commitment
control.

Entry type Description Where it is written When it is written

C BC Begin commitment control To the default journal, if When the STRCMTCTL
one is specified on the command is used.
STRCMTCTL command.
To the journal. When the first file
journaled to a journal is
opened or when an API
resource is registered for a
journal.
C SC Start commit cycle To the journal. When the first record
change occurs for the
transaction for a file
journaled to this journal1.
To the journal for an API When the QJOSJRNE API is
resource. first used with the Include
Commit Cycle Identifier key.
Journal codes D and F DDL object level entries To the journal associated When updates occur.
with the object being
updated. Only journal
entries that contain a
commit cycle identifier
represent a DDL object
level change that is part of
the transaction.
Journal code R Record level entries To the journal associated When the updates occur.
with the file being updated.
Journal code U User-created entries To the journal associated If the application program
with an API resource. uses the QJOSJRNE API is
first used with the Include
Commit Cycle Identifier key.
C CM Commit To the journal. When the commit has
completed successfully.
To the default journal. If any committable
resources are associated
with the journal.
C RB Rollback To the journal. After the rollback operation
has completed.
To the default journal. If any committable
resources are associated
with the journal.

Commitment control 17
Entry type Description Where it is written When it is written
C LW End transaction To the default journal, if When the commit or
one is specified on the rollback operation has
STRCMTCTL command. completed.
The system writes an LW
header record and one or
more detail records. These
entries are written only if
OMTJRNE(*NONE) is
specified on the
STRCMTCTL command or
if a system error occurs.
C EC End commitment control To the journal. When the End
Commitment Control
(ENDCMTCTL) command
is completed.
To a local journal that is not When a commit boundary
the default journal. is established, following the
point when all committable
resources associated with
that journal have been
removed from commitment
control.
Start of savepoint or nested To the journal. When the application
commit cycle. creates an SQL
C SB SAVEPOINT, or when the
system creates an internal
nested commit cycle to
handle a series of database
functions as a single
operation2.

Release of savepoint or To the journal. When the application

commit of nested commit releases an SQL
C SQ cycle. SAVEPOINT, or when the
system commits an internal
nested commit cycle2.

Rollback of savepoint or To the journal. When the application rolls

nested commit cycle. back an SQL SAVEPOINT,
C SU or when the system rolls
back an internal nested
commit cycle2.

18 iSeries: Commitment control

Entry type Description Where it is written When it is written
Note:
1
You can specify that the fixed-length portion of the journal entry includes transaction information by specifying the
Logical Unit of Work (*LUW) value for the Fixed-Length Data (FIXLENDTA) parameter of the Create Journal
(CRTJRN) or Change Journal (CHGJRN) command. By specifying the FIXLENDTA (*LUW) parameter, the
fixed-length portion of each C SC journal entry will contain the Logical Unit of Work ID (LUWID) of the current
transaction. Likewise for XA transactions, if you specify the FIXLENDTA(*XID) parameter, the fixed-length portion
of each C SC journal entry will contain the XID of the current transaction. The LUWID or XID can help you find all
the commit cycles for a particular transaction if multiple journals or systems are involved in the transaction.
2
These entries are sent only if you set the QTN_JRNSAVPT_MYLIB_MYJRN environment variable to *YES where
MYJRN is the journal you are using and MYLIB is the library the journal is stored in. Special value *ALL is
supported for the MYLIB and MYJRN values. You can set these variables system-wide or for a specific job. To have
the entries sent for journal MYLIB/MYJRN for just one job, use this command in that job:
v ADDENVVAR ENVVAR(QTN_JRNSAVPT_MYLIB_MYJRN) VALUE(*YES)

To have entries sent for all journals for all jobs:

v ADDENVVAR ENVVAR(’QTN_JRNSAVPT_*ALL_*ALL’) VALUE(*YES) LEVEL(*SYS)
You need to set the environment variable before you start commitment control.

Commit cycle identifier

A commit cycle is the time from one commitment boundary to the next. The system assigns a commit
cycle identifier to associate all of the journal entries for a particular commit cycle together. Each journal
that participates in a transaction has its own commit cycle and its own commit cycle identifier.

The commit cycle identifier is the journal sequence number of the C SC journal entry written for the
commit cycle. The commit cycle identifier is placed in each journal entry written during the commit cycle.
If more than one journal is used during the commit cycle, the commit cycle identifier for each journal is
different.

You can specify that the fixed-length portion of the journal entry includes transaction information by
specifying the Logical Unit of Work (*LUW) value for the Fixed-Length Data (FIXLENDTA) parameter of
the Create Journal (CRTJRN) or Change Journal (CHGJRN) command. By specifying the FIXLENDTA
(*LUW) parameter, the fixed-length portion of each C SC journal entry will contain the Logical Unit of
Work ID (LUWID) of the current transaction. Likewise for XA transactions, if you specify the
FIXLENDTA (*XID) parameter, the fixed-length portion of each C SC journal entry will contain the XID of
the current transaction. The LUWID or XID can help you find all the commit cycles for a particular
transaction if multiple journals or systems are involved in the transaction.

You can use the Send Journal Entry (QJOSJRNE) API to write journal entries for API resources. You have
the option of including the commit cycle identifier on those journal entries.

You can use the commit cycle identifier to apply or remove journaled changes to a commitment boundary
using the Apply Journaled Changes (APYJRNCHG) command or the Remove Journaled Changes
(RMVJRNCHG) command. These limitations apply:
v Most object-level changes made under commitment control are written to the journal but are not
applied or removed using the APYJRNCHG and RMVJRNCHG commands.
v The QJOSJRNE API writes user-created journal entries with a journal code of U. These entries cannot be
applied or removed using the APYJRNCHG and RMVJRNCHG commands. They must be applied or
removed with a user-written program.

Record locking
When a job holds a record lock and another job attempts to retrieve that record for update, the requesting
job waits and is removed from active processing until one of the following occurs:
v The record lock is released.

Commitment control 19
v The specified wait time ends.

More than one job can request a record to be locked by another job. When the record lock is released, the
first job to request the record receives that record. When waiting for a locked record, specify the wait
time in the WAITRCD parameter on the following create, change, or override commands:
v Create Physical File (CRTPF)
v Create Logical File (CRTLF)
v Create Source Physical File (CRTSRCPF)
v Change Physical File (CHGPF)
v Change Logical File (CHGLF)
v Change Source Physical File (CHGSRCPF)
v Override Database File (OVRDBF)

When specifying wait time, consider the following:

v If you do not specify a value, the program waits the default wait time for the process.
v For commitment definitions with transaction-scoped locks only, the job default wait time can be
overridden by a transaction lock-wait time that can be specified on:
– The xa_open API.
– A JDBC or JTA interface. Distributed transactions lists these APIs.
v If the record cannot be allocated within the specified time, a notify message is sent to the high-level
language program.
v If the wait time for a record is exceeded, the message sent to the job log gives the name of the job
holding the locked record that caused the requesting job to wait. If you experience record lock
exceptions, you can use the job log to help determine which programs to alter so they will not hold
locks for long durations.

Programs keep record locks over long durations for one of the following reasons:
v The record remains locked while the workstation user is considering a change.
v The record lock is part of a long commitment transaction. Consider making smaller transactions so a
commit operation can be performed more frequently.
v An undesired lock has occurred. For example, assume a file is defined as an update file with unique
keys, and the program updates and adds additional records to the file. If the workstation user wants to
add a record to the file, the program may attempt to access the record to determine whether the key
already exists. If it does, the program informs the workstation user that the request made is not valid.
When the record is retrieved from the file, it is locked until it is implicitly released by another read
operation to the same file, or until it is explicitly released.

Note: For more information about how to use each high-level

language interface to release record locks, see the
appropriate high-level language reference manual.
“Related information for commitment control” on page
103 has links to high-level language manuals that you can
use with commitment control.

The duration of the lock is much longer if LCKLVL(*ALL) is specified because the record that was
retrieved from the file is locked until the next commit or rollback operation. It is not implicitly released
by another read operation and cannot be explicitly released.

Another function that can put a lock on a file is the save-while-active function. The Save your server
while it is active topic has more information about the save-while-active function.

20 iSeries: Commitment control

Commitment control and independent disk pools
Independent disk pools and independent disk pool groups, can each have a separate OS/400(R) SQL
database. You can use commitment control with these databases. However, because each independent
disk pool or independent disk pool group has a separate SQL database, follow these considerations.
v “Independent disk pool considerations for commitment definitions”
v “Considerations for XA transactions” on page 22

Independent disk pool considerations for commitment definitions

You must be aware of the following are considerations for commitment definitions when you use
independent disk pools.

QRECOVERY library considerations

When you start commitment control, the “Commitment definition” on page 6 is created in the
QRECOVERY library. Each independent disk pool or independent disk pool group has its own version of
a QRECOVERY library. On an independent disk pool, the name of the QRECOVERY library is
QRCYxxxxx, where xxxxx is the number of the independent disk pool. For example, the name of the
QRECOVERY library for independent disk pool 39 is QRCY00039. Furthermore, if the independent disk
pool is part of a disk pool group, only the primary disk pool has a QRCYxxxxx library.

When you start commitment control, the commitment definition is created in the QRECOVERY library of
the independent disk pool that is associated with that job, making commitment control active on the
independent disk pool.

Set ASP Group considerations

Using the Set ASP Group (SETASPGRP) command while commitment control is active on an independent
disk pool has the following effects:
v If you switch from an independent disk pool, and resources are registered with commitment control on
the disk pool, the SETASPGRP command fails with message CPDB8EC, reason code 2, ″The thread has
an uncommitted transaction.″ This message if followed by message CPFB8E9.
v If you switch from an independent disk pool and no resources are registered with commitment control,
the commitment definitions are moved to the independent disk pool to which you are switching.
v If you switch from the system disk pool (ASP group *NONE), commitment control is not affected. The
commitment definitions stay on the system disk pool.
v If you use a notify object, the notify object must reside on the same independent disk pool or
independent disk pool group as the commitment definition.
v If you move the commitment definition to another independent disk pool or independent disk pool
group, the notify object must also reside on that other independent disk pool or independent disk pool
group. The notify object on the other independent disk pool or independent disk pool group is
updated if the commitment definition ends abnormally. If the notify object is not found on the other
independent disk pool or independent disk pool group, the update fails with message CPF8358.

Default journal considerations

The following are default journal considerations:

v If you use the default journal, the journal must reside on the same independent disk pool or
independent disk pool group as the commitment definition.
v If the default journal is not found on the other independent disk pool or independent disk pool group
when commitment control starts, the commitment control start fails with message CPF9873.
v If you move the commitment definition to another independent disk pool or independent disk pool
group, the default journal must also reside on that other independent disk pool or independent disk

Commitment control 21
 pool group. If the journal is not found on the other independent disk pool or independent disk pool
group, the commitment definition is moved, but no default journal is used from this point on.

IPL and vary off considerations

The following are IPL and vary off considerations:

v Recovery of commitment definitions residing on an independent disk pool is performed during
independent disk pool vary on processing and is similar to IPL recovery.
v Commitment definitions in an independent disk pool are not recovered during the system IPL.
v The vary off of an independent disk pool has the following effects on commitment definitions:
– Jobs associated with the independent disk pool end.
– No new commitment definitions are allowed to be created on the independent disk pool.
– Commitment definitions residing on the independent disk pool become unusable.
– Commitment definitions residing on the independent disk pool, but not attached to a job, release
transaction scoped locks.

Remote database considerations

The following are remote database considerations:

v You cannot use an LU6.2 SNA connection (protected conversations or Distributed Unit of Work
(DUW)) to connect to a remote database from an independent disk pool database. You can use
unprotected SNA conversations to connect from an independent disk pool database to a remote
database.
v When commitment control is active for a job or thread, access to data outside the independent disk
pool or disk pool group to which the commitment definition belongs is only possible remotely, as if it
were data that resides on another system. When you issue an SQL CONNECT statement to connect to
the relational database (RDB) on the independent disk pool, the system make the connection a remote
connection.
v The system disk pool and basic disk pools do not require a remote connection for read-only access to
data that resides on an independent disk pool. Likewise, an independent disk pool does not require a
remote connection for read-only access to data that resides on the system disk pool or a basic disk
pool.

Considerations for XA transactions

In the “XA transaction support for commitment control” on page 41, each database is considered a
separate resource manager. When a transaction manager wants to access two databases under the same
transaction, it must use the XA protocols to perform two-phase commit with the two resource managers.

Since each independent disk pool is a separate SQL database, in the XA environment each independent
disk pool is also considered a separate resource manager. For an application server to perform a
transaction which targets two different independent disk pools, the transaction manager must also use a
two-phase commit protocol.

For more information about independent disk pools, see the Independent disk pools topic.

Considerations and restrictions for commitment control

The following are miscellaneous considerations and restrictions for commitment control:

Database file considerations

v If you specify that a shared file be opened under commitment control, all subsequent uses of that file
must be opened under commitment control.

22 iSeries: Commitment control

v If SEQONLY(*YES) is specified for the file opened for read-only with LCKLVL(*ALL) (either implicitly
or by a high-level language program, or explicitly by the Override with Database File (OVRDBF)
command), then SEQONLY(*YES) is ignored and SEQONLY(*NO) is used.
v Record-level changes made under commitment control are recorded in a journal. These changes can be
applied to or removed from the database with the Apply Journaled Changes (APYJRNCHG) command
or the Remove Journaled Changes (RMVJRNCHG) command.
v Both before-images and after-images of the files are journaled under commitment control. If you
specify only to journal the after-images of the files, the system also automatically journals the
before-image of the file changes that occurred under commitment control. However, because the
before-images are not captured for all changes made to the files, you cannot use the RMVJRNCHG
command for these files.

Considerations for object and record level changes

v
v Object level and record level changes made under commitment control using SQL will use the
commitment definition that is currently active for the activation group that the requesting program is
running in. If neither the job level nor the activation-group-level commitment definition is active, SQL
will start an activation-group-level commitment definition. For more information regarding changes
made under commitment control using SQL, see the SQL Programming Concepts topic.

One-phase and two-phase commit considerations

v While a one-phase remote conversation or connection is established, remote conversations or
connections to other locations are not allowed. If a commitment boundary is established and all
resources are removed, the location can be changed.
v If you are using two-phase commit, you do not need to use the Submit Remote Command
(SBMRMTCMD) command to start commitment control or perform any other commitment control
operations at the remote locations. The system performs these functions for you.
v For a one-phase remote location, the COMMIT and ROLLBACK CL commands will fail if SQL is in the
call stack and the remote relational database is not on a system. If SQL is not on the call stack, the
COMMIT and ROLLBACK commands will not fail.
v For a one-phase remote location, commitment control must be started on the source system before
making committable changes to remote resources. The system automatically starts commitment control
for distributed database SQL on the source system at connection time if the SQL program is running
with the commitment control option other than *NONE. When the first remote resource is placed
under commitment control, the system starts commitment control on the target system.

Save consideration

A save operation is prevented if the job performing the save has one or more active commitment
definitions with any of the following types of committable changes:
v A record change to a file that resides in the library being saved. For logical files, all the related physical
files are checked.
v Any object-level changes within a library that is being saved.
v Any API resource that was added using the Add Commitment Resource (QTNADDCR) API and with
the Allow normal save processing field set to the default value of N.

This prevents the save operations from saving to the save media changes that are due to a partial
transaction.

Note: If you use the new save with partial transactions feature the object can be saved without ending a
commitment definition.

Commitment control 23
Object locks and record locks prevent pending changes from commitment definitions in other jobs from
being saved to the save media. This is true only for API commitment resources if locks are acquired when
changes are made to the object or objects associated with the API commitment resource.

Miscellaneous considerations and restrictions

v Before upgrading your system to a new release, all pending resynchronizations must either be
completed or canceled. See the Ensure two-phase commit integrity before you install software topic for
more details.
v The COMMIT and ROLLBACK values are shown on the WRKACTJOB Function field during a commit
or rollback. If the Function remains COMMIT or ROLLBACK for a long time, one of the following
situations might have occurred:
– A resource failure during the commit or rollback requires resynchronization. Control will not return
to the application until the resynchronization completes or is canceled.
– This system voted read-only during the commit. Control will not return to the application until the
system that initiated the commit sends data to this system.
– This system voted OK to leave out during the commit. Control will not return to the application
until the system that initiated the commit sends data to this system.

Commitment control for batch applications

Batch applications may or may not need commitment control. In some cases, a batch application can
perform a single function of reading an input file and updating a master file. However, you can use
commitment control for this type of application if it is important to start it again after an abnormal end.

The input file is an update file with a code in the records to indicate that a record was processed. This
file and any files updated are placed under commitment control. When the code is present in the input
file, it represents a completed transaction. The program reads through the input file and bypasses any
records with the completed code. This allows the same program logic to be used for normal and starting
again conditions.

If the batch application contains input records dependent on one another and contains switches or totals,
a “Commit notify object” on page 47 can be used to provide information about starting again. The values
held in the notify object are used to start processing again from the last committed transaction within the
input file.

If input records are dependent on one another, they can be processed as a transaction. A batch job can
lock a maximum of 500 000 000 records. You can reduce this limit by using a Query Options File
(QAQQINI). Use the QRYOPTLIB parameter of the Change Query Attributes (CHGQRYA) command to
specify a Query Options File for a job to use. Use the COMMITMENT_CONTROL_LOCK_LEVEL value
in the Query Options File as the lock limit for the job.

Any commit cycle that exceeds 2000 locks will probably slow down system performance noticeably.
Otherwise, the same locking considerations exist as for interactive applications, but the length of time
records are locked in a batch application may be less important than in interactive applications.

Two-phase commitment control

Two-phase commitment control ensures that committable resources on multiple systems remain
synchronized. OS/400(R) supports two-phase commit in accordance with the SNA LU 6.2 architecture. For
more detailed information about the internal protocols used by the system for two-phase commit, see the
SNA Transaction Programmer’s Reference for LU Type 6.2, GC30-3084-05. All supported releases of OS/400
support the Presumed Nothing protocols of SNA LU 6.2 and the Presumed Abort protocols of SNA LU
6.2.

24 iSeries: Commitment control

Two-phase commit is also supported using TCP/IP as a Distributed Unit of Work (DUW) DRDA(R)
protocol. To use TCP/IP DUW connections, all of the systems (both the application requester and the
application server) must be at V5R1M0 or newer. For more information about DRDA see the Open Group
Technical Standard, DRDA V2 Vol. 1: Distributed Relational Database Architecture at the Open Group Web
site

Under two-phase commit, the system performs the commit operation in two waves:
v During the prepare wave, a resource manager issues a commit request to its transaction manager. The
transaction manager informs any other resources it manages and the other transaction managers that
the transaction is ready to be committed. All the resource managers must respond that they are ready
to commit. This is called the vote.
v During the committed wave, the transaction manager that initiates the commit request decides what to
do, based on the outcome of the prepare wave. If the prepare wave completes successfully and all
participants vote ready, the transaction manager instructs all the resources it manages and the other
transaction managers to commit the transaction. If the prepare wave does not complete successfully, all
the transaction managers and resource managers are instructed to roll back the transaction.

Commit and rollback operations with remote resources

When remote resources are under commitment control, the initiator sends a commit request to all remote
agents. The request is sent throughout the transaction program network. Each agent responds with the
results of the commit operation.

If errors occur during the prepare wave, the initiator sends a rollback request to all agents. If errors occur
during the committed wave, the system attempts to bring as many locations as possible to committed
status. These attempts may result in a heuristic mixed state. See “States of the transaction for two-phase
commitment control” on page 27 for more information about the possible states.

Any errors are sent back to the initiator where they are signaled to the user. If a default journal was
specified on the Start Commitment Control (STRCMTCTL) command, C LW entries are written. If errors
occur, these entries are written, even if OMTJRNE(*LUWID) was specified. You can use these entries,
along with the error messages and the status information for the commitment definition, to attempt to
synchronize the committable resources manually.

When remote resources are under commitment control, the initiator sends a rollback request to all remote
agents. The request is sent throughout the transaction program network. Each agent responds with the
results of the rollback operation.

See the following for more information:

v “Roles in commit processing”
v “States of the transaction for two-phase commitment control” on page 27
v “Commitment definitions for two-phase commitment control” on page 30

Roles in commit processing

If a commit of a transaction involves more than one resource manager, each resource manager plays a
role in the transaction. A resource manager is responsible for committing or rolling back changes made
during the transaction. The resource managers by resource type are:

FILE
Database manager

Commitment control 25
 DDM
Database manager

DDL
Database manager

DRDA(R)
Communications transaction program

LU62
Communications transaction program

API
API exit program

The following figures shows the basic roles in a transaction. The structure shown in the figures is called a
transaction program network. The structure can be in a single-level tree and a multi-level tree.

Roles in two-phase commit processing: Single-level tree

When an application on System A issues a commit request, the resource manager on System A becomes
the initiator. For DRDA distributed unit of work over TCP/IP, the initiator is called the coordinator.

The resource managers for the other three systems (B, C, and D) become agents for this transaction. For
DRDA distributed unit of work over TCP/IP, agents are sometimes called participants.

26 iSeries: Commitment control

Roles in two-phase commit processing: Multi-level tree

If the application is using APPC communications to perform the two-phase commit, the relationship
between systems can change from one transaction to the next. The following figure shows the same
systems when an application on System B issues the commit request. This configuration is a multi-level
tree.

The roles in this figure do not apply to DRDA distributed unit of work over TCP/IP because multi-level
transactions trees are not supported.

The transaction program network has another level because System B is not communicating directly with
System C and System D. The resource manager in System A now has the roles of agent and cascaded
initiator.

To improve performance of LU6.2 two-phase transactions, the initiator may assign the role of last agent
to one of the agents. The last agent does not participate in the prepare wave. In the committed wave, the
last agent commits first. If the last agent does not commit successfully, the initiator instructs the other
agents to roll back.

For DRDA distributed unit of work over TCP/IP, the coordinator might assign the role of resync server
to a participant. The resync server is responsible is to resynchronize the other participants in the event
that there is a communications failure with the coordinator, or the coordinator has a systems failure.

States of the transaction for two-phase commitment control

A commitment definition is established at each location that is part of the transaction program network.
For each commitment definition, the system keeps track of the state of its current transaction and

Commitment control 27
previous transaction. The system uses the state to decide whether to commit or roll back if a transaction
is interrupted by a communications or system failure. If multiple locations are participating in a
transaction, the states of the transactions at each location may be compared to determine the correct
action (commit or rollback). This process of communicating between locations to determine the correct
action is called resynchronization.

The following table shows:

v The basic states that may occur during a transaction.
v Additional states that may occur.
v Whether a state requires resynchronization if the transaction is interrupted by a communications or
system failure. The possible values are:

Not needed
Each location can make the correct decision independently.

May be necessary
Each location can make the correct decision, but the initiator may need to be informed of the
decision.

Required
The state of each location must be determined before the correct decision can be made.
v Action taken by a communications or system failure.

