Migrating ML for IBM z/OS to Enterprise edition v3.2.0

The core directory structure should look similar to the following example:

|-- alnsamp
|-- bin
|-- configuration
|   |-- defaults
|   `-- 
|-- extra
|   |-- examples
|   `-- log4j2.properties
|-- iml-library
|-- iml-services
|-- iml-utilities
|-- iml-zostools
|-- nodejs
|-- spark
`-- usr
    |-- extension
    |   |-- installableApps
    |   |   `-- scoring-wola.war
    |   `-- lib
    |       `-- scoring-osgi_1.0.0.jar
    `-- servers
|-- version.txt
`-- zosmf-wf
 

|-- cics-scoring
|-- iml-onnx
|-- iml-portal
|-- imlpython
|-- node_modules
`-- pyz
|-- trust-ai
`-- version.txt
 

• The user ID that will run the migration step is granted SYSADM privileges.
• The user ID associated with the MLz Core Server has read access to the BPX.SERVER FACILITY class.

1. Set $JAVA_HOME to the directory where a version of IBM® 64-bit SDK for z/OS Java™ is installed. See Installing prerequisite hardware and software for ML for IBM z/OS for the list of MLz supported IBM Java on z/OS.
2. Set $WLP_INSTALL_DIR to the directory where Liberty installation is installed. The default is /usr/lpp/liberty_zos/current.
3. Set $IML_INSTALL_DIR to the directory where your ML for z/OS Core is installed. The default is /usr/lpp/IBM/aln/v3r2/core.
4. Set $IML_HOME to the directory that is mounted with the zFS file system you allocated for v3.2.0.
5. Set $IML_JOBNAME_PREFIX to define the prefix of the jobnames of all MLz services. See Customizing the prefix of a ML for IBM z/OS service jobname and Configuring user ID for setting up Machine Learning for IBM z/OS Enterprise Edition for more information.
6. Set $SPARK_HOME to "${IML_INSTALL_DIR}"/spark.
7. Set $AIE_INSTALL_DIR to the directory where the IBM Z Deep Neural Network Library (zDNN) is located on your z/OS system. See Configuring user ID for setting up Machine Learning for IBM z/OS Enterprise Edition for more information.
8. Verify that the LIBPATH environment variable is correctly set as shown in the following example:

   LIBPATH=$LIBPATH:"${IML_INSTALL_DIR}"/iml-services/utils

9. Remove or unset the ANACONDA_ROOT environment variable, which is no longer used.
10. Remove the ANACONDA_ROOT reference from the PATH environment variable as shown in the following example:

    PATH=$PATH:"${ANACONDA_ROOT}"/bin

11. Set $IML_INSTALL_ENT_DIR to the directory where ML for z/OS Enterprise is installed. The default is /usr/lpp/IBM/aln/v3r2/ent.
12. Verify that the PATH environment variable is correctly set as shown in the following example if you have Python runtime configured:

    PATH=$PATH:"${IML_INSTALL_ENT_DIR}"/pyz/pyz-313/bin

13. Verify that the LIBPATH environment variable is correctly set as shown in the following example if you have Python runtime configured:

    LIBPATH=$LIBPATH:"${IML_INSTALL_ENT_DIR}"/pyz/pyz-313/lib

14. Add the following environment variables if you have Python runtime configured:

    export _TAG_REDIR_ERR=txt
    export _TAG_REDIR_IN=txt
    export _TAG_REDIR_OUT=txt
    
     

15. Comment out the line source $IML_HOME/spark/conf/spark.profile if it exists.
16. Issue the source .profile command to activate the changes.

./aln-migrate.sh 

1. Run the gen-alnwmeta.sh script in the $IML_INSTALL_DIR/alnsamp directory and specify the -u option as shown in the following example:

   ./gen-alnwmeta.sh -u <version>

   where <version> is the version of MLz you upgraded from. If the upgrade is from the v3.2.0 deliverable released under APAR PH66196, specify 3.2.0.1. If the upgrade is from an earlier v3.2.0 deliverable, specify 3.2.0.

   The script generates sample JCL files ALNWMETA and ALNWMETA.037. The content of the files is the same but as their names indicate, their encoding is different, with the former in ASCII and the latter in IBM-037. Both files are stored in the $IML_HOME/alnsamp directory.

2. Customize and run the JCL job to update the metadata table and other database objects. See Creating metadata objects for the Machine Learning for IBM z/OS Enterprise Edition repository service for instructions. If the upgrade is performed from v3.2.0 or v3.1.0, the ALTER statements in the job will place the associated tablespaces in ADVISORY REORG PENDING status, resulting in a return code of 04 (RC=04).
3. If the upgrade is performed from version v3.2.0 or v3.1.0, run Db2 REORG on the tablespaces in your metadata database that are altered in the previous step. The following is a sample REORG statement:

   REORG TABLESPACE !ALNDB! <tablespace> COPYDDN SYSCOPY 
        SHRLEVEL REFERENCE
    

   where <tablespace> refers to the altered tablespaces – TSMDVERS and TSDPLOY.

   The REORG job for TSMDVERS will complete with a return code of 04 (RC=04), as they place two auxiliary tablespaces in COPY PENDING status.

   The REORG job for TSDPLOY will complete with a return code of 04 (RC=04), as they place three auxiliary tablespaces in COPY PENDING status.

4. If the upgrade is performed from version v3.2.0 or v3.1.0, run a Db2 IMAGE COPY on the tablespaces that are in COPY PENDING status. The following is a sample COPY statement:

   COPY TABLESPACE !ALNDB!.<tablespace>

When you open the Function level page, an administration dashboard service will check and display the MLz function and metadata levels on your z/OS system.

If you are using Derby as your repository, it is already up to date. However, you can fall back to <current-version> if needed.

If you are using Db2 as your repository, the current function level should be <current-version>. If the new function level is v3.2.0.26 (the latest v3.2.0 deliverable), you have the option to activate the new function level, do nothing, or fall back to <current-version>.

• Activate the new v3.2.0 function level. Under the Action column, select Activate to update your function level to v3.2.0.26. After the process completes successfully, your MLz will be running at the latest v3.2.0 deliverable. You can perform normal machine learning activities, including creating, training, deploying, and scoring models.
• Do nothing. Your MLz will be running in read-only mode. You can continue running the existing scoring workload for testing purpose, but you cannot create, train, or deploy models.
• Fall back to <current-version>. In this case, complete the following steps to revert your MLz installation and environment back to <current-version>:
  1. Stop all v3.2.0 services and processes that are currently running on z/OS, including those of scoring service. See Stopping and starting services running on ML for IBM z/OS for instructions.
  2. Switch the user profile back to the original one and issue a source .profile command to update the environment.
  3. Remove the AT-TLS policy you just added for your Liberty scoring service and Python scoring service.
  4. Start all <current-version> services and processes.

     You do not need to revert metadata level from v3.2.0.26 to <current-version>. The <current-version> code is forward compatible with v3.2.0 metadata objects.

./aln-post-migrate.sh
 

./server.sh update <name>

From the update main menu, choose option 2, and then select option 2 from the sub-menu to update the cluster host. You will be prompted to enter IP address or host name of your Sysplex scoring cluster or press <enter> to leave blank for SHAREPORT or standalone.