State name Description Resynchronization if the Action taken by a

transaction is interrupted communications or system
failure
Basic states during two-phase commit processing:
Reset (RST) From the commitment Not needed. Pending changes are rolled
boundary until a program back.
issues a request to commit
or roll back.
Prepare in Progress (PIP) The initiator has started the May be necessary. Pending changes are rolled
prepare wave. All locations back.
have not yet voted.
Prepared (PRP) This location and all Required. In doubt. Depends on the
locations further down in results of the
the transaction program resynchronization process.
network have voted to
commit. This location has
not yet received notification
from the initiator to
commit.
Commit in Progress (CIP) All locations have voted to Required. Pending changes are
commit. The initiator has committed.
started the committed Resynchronization is
wave. performed to ensure that all
locations have committed.
If a heuristic rollback is
reported by another
location, an error is
reported.

28 iSeries: Commitment control

State name Description Resynchronization if the Action taken by a
transaction is interrupted communications or system
failure
Committed (CMT) All agents have committed May be necessary. None.
and returned a reply to this
node.
Additional states during two-phase commit processing:
Last Agent Pending (LAP) If a last agent is selected, Required In doubt. Depends on the
this state occurs at the results of the
initiator between the PIP resynchronization process.
state and the CIP state. The
initiator has instructed the
last agent to commit and
has not yet received a
response.
Vote-Read-Only (VRO) This agent responded to the May be necessary. None.
prepare wave by indicating
that it has no pending
changes. If the
vote-read-only state is
permitted, this agent is not
included in the committed
wave.
Rollback Required (RBR) One of the following May be necessary. Pending changes are rolled
occurred: back.
v An agent issued a
rollback request before
the commit operation.
v A transaction failure has
occurred.
v The QTNRBRQD API
was used to place the
transaction in a rollback
required state.
The transaction program is
not allowed to perform any
additional changes under
commitment control.
Conditions that occur because of operator actions or errors:
Forced Rollback This location and all May be necessary Pending changes have
locations further down the already been rolled back.
transaction program
network, except the last
agent, have been rolled
back through operator
intervention.
Forced Commit This location and all May be necessary Pending changes have
locations further down the already been committed.
transaction program
network, except the last
agent, have committed
through operator
intervention.

Commitment control 29
State name Description Resynchronization if the Action taken by a
transaction is interrupted communications or system
failure
Heuristic Mixed (HRM) Some resource managers May be necessary The operator must perform
have committed. Some a restore operation at all
have rolled back. Operator participating locations to
intervention was used or a bring the database to a
system error occurred. consistent state.
Heuristic mixed does not
appear as a status on the
commitment definition
displays. Notification
messages are sent to the
operator.

Commitment definitions for two-phase commitment control

After you start commitment control, you can use the QTNCHGCO (Change Commitment Options) API to
change the commitment options for your transaction. Depending on your environment and your
applications, changing the commitment options can improve your system’s performance.

The following links describe the commitment options, and reasons to use them:
v “Commitment definition for two-phase commit: Allow vote read-only”
v “Commitment definition for two-phase commit: Not wait for outcome” on page 32
v “Commitment definition for two-phase commit: Indicate OK to leave out” on page 35
v “Commitment definition for two-phase commit: Not select a last agent” on page 38
v “Vote reliable affect on flow of commit processing” on page 38

If you are using a DRDA(R) distributed unit of work over TCP/IP connection, the only option that applies
is Allow vote read-only.

Commitment definition for two-phase commit: Allow vote read-only: Normally, a transaction manager
participates in both phases of commit processing. To improve the performance of commit processing, you
can set up some or all locations in a transaction to allow the transaction manager to vote read-only. If the
location has no committable changes during a transaction, the transaction manager votes read-only
during the prepare wave. The location does not participate in the committed wave. This improves overall
performance because the communication flows that normally occur during the committed wave are
eliminated during transactions in which no updates are made at one or more remote locations.

After you start commitment control, you can use the Change Commitment Options (QTNCHGCO) API to
change the Vote read-only permitted option to Y. You may want to do this if the following is true:
v One or more remote systems often do not have any committable changes for a transaction.
v A transaction does not depend on where the file cursor (next record) was set by the previous
transaction. When a location votes read-only, the application is never notified if the transaction is rolled
back. The location has committed any read operations to the database files and, thus, moved the cursor
position. The position of the file cursor is typically important only if you do sequential processing.

If your commitment definition is set up to allow vote read-only, the application waits for the next
message flow from another location.

The Vote read-only permitted option is intended for applications that are client/server in nature. If the
purpose of program A is only to satisfy requests from program I, not to do any independent work, it is
appropriate to allow the Vote read-only option for program A.

Flow of commit processing without last agent optimization when agent votes read-only

30 iSeries: Commitment control

The following figure shows the flow of messages among the application programs and the transaction
managers when an application program issues a commit instruction without last agent optimization when
the agent votes read-only. Both the initiator application program and the agent application programs are
unaware of the two-phase commit processing. The numbers in parentheses () in the figure correspond to
the numbered items in the description that follows.

Following is a description of the events for normal processing without last agent optimization when the
agent votes read-only. This describes a basic flow. The sequence of events can become much more
complex when the transaction program network has multiple levels or when errors occur.
1. Application program A does a receive request to indicate that it is ready to receive a request from
program I.
2. The initiator application (I) issues a commit instruction.
3. The transaction manager for the initiator (TM-I) takes the role of initiator for this transaction. It starts
the prepare wave by sending a prepare message to all the other locations that are participating in the
transaction.
4. The transaction managers for every other location take the role of agent (TM-A). The application
program A is notified by TM-A that a request to commit has been received. For ICF files, the
notification is in the form of the Receive Take Commit (RCVTKCMT) ICF indicator being set on.
5. The application program A responds by issuing a commit instruction (or a rollback instruction). This
is the application program’s vote.
6. If application program A has used the Change Commitment Options API (QTNCHGCO) to set the
Vote read-only permitted commitment option to Y, and no changes have been made at the agent
during the transaction, the agent (TM-A) responds to the initiator (TM-I) with a reset message. There
will be no committed wave for the agent.
7. A return is sent to the application program (A) to indicate that the transaction is complete at agent
TM-A.

Commitment control 31
8. The next time the initiator (TM-I) issues any message to the agent (TM-A), either a data flow or a
commitment instruction, TM-I causes its current transaction ID to be sent with the message. The
reason for this is that a new transaction ID may have been generated at TM-I if a communications
failure had occurred between TM-I and another system during the commit operation.
9. A return is sent to the application program (A) to indicate that the transaction is complete at agent
TM-A. The return is delayed until after the next message is received because a new transaction ID
must be received from TM-I before the next transaction can be started by application A.

For more information about two-phase commitment control see “Roles in commit processing” on page 25
and “States of the transaction for two-phase commitment control” on page 27.

Commitment definition for two-phase commit: Not wait for outcome: When a communications or
systems failure occurs during a commit operation such that resynchronization is required, the default is
to wait until the resynchronization is finished before the commit operation completes.

Note: The Not wait for outcome option does not apply if you
are using a DRDA(R) distributed unit of work over
TCP/IP connection. DRDA distributed unit of work over
TCP/IP connections never wait for outcome.

Consider changing this behavior if the following conditions are true:

v The applications that participate are independent of each other.
v Your program logic does not need the results of previous transactions to ensure that your database files
remain synchronized.

After you start commitment control, you can use the QTNCHGCO (Change Commitment Options) API to
specify that the commitment definition does not wait for the outcome of resynchronization. If you specify
N (No) for the Wait for outcome option, the system uses a database server job (QDBSRVnn) to handle
resynchronization asynchronously.

Note: These database server jobs are started during the IPL
process. If you change the options for commitment
control, this has no effect on the number of jobs that the
system starts.

This topic only refers to two values for the resolved Wait for outcome option, Y (Yes) and N (No). There
are actually two more values that you can specify, L (Yes or Inherit from Initiator) and U (No or Inherit
from Initiator). When you use these values, the actual value used during each commit operation is
resolved to Yes or No by the system. The QTNCHGCO (Change Commitment Options) API topic has
more details about these values.

Note: The initiator’s value can only be inherited by an agent if

both the initiator and the agent support presumed abort.

The wait for outcome (WFO) option does not affect normal, error-free commit processing. If an error
occurs, the WFO option determines whether the application waits for resynchronization or not, with the
following conditions:
v If the resolved WFO option is Y (Yes), the application waits for the result of the resynchronization.
v If the resolved WFO option is N (No) and a communication failure occurs during the prepare wave or
rollback of a location that supports presumed abort protocols, no resynchronization is performed and
the commitment definition is rolled back.

32 iSeries: Commitment control

v If the commitment definition is in doubt (transaction state is prepared or Last Agent Pending), the
application will wait for the result of the resynchronization regardless of the resolved WFO value. For
further information about the commitment definition being in doubt, see “States of the transaction for
two-phase commitment control” on page 27.
v If the resolved WFO option is N and neither one of conditions two or three are true, the system
attempts to resynchronize once. If it is not successful, the system signals STATUS message CPF83E6 to
the application to indicate that resynchronization is in progress.
Since CPF83E6 is a STATUS message, it only has an effect if the application is monitoring for it.
Normally, your application can treat this message as an informational message. The systems that are
participating in the transaction attempt to resynchronize the transaction until the failure is repaired.
These subsequent resynchronization attempts are performed in the database server jobs. If a
subsequent resynchronization attempt that is performed in a database server job fails, the message
CPI83D0 is sent to QSYSOPR.

Wait for outcome-Yes

In the following figure, the commitment definition for the initiator (I) uses the default value of Y (Yes) for
the Wait for outcome option. When communications between TM-I and TM-A is lost, both application A
and application I wait until the transaction is resynchronized.

Wait for outcome-No

In the following figure, the commitment definition for the initiator has the resolved WFO set to N (No).
TM-A meets condition 3 in the preceding list, while TM-I meets condition 4. Control is returned to

Commitment control 33
application I after one attempt to resynchronize with TM-A. A database server job attempts to
resynchronize. Application I never receives the return indicator when the commit request has completed
successfully. Control is not returned to the agent application (A) until after communications is
reestablished. This depends on the timing of the failure. In this case, the communications failure occurs
before the commit message is received from the initiator, leaving TM-A in doubt as to whether to commit
or rollback. When the transaction manager is in doubt, it retains control until the resynchronization is
completed, regardless of the resolved WFO value at that system.

If you want the applications at all systems to continue before resynchronization completes, you must
either change the resolved WFO option to N (No) on all systems, or set the initiator to N and the rest of
the systems to U (No or Inherit from Initiator). But remember that the resolved WFO option is ignored
when the transaction manager is in doubt as to whether to commit or rollback, and always waits until
resynchronization completes before returning control.

34 iSeries: Commitment control

When a connection is made to a remote relational database, and no protected conversations have already
been started, the system implicitly changes the Wait for outcome value to N. The reason for this is that
the performance of commit operations is improved when the Wait for outcome value is N and the
remote system supports presumed abort. This implicit change of the Wait for outcome value is only
performed for DRDA and DDM applications. APPC applications use the default Wait for outcome value
of Y unless they invoke the QTNCHGCO API to change it.

Commitment definition for two-phase commit: Indicate OK to leave out: Normally, the transaction
manager at every location in the transaction program network participates in every commit or rollback
operation. To improve performance, you can set up some or all locations in a transaction to allow the
transaction manager to indicate OK to leave out.

Commitment control 35
Note: The Indicate OK to leave out option does not apply if
you are using a DRDA(R) distributed unit of work over
TCP/IP connection.

If no communications flows are sent to the location during a transaction, the location is left out when a
commit or rollback operation is performed. This improves overall performance because the
communications flows that normally occur during the commit or rollback are eliminated during
transactions in which no data is sent to one or more remote locations.

After you start commitment control, you can use the Change Commitment Options (QTNCHGO) API to
change the OK to leave out option to Y (Yes). You may want to do this if one or more remote systems
often are not involved in a transaction.

If your commitment definition is set up to indicate OK to leave out, the application waits for the next
message flow from another location.

The OK to leave out option is intended for applications that are client/server in nature. If the only
purpose of program A is to satisfy requests from program I and not do any independent work, then it is
appropriate to allow the OK to leave out option for program A.

Flow of commit processing without last agent optimization when agent votes OK to leave out

The following figure shows the flow of messages among the application programs and the transaction
managers when an application program issues a commit instruction without last agent optimization when
the agent indicates OK to leave out. Both the initiator application program and the agent application
programs are unaware of the two-phase commit processing. The numbers in parentheses () in the figure
correspond to the numbered items in the description that follows.

36 iSeries: Commitment control

Following is a description of the events for normal processing without last agent optimization when the
agent votes OK to leave out. This describes a basic flow. The sequence of events can become much more
complex when the transaction program network has multiple levels or when errors occur.
1. Application program A does a receive request to indicate that it is ready to receive a request from
program I.
2. The initiator application (I) issues a commit instruction.
3. The transaction manager for the initiator (TM-I) takes the role of initiator for this transaction. It starts
the prepare wave by sending a prepare message to all the other locations that are participating in the
transaction.
4. The transaction managers for every other location take the role of agent (TM-A). The application
program A is notified by TM-A that a request to commit has been received. For ICF files, the
notification is in the form of the Receive Take Commit (RCVTKCMT) ICF indicator being set on.
5. The application program A responds by issuing a commit instruction (or a rollback instruction). This
is the application program’s vote.
6. If application program A has used the Change Commitment Options API (QTNCHGCO) to set the
OK to leave out commitment option to Y, an indicator is sent when the agent (TM-A) responds to
the initiator (TM-I) with a request commit message.

Commitment control 37
Note: Any change to the OK to leave out commitment option
does not take effect until the next successful commit
operation.

7. When the initiator (TM-I) receives all the votes, the TM-I sends a commit message. This starts the
committed wave.
8. Each agent (TM-A) commits and responds with a reset message.
9. A return is sent to the application program (I) to indicate that the transaction is complete at the
initiator.
10. Any number of transactions may occur on TM-I, none of which require changes to TM-A or data
from TM-A. TM-A is not included in these transactions.
11. The next time the initiator (TM-I) issues a message to the agent (A), a new transaction ID is sent
with the message. If the initiator performs any commit or rollback operations before sending a
message to the agent, no messages are sent to the agent during those operations (the agent is ’left
out’ of those commits or rollbacks). Because one or more transactions may have been committed or
rolled back at the initiator while the agent was left out, the initiator must communicate its current
transaction ID when the next message is sent to the agent.
12. A return is sent to the application program (A) to indicate that the original commit is complete and
that it is participating in the current transaction.

Commitment definition for two-phase commit: Not select a last agent: By default, the transaction
manager for the initiator is free to select any agent as a last agent during a commit operation.

Note: The Not select last agent option does not apply if you are
using a DRDA(R) distributed unit of work over TCP/IP
connection.

In case of a multi-level tree, any agent selected as a last agent by its initiator is also free to select a last
agent of its own. Performance is improved when a last agent is selected during the commit operation
because two communications flows are eliminated between an initiator and its last agent (the prepare
phase is eliminated for these systems).

However, when the initiator sends the request commit to its last agent, it must wait until it has received
the last agent’s vote before it can continue. This is regardless of the Wait for outcome value for the
commitment definition. During normal, error-free commit processing, this is not an issue. But, if an error
occurs during this window, the initiator cannot continue until resynchronization completes. If the initiator
application is handling requests from a user at a terminal, this can be a usability consideration.

You must consider whether the improved performance during normal commit operations is more
important than the impact on usability when such an error occurs. Note that if the error occurs before the
request commit is sent to the last agent, the LUW will immediately roll back and the initiator will not
wait. Therefore, the window when an error can cause the initiator to wait is quite small, so such an error
is rare.

If you decide that the usability impact is not worth the improved performance, you can change your
commitment definitions to not select a last agent. After you start commitment control, you can use the
Change Commitment Options (QTNCHGCO) API to change the Last agent permitted option to N.

Vote reliable affect on flow of commit processing: Vote reliable is an optimization that improves
performance by returning earlier to the initiator application after a commit operation and eliminating one
message during a commit operation. There is no explicit vote reliable optimization for DRDA(R)
distributed unit of work over TCP/IP. However, OS/400(R) never requests a reset (forget) confirmation for
TCP/IP connections. Therefore a reset (forget) is always implied for TCP/IP connections.

38 iSeries: Commitment control

After you start commitment control, you can use the Change Commitment Options (QTNCHGCO) API to
change the Accept vote reliable option to Y.

Vote reliable can be thought of as a promise by an agent to its initiator that no heuristic decisions will be
made at the agent if communications failure occurs while the agent is in doubt. An agent that is using the
vote reliable optimization sends an indicator to the initiator during the prepare wave of the commit. If
the initiator is also using the vote reliable optimization, it then sends an indicator to the agent that no
reset is required in response to the commit message. This eliminates the reset message, and allows the
transaction manager to return to the application at the initiator as soon as the commit message is sent.

Consider using the vote reliable optimization if the following conditions are true:
v It is unlikely that a heuristic decision will be made at an in doubt agent in the event of a systems or
communications failure unless the failure cannot be repaired.
v Your program logic does not need the results of previous transactions to ensure that your database files
remain synchronized.

The vote reliable optimization will be used by OS/400 only if all the following conditions are true:
v The initiator and agent locations support the presumed abort level of commitment control.
v The initiator location accepts the vote reliable indication from the agent. On OS/400 initiators, this
depends on the value of two commitment options:
– The value of the Wait for outcome commitment option must be No (Yes is the default).
– The value of the Accept vote reliable commitment option must be Yes (Yes is the default).
v The agent location votes reliable during the prepare wave. OS/400 agents always vote reliable. This is
because heuristic decisions can be made only through a manual procedure that warns of the possible
negative side effects of making a heuristic decision.

Flow of commit processing with vote reliable optimization

The following figure shows the flow of messages among the application programs and the transaction
managers when the vote reliable optimization is used. Both the initiator application program and the
agent application programs are unaware of the two-phase commit processing. The numbers in
parentheses () in the figure correspond to the numbered items in the description that follows.

Commitment control 39
Following is a description of the events for normal processing without last agent optimization when the
agent votes reliable. This describes a basic flow. The sequence of events can become much more complex
when the transaction program network has multiple levels or when errors occur.
1. Application program A does a receive request to indicate that it is ready to receive a request from
program I.
2. The initiator application (I) issues a commit instruction.
3. The transaction manager for the initiator (TM-I) takes the role of initiator for this transaction. It starts
the prepare wave by sending a prepare message to all the other locations that are participating in the
transaction.
4. The transaction managers for every other location take the role of agent (TM-A). The application
program A is notified by TM-A that a request to commit has been received. For ICF files, the
notification is in the form of the Receive Take Commit (RCVTKCMT) ICF indicator being set on.
5. The application program A responds by issuing a commit instruction (or a rollback instruction). This
is the application program’s vote.
6. The agent (TM-A) responds to the initiator (TM-I) with a request commit message. OS/400 systems
send a vote reliable indicator with the request commit.
7. When the initiator (TM-I) receives all the votes, the TM-I sends a commit message. If the Wait for
outcome commitment option is N (No) and the Accept vote reliable commitment option is Y (Yes), a
no reset indicator is sent with the commit message. This tells the agent that no reset message is
required in response to the commit.
8. The transaction is complete. A return is sent to the application programs (I and A). This return
indicates that the commit operation was successful. If a heuristic damage occurs at system A due to a

40 iSeries: Commitment control

 heuristic decision being made before the committed message is received, application I would not be
informed. Instead, a message would be sent to the QSYSOPR message queue. However, application A
would receive the heuristic damage indication.
9. The next time the agent (TM-A) sends any message to the initiator (TM-I), either a data flow or a
commitment instruction, an implied reset indicator is sent with the message to inform TM-I that
TM-A completed the commit successfully. The reason for this is that TM-I must retain information
about the completed transaction until it has confirmed that TM-A successfully received the commit
message in step 7.

XA transaction support for commitment control

DB2(R) UDB for iSeries(TM) can participate in X/Open global transactions. The Open Group has defined
an industry standard model for transactional work that allows changes made against unrelated resources
to be part of a single global transaction. An example of this is changes to databases that are provided by
two separate vendors. This model is called the X/Open Distributed Transaction Processing model. The
following publications describe the X/Open Distributed Transaction Processing model in detail:
v X/Open Guide, February 1996, Distributed Transaction Processing: Reference Model, Version 3
(ISBN:1-85912-170-5, G504), The Open Group.
v X/Open CAE Specification, December 1991, Distributed Transaction Processing: The XA Specification
(ISBN:1-872630-24-3, C193 or XO/CAE/91/300), The Open Group.
v X/Open CAE Specification, April 1995, Distributed Transaction Processing: The TX (Transaction
Demarcation) Specification (ISBN:1-85912-094-6, C504), The Open Group.

Be familiar with the information in these books, particularly the XA Specification, before attempting to
use the XA transaction support provided by DB2 UDB for iSeries. You can find these books at the Open
Group Web site

There are five components to the DTP model:

Application Program (AP)

Implements the required function of the user by specifying a sequence of operations that involves
resources such as databases. It defines the start and end of global transactions, accesses resources
within transaction boundaries, and normally makes the decision whether to commit or roll back
each transaction.

Transaction Manager (TM)

Manages global transactions and coordinates the decision to start them, and commit them or roll
them back in order to ensure atomic transaction completion. The TM also coordinates recovery
activities with the RMs after a component fails.

Resource Manager (RM)

Manages a defined part of the computer’s shared resources, such as a database management system.
The AP uses interfaces defined by each RM to perform transactional work. The TM uses interfaces
provided by the RM to carry out transaction completion.

Communications Resource Manager (CRM)

Allows an instance of the model to access another instance either inside or outside the current TM
domain. CRMs are outside the scope of DB2 UDB for iSeries and are not discussed here.

Commitment control 41
 Communication Protocol
The protocols used by CRMs to communicate with each other. This is outside the scope of DB2 UDB
for iSeries and is not discussed here.

The XA Specification is the part of the DTP model that describes a set of interfaces that is used by the TM
and RM components of the DTP model. DB2 UDB for iSeries implements these interfaces as a set of
UNIX(R) style APIs and exit programs. See XA APIs for detailed documentation of these APIs and for
more information about how to use DB2 UDB for iSeries as an RM.

iSeries Navigator and XA transactions

iSeries Navigator supports the management of XA transactions as Global Transactions

A Global transaction may contain changes both outside and within DB2 UDB for iSeries. A global
transaction is coordinated by an external Transaction Manger using the Open Group XA architecture, or
another similar architecture. An application commits or rolls back a global transaction using interfaces
provided by the Transaction Manager. The Transaction Manager uses commit protocols defined by the XA
architecture, or another architecture, to complete the transaction. DB2 UDB for iSeries acts as an XA
Resource Manager when participating in a global transaction. There are two types of global transactions:
v Transaction-scoped locks: Locks acquired on behalf of the transaction are scoped to the transaction.
The transaction can move from one job or thread to another.
v Job-scoped locks: Locks acquired on behalf of the transaction are scoped to the job. The transaction
cannot move from the job that started it.

If you are running XA transactions against a database that resides on the local system, use the XA APIs
for transaction-scoped locks. These APIs have fewer restrictions than the XA APIs for job-scoped locks,
and provide better performance in the following situations:
v If multiple SQL connections are ever used to work on a single XA transaction branch.
v If a single SQL connection is used to work on multiple, concurrent XA transaction branches.

In these situations, a separate job must be started to run XA transaction branches when you use the XA
APIs for Job Scoped Locks.

If you are running against a database that resides on a remote system, you must use the XA APIs for Job
Scoped Locks.

Considerations for XA transactions

Understand the following considerations and restrictions before using DB2 UDB for iSeries as a RM. The
term ″thread″ refers to either a job that is not thread capable, or a single thread within a thread capable
job.

The following considerations apply to both transactions with transaction-scoped locks and transactions
with job-scoped locks unless noted otherwise.

DB2 UDB for iSeries considerations

v XA transactions may only be performed in jobs that are running in SQL server mode. One effect of this
is that applications are limited to SQL interfaces when making changes to DB2 UDB for iSeries during
an XA transaction. If the db2xa_open() API is used in a job that is not already running in SQL server
mode, SQL server mode is implicitly started. Refer to “SQL server mode and thread-scoped
transactions for commitment control” on page 45

42 iSeries: Commitment control

v Any errors that are detected by DB2 UDB for iSeries during the XA API invocations are reported
through return codes by the XA specification. Diagnostic messages are left in the job log when the
meaning of the error may not be clear from the return code alone.

Embedded SQL considerations

v In order to use a Structured Query Language (SQL) connection for XA transactions, you must use the
db2xa_open() application programming interface (API) before the SQL connection is made. The
relational database that will be connected to must be passed to the db2xa_open() API by the Xainfo
parameter. The user profile and password to be used in the job that the connection is routed to may be
passed to the db2xa_open() API. If it is not passed, the profile will default to the one that was specified
or defaulted during the connection attempt.
v If embedded SQL is used to perform XA transactions, the work performed for each connection is
routed to a different job, even if the connections are made in the same thread. This is different than
SQL server mode without XA, where work performed for all connections in a single thread is routed to
the same job. This is because the XA specification requires a separate prepare, commit or rollback call
for each resource manager instance.
Note: The following consideration applies only to transactions with job-scoped locks.
v If embedded SQL is used to perform XA transactions, only one connection per relational database can
be made per thread. Whenever the thread is not actively associated with a transaction branch, work
requested over one of the thread’s connections will cause the RM to use the TM’s ax_reg() exit
program to determine whether the work is to start, resume or join a transaction branch.
If the work is to start a transaction branch, it is performed over that thread’s connection to the
corresponding relational database.
If the work is to join a transaction branch, it is rerouted over the connection to the corresponding
relational database that was made in the thread that started the transaction branch. Note that the
system does not enforce that the user profile for that connection is the same as the one for the
connection of the joining thread. The TM is responsible to ensure that this is not a security concern.
Typical TMs use the same user profile for all connections. This user profile is authorized to all data
that is managed by the TM. Further security of access to this data is managed by the TM or AP instead
of using the standard iSeries security techniques.
v If the work is to resume a transaction branch, the connection that is used depends on whether the
suspended transaction branch association was established by starting or joining the transaction branch.
Subsequent work is performed over the same connection until the db2xa_end() API is used to suspend
or end the thread’s association with that transaction branch.

CLI considerations
v If the CLI is used to perform XA transactions, more than one connection may be made in the same
thread after the db2xa_open() API is used. The connections may be used in other threads to perform
XA transactions, as long as those other threads first use the db2xa_open() API with the same Xainfo
parameter value.
Note: The following consideration applies only to transactions with job-scoped locks.
v If the CLI is used to perform XA transactions, the connection that is used to start a transaction branch
must be used for all work on that transaction branch. If another thread is to join the transaction
branch, the connection handle for the connection used to start the transaction branch must be passed to
the joining thread so that it can perform work over that same connection. Likewise, if a thread is to
resume the transaction branch, the same connection must be used.
Note: The following applies to transactions with transaction-scoped locks and job-scoped locks.
Since CLI connection handles cannot be used in a different job, the join function is limited to threads
running in the same job that started the transaction branch when the CLI is used.

Remote relational database considerations

Commitment control 43
Note: These considerations for a remote relational database apply only to transactions with job-scoped
locks.
v XA connections to a remote relational database are supported only if the relational database resides on
a system that supports Distributed Unit of Work (DUW) DRDA(R) connections. This includes systems
running DRDA over SNA LU6.2 conversations. This also includes systems using V5R1 when running
DRDA using TCP/IP connections.
v Before using the XA join function, the db2xa_open() API must be used in the joining thread. The same
relational database name and RMID must be specified on the db2xa_open() API in both the thread that
started the transaction branch and the joining thread. If the transaction branch is active when a join is
attempted, the joining thread is blocked. The joining thread remains blocked until the active thread
suspends or ends its association with the transaction branch.

Recovery consideration
v The manual heuristic commit and rollback support that is provided for all commitment definitions can
be used if it becomes necessary to force a transaction branch to commit or roll back while it is in a
prepared state. See “When to force commits and rollbacks and when to cancel resynchronization” on
page 101 for details.

Transaction branch considerations

v Information about XA transaction branches is shown as part of the commitment control information
displayed by iSeries Navigator and the Work with Job (WRKJOB), Display Job (DSPJOB), and Work
with Commitment Definition (WRKCMTDFN) commands. The TM name, transaction branch state,
transaction identifier and branch qualifier are all shown. The commitment definitions related to all
currently active XA transactions can be displayed by using the command WRKCMTDFN JOB(*ALL)
STATUS(*XOPEN) or by displaying the Global Transactions in iSeries Navigator.

Note: The following item applies only to transactions with job-scoped locks.
v If an association between a thread and an existing transaction branch is suspended or ended using the
db2xa_end() API, the thread may start a new transaction branch. If the connection used to start the
new transaction branch was used earlier to start a different transaction branch, and the thread’s
association with that transaction branch has been ended or suspended by the db2xa_end() API, a new
SQL server job may be started. A new SQL server job is needed only if the first transaction branch has
not yet been completed by the db2xa_commit() or db2xa_rollback() API. In this case, another
completion message SQL7908 is sent to the job log identifying the new SQL server job, just as the
connection’s original SQL server job was identified when the connection was established. All SQL
requests for the new transaction branch are routed to the new SQL server job. When the transaction
branch is completed by the db2xa_commit() or db2xa_rollback() API, the new SQL server job is
recycled and returned to the prestart job pool.
v A transaction branch will be marked Rollback Only by the system when the following situations occur:
– A thread ends when it is still associated with the transaction branch.
– The db2xa_close() API is used in a thread that has an active association with the transaction branch.
v A transaction branch will be rolled back by the system if any threads are still associated with it when
any of the following situations occur:
– The connection that is related to the transaction branch is ended.
– The job that started the transaction branch is ended.
– The system fails.
v There is one situation where a transaction branch will be rolled back by the system, regardless of
whether there are still associated threads. This occurs when the SQL server job that the connection’s
work is being routed to is ended. This can only happen when the End Job (ENDJOB) CL command is
used against that job.

Note: The following item applies only to transactions with job-scoped locks.

44 iSeries: Commitment control

v A transaction branch will not be affected if no threads have an active association with it when any of
the following situations occur. The TM may commit or roll back the transaction branch from any
thread that has used the db2xa_open() API with the same Xainfo parameter value that was specified in
the thread that started the transaction branch.
– The connection that is related to the transaction branch is ended.
– A thread or job that performed work for the transaction branch, but not longer has an active
association with it, uses the db2xa_close() API.
– A thread or job that performed work for the transaction branch, but no longer has an active
association with it, uses the db2xa_close() API.
– The system fails. In this case, the transaction branch is not affected only if it is in prepared state. If it
is in idle state, the system rolls it back.

SQL server mode and thread-scoped transactions for commitment

control
Commitment definitions with job-scoped locks are normally scoped to an activation group. If a job is
multi-threaded, all threads in the job have access to the commitment definition and changes made for a
particular transaction may be spread across multiple threads. That is to say, all threads whose programs
run in the same activation group participate in a single transaction.

There are cases where it is desirable for transactional work to be scoped to the thread, rather than an
activation group. In other words, each thread would have its own commitment definition and
transactional work for each commitment definition would be independent of work performed in other
threads.

This is supported by DB2(R) UDB for iSeries(TM) by using the Change Job (QWTCHGJB) API to change
the job to run in SQL server mode. When an SQL connection is requested in SQL server mode, it is
routed to a separate job. All subsequent SQL operations that are performed for that connection are also
routed to that job. When the connection is made, completion message SQL7908 is sent to the SQL server
mode job’s job log indicating which job the SQL requests are being routed to. The commitment definition
is owned by the job that is indicated in this message. If errors occur, it may be necessary to look at the
job logs for both jobs to understand the source of the problem because no real work is being done in the
job performing the SQL statements.

When running in SQL server mode, only SQL interfaces may be used to perform work under
commitment control. Embedded SQL or Call Level Interface (CLI) may be used. All connections made
through embedded SQL in a single thread are routed to the same back-end job. This allows a single
commit request to commit the work for all the connections, just as it would in a job that is not running in
SQL server mode. Each connection made through the CLI is routed to a separate job. The CLI requires
work that is performed for each connection to be committed or rolled back independently.

You cannot perform the following operations under commitment control when running in SQL server
mode:
v Record changes that are made with interfaces that are not SQL interfaces
v Changes to DDM files
v Changes to API commitment resources

You cannot start commitment control directly in a job running in SQL server mode. For more information
about SQL server mode, see the following pages in the Database topic:
v Run DB2 CLI in SQL server mode
v Starting DB2 CLI in SQL Server Mode
v Restrictions for running DB2 CLI in server mode

Commitment control 45
Start commitment control
To start commitment control, use the STRCMTCTL (Start Commitment Control) Command.

Note: Commitment control does not need to be started by SQL applications. SQL implicitly starts
commitment control at connect time when the SQL isolation level is not *NONE.

When you use the STRCMTCTL command you can specify these things:

These parameters are explained in the topics that follow.

Commit lock-level
Specify the lock-level with the LCKLVL parameter on the STRCMTCTL command. The level you
specify becomes the default level of record locking for database files that are opened and placed
under commitment control for the commitment definition. See “Commit lock level” on page 48 for
more information.

Commit notify object

Use the NTFY parameter to specify the notify object. A notify object is a message queue, data area,
or database file that contains information identifying the last successful transaction completed for a
particular commitment definition if that commitment definition did not end normally. See Commit
notify object for more information

Commit scope parameter

Use the CMTSCOPE parameter to specify commit scope. When commitment control is started, the
system creates a commitment definition. The commit scope parameter identifies the scope for the
commitment definition. The default is to scope the commitment definition to the activation group of
the program making the start commitment control request. The alternative scope is to the job.

Default journal parameter

You can specify a default journal when you start commitment control. You might use a default
journal for these reasons:
v You want to capture transaction journal entries. These entries can assist you in analyzing the
history of what resources are associated with a transaction. They are not used for applying and
removing journaled changes. The omit journal entries (OMTJRNE) parameter determines whether
the system writes transaction entries.
v You want to improve performance for jobs that close files and open them again within a routing
step. If you close all the files assigned to a journal that is not the default journal, all the system
information about the journal is removed from the routing step. If a file that is assigned to that
journal is opened later, all the information about the journal must be created again. The system
keeps information about the default journal with the commitment definition, whether any
resources that are assigned to the journal are active.

Commit text parameter

Use the TEXT parameter to identify the specific text to be associated with a commitment definition
when displaying information about the commitment definitions started for a job. If no text is
specified, the system provides a default text description.

Omit journal entries parameter

If you specify a default journal to improve performance, you can use the OMTJRNE parameter to
prevent the system from writing transaction journal entries. Having the system write transaction
entries significantly increases the size of your journal receiver and degrades performance during
commit and rollback operations.

46 iSeries: Commitment control

 Transaction entries can be useful when you are setting up and testing either your commitment
control environment or a new application.

Transaction entries are written to the default journal regardless of the value of the OMTJRNE
parameter under these conditions:
v A system error occurs during a commit or rollback operation.
v A manual change is made to a resource that participated in a transaction, and the change caused
a heuristic mixed condition. See “States of the transaction for two-phase commitment control” on
page 27 for a description of the heuristic mixed condition. This type of manual change is called a
“When to force commits and rollbacks and when to cancel resynchronization” on page 101.

You can use the information about what resources participated in the transaction to determine what
action to take in these situations.

You can use the Journal entry information finder to show the layouts for the entry-specific data for
transaction (commitment control) journal entries.

The following have more information about starting commitment control:

v Commit notify object
v “Commit lock level” on page 48

Commit notify object

A notify object is a message queue, data area, or database file that contains information identifying the
last successful transaction completed for a particular commitment definition if that commitment
definition did not end normally. The information used to identify the last successful transaction for a
commitment definition is given by the commit identification that associates a commit operation with a
specific set of committable resource changes.

The commit identification of the last successful transaction for a commitment definition is placed in the
notify object only if the commitment definition does not end normally. This information can be used to
help determine where processing for an application ended so that the application can be started again.

For independent disk pools, the notify object must reside on the same independent disk pool or
independent disk pool group as the commitment definition. If you move the commitment definition to
another independent disk pool or independent disk pool group, the notify object must also reside on that
other independent disk pool or independent disk pool group. The notify object on the other independent
disk pool or independent disk pool group is updated if the commitment definition ends abnormally. If
the notify object is not found on the other independent disk pool or independent disk pool group, the
update fails with message CPF8358.

If journaled resources participate in the current transaction and a commit operation is performed with a
commit identification, the commit identification is placed in the commit journal entry (journal code and
entry type of C CM) that identifies that particular transaction as being committed. A commit journal
entry containing the commit identification is sent to each journal associated with resources that
participated in the transaction.

The following table shows how you specify the commit identification and its maximum size. If the
commit identification exceeds its maximum size, it is truncated when it is written to the notify object.

Language Operation Maximum characters in commit identification

CL COMMIT command 30001
ILE RPG* COMIT operation code 40001
PLI. PLICOMMIT subroutine 40001

Commitment control 47
Language Operation Maximum characters in commit identification
ILE C* _Rcommit function 40001
ILE COBOL* COMMIT verb Not supported
SQL COMMIT statement Not supported
Note:
1
If the notify object is a data area, the maximum size is 2000 characters.

When a notify object is updated with the commit identification, it is updated as follows:

Database file
If a database file is used as the notify object, the commit identification is added to the end of the
file. Any existing records will be left in the file. Because several users or jobs can be changing
records at the same time, each commit identification in the file contains unique information to
associate the data with the job and commitment definition that failed. The file that serves can be
journaled

Data area
If a data area is used as the notify object, the entire content of the data area is replaced when the
commit identification is placed in the data area. If more than one user or job is using the same
program, only the commit identification from the last commitment definition that did not end
normally will be in the data area. Consequently, a single data area notify object may not produce the
correct information for starting the application programs again. To solve this problem, use a
separate data area for each commitment definition for each workstation user or job.

Message queue
If a message queue is used as a notify object, message CPI8399 is sent to the message queue. The
commit identification is placed in the second-level text for message CPI8399. As with using a
database file for the notify object, the contents of each commit identification uniquely identify a
particular commitment definition for a job so that an application program can be started again.

See “Example: Use a notify object to start an application” on page 81 for examples of using a notify
object.

Commit lock level

The value you specify for the LCKLVL parameter on the Start Commitment Control (STRCMTCTL)
command becomes the default level of record locking for database files that are opened and placed under
commitment control for the commitment definition. The default level of record locking cannot be
overridden when opening local database files. However, database files accessed by SQL use the current
SQL isolation level in effect at the time of the first SQL statement issued against it. “Considerations and
restrictions for commitment control” on page 22 has states considerations for object and record level
changes

The lock level must be specified with respect to your needs, the wait periods allowed, and the release
procedures used most often.

The following descriptions apply only to files that are opened under commitment control:

*CHG Lock Level

Use this value if you want to protect changed records from changes by other jobs running at the
same time. For files that are opened under commitment control, the lock is held for the duration of
the transaction. For files not opened under commitment control, the lock on the record is held only
from the time the record is read until the update operation is complete.

48 iSeries: Commitment control

 *CS Lock Level
Use this value to protect both changed and retrieved records from changes by other jobs running at
the same time. Retrieved records that are not changed are protected only until they are released, or
a different record is retrieved.

The *CS lock level ensures that other jobs are not able to read a record for update that this job has
read. In addition, the program cannot read records for update that have been locked with a record
lock type of *UPDATE in another job until that job accesses a different record.

*ALL Lock Level

Use this value to protect changed records and retrieved records that are under commitment control
from changes by other jobs running under commitment control at the same time. Records that are
retrieved or changed are protected until the next commit or rollback operation.

The *ALL lock level ensures that other jobs are not able to access a record for update that this job
has read. This is different from normal locking protocol. When the lock level is specified as *ALL,
even a record that is not read for update cannot be accessed if it is locked with a record lock type of
*UPDATE in another job.

The following table shows the duration of record locks for files under and not under commitment
control.

Request LCKLVL parameter Duration of lock Lock type

Read-only No commitment control No lock None
*CHG No lock None
*CS From read to next read, *READ
commit, or rollback
*ALL From read to commit or *READ
rollback
Read for update then No commitment control From read to update or *UPDATE
update or delete1 delete
*CHG From read to update or *UPDATE
delete
Then from update or delete *UPDATE
to next commit or rollback2
*CS From read to update or *UPDATE
delete
Then from update or delete *UPDATE
to next commit or rollback2
*ALL From read to update or *UPDATE
delete
Then from update or delete
to next commit or rollback2

Commitment control 49
Request LCKLVL parameter Duration of lock Lock type
Read for update then No commitment control From read to release *UPDATE
release1
*CHG From read to release *UPDATE
*CS From read to release, *UPDATE
commit, or rollback
Then from release to next *UPDATE
read, commit, or rollback
*ALL From read to release, *UPDATE
commit, or rollback
Then from release to next
commit or rollback
Add No commitment control No lock None
*CHG From add to commit or *UPDATE
rollback
*CS From add to commit or *UPDATE
rollback
*ALL From add to commit or *UPDATE
rollback
Write direct No commitment control For duration of write direct *UPDATE
*CHG From write direct to *UPDATE
commit or rollback
*CS From write direct to *UPDATE
commit or rollback
*ALL From write direct to *UPDATE
commit or rollback
Note:
1
If a commit or rollback operation is performed after a read-for-update operation, but before the record is updated,
deleted, or released, the record is unlocked during the commit or rollback operation. The protection on the record is
lost as soon as the commit or rollback completes.
2
If a record is deleted but the commit or rollback has not yet been issued for the transaction, the deleted record does
not remain locked. If the same or a different job attempts to read the deleted record by key, the job receives a record
not found indication. However, if a unique keyed access path exists over the file, another job is prevented from
inserting or updating a record with the same unique key value as that of the deleted record until the transaction is
committed.

A record lock type of *READ is obtained on records that are not read for update when the lock level is
*CS or *ALL. This type of lock prevents other jobs from reading the records for update but does not
prevent the records from being accessed from a read-only operation.

A record lock type of *UPDATE is obtained on records that are updated, deleted, added, or read for
update. This type of lock prevents other jobs from reading the records for update, and prevents jobs
running under commitment control with a record lock level of *CS or *ALL from accessing the records for
even a read-only operation.

Programs that are not using commitment control can read records locked by another job, but cannot read
records for update, regardless of the value specified for the LCKLVL parameter.

The lock level specified for a commitment definition when commitment control is started for an
activation group or for the job applies only to opens associated with that particular commitment
definition.

50 iSeries: Commitment control

Note: The *CS and *ALL lock-level values protect you from
retrieving a record that currently has a pending change
from a different job. However, the *CS and *ALL
lock-level values do not protect you from retrieving a
record using a program running in one activation group
that currently has a pending change from a program
running in a different activation group within the same
job.

Within the same job, a program can change a record that has already been changed within the current
transaction as long as the record is accessed again using the same commitment definition. When using
the job-level commitment definition, the access to the changed record can be made from a program
running within any activation group that is using the job-level commitment definition.

End commitment control

You can end commitment control for either the job-level or activation-group-level commitment definition
using the End Commitment Control (ENDCMTCTL) command. Issuing the ENDCMTCTL command
indicates to the system that the commitment definition in use by the program making the request is to be
ended. The ENDCMTCTL command ends only one commitment definition for the job and all other
commitment definitions for the job remain unchanged.

If the activation-group-level commitment definition is ended, then programs running within that
activation group can no longer make changes under commitment control, unless the job-level
commitment definition is already started for the job. If the job-level commitment definition is active, then
it is immediately made available for use by the programs running within the activation group that just
ended commitment control.

If the job-level commitment definition is ended, then any program running within the job that was using
the job-level commitment definition can no longer make changes under commitment control without first
starting commitment control again with the STRCMTCTL command.

Before issuing the ENDCMTCTL command, the following must be satisfied for the commitment
definition to be ended:
v All files opened under commitment control for the commitment definition to be ended must first be
closed. When ending the job-level commitment definition, this includes all files opened under
commitment control by any program running in any activation group that is using the job-level
commitment definition.
v All API commitment resources for the commitment definition to be ended must first be removed using
the QTNRMVCR API. When ending the job-level commitment definition, this includes all API
commitment resources added by any program running in any activation group that is using the
job-level commitment definition.
v A remote database associated with the commitment definition to be ended must be disconnected.
v All protected conversations associated with the commitment definition must be ended normally using
the correct synchronization level.

If commitment control is being ended in an interactive job and one or more committable resources
associated with the commitment definition have pending changes, inquiry message CPA8350 is sent to
the user asking whether to commit the pending changes, roll back the pending changes, or cancel the
ENDCMTCTL request.

If commitment control is being ended in a batch job, and one or more closed files associated with the
commitment definition to be ended still have pending changes, the changes are rolled back and a
message is sent:

Commitment control 51
v CPF8356 if only local resources are registered
v CPF835C if only remote resources are registered
v CPF83E4 if both local and remote resources are registered

If a notify object is defined for the commitment definition being ended it may be updated.

When an activation group that has an API registered as the last agent is ending, the exit program for the
API is called to receive the commit or rollback decision. In this case, even though the activation group is
ending normally, a rollback request can still be returned from the API exit program. Thus, the implicit
commit operation might not be performed.

After the commitment definition has successfully ended, all the necessary recovery, if any, has been
performed. No additional recovery is performed for the commitment resources associated with the
commitment definition just ended.

After the commitment definition is ended, the job-level or activation-group-level commitment definition
can then be started again for the programs running within the activation group. The job-level
commitment definition may be started only if it is not already started for the job.

Although commitment definitions can be started and ended many times by the programs that run within
an activation group, the amount of system resources required for the repeated start and end operations
can cause a decrease in job performance and overall system performance. Therefore, it is recommended
that a commitment definition be left active if a program to be called later will use it.

For more information for on how the system updates the notify object, see “Updates to the notify object”
on page 60.

System-initiated end of commitment control

The system can end commitment control, or perform an implicit commit or rollback operation. Sometimes
the system-initiated end of commitment control is normal. Other times, commitment control ends with an
abnormal system or job end.

The following pages describe the situations in which the system ends commitment control ends implicitly,
and what actions, if any, to take:
v “Commitment control during activation group end”
v “Implicit commit and rollback operations” on page 53
v “Commitment control during normal routing step end” on page 57
v “Commitment control during abnormal system or job end” on page 57
v “Updates to the notify object” on page 60
v “Commitment control recovery during initial program load after abnormal end” on page 58

Commitment control during activation group end

The system automatically ends an activation-group-level commitment definition when an activation
group ends. If pending changes exist for an activation-group-level commitment definition and the
activation group is ending normally, the system performs an implicit commit operation for the
commitment definition before it is ended. Otherwise, an implicit rollback operation is performed for the
activation-group-level commitment definition before being ended if the activation group is ending
abnormally, or if errors were encountered by the system when closing any files opened under
commitment control scoped to the activation group.

52 iSeries: Commitment control

Note: An implicit commit or rollback operation is never
performed during activation-group end processing for the
*JOB or *DFTACTGRP commitment definitions. This is
because the *JOB and *DFTACTGRP commitment
definitions are never ended due to an activation group
ending. Instead, these commitment definitions are either
explicitly ended with an ENDCMTCTL command or
ended by the system when the job ends.

The system automatically closes any files scoped to the activation group when the activation group ends.
This includes any database files scoped to the activation group opened under commitment control. The
close for any such file occurs before any implicit commit operation that may be performed for the
activation-group-level commitment definition. Therefore, any records that reside in an I/O buffer are first
forced to the database before any implicit commit operation is performed.

As part of the implicit commit or rollback operation that may be performed, a call is made to the API
commit and rollback exit program for each API commitment resource associated with the
activation-group-level commitment definition. The exit program must complete its processing within 5
minutes. After the API commit and rollback exit program is called, the system automatically removes the
API commitment resource.

If an implicit rollback operation is performed for a commitment definition that is being ended due to an
activation group being ended, then the notify object, if one is defined for the commitment definition, may
be updated. See “Updates to the notify object” on page 60 for more information regarding the updating
of a notify object by the system.

Implicit commit and rollback operations

Typically, a commit or rollback operation is initiated from an application program using one of the
available programming languages that supports commitment control. These types of commit and rollback
operations are known as explicit commit and rollback requests. However, in some instances the system
initiates a commit or rollback operation for a commitment definition. Commit and rollback operations
initiated by the system are known as implicit commit and rollback requests.

The following two tables show what the system does when certain events occur related to a commitment
definition that has pending changes. A commitment definition has pending changes if any of the
following is true:
v Any committable resource has been updated.
v A database file opened under commitment control has been read because reading a file changes the file
position.
v The commitment definition has an API resource. Because changes to API resources are done by a user
program, the system must assume that all API resources have pending changes.

The C CM (commit operation) journal entry and C RB (rollback operation) journal entry indicate whether
the operation was explicit or implicit.

The following table shows the actions the system takes when a job ends, either normally or abnormally,
based on the following:
v The state of the transaction.
v The action-if-end job value for the commitment definition.
v Whether an API resource is the last agent.

Commitment control 53
 Commit or rollback
State Last agent API Action if Endjob1 option operation
RST N/A N/A If the commitment
definition is not associated
with an X/Open global
transaction, an implicit
rollback is performed.

If the commitment
definition is associated with
an X/Open global
transaction, the following
occurs:
v If the transaction branch
state is not Active (S1),
no action is performed
and the transaction
branch is left in the same
state.
v If the transaction branch
state is Active (S1), an
implicit rollback is
performed.
PIP N/A N/A If the commitment
definition is not associated
with an X/Open global
transaction, an implicit
rollback is performed.

If the commitment
definition is associated with
an X/Open global
transaction, the transaction
branch is in the Idle (S2)
state, and it is left in the
Idle (S2) state.
PRP N/A WAIT If the commitment
definition is not associated
with an X/Open2 global
transaction, the following
occurs:
v Resynchronization is
started to receive the
decision from the
initiator of the commit
operation.
v The returned decision to
commit or rollback is
performed. It is
considered an explicit
operation.

54 iSeries: Commitment control

 Commit or rollback
State Last agent API Action if Endjob1 option operation
PRP N/A C If the commitment
definition is not associated
with an X/Open2 global
transaction, an implicit
commit operation is
performed.
R If the commitment
definition is not associated
with an X/Open global
transaction, an implicit
rollback operation is
performed.

If the commitment
definition is associated with
an X/Open global
transaction, the following
occurs:
v If the job that started the
transaction ends, the
transaction is left in a
prepared state until the
XA TM either commits it
or rolls it back. The XA
transaction branch state
will be left at Prepared
(S3) in this case.
v If the SQL server job that
the transaction’s work is
being routed to is ended,
a forced rollback is
implicitly performed. The
XA transaction branch
state will be changed to
Heuristically Completed
(S5) in this case.
CIP N/A N/A An explicit commit
operation is performed.
LAP NO WAIT 1. Resynchronization to the
last agent is used to
retrieve the decision to
commit or to roll back.
2. The returned decision to
commit or to roll back is
performed. It is considered
an explicit operation.
LAP YES WAIT 1. The last agent API is
called to retrieve the
commit or rollback
decision.
2. The commit or rollback
operation is performed. It is
considered an explicit
operation.

Commitment control 55
 Commit or rollback
State Last agent API Action if Endjob1 option operation
LAP N/A C An implicit commit
operation is performed.
R An implicit rollback
operation is performed.
CMT N/A N/A A commit operation has
already completed for this
commitment definition and
any downstream locations.
The commit operation is
complete.
VRO N/A N/A The local and remote
agents voted to read-only.
All downstream agents
must also have voted to
read-only. No action is
required.
RBR N/A N/A A rollback operation is
required. An explicit
rollback operation is
performed.
Note:
1
You can change the Action if Endjob option with the Change Commitment Options (QTNCHGCO) API.

2
If the commitment definition is associated with an X/Open global transaction, the following occurs:
v If the job that started the transaction ends, the transaction is left in a prepared state until the XA TM either
commits it or rolls it back. The XA transaction branch state will be left at Prepared (S3) in this case.
v For transaction-scoped locks only, if the SQL server job that the transaction’s work is being routed to is ended, a
forced rollback is implicitly performed. The XA transaction branch state will be changed to Heuristically
Completed (S5) in this case.

The following table shows the actions the system takes when an activation group ends and applies only
to transactions with job-scoped locks. The system actions are based on the following:
v The state of the transaction. (It is always reset (RST) when an activation group ends).
v How the activation group ends-normally or abnormally.
v Whether an API resource is the last agent.

Note: If an API resource is registered as the last agent, this

gives control of the commit or rollback decision to the
last agent. The result of the decision is considered an
explicit operation

Commit or rollback
State Last agent API Type of end operation
RST No Normal An implicit commit
operation is performed. If
protected conversations
exist, the commitment
definition will become the
root initiator of the commit
operation.

56 iSeries: Commitment control

 Commit or rollback
State Last agent API Type of end operation
RST No Abnormal An implicit rollback is
performed.
RST Yes Normal The API exit program is
called. The commit or
rollback operation is
determined by the API.
RST Yes Abnormal The API exit program is
called. The commit or
rollback operation is
determined by the API.

Commitment control during normal routing step end

The system ends all commitment definitions for a job when a routing step is normally ended.

Note: The following applies only to commitment definitions

with job-scoped locks.

A routing step ends normally by one of the following:

v A normal end for a batch job.
v A normal sign-off for an interactive job.
v The Reroute Job (RRTJOB), Transfer Job (TFRJOB), or Transfer Batch Job (TFRBCHJOB) command ends
the current routing step and starts a new routing step.

Any other end of a routing step is considered abnormal and is recognized by a nonzero completion code
in job completion message CPF1164 in the job log.

Before ending a commitment definition during routing step end, the system performs an implicit rollback
operation if the commitment definition has pending changes. This includes calling the API commit and
rollback exit program for each API commitment resource associated with the commitment definition. The
exit program must complete its processing within 5 minutes. After the API commit and rollback exit
program is called, the system automatically removes the API commitment resource.

If a notify object is defined for the commitment definition, it may be updated. See “Updates to the notify
object” on page 60 for more information about the updating of a notify object by the system.

Commitment control during abnormal system or job end

This topic applies only to commitment definitions with job-scoped locks. The system ends all
commitment definitions for a job when the job ends abnormally. These commitment definitions are ended
during the end job processing. If the system ends abnormally, the system ends all commitment definitions
that were started and being used by all active jobs at the time of the abnormal system end. These
commitment definitions are ended as part of the database recovery processing that is performed during
the next IPL after the abnormal system end.

Attention:
The recovery for commitment definitions refers to an abnormal end for the system or a job due to a
power failure, a hardware failure, or a failure in the operating system or licensed internal code. You
must not use the End Job Abnormal (ENDJOBABN) command to force a job to end abnormally. The
abnormal end can result in pending changes for active transactions for the job you are ending to be
partially committed or rolled back. The next IPL might attempt recovery for any partial transactions
for the job ended with the ENDJOBABN command.

Commitment control 57
 The outcome of commitment control recovery that the system performs during an IPL for a job that
you end with the ENDJOBABN command is uncertain. This uncertainty is because all locks for
commitment resources are released when the job is ended abnormally. Any pending changes due to
partial transactions are made available to other jobs. These pending changes can then cause other
application programs to make additional erroneous changes to the database. Likewise, any ensuing
IPL recovery that is performed later can adversely affect the changes made by applications after the
job was ended abnormally. For example, an SQL table may be dropped during IPL recovery as the
rollback action for a pending create table. However, other applications may have already inserted
several rows into the table after the job was ended abnormally.

The system performs the following for commitment definitions being ended during an abnormal job end
or during the next IPL after an abnormal system end:
v Before ending a commitment definition, the system performs an implicit rollback operation if the
commitment definition has pending changes, unless processing for the commitment definition was
interrupted in the middle of a commit operation. If ended in the middle of a commit operation, the
transaction may be rolled back, resynchronized, or committed, depending on its state. See “Implicit
commit and rollback operations” on page 53. The processing to perform the implicit rollback operation
or to complete the commit operation includes calling the API commit and rollback exit program for
each API commitment resource associated with the commitment definition. After the API commit and
rollback exit program is called, the system automatically removes the API commitment resource.

Attention:
Ending the job while a transaction is in doubt (transaction state is LAP or PRP) can cause
inconsistencies in the database (changes might be committed on one or more systems and rolled
back on other systems).
– If the Action if Endjob commitment option is COMMIT, changes on this system are committed if
the job is ended, without regard to whether changes on the other systems participating in the
transaction are committed or rolled back.
– If the Action if Endjob commitment option is ROLLBACK, changes on this system are rolled
back if the job is ended, without regard to whether changes on the other systems participating
in the transaction are committed or rolled back.
– If the Action if Endjob commitment option is WAIT, the job will not end until resynchronization
completes to the system that owns the commit or rollback decision. To make the job end before
resynchronization is complete, a heuristic decision must be made and resynchronization must
be canceled.
Ending the job or system abnormally during a long-running rollback is not recommended. This
will cause another rollback to occur as the job ends (or during the next IPL if the system is
ended). The subsequent rollback will repeat the work performed by the original rollback and take
significantly longer to run.
v If a notify object is defined for the commitment definition, it may be updated. See “Updates to the
notify object” on page 60 for more information about the system updating a notify object.

If a process ends before commitment control is ended and protected conversations are still active, the
commitment definition may be required to commit or roll back. The action is based on the State option
and the Action if end job option for the commitment definition.

Commitment control recovery during initial program load after

abnormal end
When you perform an initial program load (IPL) after your system ends abnormally, the system attempts
to recover all the commitment definitions that were active when the system ended. Likewise, when you
vary on an independent disk pool, the system attempts to recover all the commitment definitions related
to that independent disk pool that were active when it was varied off or ended abnormally. The recovery

58 iSeries: Commitment control

is performed by database server jobs that are started by the system during IPL. Database server jobs are
started by the system to handle work that cannot or must not be performed by other jobs.

The database server jobs are named QDBSRVnn, where nn is a two-digit number. The number of database
server jobs depends on the size of your system. Likewise, the name of the database server job for an
independent disk pool or independent disk pool group is QDBSxxxVnn, where xxx is the independent
disk pol number and nn is a two-digit number. For example, QDBS035V02 can be the name of the
database server job for independent disk pool 35.

“States of the transaction for two-phase commitment control” on page 27 shows the actions that the
system takes, depending on the state of the transaction when the failure occurred. For two states, PRP
and LAP, the system action is in doubt.

Note: v The following applies only to commitment definitions

with job-scoped locks.
v The transaction manager recovers commitment
definitions associated with XA transactions (whether
their locks are job-scoped or transaction-scoped) using
XA APIs, not the resynchronization process described
in this topic.

The system cannot determine what to do until it performs resynchronization with the other locations that
participated in the transaction. This resynchronization is performed after the IPL or vary on operation
completes.

The system uses the database server jobs to perform this resynchronization. The commitment definitions
that need to be recovered are associated with the database server jobs. During the IPL, the system
acquires all record locks and other object locks that were held by the commitment definition before the
system ended. These locks are necessary to protect the local commitment resources until
resynchronization is complete and the resources can be committed or rolled back.

Messages are sent to the job logs of the database server jobs to indicate the status of resynchronization
with the remote locations. If the transaction is in doubt, resynchronization must be completed with the
location that owns the decision for the transaction before local resources can be committed or rolled back.

When the decision for a transaction is made, the following messages may be sent to the job log for the
database server job.

CPI8351
&1 pending changes being rolled back

CPC8355
Post-IPL recovery of commitment definition &8 for job &19/&18/&17 completed.

CPD835F
IPL recovery of commitment definition &8 for job &19/&18/&17 failed.

Other messages related to the recovery may also be sent. These messages are sent to the history (QHST)
log. If errors occur, messages are also sent to the QSYSOPR message queue.

You can determine the progress of the recovery by using iSeries(TM) Navigator, by displaying the job log
for the database server job, or by using the Work with Commitment Definitions (WRKCMTDFN)
command. Although iSeries Navigator and the Work with Commitment Definitions display allow you to
force the system to commit or roll back, you must use this only as a last resort. If you anticipate that all

Commitment control 59
of the locations that participated in the transaction will eventually be returned to operation, you must
allow the systems to resynchronize themselves. This ensures the integrity of your databases.

Updates to the notify object

For purposes of the notify object, the following are considered uncommitted changes:
v An update to a record that is made under commitment control.
v A record that is deleted under commitment control.
v An object level change that is made to a local DDL object under commitment control.
v A read operation performed for a database file that was opened under commitment control. This is
because file position is brought back to the last commitment boundary when a rollback operation is
performed. If you perform a read operation under commitment control, the file position is changed
and therefore, an uncommitted change then exists for the commitment definition.
v A commitment definition with one of the following resources that are added is always considered to
have uncommitted changes:
– An API commitment resource
– A remote Distributed Relational Database Architecture(TM) (DRDA(R)*) resource
– A Distributed Database Management Architecture (DDM) resource
– An LU 6.2 resource
This is because the system does not know when a real change is made to the object or objects that are
associated with these types of resources. “Types of committable resources” on page 13 has more
information about how you add and work with these types of resources.

The system makes updates to the notify object and are based on the following ways that a commitment
definition can end:
v If a job ends normally and no uncommitted changes exist, the system does not place the commit
identification of the last successful commit operation in the notify object.
v If an implicit commit operation is performed for an activation-group-level commitment definition when
the activation group is ended, the system does not place the commit identification of the last successful
commit operation in the notify object.

Note: Implicit commit operations are never performed for the

*DFTACTGRP or *JOB commitment definition

v If the system, job, or an activation group ends abnormally before the first successful commit operation
for a commitment definition, the system does not update the notify object because there is no last
commit identification. To differentiate between this condition and a normal program completion, your
program must update the notify object with a specific entry prior to completing the first successful
commit operation for the commitment definition.
v If an abnormal job end or an abnormal system end occurs after at least one successful commit
operation, the system places the commit identification of that commit operation in the notify object. If
the last successful commit operation did not specify a commit identification, then the notify object is
not updated. For an abnormal job end, this notify object processing is performed for each commitment
definition that was active for the job. For an abnormal system end, this notify object processing is
performed for each commitment definition that was active for all jobs on the system.
v The system updates the notify object with the commit identification of the last successful commit
operation for that commitment definition if all of the following occur:
– A non-default activation group ends.
– An implicit rollback operation is performed for the activation-group-level commitment definition.
– At least one successful commit operation has been performed for that commitment definition.
If the last successful commit operation did not specify a commit identification, then the notify object is
not updated. An implicit rollback operation is performed for an activation-group-level commitment

60 iSeries: Commitment control

 definition if the activation group is ending abnormally or errors occurred when closing the files opened
under commitment control that were scoped to that activation group. For more information about
scoping database files to activation groups and how activation groups can be ended, see the reference
book for the ILE language that you are using.
v If uncommitted changes exist when a job ends normally and at least one successful commit operation
has been performed, the commit identification of the last successful commit operation is placed in the
notify object and the uncommitted changes are rolled back. If the last successful commit operation did
not specify a commit identification, then the notify object is not updated. This notify object processing
is performed for each commitment definition that was active for the job when the job ended. For more
information about the functions that are done during the normal ending of a job, see the topic
“Commitment control during normal routing step end” on page 57.
v If uncommitted changes exist when the ENDCMTCTL command is run, the notify object is updated
only if the last successful commit operation specified a commit identification:
– For a batch job, the uncommitted changes are rolled back and the commit identification of the last
successful commit operation is placed in the notify object.
– For an interactive job, if the response to inquiry message CPA8350 is to rollback the changes, the
uncommitted changes are rolled back and the commit identification of the last successful commit
operation is placed in the notify object.
– For an interactive job, if the response to inquiry message CPA8350 is to commit the changes, the
system prompts for a commit identification to use and the changes are committed. The commit
identification that is entered on the prompt display is placed in the notify object.
– For an interactive job, if the response to inquiry message CPA8350 is to cancel the ENDCMTCTL
request, the pending changes remain and the notify object is not updated.

Manage transactions and commitment control

Following are tasks that you can do to manage commitment control.

“Display commitment control information”

This information has tasks that can do to display information about all transactions on the system
and display information about jobs associated with transactions.

“Optimize performance for commitment control” on page 63

This information has tasks that you can do to minimize the effect that commitment control has on
system performance.

Display commitment control information

You can use iSeries(TM) Navigator to display information about all transactions (logical units of work) on
the system. You can also look at information about the job, if any, associated with a transaction.

Note: These display operations do not display the isolation

level for SQL applications.

To display information proceed as follows:

1. In the iSeries Navigator window, expand the server you want to use.
2. Expand Databases.
3. Expand the system you want to work with.
4. Expand Transactions.

Note: To view a transaction that is associated with an X/Open

global transaction, expand Global Transactions. To view
a DB2(R) UDB managed transaction expand Database
Transactions.

Commitment control 61
5. Expand Global Transactions or Database Transactions.

This display shows the following:

v Unit of Work ID
v Unit of Work State
v Job
v User
v Number
v Resynchronization in Progress
v Commitment Definition

Online help provides information about all the status displays and the fields on each display.

You can also use iSeries Navigator to display the following information:
v “Display locked objects for a transaction”
v “Display jobs associated with a transaction”
v “Display resource status of a transaction”
v “Display transaction properties” on page 63

Display locked objects for a transaction

You can display locked objects for global transactions with transaction-scoped locks only.

To display locked objects for a transaction:

1. In the iSeries(TM) Navigator window, expand the server you want to use.
2. Expand Databases.
3. Expand the system you want to work with
4. Expand Transactions.
5. Expand Global Transactions.
6. Right-click the transaction that you want to work with and select Locked Objects.

Display jobs associated with a transaction

To display jobs associated with a transaction:
1. In the iSeries(TM) Navigator window, expand the server you want to use.
2. Expand Databases.
3. Expand the system you want to work with
4. Expand Transactions.
5. Expand Global Transactions or Database Transactions.
6. Right-click the transaction that you want to work with and select Jobs.

For database transactions and global transactions with job-scoped locks, a list of the jobs associated with
the transaction is displayed.

For global transactions with transaction-scoped locks, a list of jobs with this transaction object attached or
waiting for this transaction object to be attached is displayed

Display resource status of a transaction

To display resource status of a transaction:
1. In the iSeries(TM) Navigator window, expand the server you want to use.
2. Expand Databases.

62 iSeries: Commitment control

3. Expand the system you want to work with
4. Expand Transactions.
5. Expand Global Transactions or Database Transactions.
6. Right-click the transaction that you want to work with and select Resource status.

Display transaction properties

To display transaction properties:
1. In the iSeries(TM) Navigator window, expand the server you want to use.
2. Expand Databases.
3. Expand the system you want to work with
4. Expand Transactions.
5. Expand Global Transactions or Database Transactions.
6. Right-click the transaction that you want to work with and select Properties.

Optimize performance for commitment control

Using commitment control requires resources that can affect system performance. Several factors affect
system performance regarding commitment control. Following are a factor that does not affect
performance, factors that degrade performance, and factors that improve performance.

Factor that does not affect performance

Opening a file
If you open a file without specifying the commit open option, no additional system resource is used
even if a commitment definition has been started. For more information about specifying the
commit open option, see the appropriate high-level language reference manual.

Factors that degrade performance

Journaling
Journaling a file requires system resources. However, in most cases journaling performs better with
commitment control than without commitment control. If you specify only after-images,
commitment control changes this to both before-images and after-images while commitment control
is in effect. Typically this is a space, not a performance, consideration. See the Journal management
topic for more information about journaling.

Commit operation
If any changes were made to journaled resources during the transaction, each commit of a
transaction adds two entries to each journal related to those resources. The number of entries added
can increase significantly for a large volume of small transactions. You may want to place the
journal receivers in a separate disk pool from the journals.

Rollback operation
Because commitment control must rollback the pending changes recorded in the database,
additional system resources are required whenever a rollback occurs. Also, if record changes are
pending, a rollback operation causes additional entries to be added to the journal.

Start Commitment Control (STRCMTCTL) and End Commitment Control (ENDCMTCTL)

commands
Additional overhead is incurred by the system each time a commitment definition is started and
ended using the STRCMTCTL and ENDCMTCTL commands respectively. Avoid using the
STRCMTCTL and ENDCMTCTL commands for each transaction. Use them only when necessary.
You can establish a commitment definition at the beginning of an interactive job and use it for the
duration of the job.

Commitment control 63
 Using more than one journal for commitment control transactions
With two-phase commit, files that are opened under commitment control can be journaled to more
than one journal. However, using more than one journal takes additional system resources to
manage the commitment definition. Using more than one journal can also make recovery more
complicated.

Record locking
Record locking can affect other applications. The number of records locked within a particular job
increases the overall system resources used for the job. Applications needing to access the same
record must wait for the transaction to end.

Requesting SEQONLY(*YES)
If you request the SEQONLY(*YES) option (by using the OVRDBF command or the application
program implicitly attempts to use SEQONLY(*YES)) and the file is opened for input only under
commitment control with LCKLVL(*ALL), the option is changed to SEQONLY(*NO). This option can
affect the performance of input files because records will not be blocked.

Requesting a record-level change for a database file when save-while-active processing is active
A request to make a record-level change under commitment control for a database file may be
delayed if the commitment definition is at a commitment boundary and a save-while-active
operation is running in a different job. This can happen when a file is journaled to the same journal
as some of the objects on the save request.

Note: The Status column on the Work with Active Jobs

(WRKACTJOB command) display shows CMTW (commit
wait) when a job is being held due to save-while-active
checkpoint processing.

Committing or rolling back changes when save-while-active processing is active

A commit or rollback operation may be delayed at a commitment boundary when a
save-while-active operation is running in a different job. This can happen when an API commitment
resource was previously added to the commitment definition, unless the API resource was added
using the QTNADDCR API and the Allow normal save processing field has a value of Y.

Because the job is held during the commit or rollback request, and because a commit or rollback
request can be performed only for a single commitment definition at a time, jobs with more than
one commitment definition with API commitment resources added can prevent a save-while-active
operation from completing.

Note: If you use the new save with partial transactions feature the object can be saved without
ending a commitment definition.

Requesting an object-level change when save-while-active processing is active

A request to make an object-level change under commitment control may be delayed if the
commitment definition is at a commitment boundary and a save-while-active operation is running
in a different job. This can happen when an object-level change is made while the save-while-active
operation is running against the library the object is in. For example, the create SQL table operation
under commitment control for table MYTBL in library MYSQLLIB may be delayed when a
save-while-active operation is running against library MYSQLLIB.

64 iSeries: Commitment control

Note: If the wait time exceeds 60 seconds, inquiry message
CPA8351 is sent to ask the user whether to continue
waiting or cancel the operation.

Adding an API resource using the QTNADDCR API

A request to add an API commitment resource using the QTNADDCR API may be delayed if all
commitment definitions for the job are at a commitment boundary and a save-while-active operation
is running in a different job.

Note:
1. If the wait time exceeds 60 seconds, inquiry message CPA8351 is sent to ask the user whether to
continue waiting or cancel the operation.
2. This does not apply to API resources that were added using the QTNADDCR API if the Allow
normal save processing field has a value of Y.

Factors that improve performance

Using a default journal

Using a default journal can help performance if you close and reopen all files under commitment
control while the commitment definition is active. However, using a default journal with
OMTJRNE(*NONE) degrades the performance of commit and rollback operations.

Selecting a last agent

Performance is enhanced when a last agent is selected because fewer interactions between the
system and the last agent are required during a commit operation. However, if a communications
failure occurs during a commit operation, the commit operation will not complete until
resynchronization completes regardless of the value of the wait for outcome option. Such a failure is
rare but this option allows the application writer to consider the negative impact of causing the user
to wait indefinitely for the resynchronization to complete when a failure does occur. Weigh this
against the performance enhancement that is provided by last agent optimization during successful
commit operations. This consideration is generally more significant for interactive jobs than for
batch jobs.

The default is that a last agent is permitted to be selected by the system but the user can modify
this value using the QTNCHGCO API.

Not using the wait for outcome option

When remote resources are under commitment control, performance is improved when the Wait for
Outcome option is set to N (No) and all remote systems support presumed abort. The Wait for
Outcome option is set to N by the system for DRDA(R) and DDM application when the first
connection is made to a remote system. APPC applications must explicitly set the Wait for Outcome
option, or the default value of Y will be used.

Selecting the OK to Leave Out option

Performance is improved when the OK to Leave Out option is selected. For further information
about this option, see “Commitment definition for two-phase commit: Indicate OK to leave out” on
page 35.

Selecting the Vote Read Only option

Performance is improved when the Vote Read Only option is selected. For further information about
this option, see “Commitment definition for two-phase commit: Allow vote read-only” on page 30.

Tasks that you can do to improve performance are:

v “Minimize locks” on page 66

Commitment control 65
v “Manage transaction size” on page 67

Minimize locks
A typical way to minimize record locks is to release the record lock. (This technique does not work if
LCKLVL(*ALL) has been specified.) For example, a single file maintenance application typically does the
following:
v Displays a prompt for a record identification to be changed.
v Retrieves the requested record.
v Displays the record.
v Allows the workstation user to make the change.
v Updates the record.

In most cases, the record is locked from the access of the requested record through the update. The record
wait time may be exceeded for another job that is waiting for the record. To avoid locking a record while
the workstation user is considering a change, release the record after it is retrieved from the database
(before the record display appears). You then need to access the record again before updating. If the
record was changed between the time it was released and the time it was accessed again, you must
inform the workstation user. The program can determine if the record was changed by saving one or
more fields of the original record and comparing them to the fields in the same record after it is
retrieved, as follows:
v Use an update count field in the record and add 1 to the field just before an update. The program
saves the original value and compares it to the value in the field when the record is retrieved again. If
a change has occurred, the workstation user is informed and the record appears again. The update
count field is changed only if an update occurs. The record is released while the workstation user is
considering a change. If you use this technique, you must use it in every program that updates the file.
v Save the contents of the entire data record and compare it to the record the next time it is retrieved.

In both cases above, the sequence of operations prevents the simple use of externally described data in
RPG where the same field names are used in the master record and in the display file. Using the same
field names (in RPG) does not work because the workstation user’s changes are overlaid when the record
is retrieved again.

You can solve this problem by moving the record data to a data structure or continue to use externally
described data if you use the DDS keyword RTNDTA. The RTNDTA keyword allows your program to
reread data on the display without the operating system having to move data from the display to the
program. This allows the program to do the following:
1. Prompt for the record identification.
2. Retrieve the requested record from the database.
3. Release the record.
4. Save the field or fields used to determine if the record was changed.
5. Display the record and wait for the workstation user to respond.

If the workstation user changes the record on the display, the program uses the following sequence:
1. Retrieves the record from the database again.
2. Compares the saved fields to determine if the database record has been changed. If it has been
changed, the program releases the record and sends a message when the record appears.
3. Retrieves the record from the display by running a read operation with the RTNDTA keyword and
updates the record in the database record.
4. Proceeds to the next logical prompt because there are no additional records to be released if the
workstation user cancels the request.

66 iSeries: Commitment control

LCKLVL(*CHG) and LCKLVL(*CS) work in this situation. If LCKLVL(*ALL) is used, you must release the
record lock by using a commit or rollback operation.

For more information about locks see “Detect deadlocks” on page 100.

Manage transaction size

For this discussion, a transaction is interactive. (Commitment control can also be used for batch
applications, which often can be considered a series of transactions. Many of the same considerations
apply to batch applications, which are discussed in “Commitment control for batch applications” on page
24.)

You can lock a maximum of 500 000 000 records during a transaction for each journal associated with the
transaction. You can reduce this limit by using a Query Options File (QAQQINI). Use the QRYOPTLIB
parameter of the Change Query Attributes (CHGQRYA) command to specify a Query Options File for a
job to use. Use the COMMITMENT_CONTROL_LOCK_LEVEL value in the Query Options File as the
lock limit for the job.

When choosing the lock level for your records, consider the size of your transactions. Use size to
determine how long records are locked before a transaction ends. You must decide if a commit or
rollback operation for commitment control is limited to a single use of the Enter key, or if the transaction
consists of many uses of the Enter key.

Note: The shorter the transaction, the earlier the job waiting to
start save-while-active checkpoint processing can continue
and complete.

For example, for an order entry application, a customer might order several items in a single order
requiring an order detail record and an inventory master record update for every item in the order. If the
transaction is defined as the entire order and each use of the Enter key orders an item, all records
involved in the order are locked for the duration of the entire order. Therefore, often used records (such
as inventory master records) might be locked for long periods of time, preventing other work from
progressing. If all items are entered with a single Enter key using a subfile, the duration of the locks for
the entire order is minimized.

In general, the number and duration of locks must be minimized so several workstation users can access
the same data without long waiting periods. You can do this by not holding locks while the user is
entering data on the display. Some applications may not require more than one workstation user
accessing the same data. For example, in a cash posting application with many open item records per
customer, the typical approach is to lock all the records and delay them until a workstation user
completes posting the cash for a given receipt.

If the workstation user presses the Enter key several times for a transaction, it is possible to perform the
transaction in a number of segments. For example:
v The first segment is an inquiry in which the workstation user requests the information.
v The second segment is a confirmation of the workstation user’s intent to complete the entire
transaction.
v The third segment is retrieval and update of the affected records.

This approach allows record locking to be restricted to a single use of Enter.

This inquiry-first approach is normally used in applications where a decision results from information
displayed. For example, in an airline reservation application, a customer may want to know what flight
times, connecting flights, and seating arrangements are available before making a decision on which
flight to take. Once the customer makes a decision, the transaction is entered. If the transaction fails (the

Commitment control 67
flight is now full), the rollback function can be used and a different request entered. If the records were
locked from the first inquiry until a decision is made, another reservation clerk would be waiting until
the other transaction is complete.

Scenarios and examples: Commitment control

The following are a scenario and examples of commitment control. The scenario shows at a high level
how the JKL toy company implements a commitment control to track transactions on their local database.

The examples provide sample code for commitment control. The Practice problem is an RPG program
that implements commitment control. It includes a logic flow that shows what is happening each step of
the way.

The three following examples show example uses of commitment control to start applications after an
abnormal system end.

Scenario
v “Scenario: Commitment control”

Examples
v “Practice problem for commitment control” on page 71
v Logic flow for the practice problem for commitment control
v “Example: Use a transaction logging file to start an application” on page 77
v “Example: Use a notify object to start an application” on page 81
v “Example: Use a standard processing program to start an application” on page 87

Note: Read the Code example disclaimer for important legal information.

Scenario: Commitment control

The JKL Toy Company uses commitment control to protect the database records for manufacturing and
inventory. This scenario, shows at a high level how JKL toy company uses commitment control as a part
is transferred from their inventory department to their manufacturing department.

See Scenario: Journal management for a description of JKL Toy Company’s network environment. The
scenario that follows shows how commitment control works on their production server, JKLPROD.

This scenario illustrates the advantages of using commitment control in two examples. The first example
shows how the company’s inventory program, Program A, might work without commitment control, and
the possible problems that can occur. The second example shows how the program works with
commitment control.

JKL Toy Company uses an inventory application program, Program A, on their server JKLPROD.
Program A uses two records. One record tracks items that are stored in the stock room. Another record
keeps track of items that are removed from the stock room, and used in production.

Program A without commitment control

Assume that the following application program does not use commitment control. The system locks
records read for updating. The following steps describe how the application program tracks a diode as it
is removed from the stock room and transferred to checking account:
v Program A locks and retrieves the stock room record. (This action might require a wait if the record is
locked by another program.)

68 iSeries: Commitment control

v Program A locks and retrieves the production record. (This might also require a wait.) Program A now
has both records locked, and no other program can change them.
v Program A updates the stock room record. This causes the record to be released so it is now available
to be read for update by any other program.
v Program A updates the production record, which causes the record to be released so it is now available
to be read for update by any other program.

Without using commitment control, a problem needs to be solved to make this program work properly in
all circumstances. For example, a problem occurs if program A does not update both records because of a
job or system failure. In this case, the two files are not consistent — diodes are removed from the stock
room record, but they are not added to the production record. Using commitment control allows you to
ensure that all changes involved in the transaction are completed or that the files are returned to their
original state if the processing of the transaction is interrupted.

Program A with commitment control

If commitment control is used, the preceding example is changed as follows:

1. Commitment control is started.
2. Program A locks and retrieves the stock room record. (This action can require a wait if the record is
locked by another program.)
3. Program A locks and retrieves the production record. (This can also require a wait.) Program A now
has both records locked, and no other program can change them.
4. Program A updates the stock room record, and commitment control keeps the lock on the record.
5. Program A updates the production record, and commitment control keeps the lock on the record.
6. Program A commits the transaction. The changes to the stock room record and the production record
are made permanent in the files. The changes are recorded in the journal, which assumes they will
appear on disk. Commitment control releases the locks on both records. The records are now available
to be read for update by any other program.

Because the locks on both records are kept by commitment control until the transaction is committed, a
situation cannot arise in which one record is updated and the other is not. If a routing step or system
failure occurs before the transaction is committed, the system removes (rolls back) the changes that have
been made so that the files are updated to the point where the last transaction was committed.

For each routing step in which files are to be under commitment control, the steps shown in the
following figure occur:

Commitment control 69
The operations that are performed under commitment control are journaled to the journal. The start
commitment control journal entry appears after the first file open entry under commitment control. This
is because the first file open entry determines what journal is used for commitment control. The journal
entry from the first open operation is then used to check subsequent open operations to ensure all files
are using the same journal.

When a job failure or system failure occurs, the resources under commitment control are updated to a
commitment boundary. If a transaction is started but is not completed before a routing step ends, that
transaction is rolled back by the system and does not appear in the file after the routing step ends. If the
system abnormally ends before a transaction is completed, that transaction is rolled back by the system
and does not appear in the file after a subsequent successful initial program load (IPL) of licensed
internal code. Anytime a rollback occurs, reversing entries are placed in the journal.

For example, assume JKL company has 100 diodes in stock. Manufacturing takes out 20 from stock, for a
new balance of 80. The database update causes both before-image (100) and after-image (80) journal
entries.

Assume the system abnormally ended after journaling the entries, but before reaching the commitment
point or rollback point. After the IPL, the system reads the journal entry and updates the corresponding
database record. This update produces two journal entries that reverse the update: the first entry is the
before-image (80) and the second entry is the after-image (100).

70 iSeries: Commitment control

When the IPL is successfully completed after the abnormal end, the system removes (or rolls back) any
database changes that are not committed. In the preceding example, the system removes the changes
from the stock room record because a commit operation is not in the journal for that transaction. In this
case, the before-image of the stock room record is placed in the file. The journal contains the rolled back
changes, and an indication that a rollback operation occurred.

Practice problem for commitment control

This practice problem will assist you in understanding commitment control and its requirements. The
following steps assume you are familiar with the OS/400(R) licensed program and the data file utility
(DFU), and have read this topic. The Logic flow will help you further understand this practice program
for commitment control.

Note: Read the Code example disclaimer for important legal information.

Before beginning this problem, do the following:

v Create a special library for this practice problem. In the instructions, the library is called CMTLIB.
Substitute the name of your library where you see CMTLIB.
v Create source files and a job description.

Perform the following steps:

1. Create a physical file named ITMP (item master file). The data description specification (DDS) for
this file is:
10 A R ITMR
20 A ITEM 2
30 A ONHAND 5 0
40 A K ITEM
2. Create a physical file named TRNP (transaction file). This file is used as a transaction log file. The
DDS for this file is:
10 A R TRNR
20 A QTY 5 0
30 A ITEM 2
40 A USER 10
3. Create a logical file named TRNL (transaction logical). This file is used to assist in starting the
application again. The USER field is the type LIFO sequence. The DDS for this file is:
10 LIFO
20 A R TRNR PFILE (TRNP)
30 A K USER
4. Enter the STRDFU command, and create a DFU application named ITMU for the ITMP file. Accept
the defaults offered by DFU during the application definition.
5. Type the command CHGDTA ITMU and enter the following records for the ITMP file:

Item On hand
AA 450
BB 375
CC 4000

6. End the program using F3. This entry provides some data against which the program will operate.
7. Create the CL program Item Process (ITMPCSC) as follows:
PGM
DCL &USER *CHAR LEN(10)
RTVJOBA USER(&USER)
CALL ITMPCS PARM(&USER)
ENDPGM
This is the control program that calls the ITMPCS program. It retrieves the user name and passes it
to the processing program. This application assumes that unique user names are used.

Commitment control 71
 8. Create a display file named ITMPCSD from the DDS as follows.
There are two formats, the first for the basic prompt display and the second to allow the operator to
review the last transaction entered. This display file is used by the ITMPCS program.
SEQNBR *... ... 1 ... ... 2 ... ... 3 ... ... 4 ... ... 5 ... ... 6 ... ... 7 ..

1.00 A R PROMPT
2.00 A CA03(93 ’End of program’)
3.00 A CA04(94 ’Review last’)
4.00 A SETOFF(64 ’No rcd to rvw’)
5.00 A 1 2’INVENTORY TRANSACTIONS’
6.00 A 3 2’Quantity’
7.00 A QTY 5 0I +1
8.00 A 61 ERRMSG(’Invalid +
9.00 A quantity’ 61)
10.00 A +5’ITEM’
11.00 A ITEM 2 I +1
12.00 A 62 ERRMSG(’Invalid +
13.00 A Item number’ 62)
14.00 A 63 ERRMSG(’Rollback +
15.00 A occurred’ 63)
16.00 A 64 24 2’CF4 was pressed and +
17.00 A there are no +
18.00 A transactions for +
19.00 A this user’
20.00 A DSPATR(HI)
21.00 A 23 2’CF4 Review last +
22.00 A transaction’
23.00 A R REVW
24.00 A 1 2’INVENTORY TRANSACTIONS’
25.00 A +5’REVIEW LAST TRANSACTION’
26.00 A 3 2’Quantity’
27.00 A QTY 5 0 +1EDTCDE(Z)
28.00 A +5’Item’
29.00 A ITEM 2 +1
9. Study the logic flow provided in Logic flow for the practice program for commitment control.
10. Enter the STRSEU command and type the source as follows:
SEQNBR *... ... 1 ... ... 2 ... ... 3 ... ... 4 ... ... 5 ... ... 6 ... ... 7 ..

1.00 FITMP UF E K DISK

2.00 F* KCOMIT
3.00 FTRNP O E DISK
4.00 F* KCOMIT
5.00 FTRNL IF E K DISK
6.00 F TRNR KRENAMETRNR1
7.00 FITMPCSD CF E WORKSTN
8.00 C* Enter parameter with User name for -TRNP- file
9.00 C *ENTRY PLIST
10.00 C PARM USER 10
11.00 C LOOP TAG
12.00 C EXFMTPROMPT
13.00 C* Check for CF3 for end of program
14.00 C 93 DO End of Pgm
15.00 C SETON LR
16.00 C RETRN
17.00 C END
18.00 C* Check for CF4 for review last transaction
19.00 C 94 DO Review last
20.00 C* Check for existence of a record for this user in -TRNL- file
21.00 C USER CHAINTRNR1 64 Not found
22.00 C 64 GOTO LOOP
23.00 C EXFMTREVW
24.00 C GOTO LOOP
25.00 C END
26.00 C* Access Item record
27.00 C ITEM CHAINITMR 62 Not found

72 iSeries: Commitment control

 28.00 C* Handle -not found- Condition
29.00 C 62 GOTO LOOP
30.00 C* Does sufficient quantity exist
31.00 C ONHAND SUB QTY TEST 50 61 Minus
32.00 C* Handle insufficient quantity
33.00 C 61 DO
34.00 C* Release Item record which was locked by the CHAIN for update
35.00 C EXCPTRLSITM
36.00 C GOTO LOOP
37.00 C END
38.00 C* Change ONHAND and update the Item record
39.00 C Z-ADDTEST ONHAND
40.00 C UPDATITMR
41.00 C* Test for Special Simulation Conditions
42.00 C ITEM IFEQ ’CC’
43.00 C* Simulate program need for rollback
44.00 C QTY IFEQ 100
45.00 C SETON 63 Simult Rlbck
46.00 C* ROLBK
47.00 C GOTO LOOP
48.00 C END
49.00 C* Simulate an abnormal program cancellation by Div by zero
50.00 C* Operator Should respond -C- to inquiry message
51.00 C QTY IFEQ 101
52.00 C Z-ADD0 ZERO 30
53.00 C TESTZ DIV ZERO TESTZ 30 Msg occurs
54.00 C END
55.00 C* Simulate an abnormal job cancellation by DSPLY.
56.00 C* Operator Should System Request to another job
57.00 C* and cancel this one with OPTION(*IMMED)
58.00 C QTY IFEQ 102
59.00 C ’CC=102’ DSPLY Msg occurs
60.00 C END
61 00 C END ITEM=CC
62.00 C* Write the -TRNP- file
63 00 C WRITETRNR
64.00 C* Commit the update to -ITMP- and write to -TRNP-
65.00 C* COMIT
66.00 C GOTO LOOP
67.00 OITMR E RLSITM
11. Enter the CRTRPGPGM command to create program ITMPCS from the source entered in the
previous step.
12. Type the command CALL ITMPCSC, press Enter, and press F4. A message states that there are no
entries for this operator.
13. Enter the following data to see if the program operates correctly:

Quantity Item
3 AA
4 BB

14. Press F4. The review display shows the BB item last entered. Enter the following data:

Quantity Item
5 FF (Invalid item number message occurs.)
9000 BB (Insufficient quantity error message occurs.)
100 CC (Rollback message occurs.)
102 CC (RPG DSPLY operation must occur. Press the Enter
key.)

Commitment control 73
Quantity Item
101 CC (The program must display an inquiry message
stating that a divide by zero condition has occurred or
end, depending on the setting of job attribute
INQMSGRPY. If the inquiry message appears, enter C to
cancel the RPG program and then C to cancel the CL
program on the subsequent inquiry. This simulates an
unexpected error condition.)

15. Type the Display Data command DSPDTA ITMP.

See if the records AA and BB have been updated correctly. The values must be AA = 447, BB = 371,
and CC = 3697. Note that the quantities subtracted from CC occurred, but the transaction records
were not written.
16. Create a journal receiver for commitment control. Use the Create Journal Receiver (CRTJRNRCV)
command to create a journal receiver called RCVR1 in the CMRLIB library. Specify a threshold of at
least 5000KB. A larger threshold is recommended if your system has sufficient space in order to
maximize the time between generation of new journal receivers to minimize the performance
impacts of too frequent change journals
17. Create a journal for commitment control. Use the Create Journal (CRTJRN) command to create a
journal called JRNTEST in the CMTLIB library. Because this journal is used only for commitment
control, specify MNGRCV(*SYSTEM) DLTRCV(*YES). For the JRNRCV parameter, specify the journal
receiver that you created in step 16 (page 74).
18. Use the Start Journal Physical File (STRJRNPF) command with the parameters FILE(CMTLIB/ITMP
CMTLIB/TRNP) JRN(CMTLIB/JRNTEST) to journal the files to be used for commitment control.
The IMAGES parameter uses a default of *AFTER, meaning only after-image changes of the records
appear in the journal. The files ITMP and TRNP have now started journaling.
Normally, you save the files after starting journaling. You cannot apply journaled changes to a
restored file that does not have the same JID as the journal entries. Because this practice problem
does not require you to apply journaled changes, you can skip saving the journaled files.
19. Type the command CALL ITMPCSC and enter the following transactions:

Quantity Item
5 AA
6 BB

End the program by pressing F3.

20. Type the Display Journal command: DSPJRN CMTLIB/JRNTEST.
Note the entries appearing in the journal. The same sequence of entries (R UP = update of ITMP
followed by R PT = record added to TRNP) occurs in the journal as was performed by the program.
This is because a logical file is defined over the physical file TRNP and the system overrides the
RPG default. If no logical file existed, the RPG assumption of SEQONLY(*YES) is used, and a block
of PT entries appear because the records are kept in the RPG buffer until the block is full.
21. Change the CL program ITMPCSC as follows (the new statements are shown with an asterisk):
PGM
DCL &USER *CHAR LEN(10)
RTVJOBA USER(&USER)
* STRCMTCTL LCKLVL(*CHG)
CALL ITMPCS PARM(&USER)
* MONMSG MSGID(RPG9001) EXEC(ROLLBACK)
* ENDCMTCTL
ENDPGM
The STRCMTCTL command sets up the commitment control environment. The LCKLVL word
specifies that records read for update but not updated can be released during the transaction. The

74 iSeries: Commitment control

 MONMSG command handles any RPG escape messages and performs a ROLLBACK in case the
RPG program abnormally ends. The ENDCMTCTL command ends the commitment control
environment.
22. Delete the existing ITMPCSC program and create it again.
23. Change the RPG program to remove the comment symbols at statements 2.00, 4.00, 46.00, and 65.00.
The source is now ready for use with commitment control.
24. Delete the existing ITMPCS program and create it again. The program is now ready to operate under
commitment control.
25. Type the command CALL ITMPCSC and the following transactions:

Quantity Item
7 AA
8 BB

26. Use System Request and request the option to display the current job. When the Display Job display
appears, select option 16 to request the display of the commitment control status.
Note the values on the display. There must be two commits because two commit statements were
run in the program.
27. Press F9 to see a list of the files under commitment control and the amount of activity for each file.
28. Return to the program and end it by pressing F3.
29. Type DSPJRN CMTLIB/JRNTEST and note the entries for the files and the special journal entries for
commitment control:

C BC STRCMTCTL command occurred.

C SC Start commit cycle. This occurs whenever the first
database operation in the transaction causes a record to
be inserted, updated, or deleted as part of commitment
control.
C CM Commit operation has occurred.
C EC ENDCMTCTL command occurred.

The commitment control before-images and after-images (R UB and R UP types) automatically occur
even though you had originally requested IMAGES(*AFTER) for the journal.
30. Type the command CALL ITMPCSC and the following transactions:

Quantity Item
12 AA
100 CC (This is the condition to simulate the need for an
application use of rollback. The CC record in the ITMP
file, which was updated by RPG statement 40.00 is rolled
back.)

31. Press F4 to determine the last transaction entered.

The last committed transaction is the entry for item AA.
32. Use System Request and request the Display Current Job option. When the Display Job display
appears, request the display of the commitment control status.
Note the values on the display and how they have been changed by the rollback.
33. Return to the program.
34. Return to the basic prompt display and end the program by pressing F3.
35. Type the command DSPJRN CMTLIB/JRNTEST.

Commitment control 75
 Note the additional entries that appear in the journal for the use of the rollback entry (C RB entry).
When the ITMP record is rolled back, three entries are placed in the journal. This is because any
change to the database file under commitment control produces a before (R BR) and after (R UR)
entry.
36. Display the entries with journal code R and these entry types: UB, UP, BR, and UR. Use option 5 to
display the full entries. Because the Quantity field is in packed decimal, use F11 to request a hex
display. Note the following:
v The on-hand value of the ITMP record in the UB record
v How the on-hand value is reduced by the UP record
v How the BR record is the same as the UP record
v How the UR record returns the value as originally displayed for the UB record
The last entry is the RB entry for the end of the rollback.
37. Type the command CALL ITMPCSC, press Enter, and press F4. Note the last transaction entered.
38. Type the following transactions:

Quantity Item
13 AA
101 CC (This is the condition to simulate an unexpected error
condition, which causes the program to end. The
simulation occurs by dividing a field by 0. The program
will display an inquiry message or end, depending on the
setting of the job attribute INQMSGRPY. If the inquiry
message appears, enter C to end the program. Because
the CL program was changed to monitor for RPG
program errors, the second inquiry which occurred does
not occur.)

39. Type the command DSPJRN CMTLIB/JRNTEST.

The same type of rollback handling has occurred, but this time the rollback was caused by the EXEC
parameter of the MONMSG command in the CL program instead of the RPG program. Display the
two RB entries to see which program caused them.
40. Type the command WRKJOB and write down the fully qualified job name to be used later.
41. Type the command CALL ITMPCSC and enter the following transaction:

Quantity Item
14 AA
102 CC (The RPG DSPLY operation must occur to the external
message queue. Use the System Request key and select
option 1 on the system request menu to transfer to a
secondary job.)

42. Sign on to the second job and reestablish your environment.

43. Type the command ENDJOB and specify the fully qualified job name identified earlier and
OPTION(*IMMED). This simulates an abnormal job or system end.
44. Wait about 30 seconds, type the command CALL ITMPCSC and press F4.
Note the last committed transaction. It must be the AA item entered earlier.
45. Return to the basic prompt display and end the program by pressing F3.
46. Type the command DSPJRN CMTLIB/JRNTEST.
The same type of rollback handling has occurred, but this time the rollback was caused by the
system instead of one of the programs. The RB entry was written by the program QWTPITPP, which
is the work management abnormal end program.

76 iSeries: Commitment control

You have now used the basic functions of commitment control. You can proceed with commitment
control on your applications or try some of the other functions such as:
v Using a notify object
v Locking records that are only read with LCKLVL(*ALL)
v Locking multiple records in the same file with LCKLVL(*ALL)

Example: Use a transaction logging file to start an application

This example provides sample code and instructions of how to use a transaction logging file to start an
application after an abnormal end.

Note: Read the Code example disclaimer for important legal information.

A transaction logging file is used to start an application again after a system or job failure when a notify
object is not used. A transaction logging file is often used in interactive applications to summarize the
effects of a transaction.

For example, in an order entry application, a record is typically written to a transaction logging file for
each item ordered. The record contains the item ordered, the quantity, and the price. In an accounts
payable application, a record is written to a transaction logging file for each account number that is to
receive a charge. This record normally contains such information as the account number, the amount
charged, and the vendor.

In many of the applications where a transaction logging file already exists, a workstation user can request
information about the last transaction entered. By adding commitment control to the applications in
which a transaction logging file already exists, you can:
v Ensure that the database files are updated to a commitment boundary.
v Simplify the starting of the transaction again.

You must be able to uniquely identify the workstation user if you use a transaction logging file for
starting applications again under commit control. If unique user profile names are used on the system,
that profile name can be placed in a field in the transaction logging record. This field can be used as the
key to the file.

The following examples assume that an order inventory file is being used to perform transactions and
that a transaction logging file already exists. The program does the following:
1. Prompts the workstation user for a quantity and item number.
2. Updates the quantity in the production master file (PRDMSTP).
3. Writes a record to the transaction logging file (ISSLOGL).

If the inventory quantity on hand is insufficient, the program rejects the transaction. The workstation user
can ask the program where the data entry was interrupted, because the item number, description,
quantity, user name, and date are written to the transaction logging file.

DDS for physical file PRDMSTP

SEQNBR *... ... 1 ... ... 2 ... ... 3 ... ... 4 ... ... 5 ... ... 6 ... ... 7

1.00 A R PRDMSTR TEXT(’Master record’)

2.00 A PRODCT 3 COLHDG(’Product’ ’Number’)
3.00 A DESCRP 20 COLHDG(’Description’)
4.00 A ONHAND 5 0 COLHDG(’On Hand’ ’Amount’)
5.00 A EDTCDE(Z)
6.00 A K PRODCT

DDS for physical file ISSLOGP used by ISSLOGP

Commitment control 77
SEQNBR *... ... 1 ... ... 2 ... ... 3 ... ... 4 ... ... 5 ... ... 6 ... ... 7

1.00 A R ISSLOGR TEXT(’Product log record’)

2.00 A PRODCT 3 COLHDG(’Product’ ’Number’)
3.00 A DESCRP 20 COLHDG(’Description’)
4.00 A QTY 3 0 COLHDG(’Quantity’)
5.00 A EDTCDE(Z)
6.00 A USER 10 COLHDG(’User’ ’Name’)
7.00 A DATE 6 0 EDTCDE(Y)
8.00 A COLHDG(’Date’)

DDS for logical file ISSLOGL

SEQNBR *... ... 1 ... ... 2 ... ... 3 ... ... 4 ... ... 5 ... ... 6 ... ... 7

1.00 A LIFO
2.00 A R ISSLOGR PFILE(ISSLOGP)
3.00 A K USER

DDS for display file PRDISSD used in the program

SEQNBR *... ... 1 ... ... 2 ... ... 3 ... ... 4 ... ... 5 ... ... 6 ... ... 7 ..

1.00 A REF(ISSLOGP)
2.00 A R PROMPT
3.00 A CA03(98 ’End of program’)
4.00 A CA02(97 ’Where am I’)
5.00 A 1 20’ISSUES PROCESSING’
6.00 A 3 2’Quantity’
7.00 A QTY R I +1
8.00 A 62 ERRMSG(’Not enough +
9.00 A Qty’ 62)
10.00 A +6’Product’
11.00 A PRODCT R I +1
12.00 A 61 ERRMSG(’No Product +
13.00 A record found’ 62)
14.00 A 55 15 2’No Previous record exists’
15.00 A 24 2’CF2 Last transaction’
16.00 A R RESTART
17.00 A 1 20’LAST TRANSACTION +
18.00 A INFORMATION’
19.00 A 5 2’Product’
20.00 A PRODCT R +1
21.00 A 7 2’Description’
22.00 A DESCRP R +1
23.00 A 9 2’Qty’
24.00 A QTY R +1REFFLD(QTY)

This process is outlined the Program flow.

Program flow

78 iSeries: Commitment control

Link to desciption

The RPG COMMIT operation code is specified after the PRDMSTP file is updated and the record is
written to the transaction logging file. Since each prompt to the operator represents a boundary for a new
transaction, the transaction is considered a single Enter transaction.

The user name is passed to the program when it is called. The access path for the transaction logging file
is defined in last-in-first-out (LIFO) sequence so the program can easily access the last record entered.

The workstation user can start the program again after a system or job failure by using the same function
that identified where data entry was stopped. No additional code needs to be added to the program. If

Commitment control 79
you are currently using a transaction logging file but are not using it to find out where you are, add the
user name to the transaction logging file (assuming user names are unique) and use this approach in the
program.

The following shows the RPG program used. Statements required for commitment control are marked
with arrows (==>).

RPG Program
SEQNBR *... ... 1 ... ... 2 ... ... 3 ... ... 4 ... ... 5 ... ... 6 ... .. 7 ..
=>1.00 FPRDMSTP UP E K DISK KCOMIT
=>2.00 FISSLOGL IF E K DISK KCOMIT
3.00 PRDISSD CP E WORKSTN
4.00 *ENTRY PLIST
5.00 PARM USER 10
6.00 C*
7.00 C* Initialize fields used in Trans Log Rcd
8.00 C*
9.00 C MOVE UDATE DATE
10.00 C*
11.00 C* Basic processing loop
12.00 C*
13.00 C LOOP TAG
14.00 C EXFMTPROMPT
15.00 C 98 GOTO END End of pgm
16.00 C 97 DO Where am I
17.00 C EXSR WHERE
18.00 C GOTO LOOP
19.00 C END
20.00 C PRODCT CHAINPRDMSTR 61 Not found
21.00 C 61 GOTO LOOP
22.00 C ONHAND SUB QTY TEST 50 62 Less than
23.00 C 62 DO Not enough
24.00 C EXCPTRLSMST Release lock
25.00 C GOTO LOOP
26.00 C END
27.00 C*
28.00 C* Update master record and output the Transaction Log Record
29.00 C*
30.00 C Z-ADDTEST ONHAND
31.00 C UPDATPRDMSTR
32.00 C WRITEISSLOGR
=>33.00 C COMIT
34.00 C GOTO LOOP
35.00 C*
36.00 C* End of program processing
37.00 C*
38.00 C END TAG
39.00 C SETON LR
40.00 C*
41.00 C* WHERE subroutine for "Where am I" requests
42.00 C*
43.00 C WHERE BEGSR
44.00 C USER CHAINISSLOGL 55 Not found
45.00 C N55 EXFMTRESTART
46.00 C ENDSR
47.00 OPRDMSTR E RLSMST

CL program used to call RPG program PRDISS

SEQNBR *... ... 1 ... ... 2 ... ... 3 ... ... 4 ... ... 5 ... ... 6 ... ... 7 ..

1.00 PGM
2.00 DCL &USER *CHAR LEN(10)
3.00 STRCMTCTL LCKLVL(*CHG)
4.00 RTVJOBA USER(&USER)

80 iSeries: Commitment control

 5.00 CALL PRDISS PARM(&USER)
6.00 MONMSG MSGID(RPG900l) EXEC(ROLLBACK)
7.00 ENDCMTCTL
8.00 ENDPGM

To use commitment control in this program, a lock level of *CHG would normally be specified. The
record is locked by the change until a commit operation is run. Note that if there is an insufficient
quantity of inventory, the record is explicitly released. (If the record were not explicitly released in the
program, it would be released when the next record is read for update from the file.)

In this example, there is no additional advantage to using the lock level *ALL. If *ALL were used, a
rollback or commit operation must be used to release the record when an insufficient quantity existed.

The previous code is a CL program that calls the RPG program PRDISS. Note the use of
STRCMTCTL/ENDCMTCTL commands. The unique user name is retrieved (RTVJOBA command) and
passed to the program. The use of the MONMSG command to cause a rollback is described in “Example:
Use a standard processing program to start an application” on page 87.

Example: Use a notify object to start an application

When a program is started after an abnormal end, it can look for an entry in the notify object. If one
exists, the program can start a transaction again. After the transaction has been started again, the notify
object is cleared by the program to prevent it from starting the same transaction yet another time.

Following are ways you can use a notify object:

v If the commit identification is placed in a database file, query this file to determine where to start each
application or workstation job again.
v If the commit identification is placed in a message queue for a particular workstation, a message can
be sent to the work station users when they sign on to inform them of the last transaction committed.
v If the commit identification is placed in a database file that has a key or user name, the program can
read this file when it is started. If a record exists in the file, start the program again. The program can
send a message to the workstation user identifying the last transaction committed. Any recovery is
performed by the program. If a record existed in the database file, the program deletes that record at
the end of the program.
v For a batch application, the commit identification can be placed in a data area that contains totals,
switch settings, and other status information necessary to start the application again. When the
application is started, it accesses the data area and verifies the values stored there. If the application
ends normally, the data area is set up for the next run.
v For a batch application, the commit identification can be sent to a message queue. A program that is
run when the application is started can retrieve the messages from the queue and start the programs
again.

There are several techniques for starting your applications again depending on your application needs. In
choosing the technique, consider the following:
v When there are multiple users of a program at the same time, a single data area cannot be used as the
notify object because after an abnormal system end, the commit identification for each user would
overlay each other in the data area.
v Your design for deleting information in the notify object must handle the situation when a failure
occurs immediately following use of the information:
– If information is deleted immediately, it would not exist if another failure occurs before processing
the interrupted transaction.
– The information in the notify object must not be deleted until the successful processing of the
interrupted transaction. In this case, more than one entry will exist in the notify object if it is a
database file or message queue.
– The program must access the last record if there is more than one entry.
Commitment control 81
v A notify object cannot be used to provide the work station user with the last transaction committed
because the notify object is updated only if a system or job failure occurs or if uncommitted changes
exist at the normal end of a job.
v If information is displayed to the workstation user, it must be meaningful. To accomplish this may
require that the program translate codes kept in the notify object into information that will help the
user start again.
v Information for starting again must be displayed if the work station user needs it. Additional logic in
the program is required to prevent information from being displayed again when it is no longer
meaningful.
v A single notify object and a standard processing program can provide a starting again function if the
notify object is a database file. This standard processing program is called by the programs that require
the ability to start again to minimize the changes to each individual program.

For example code for using a notify object, see the following:

Note: Read the Code example disclaimer for important legal information.
v “Example: Unique notify object for each program”
v “Example: Single notify object for all programs” on page 87

Example: Unique notify object for each program

This topic provides sample code and instructions to use a unique notify object to restart each program.

Using a single, unique notify object for each job allows use of an externally described commit
identification even though there may be multiple users of the same program. In the following examples, a
database file is used as a notify object and it is used only by this program.

The program has two database files (PRDMSTP and PRDLOCP) that must be updated for receipts to
inventory. The display file used by the program is named PRDRCTD. A database file, PRDRCTP, is used
as the notify object. This notify object is defined to the program as a file and is also used as the definition
of a data structure for the notify function.

See DDS for physical file PRDMSTP (page 77) for to see the DDS for physical file PRDMSTP.

Note: Read the Code example disclaimer for important legal information.

DDS for physical file PRDLOCP

SEQNBR *... ... 1 ... ... 2 ... ... 3 ... ... 4 ... ... 5 ... ... 6 ... ... 7

1.00 A R PRDLOCR TEXT(’Location record’)

2.00 A PRODCT 3 COLHDG(’Product’ ’Number’)
3.00 A LOCATN 6 COLHDG(’Location’)
4.00 A LOCAMT 5 0 COLHDG(’Location’ ’Amount’)
5.00 A EDTCDE(Z)
6.00 A K PRODCT
7.00 A K LOCATN

DDS for display file PRDRCTD

SEQNBR *... ... 1 ... ... 2 ... ... 3 ... ... 4 ... ... 5 ... ... 6 ... ... 7 ..

1.00 A REF(PRDMSTP)
2.00 A R PROMPT
3.00 A CA03(98 ’End of program’)
4.00 A SETOFF(71 ’RESTART’)
5.00 A 1 20’PRODUCT RECEIPTS’
6.00 A 3 2’Quantity’
7.00 A QTY 3 OI +1
8.00 A +6’Product’

82 iSeries: Commitment control

 9.00 A PRODCT R I +1
10.00 A 61 ERRMSG(’No record +
11.00 A found in the +
12.00 A master file’ 62)
13.00 A +6’Location’
14.00 A LOCATN R I +1REFFLD(LOCATN PRDLOCP)
15.00 A 62 ERRMSG(’No record +
16.00 A found in the +
17.00 A location file’ 62)
18.00 A 9 2’Last Transaction’
19.00 A 71 +6’This is restart +
20.00 A information’
21.00 A DSPATR(HI BL)
22.00 A 12 2’Quantity’
23.00 A 12 12’Product’
24.00 A 12 23’Location’
25.00 A 12 35’Description’
26.00 A LSTPRD R 14 15REFFLD(PRODCT)
27.00 A LSTLOC R 14 26REFFLD(LOCATN *SRC)
28.00 A LSTQTY R 14 5REFFLD(QTY *SRC)
29.00 A EDTCDE(Z)
30.00 A LSTDSC R 14 35REFFLD(DESCRP)

DDS for notify object and externally described data structure (PRDRCTP)
SEQNBR *... ... 1 ... ... 2 ... ... 3 ... ... 4 ... ... 5 ... ... 6 ... ... 7 ..

1.00 A LIFO
2.00 A REF(PRDMSTP)
3.00 A R PRDRCTR
4.00 A USER 10
5.00 A PRODCT R
6.00 A DESCRP R
7.00 A QTY 3 0
8.00 A LOCATN R REFFLD(LOCATN PRDLOCP)
9.00 A K USER

The program processes the notify object as follows:

v At the beginning, the program randomly processes the notify object and displays a record if it exists
for the specific key:
– If multiple records exist, the last record for this key is used because the PRDRCTP file is in LIFO
sequence.
– If no record exists, a transaction was not interrupted so it is not necessary to start again.
– If the program fails before the first successful commit operation, it does not consider that starting
again is required.
v The routine to clear the notify object occurs at the end of the program:
– If there were multiple failures, the routine can handle deletion of multiple records in the notify
object.
– Although the system places the commit identification in a database file, the commit identification
must be specified as a variable in the RPG program.
– Because RPG allows a data structure to be externally described, a data structure is a convenient way
of specifying the commit identification. In this example, the data structure uses the same external
description that the database file used as the notify object.

The processing for this program prompts the user for a product number, a location, and a quantity:
v Two files must be updated:
– Product master file (PRDMSTP)
– Product location file (PRDLOCP)
v A record in each file must exist before either is updated.

Commitment control 83
v The program moves the input fields to corresponding last fields after each transaction is successfully
entered. These last fields are displayed to the operator on each prompt as feedback for what was last
entered.
v If information for starting again exists, it is moved to these last fields and a special message appears on
the display.

This process is outlined in the following figure. The user name is passed to the program to provide a
unique record in the notify object.

Program flow

84 iSeries: Commitment control

Link to description

The following is the RPG source code for this example. The notify object (file PRDRCTP) is used as a
normal file at the beginning and end of the program, and is also specified as the notify object in the CL
(STRCMTCTL command) before calling the program.

RPG source

Commitment control 85
SEQNBR *... ... 1 ... ... 2 ... ... 3 ... ... 4 ... ... 5 ... ... 6 ... ... 7 ..

1.00 FPRDMSTP UF E K DISK KCOMIT

2.00 FPRDLOCP UF E K DISK KCOMIT
3.00 FPRDRCTD CF E WORKSTN
4.00 F*
5.00 F* The following file is the specific notify object for this pgm.
6.00 F* It is accessed only in a restart situation and at the
7.00 F* end of the program to delete any records. The records
8.00 F* are written to the notify object by Commitment Control.
9.00 F*
10.00 FPRDRCTP UF E K DISK
11.00 ICMTID E DSPRDRCTP

12.00 C *ENTRY PLIST

13.00 C PARM USER10 10
14.00 C MOVE USER10 USER
15.00 C*
16.00 C* Check for restart information - get last rcd per user
17.00 C* PRDRCTP file access path is in LIFO sequence
18.00 C*
19.00 C USER CHAINPRDRCTR 20 Not found
20.00 C N20 DO Restart
21.00 C EXSR MOVLST Move to last
22.00 C SETON 71 Restart
23.00 C END
24.00 C*
25.00 C* Basic processing loop
26.00 C*
27.00 C L00P TAG
28.00 C EXFMTPROMPT
29.00 C 98 GOTO END End of pgm
30.00 C PRODCT CHAINPRDMSTR 61 Not found
31.00 C 61 GOTO L00P
32.00 C KEY KLIST
33.00 C KFLD PRODCT
34.00 C KFLD LOCATN
35.00 C KEY CHAINPRDLOCR 62 Not found
36.00 C 62 DO
37.00 C EXCPTRLSMST Release lck
38.00 C GOTO L00P
39.00 C END
40.00 C ADD QTY ONHAND Add
41.00 C ADD QTY LOCAMT
42.00 C UPDATPRDMSTR Update
43.00 C UPDATPRDLOCR Update
44.00 C*

45.00 C* Commit and move to previous fields

46.00 C*
47.00 C CMTID COMIT
48.00 C EXSR MOVLST Move to last
49.00 C GOTO L00P
50.00 C*
51.00 C* End of program processing
52.00 C*
53.00 C END TAG
54.00 C SETON LR
55.00 C*56.00 C* Delete any records in the notify object
57.00 C*
58.00 C DLTLP TAG
59.00 C USER CHAINPRDRCTR 20 Not found
60.00 C N20 DO
61.00 C DELETPRDRCTR Delete
62.00 C GOTO DLTLP
63.00 C END
64.00 C*

86 iSeries: Commitment control

 65.00 C* Move to -Last Used- fields for operator feedback
66.00 C*
67.00 C MOVLST BEGSR
68.00 C MOVE PRODCT LSTPRD
69.00 C MOVE LOCATN LSTLOC
70.00 C MOVE QTY LSTQTY
71.00 C MOVE DESCRP LSTDSC
72.00 C ENDSR
73.00 OPRDMSTR E RLSMST

Example: Single notify object for all programs

Using a single notify object for all programs is advantageous because all information required to start
again is in the same object and a standard approach to the notify object can be used in all programs. In
this situation, use a unique combination of user and program identifications to make sure that the
program accesses the correct information when it starts again.

Because the information required to start again may vary from program to program, do not be use an
externally described data structure for the commit identification. If a single notify object is used, the
preceding program can describe the data structure within the program rather than externally. For
example:
1 10 USER
11 20 PGMNAM
21 23 PRODCT
24 29 LOCATN
30 49 DESC
50 51 0 QTY
52 220 DUMMY

In each program that uses this notify object, the information specified for the commit identification would
be unique to the program (the user and program names are not unique). The notify object must be large
enough to contain the maximum information that any program would place in the commit identification.

“Example: Unique notify object for each program” on page 82 provides more examples about using notify
objects.

Example: Use a standard processing program to start an application

A standard processing program is one way to start your application again using one database file as the
notify object for all applications. This approach assumes that user profile names are unique by user for all
applications using the standard program.

For this approach, the physical file NFYOBJP is used as the notify object and defined as:
Unique user profile name 10 characters
Program identification 10 characters
Information for
starting again Character field
(This must be large
enough to contain the maximum
amount of information for starting
programs again that require
information for starting again.
This field is required by
the application programs.
In the example, it is
assumed to be a length of 200.)

The file is created with SHARE(*YES). The first two fields in the file are the key to the file. (This file can
also be defined as a data structure in RPG programs.)

Note: Read the Code example disclaimer for important legal information.

Commitment control 87
This the following provide example code for a standard processing program:
v “Example: Code for a standard processing program”
v “Example: Code for a standard commit processing program” on page 89
v “Example: Use a standard processing program to decide whether to restart the application” on page 92

Example: Code for a standard processing program

Following is an example of using a standard processing program. The application shown in the following
code example performs as follows:
1. The application program receives the user name in a parameter and uses it with the program name as
a unique identifier in the notify object.
2. The application program passes a request code of R to the standard commit processing program,
which determines if a record exists in the notify object.
3. If the standard commit processing program returns a code of 1, a record was found and the
application program presents the information needed to start again to the user.
4. The application program proceeds with normal processing.
5. When a transaction is completed, values are saved for reference so the workstation user can see what
was done for the previous transaction.
The information saved is not provided by the notify object because the notify object is updated only if
a job or system failure occurs.

See Processing flow for the flow of this program.

Note: Read the Code example disclaimer for important legal information.

Application program example

SEQNBR *... ... 1 ... ... 2 ... ... 3 ... ... 4 ... ... 5 ... ... 6 ... ... 7 ..

1.00 FPRDMSTP UF E K DISK KCOMIT

2.00 FPRDLOCP UF E K DISK KCOMIT
3.00 FPRDRCTD CF E WORKSTN
4.00 F*
5.00 F* The following is a compile time array which contains the
6.00 F* restart information used in the next example
7.00 F*
8.00 E RTXT 50 50 1 Restart text
9.00 I*
10.00 I* Data structure used for info passed to notify object
11.00 I*
12.00 ICMTID DS
13.00 I 1 10 USER
14.00 I 11 20 PGMNAM
15.00 I 21 23 PRODCT
16.00 I 24 29 LOCATN
17.00 I 30 49 DESCRP
18.00 I P 50 510QTY
19.00 I 52 170 DUMMY
20.00 I 171 220 RSTART
21.00 C *ENTRY PLIST
22.00 C PARM USER10 10
23.00 C*
24.00 C* Initialize fields used to communicate with std program
25.00 C*
26.00 C MOVE USER10 USER
27.00 C MOVEL’PRDRC2’ PGMNAM
28.00 C MOVE ’R’ RQSCOD Read Rqs
29.00 C CALL ’STDCMT’
30.00 C PARM RQSCOD 1
31.00 C PARM RTNCOD 1
32.00 C PARM CMTID 220 Data struct

88 iSeries: Commitment control

33.00 C RTNCOD IFEQ ’1’ Restart
34.00 C EXSR MOVLST Move to last
35.00 C SETON 71 Restart
36.00 C END
37.00 C*
38.00 C* Initialize fields used in notify object
39.00 C*
40.00 C MOVEARTXT,1 RSTART Move text
41.00 C*
42.00 C* Basic processing loop
43.00 C*
44.00 C LOOP TAG
45.00 C EXFMTPROMPT
46.00 C 98 GOTO END
47.00 C PRODCT CHAINPRDMSTR 61 Not found
48.00 C 61 GOTO LOOP
49.00 C KEY KLIST
50.00 C KFLD PRODCT
51.00 C KFLD LOCATN
SEQNBR *... ... 1 ... ... 2 ... ... 3 ... ... 4 ... ... 5 ... ... 6 ... ... 7 ..

52.00 C KEY CHAINPRDLOCR 62 Not found

53.00 C 62 DO
54.00 C EXCPTRLSMST Release lck
55.00 C GOTO LOOP
56.00 C END
57.00 C ADD QTY ONHAND Add
58.00 C ADD QTY LOCAMT
59.00 C UPDATPRDMSTR Update
60.00 C UPDATPRDLOCR Update
61.00 C*
62.00 C* Commit and move to previous fields
63.00 C*
64.00 C CMTID COMIT
65.00 C EXSR MOVLST Move to last
66.00 C GOTO LOOP
67.00 C* End of program processing
68.00 C*
69.00 C END TAG
70.00 C MOVE ’D’ RQSCOD Dlt Rqs
71.00 C CALL ’STDCMT’
72.00 C PARM RQSCOD
73.00 C PARM RTNCOD
74.00 C PARM CMTID
75.00 C SETON LR
76.00 C*
77.00 C* Move to -Last Used- fields for operator feedback
78.00 C*
79.00 C MOVLST BEGSR
80.00 C MOVE PRODCT LSTPRD
81.00 C MOVE LOCATN LSTLOC
82.00 C MOVE DESCRP LSTDSC
83.00 C MOVE QTY LSTQTY
84.00 C ENDSR
85.00 OPRDMSTR E RLSMST
86.00 ** RTXT Restart Text
87.00 Inventory Menu - Receipts Option

Example: Code for a standard commit processing program

The standard commit (STDCMT) processing program performs the functions required to communicate
with a single notify object used by all applications. While the commitment control function automatically
writes an entry to the notify object, a user-written standard program must process the notify object. The
standard program simplifies and standardizes the approach.

Commitment control 89
The program is written to verify the parameters that were passed and perform the appropriate action as
follows:

O=Open
The calling program requests the notify object file be kept open on return. Because the notify object
is opened implicitly by the RPG program, the program must not close it. Indicator 98 is set so the
program returns with LR off to keep the program’s work areas and leaves the notify object open so
it can be called again without excess overhead.

C=Close
The calling program has determined it no longer needs the notify object and requests a close.
Indicator 98 is set off to allow a full close of the notify object.

R=Read
The calling program requests that a record with matching key fields be read and passed back. The
program uses the passed key fields to attempt to retrieve a record from NFYOBJP. If duplicate
records exist for the same key, the last record is returned. The return code is set accordingly and, if
the record existed, it is passed back in the data structure CMTID.

W=Write
The calling program requests a record to be written to the notify object to allow the calling program
to start again the next time it is called. The program writes the contents of the passed data as a
record in NFYOBJP.

D=Delete
The calling program requests that records for this matching key be deleted. This function is typically
performed at the successful completion of the calling program to remove any information about
starting again. The program attempts to delete any records for passed key fields. If no records exist,
a different return code is passed back.

S=Search
The calling program requests a search for a record for a particular user regardless of which program
wrote it. This function is used in the program for sign-on to indicate that starting again is required.
The program uses only the user name as the key to see if records exist. The return code is set
appropriately, and the contents of the last record for this key (if it exists) are read and passed back.

Note: Read the Code example disclaimer for important legal information.

The following shows the standard commit processing program, STDCMT.

Standard commit processing program

SEQNBR *... ... 1 ... ... 2 ... ... 3 ... ... 4 ... ... 5 ... ... 6 ... ... 7 ..

1.00 FNFYOBJP UF E K DISK A

2.00 ICMTID DS
3.00 I 1 10 UNQUSR
4.00 I 11 20 UNQPGM
5.00 I 21 220 BIGFLD
6.00 C *ENTRY PLIST
7.00 C PARM RQSCOD 1
8.00 C PARM RTNCOD 1
9.00 C PARM CMTID 220
10.00 C UNQUSR CABEQ*BLANKS BADEND H1 Invalid
11.00 C UNQPGM CABEQ*BLANKS BADEND H2 Invalid
12.00 C*
13.00 C* ’O’ for Open
14.00 C*
15.00 C RQSCOD IFEQ ’O’ Open
16.00 C SETON 98 End LR

90 iSeries: Commitment control

17.00 C GOTO END
18.00 C END
19.00 C*
20.00 C* ’C’ for Close
21.00 C*
22.00 C RQSCOD IFEQ ’C’ Close
23.00 C SETOF 98
24.00 C GOTO END
25.00 C END
26.00 C*
27.00 C* ’R’ for Read - Get last record for the key
28.00 C*
29.00 C RQSCOD IFEQ ’R’ Read
30.00 C KEY KLIST
31.00 C KFLD UNQUSR
32.00 C KFLD UNQPGM
33.00 C KEY CHAINNFYOBJR 51 Not found
34.00 C 51 MOVE ’0’ RTNCOD
35.00 C 51 GOTO END
36.00 C MOVE ’1’ RTNCOD Found
37.00 C LOOPl TAG
38.00 C KEY READENFYOBJR 20 EOF
39.00 C 20 GOTO END
40.00 C GOTO LOOP1
41.00 C END
42.00 C*
43.00 C* ’W’ FOR Write
44.00 C*
45.00 C RQSCOD IFEQ ’W’ Write
46.00 C WRITENFYOBJR
47.00 C GOTO END
48.00 C END
49.00 C*
50.00 C* ’D’ for Delete - Delete all records for the key
51.00 C*
52.00 C RQSCOD IFEQ ’D’ Delete
53.00 C KEY CHAINNFYOBJR 51 Not found
54.00 C 51 MOVE ’0’ RTNCOD
55.00 C 51 GOTO END
56.00 C MOVE ’1’ RTNCOD Found
57.00 C LOOP2 TAG
58.00 C DELETNFYOBJR
59.00 C KEY READENFYOBJR 20 EOF
60.00 C N20 GOTO LOOP2
61.00 C GOTO END
62.00 C END
63.00 C*
64.00 C* ’S’ for Search for the last record for this user
65.00 C* (Ignore the -Program- portion of the key)
66.00 C*
67.00 C RQSCOD IFEQ ’S’ Search
68.00 C UNQUSR SETLLNFYOBJR 20 If equal
69.00 C N20 MOVE ’0’ RTNCOD
70.00 C N20 GOTO END
71.00 C MOVE ’1’ RTNCOD Found
72.00 C LOOP3 TAG
73.00 C UNQUSR READENFYOBJR 20 EOF
74.00 C N20 GOTO LOOP3
75.00 C GOTO END
76.00 C END
77.00 C*
78.00 C* Invalid request code processing
79.00 C*
80.00 C SETON H2 Bad RQS code
81.00 C GOTO BADEND
82.00 C*
83.00 C* End of program processing

Commitment control 91
 84.00 C*
85.00 C END TAG
86.00 C N98 SETON LR
87.00 C RETRN
88.00 C* BADEND tag is used then fall thru to RPG cycle error return
89.00 C BADEND TAG

Example: Use a standard processing program to decide whether to restart the

application
This topic provides example CL code for using a standard processing program to decide whether to
restart an application after an abnormal IPL.

The initial program can call the standard commit processing program to determine if it is necessary to
start again. The workstation user can than decide whether to start again.

The initial program passes a request code of S (search) to the standard program, which searches for any
record for the user. If a record exists, the information for starting again is passed to the initial program
and the information is displayed to the workstation user.

The commit identification in the notify object contains information that the initial program can display
identifying what program needs to be started again. For example, the last 50 characters of the commit
identification can be reserved to contain this information. In the application program, this information
can be in a compile time array and moved to the data structure in an initialization step. “Example: Code
for a standard commit processing program” on page 89 shows how to include this in the application
program.

The following is an example of an initial program that determines if a record exists in the notify object.

Note: Read the Code example disclaimer for important legal information.

Initial program example

SEQNBR *... ... 1 ... ... 2 ... ... 3 ... ... 4 ... ... 5 ... ... 6 ... ... 7

1.00 PGM
2.00 DCLF CMTINLD
3.00 DCL &RQSCOD *CHAR LEN(1) VALUE(S) /* Search */
4.00 DCL &RTNCOD *CHAR LEN(1)
5.00 DCL &CMTID *CHAR LEN(220)
6.00 DCL &USER *CHAR LEN(10)
7.00 DCL &INFO *CHAR LEN(50)
8.00 RTVJOBA USER(&USER)
9.00 CHGVAR &CMTID (&USER *CAT XX)
10.00 /* The XX is required to prevent a blank Pgm nam */
11.00 CALL STDCMT PARM(&RQSCOD &RTNCOD &CMTID)
12.00 IF (&RTNCOD *EQ ’1’) DO /* RESTART REQD */
13.00 CHGVAR &INFO %SST(&CMTID 171 50)
14.00 SNDRCVF RCDFMT(RESTART)
15.00 ENDDO
16.00 /* */
17.00 /* Enter normal initial program statements */
18.00 /* or -TFRCTL- to first menu program */
19.00 /* */
20.00 ENDPGM

Troubleshoot transactions and commitment control

The following pages provide tasks that you need to do to troubleshoot commitment control:

92 iSeries: Commitment control

 “Commitment control errors”
This information describes conditions that create errors, lists commitment control errors, and
provides methods to deal with errors.

“Detect deadlocks” on page 100

The goal of this task is to find deadlock conditions.

“Recover transactions after communications failure” on page 101

The goal of this task is to handle transactions performing work on a remote system after the
communication with that system fails.

“When to force commits and rollbacks and when to cancel resynchronization” on page 101
This information describes when and how to force a rollback or commit, and when to cancel
resynchronization

“End a long-running rollback” on page 102

This information describes how to end long-running rollbacks that consume critical processor time,
lock resources, or take up storage space.

Commitment control errors

When you use commitment control, it is important to understand which conditions cause errors and
which do not. In general, errors occur when commitment control functions are used inconsistently, such
as running an End Commitment Control (ENDCMTCTL) command when files that use the commitment
definition are still open.

Errors During Commit Processing

If a communications or system failure occurs during a commit operation, resynchronization may need to
be performed to ensure that the transaction managers keep the data consistent on all the systems
involved in the transaction. The behavior of the resynchronization and how it affects the commit
operation depends on these factors :
v The Wait for outcome commitment option. Refer to “Commitment definition for two-phase commit:
Not wait for outcome” on page 32 for details.
v The state of the transaction. Refer to “States of the transaction for two-phase commitment control” on
page 27 for details.

If the failure is catastrophic such that it can never be repaired, or it cannot be repaired in a timely
manner, the system operators for other systems involved in the transaction must make a heuristic
decision. The heuristic decision commits or rolls back the changes made on that system during the
transaction. If the failure is repaired after such a decision, and the resynchronization detects that the
decision caused data integrity problems, message CPD83D9 or CPD83E9 is sent to the QSYSOPR message
queue.

The following information provides more details about working with commitment control errors:
v “Error conditions”
v “Non error conditions” on page 95
v “Error messages to monitor for during commitment control” on page 95
v “Monitor for errors after a CALL command” on page 98
v “Failure of normal commit or rollback processing” on page 99

Error conditions
If an error occurs, an escape message is sent that you can monitor for in a program. The following are
some typical errors related to commitment control:
v Consecutive STRCMTCTL commands are run without an intervening ENDCMTCTL command.
Commitment control 93
v Files are opened under commitment control, but no STRCMTCTL command was run.
This is not an error condition for programs that run within an activation group that are to use the
job-level commitment definition. The job-level commitment definition can be started only by a single
program, but when started by a program, the job-level commitment definition is used by any program
running in any activation group that is not using an activation-group-level commitment definition.
Programs that run within an activation group that are to use the activation-group-level commitment
definition must first start the activation-group-level commitment definition with the STRCMTCTL
command.
v Files that are opened for output under commitment control are not journaled.
v The first open operation of a shared file places the file under commitment control, but subsequent open
operations of the same shared file do not.
v The first open operation of a shared file does not place the file under commitment control, but
subsequent open operations of the same shared file do.
v The record lock limit for the job is reached in a single transaction.
v The program issues a read operation, a commit operation, and a change to the same record. The read
operation must be issued again after the commit operation because the commit operation has freed the
lock on the record.
v For a one-phase location, resources placed under commitment control do not reside at the same
location as resources already under commitment control for the commitment definition.
v Uncommitted changes exist when an ENDCMTCTL command is issued.
This is not an error condition for the ENDCMTCTL command if all files are closed, any remote
database is disconnected, and no API commitment resource is still associated with the commitment
definition to be ended.
v A commit, rollback, or ENDCMTCTL command is run, and a STRCMTCTL command was not run.
This is not an error condition for programs that run within an activation group and the job-level
commitment definition is active. The job-level commitment definition can be started only by a single
program, but when started by a program, the job-level commitment definition is used by any program
running in any activation group that is not using an activation-group-level commitment definition.
Programs that run within an activation group and are to use the activation-group-level commitment
definition must first start the activation-group-level commitment definition with the STRCMTCTL
command.
v An ENDCMTCTL command is run with files still open under commitment control for the commitment
definition.
v A job performing a save operation has one or more commitment definitions that are not at a
commitment boundary.
v A save-while-active operation ended because other jobs with committable resources did not reach a
commitment boundary in the time specified for the SAVACTWAIT parameter.
v A save-while-active process was not able to continue because of API committable resources being
added to more than one commitment definition for a single job.
v More than 1023 commitment definitions exist for a single job.
v The conversation to a remote location is lost due to a resource failure. This may cause the transaction
to be rolled back.
v A one-phase resource that is opened for update is present at a node that did not initiate the commit
operation. You must remove either the resource or the node that initiated the commit request.
v A commit operation is requested while the transaction is in rollback required (RBR) state. A rollback
operation must be done.
v An API exit program issues a commit request or a rollback request.
v A trigger program issues a commit request or a rollback request for the commitment definition under
which the trigger program was called.

94 iSeries: Commitment control

The trigger program can start a separate commitment definition and issue a commit or rollback request
for that definition.

Non error conditions

You might expect these situations to cause an error message. However, commitment control allows these
situations. The following are some situations for commitment control in which no errors occur:
v A commit or rollback operation is run and no resources are under commitment control. This allows
you to include commit or rollback operations in your program without considering whether there are
resources under commitment control. It also allows you to specify a commit identification before
making any committable changes.
v A commit or rollback operation is run and there are no uncommitted resource changes. This allows you
to include commit and rollback operations within your program without considering whether there are
uncommitted resource changes.
v A file under commitment control is closed and uncommitted records exist. This situation allows
another program to be called to perform the commit or rollback operation. This occurs regardless of
whether the file is shared. This function allows a subprogram to make database changes that are part
of a transaction involving multiple programs.
v A job ends, either normally or abnormally, with uncommitted changes for one or more commitment
definitions. The changes for all commitment definitions are rolled back.
v An activation group ends with pending changes for the activation-group-level commitment definition.
If the activation group is ending normally and there are no errors encountered when closing any files
opened under commitment control scoped to the same activation group that is ending, an implicit
commit is performed by the system. Otherwise, an implicit rollback is performed.
v A program accesses a changed record again that has not been committed. This allows a program to:
– Add a record and update it before specifying the commit operation.
– Update the same record twice before specifying the commit operation.
– Add a record and delete it before specifying the commit operation.
– Access an uncommitted record again by a different logical file (under commitment control).
v You specify LCKLVL(*CHG or *CS) on the STRCMTCTL command and open a file with a commit
operation for read-only. In this case, no locks occur on the request. It is treated as if commitment
control is not in effect, but the file does appear on the WRKJOB menu option of files under
commitment control.
v You issue the STRCMTCTL command and do not open any files under commitment control. In this
situation, any record-level changes made to the files are not made under commitment control.

Error messages to monitor for during commitment control

Several different error messages can be returned by the commit or rollback operations or sent to the job
log, depending on the type of message and when the error occurred.

These messages can occur during the following:

v Normal commit or rollback processing
v Commit or rollback processing during job process end
v Commit or rollback processing during activation group end

You cannot monitor for any of the following messages during activation group end or job process end.
Also, you can only monitor for CPFxxxx messages. CPDxxxx messages are always sent as diagnostic
messages, which cannot be monitored. Any errors encountered when ending an activation-group-level
commitment definition during activation group end or any commitment definition during job end are left
in the job log as diagnostic messages.

Error messages related to commitment control to look for are as follows:

Commitment control 95
 CPD8351
Changes may not have been committed.

CPD8352
Changes not committed at remote location &3.

CPD8353
Changes to relational database &1 may not have been committed.

CPD8354
Changes to DDM file &1 may not have been committed.

CPD8355
Changes to DDL object &1 may not have been committed.

CPD8356
Rolled back changes may have been committed.

CPD8358
Changes to relational database &1 may not have been rolled back.

CPD8359
Changes to DDM file &1 may not have been rolled back.

CPD835A
Changes to DDL object &3 may not have been rolled back.

CPD835C
Notify object &1 in &2 not updated.

CPD835D
DRDA(R) resource does not allow SQL cursor hold.

CPF835F
Commit or rollback operation failed.

CPD8360
Members or files or both were already deallocated.

CPD8361
API exit program &1 failed during commit.

CPD8362
API exit program &1 failed during roll back.

CPD8363
API exit program &1 ended after &4 minutes during commit.

CPD8364
API exit program &1 ended after &4 minutes during rollback.

CPD836F
Protocol error occurred during commitment control operation.

CPD83D1
API resource &4 cannot be last agent.

96 iSeries: Commitment control

CPD83D2
Resource not compatible with commitment control.

CPD83D7
Commit operation changed to rollback.

CPD83D9
A heuristic mixed condition occurred.

CPF83DB
Commit operation resulted in rollback.

CPD83DC
Action If Problems Used to determine commit or rollback operation; reason &2.

CPD83DD
Conversation ended; reason &1.

CPD83DE
Return information not valid.

CPD83EC
API exit program &1 voted rollback.

CPD83EF
Rollback operation started for next logical unit of work.

CPF8350
Commitment definition not found.

CPF8355
ENDCMTCTL not allowed. Pending changes active.

CPF8356
Commitment control ended with &1 local changes not committed.

CPF8358
Notify object &1 in &2 not updated.

CPF8359
Rollback operation failed.

CPF835A
End of commitment definition &1 canceled.

CPF835B
Errors occurred while ending commitment control.

CPF835C
Commitment control ended with remote changes not committed.

CPF8363
Commit operation failed.

CPF8364
Commitment control parameter value is not valid. Reason code &3.

Commitment control 97
 CPF8367
Cannot perform commitment control operation.

CPF8369
Cannot place API commitment resource under commitment control; reason code &1.

CPF83D0
Commitment operation not allowed.

CPF83D2
Commit complete == Resynchronization in progress has been returned.

CPF83D3
Commit complete == Heuristic Mixed has been returned.

CPF83D4
Logical unit of work journal entry not sent.

CPF83E1
Commit operation failed due to constraint violation.

CPF83E2
Rollback operation required.

CPF83E3
Requested nesting level is not active.

CPF83E4
Commitment control ended with resources not committed.

CPF83E6
Commitment control operation completed with resynchronization in progress.

CPF83E7
Commit or rollback of X/Open global transaction not allowed.

Monitor for errors after a CALL command

When a program that uses commitment control is called, monitor for unexpected errors and perform a
rollback operation if an error occurs. For example, uncommitted records can exist when a program
encounters an unexpected error such as an RPG divide by zero error. Depending on the status of the
inquiry message reply (INQMSGRPY) parameter for a job, the program sends an inquiry message or
performs a default action. If the operator response or the default action ends the program, uncommitted
records still exist waiting for a commit or rollback operation.

If another program is called and causes a commit operation, the partially completed transaction from the
previous program is committed.

To prevent partially completed transactions from being committed, monitor for escape messages after the
CALL command. For example, if it is an RPG program, use the following coding:
CALL RPGA
MONMSG MSGID(RPG9001)
EXEC(ROLLBACK) /*Rollback if pgm is canceled*/

If it is a COBOL program:

98 iSeries: Commitment control

CALL COBOLA
MONMSG MSGID(CBE9001)
EXEC(ROLLBACK) /*Rollback if pgm is canceled*/

Failure of normal commit or rollback processing

Errors may occur at any time during commit or rollback processing. The following table divides this
processing into four situations. The middle column describes the actions taken by the system when it
encounters errors during each situation. The third column suggests what you or your application must do
in response to the messages. These suggestions are consistent with the way commitment control
processing is handled by the system.

Situation Commit or Rollback Processing Suggested Action

Record-level I/O commit v If the error occurs during the prepare wave, the Monitor for messages;
fails transaction is rolled back and message CPF83DB is handle as desired
sent.
v If the error occurs during the committed wave,
commit processing continues to commit as many
remaining resources as possible. Message CPF8363 is
sent at the end of commit processing.
Object-level or commit and v If the error occurs during the prepare wave, the Monitor for messages;
rollback exit program for transaction is rolled back and message CPF83DB is handle as desired
API commitment resource sent.
fails during commit
v If the error occurs during the committed wave,
processing continues to commit or roll back as many
remaining resources as possible. One of the following
messages is returned, depending on the commitment
resource type:
– CPD8353
– CPD8354
– CPD8355
– CPD8361
Message CPF8363 is sent at the end of commit
processing.
Record-level I/O rollback 1. Returns CPD8356 Monitor for messages;
fails handle as desired
2. Attempt to continue processing to rollback
object-level or API commitment resources
3. Returns CPF8359 at end of processing
Object-level or commit and 1. Returns one of the following messages depending on Monitor for messages;
rollback exit program for the commitment resource type: handle as desired
API commitment resources
v CPD8358
fails during rollback
v CPD8359
v CPD835A
v CPD8362
2. Continues processing
3. Returns CPF8359 at end of processing

Commit or Rollback Processing During Job End

All of the situations described in the previous table also apply when a job is ending except that one of
the following messages is sent:
v CPF8356 if only local resources are registered

Commitment control 99
v CPF835C if only remote resources are registered
v CPF83E4 if both local and remote resources are registered

In addition, one of two messages may appear specific to job completion if a commit and rollback exit
program for an API committable resource has been called. If the commit and rollback exit program does
not complete within 5 minutes, the program is canceled, a diagnostic message CPD8363 (for commit) or
CPD8364 (for rollback) is sent, and the remainder of the commit or rollback processing continues.

Commit or Rollback Processing During IPL

All of the situations described in the previous table also apply during IPL recovery for commitment
definitions except that message CPF835F is sent instead of message CPF8359 or CPF8363. Messages that
get sent for a particular commitment definition may appear in the job log for one of the QDBSRVxx jobs
or the QHST log. In the QHST log, message CPI8356 indicates the beginning of IPL recovery for a
particular commitment definition. Message CPC8351 indicates the end of IPL recovery for a particular
commitment definition and any other messages regarding the recovery of that commitment definition is
found between those two messages.

One of two messages may appear specific to a commitment definition if a commit and rollback exit
program for an API committable resource has been called. If the commit and rollback exit program does
not complete within 5 minutes, the program is canceled, a diagnostic message CPD8363 (for commit) or
CPD8364 (for rollback) is sent, and the remainder of the commit or rollback processing continues.

Detect deadlocks
A deadlock condition can occur when a job holds a lock on an object, object A, and is waiting to obtain a
lock on another object, object B. At the same time, another job or transaction currently holds a lock on
object B and is waiting to obtain a lock on object A.

Do the following steps to find out if a deadlock condition has occurred and fix it if it has:
1. Locate the suspended job in the list of active jobs. Determining the status of a job to determine if a job
is suspended.
2. Look at the objects the job is waiting to lock. For transactions with transaction-scoped locks see
“Display locked objects for a transaction” on page 62 for steps. For transactions with job-scoped locks
see Details: active job properties.
3. For all the objects the job is waiting to lock, look at the list of lock holders (transactions or jobs) and
try to find a conflicting lock corresponding to the level requested by the suspended job.
4. If a transaction is holding a conflicting lock, “Display jobs associated with a transaction” on page 62
and see if one of them is waiting to lock.
5. Determine if this waiting job is trying to lock one of the objects locked by the initial suspended job.
When you find the job that is trying to lock on of the objects locked by the initial suspended job, you
can identify the objects in question as the trouble spots.
6. Investigate the transaction in order to determine the appropriate course of action.
a. Look at the transaction properties to find out what application initiated it and then look at the
application code.
b. Or trace the transaction’s actions up to this point by finding the Commit cycle ID in the
transaction properties and then searching in a journal for entries containing this ID. To do this,
you can use the Retrieve Journal Entry (RTVJRNE) command and specify the CMTCYCID
parameter.
c. After obtaining relevant information the user may choose to “When to force commits and rollbacks
and when to cancel resynchronization” on page 101 operation.

100 iSeries: Commitment control

Recover transactions after communications failure
In case of a communications failure, the system typically completes the resynchronization with any
remote system automatically. However, if the failure is catastrophic such that the communications will
never be reestablished to the remote system (if, for instance, the communication line is cut), you must
cancel resynchronization and restore transactions yourself. The transactions also might be holding locks
that need to be released.
1. In iSeries(TM) Navigator “Display commitment control information” on page 61 for the transaction
with which you are working.
2. Find the transaction of interest that is trying to resynchronize with the remote system. The
Resynchronization in Progress field for that transaction is set to yes.
3. Look for transactions that had a connection to the remote system by checking the resource Status for
individual transactions.
4. After identifying transactions, depending on the state of the transaction you might must “When to
force commits and rollbacks and when to cancel resynchronization.”
5. You can make the decision to commit or rollback after you investigate the transaction properties.
v You can use the Unit of Work ID to find other parts of the transaction on other systems.
v You can also determine to commit or rollback from the state of transaction. For example, if a
database transaction was performing two-phase commit during communication failure and its state
after the failure is ″prepared″ or ″last agent pending″, you might choose to force commit on the
transaction.
6. After forcing a commit or rollback on the transactions in doubt, stop resynchronization on the failed
connection for the identified transactions.

For more information about recovery see “When to force commits and rollbacks and when to cancel
resynchronization.”

When to force commits and rollbacks and when to cancel

resynchronization
The decision to force a commit, roll back, or cancel resynchronization is called a heuristic decision. A
heuristic decision is the action that you take when you force the system to commit or to roll back a
transaction. When you make a heuristic decision, the state of the transaction becomes heuristic mixed if
your decision is inconsistent with the results of the other locations in the transaction. You must take
responsibility for determining the action taken by all the other locations that participated in the
transaction and resynchronizing the database records.

Before you make a heuristic decision, gather as much information as you can about the transaction.
Display the jobs that are associated with the commitment definition and make a record of what journals
and files are involved. You can use this information later if you need to display journal entries and apply
or remove journaled changes manually.

The best place to find out information about a transaction is to look at the location that was the initiator
for that transaction. However, the decision to commit or roll back may be owned by an API resource or
by a last agent.

If an API resource was registered as a last agent resource, the final decision of whether to commit or roll
back is owned by the API resource. You need to look at information about the application and how it
uses the API resource to determine whether to commit or to roll back.

If the transaction has a last agent selected, the last agent owns the decision to commit or roll back. Look
at the status of the last agent for information about the transaction.

When you must make heuristic decisions or cancel resynchronization due to a system or communications
failure that cannot be repaired, you can find all transactions in doubt by using the following:

Commitment control 101

1. In iSeries(TM) Navigator, expand the system you want to work with.
2. Expand Databases and the local database for the system.
3. Expand Transactions.
4. Expand Database Transactions or Global Transactions

In this display you can see the commitment definition, resynchronization status, the current unit of work
ID, and the current unit of work state for each transaction. Look for transactions with the following:
v Transactions with a Logical Unit of Work State of Prepared or Last Agent Pending.
v Transactions that show Resynchronization in Progress status of yes.

To work with the job that is participating in the transaction on this system right-click the transaction and
select job.

When you right-click the transaction, you can also select Force Commit, Force Rollback, or Cancel
Resynchronization.

Before making a heuristic decision or canceling resynchronization, you might want to check the status of
the jobs on other systems associated with the transaction. Checking the jobs on remote systems might
help you avoid decisions that cause database inconsistencies between systems.
1. Right-click the transaction you want to work with.
2. Select Resource Status
3. In the Resource Status dialog, select the Conversation tab for SNA connections; select Connection for
TCP/IP connections.

Each conversation resource represents a remote system that is participating in the transaction. On the
remote systems, you can use iSeries Navigator to see the transactions associated with the transaction.

The base portion of the unit of work ID is the same on all the systems. When you “Display commitment
control information” on page 61 on the remote system, look for the base portion of the unit of work ID
that is the same on the local system.

For example, if the unit of work ID on the local system starts with: APPN.RCHASL97.X’112233445566, look
for the unit of work ID on the remote system that also starts with APPN.RCHASL97.X’112233445566.

End a long-running rollback

A rollback operation removes all changes made within a transaction since the previous commit operation
or rollback operation. During a rollback operation, the system also releases locks related to the
transaction. If the system contains thousands of transactions, the system can take hours to complete a
rollback operation. These long-running rollbacks can consume critical processor time, lock resources or
take up storage space.

Before you end a long-running rollback, you need to know which commitment definitions are being
rolled back and what state the commitment definitions are in. The State field for commitment definitions
that are rolling back is set to ROLLBACK IN PROGRESS.

Use the Work with Commitment Definitions (WRKCMTDFN) command to check the status of a rollback
by following these steps:
1. Type WRKCMTDFN JOB(*ALL) from the character-based interface.
2. Type F11 to display the State field.

102 iSeries: Commitment control

If you end a long-running rollback, files that were changed during the transaction will be left with partial
transactions. You must not end a rollback if your files cannot have partial transactions. To see which files
were changed during the transaction, choose option 5 to display status from the WRKCMTDFN list. Press
F6 to display resource status and select Record Level.

You must have All Object (*ALLOBJ) special authority to end a long-running rollback. To end a
long-running rollback, follow these steps:
1. Type WRKCMTDFN JOB(*ALL) from the character-based interface.
2. Type option 20 (End rollback) on the commitment definition you want to end.

Files with partial transactions have the Partial Transactions Exist, Rollback Ended field set to *YES in the
output from the Display File Description (DSPFD) command. You must remove partial transactions before
the file can be used. You can remove partial transactions by deleting the file and restoring the file from a
prior save. If you do not have a prior save, you can use the Change Journaled Object (CHGJRNOBJ)
command to reset the Partial Transaction Exists state so you can open the file. Using the CHGJRNOBJ
requires you to edit the file to bring the file to a consistent state. You must use the CHGJRNOBJ
command only if no prior save is available.

Disable the ability to end a long-running rollback

Users with *ALLOBJ special authority can end rollbacks by default. If you want to restrict users who
have *ALLOBJ special authority from ending rollbacks, you can do this by creating data area
QGPL/QTNNOENDRB.

For more information about data areas see the Create Data Area (CRTDTAARA) command.

Related information for commitment control

The following are the iSeries(TM) manuals and IBM(R) Redbooks(TM) (in PDF format), Web sites, and
Information Center topics that relate to the Commitment control topic. You can view or print any of the
PDFs.

Manuals
v See the COBOL/400(R) User’s Guide

(5.8 MB) on the V5R1 Supplemental Manuals Web site.

v See the RPG/400(R) User’s Guide

(2 MB) on the V5R1 Supplemental Manuals Web site.

Redbooks
v Connecting WebSphere(R) to DB2(R) UDB Server

(5 MB)
v Advanced Functions and Administration on DB2 Universal Database(TM) for iSeries

Commitment control 103

 (5.4 MB)
v Stored Procedures and Triggers on DB2 Universal Database for iSeries

(4.8 MB)
v Striving for Optimal Journal Performance on DB2 Universal Database for iSeries

(3.1 MB)

Web site

The Open Group (www.opengroup.org)

Other information
v Database programming
v SQL programming concepts
v XA APIs
v Journal management

To save a PDF on your workstation for viewing or printing:

1. Right-click the PDF in your browser (right-click the link above).
2. Click Save Target As...
3. Navigate to the directory in which you want to save the PDF.
4. Click Save.

If you need Adobe Acrobat Reader to view or print these PDFs, you can download a copy from the
Adobe Web site (www.adobe.com/products/acrobat/readstep.html)

104 iSeries: Commitment control

Appendix. Notices
This information was developed for products and services offered in the U.S.A.

IBM(R) may not offer the products, services, or features discussed in this document in other countries.
Consult your local IBM representative for information about the products and services currently available
in your area. Any reference to an IBM product, program, or service is not intended to state or imply that
only that IBM product, program, or service may be used. Any functionally equivalent product, program,
or service that does not infringe any IBM intellectual property right may be used instead. However, it is
the user’s responsibility to evaluate and verify the operation of any non-IBM product, program, or
service.

IBM may have patents or pending patent applications covering subject matter described in this
document. The furnishing of this document does not give you any license to these patents. You can send
license inquiries, in writing, to:
IBM Director of Licensing
IBM Corporation
500 Columbus Avenue
Thornwood, NY 10594-1785
U.S.A.

For license inquiries regarding double-byte (DBCS) information, contact the IBM Intellectual Property
Department in your country or send inquiries, in writing, to:
IBM World Trade Asia Corporation
Licensing
2-31 Roppongi 3-chome, Minato-ku
Tokyo 106, Japan

The following paragraph does not apply to the United Kingdom or any other country where such
provisions are inconsistent with local law: INTERNATIONAL BUSINESS MACHINES CORPORATION
PROVIDES THIS PUBLICATION “AS IS” WITHOUT WARRANTY OF ANY KIND, EITHER EXPRESS
OR IMPLIED, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
NON-INFRINGEMENT, MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE. Some
states do not allow disclaimer of express or implied warranties in certain transactions, therefore, this
statement may not apply to you.

This information could include technical inaccuracies or typographical errors. Changes are periodically
made to the information herein; these changes will be incorporated in new editions of the publication.
IBM may make improvements and/or changes in the product(s) and/or the program(s) described in this
publication at any time without notice.

Any references in this information to non-IBM Web sites are provided for convenience only and do not in
any manner serve as an endorsement of those Web sites. The materials at those Web sites are not part of
the materials for this IBM product and use of those Web sites is at your own risk.

IBM may use or distribute any of the information you supply in any way it believes appropriate without
incurring any obligation to you.

Licensees of this program who wish to have information about it for the purpose of enabling: (i) the
exchange of information between independently created programs and other programs (including this
one) and (ii) the mutual use of the information which has been exchanged, should contact:

© Copyright IBM Corp. 2003, 2005 105

IBM Corporation
Software Interoperability Coordinator, Department 49XA
3605 Highway 52 N
Rochester, MN 55901
U.S.A.

Such information may be available, subject to appropriate terms and conditions, including in some cases,
payment of a fee.

The licensed program described in this information and all licensed material available for it are provided
by IBM under terms of the IBM Customer Agreement, IBM International Program License Agreement, or
any equivalent agreement between us.

Any performance data contained herein was determined in a controlled environment. Therefore, the
results obtained in other operating environments may vary significantly. Some measurements may have
been made on development-level systems and there is no guarantee that these measurements will be the
same on generally available systems. Furthermore, some measurements may have been estimated through
extrapolation. Actual results may vary. Users of this document should verify the applicable data for their
specific environment.

Information concerning non-IBM products was obtained from the suppliers of those products, their
published announcements or other publicly available sources. IBM has not tested those products and
cannot confirm the accuracy of performance, compatibility or any other claims related to non-IBM
products. Questions on the capabilities of non-IBM products should be addressed to the suppliers of
those products.

All statements regarding IBM’s future direction or intent are subject to change or withdrawal without
notice, and represent goals and objectives only.

All IBM prices shown are IBM’s suggested retail prices, are current and are subject to change without
notice. Dealer prices may vary.

This information is for planning purposes only. The information herein is subject to change before the
products described become available.

This information contains examples of data and reports used in daily business operations. To illustrate
them as completely as possible, the examples include the names of individuals, companies, brands, and
products. All of these names are fictitious and any similarity to the names and addresses used by an
actual business enterprise is entirely coincidental.

COPYRIGHT LICENSE:

This information contains sample application programs in source language, which illustrate programming
techniques on various operating platforms. You may copy, modify, and distribute these sample programs
in any form without payment to IBM, for the purposes of developing, using, marketing or distributing
application programs conforming to the application programming interface for the operating platform for
which the sample programs are written. These examples have not been thoroughly tested under all
conditions. IBM, therefore, cannot guarantee or imply reliability, serviceability, or function of these
programs. You may copy, modify, and distribute these sample programs in any form without payment to
IBM for the purposes of developing, using, marketing, or distributing application programs conforming
to IBM’s application programming interfaces.

Each copy or any portion of these sample programs or any derivative work, must include a copyright
notice as follows:

(C) (your company name) (year). Portions of this code are derived from IBM Corp. Sample Programs. (C)
Copyright IBM Corp. _enter the year or years_. All rights reserved.

106 iSeries: Commitment control

If you are viewing this information softcopy, the photographs and color illustrations may not appear.

Programming Interface Information

This Commitment control topic documents intended Programming Interfaces that allow the customer to
write programs to obtain the services of i5/OS.

Trademarks
The following terms are trademarks of International Business Machines Corporation in the United States,
other countries, or both:
Application System/400
AS/400
e (logo)
IBM
iSeries
Operating System/400
OS/400
400
Websphere
RPG/400
Redbooks
Integrated Language Environment
DRDA
Distributed Relational Database Architecture
DB2 Universal Database
DB2
COBOL/400

Other company, product, and service names may be trademarks or service marks of others.

Terms and conditions for downloading and printing publications

Permissions for the use of the publications you have selected for download are granted subject to the
following terms and conditions and your indication of acceptance thereof.

Personal Use: You may reproduce these Publications for your personal, noncommercial use provided that
all proprietary notices are preserved. You may not distribute, display or make derivative works of these
Publications, or any portion thereof, without the express consent of IBM.

Commercial Use: You may reproduce, distribute and display these Publications solely within your
enterprise provided that all proprietary notices are preserved. You may not make derivative works of
these Publications, or reproduce, distribute or display these Publications or any portion thereof outside
your enterprise, without the express consent of IBM.

Except as expressly granted in this permission, no other permissions, licenses or rights are granted, either
express or implied, to the Publications or any information, data, software or other intellectual property
contained therein.

IBM reserves the right to withdraw the permissions granted herein whenever, in its discretion, the use of
the Publications is detrimental to its interest or, as determined by IBM, the above instructions are not
being properly followed.

You may not download, export or re-export this information except in full compliance with all applicable
laws and regulations, including all United States export laws and regulations. IBM MAKES NO
GUARANTEE ABOUT THE CONTENT OF THESE PUBLICATIONS. THE PUBLICATIONS ARE

Appendix. Notices 107

PROVIDED ″AS-IS″ AND WITHOUT WARRANTY OF ANY KIND, EITHER EXPRESSED OR IMPLIED,
INCLUDING BUT NOT LIMITED TO IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
FOR A PARTICULAR PURPOSE.

All material copyrighted by IBM Corporation.

By downloading or printing a publication from this site, you have indicated your agreement with these
terms and conditions.

Code example disclaimer

IBM grants you a nonexclusive copyright license to use all programming code examples from which you
can generate similar function tailored to your own specific needs.

All sample code is provided by IBM for illustrative purposes only. These examples have not been
thoroughly tested under all conditions. IBM, therefore, cannot guarantee or imply reliability, serviceability,
or function of these programs.

All programs contained herein are provided to you ″AS IS″ without any warranties of any kind. The
implied warranties of non-infringement, merchantability and fitness for a particular purpose are expressly
disclaimed.

108 iSeries: Commitment control




Printed in USA