An Overview of Session Recovery in PowerCenter

Content
Recovery allows the user to restart a failed session and complete it as if the session had run without pause.When the Integration Service runs a session in recovery mode, it continues to commit data from the point of the last successfulcommit.This article explains the process of recovery and is aimed at users who will be incorporating this feature for recovering their sessions.
Requirements for Recovery Run
For a session to run in recovery mode, the following criteria need to be satisfied.Otherwise an attempt to perform a recovery session run will fail.

• The session is enabled for recovery. A Recovery Strategy (under Edit Session >Properties > General Options) must beselected for every individual session task that needs to be recovered (in previous versions it the Enable Recovery option wasunder Config Object tab in Session Properties).Sessions upgraded from a previous version of PowerCenter have recovery disabled by default.



• The workflow/session is run in recovery mode. This can be done using the workflow run options Recover Task or RecoverWorkflow From Task. These options are available in Workflow Monitor or Workflow Manager or using pmcmd.



• The previous session run did not succeed. This includes normal and recovery mode session runs.



• The number of partitions is not changed between the normal and recovery run.



• The recovery table created at the target database is not dropped or changed by the user.



• The OS platform is not changed between normal and recovery run.
  Recovery Process
  When you run a session enabled for recovery, the Integration Service creates and stores session status information in recoverytables and cache files. If the session fails and you recover the session, the Integration Service uses this status information todetermine where the previous session failed and where to continue the session. The type of recovery the Integration Serviceperforms depends on the target type. You can also choose to run part of the workflow when you recover the failed session.
  
  Recovery Tables
  When you enable recovery for a session that writes to a relational target, the Integration Service creates two tables at the targetdatabase, PM_RECOVERY and PM_TGT_RUN_ID. The Integration Service updates these tables during regular session runs with statusinformation about the target load. When the Integration Service runs a failed session in recovery mode, it uses the information in therecovery tables to determine where the previous session failed, and then continues the session from that point. Do not edit or deleteinformation in the recovery tables.
  PM_RECOVERY
  This table records the target load information during the session run per target instance per partition. In previous releases, this tablewas OPB_SRVR_RECOVERY. PM_RECOVERY saves the following information:



• Repository ID. This allows multiple repositories to use the same recovery table.



• Workflow ID



• Folder ID



• Session instance ID



• Workflow run ID



• Target instance ID. The Integration Service creates one row per target instance



• Partition ID. The Integration Service creates one row per partition



• Row count. Stores the last committed number of rows



• Check point. Used with GMD sources only.
  PM_TGT_RUN_ID
  PM_TGT_RUN_ID is for internal use. It stores the latest unique number the Integration Service uses to identify the target instance.This table has a single field, LAST_TGT_RUN_ID.
  Recovery Cache Files
  As a result of GMD integration, the Integration Service also creates a pmgmd cache file in $PMCacheDir when a normal session that isenabled for recovery fails. The Integration Service creates the cache file in the following format:

pmgmd_metadata_ _ _ _ _ _ _ _ _ .dat

Here is a sample cache file name:
pmgmd_metadata_7661f424_379f_11d7_947e_f63b53abfef7_103_2_102_0_0_0_1_1.dat

• The Integration Service creates cache files for both relational and non-relational targets. The session will fail if either the recoverytables or the pmgmd cache files cannot be created due to insufficient privileges.When a failed session is run in recovery mode, the Integration Service uses the pmgmd cache files when it recovers the session. Donot edit or delete the cache files before the recovery run.
  Recovery Types
  Internally, recovery can be divided into two types: incremental recovery and full load recovery.In incremental load recovery, the Integration Service uses the ROW_COUNT information in the PM_RECOVERY table to determine thenumber of rows committed to the target. When the session is recovered, previously loaded rows will be skipped. Thus, incrementalrecovery ensures that there are no duplicates loaded to the target. The Integration Service performs incremental recovery forrelational, normal load targets.In full load recovery, data is loaded from the beginning during the recovery run. The Integration Service performs full load recoveryfor relational bulk load, flat file (including regular, MQSeries, and external loader), XML, and SAP BW targets. Recovery for nonrelationaltargets is done using the pmgmd file created at the $PMCacheDir cache directory.
  Recovery and Truncation
  During the recovery session run, the Integration Service ignores the Truncate Target Table option for relational normal load targets,since the load will continue from the last commit point. For relational bulk load targets the Integration Service allows the TruncateTarget Table option in both normal and recovery run.
  At the end of a successful session run the Integration Service deletes all recovery information in the target database and in$PMCacheDir. The Integration Service always resets the recovery information when it runs the session normally, regardless ofwhether the previous session run succeeded or failed.
  Recovery Scenarios
  Recovery can be performed under the following scenarios:



• The workflow suspends. If the workflow suspends due to session failure, you can recover the failed session and resume theworkflow.



• The workflow fails. If the workflow fails as a result of session failure, you can recover the session and run the rest of theworkflow.



• The workflow completes, but a session fails. If the workflow completes, but a session fails, you can recover the session alonewithout running the rest of the workflow.

In order to run a recovery session, the following options are available in the Workflow Manager:

• Recover Task. Use this option when you want to recover a single session.



• Recover Workflow From Task. Use this option when you want to recover the session and continue the workflow.The following options are available in the Workflow Monitor



• Resume/Recover. Use this option to resume a suspended workflow in recovery mode.



• Recover Workflow From Task. Use this option when you want to recover the session and continue the workflow.



• Recover Task. Use this option when you want to recover a single session.

You can also use one of the following pmcmd commands to run the session in recovery mode when you specify -recovery

• pmcmd startworkflow



• pmcmd resumeworkflow



• pmcmd resumeworklet



• pmcmd starttask
  Unrecoverable Sessions
  PowerCenter does not support session recovery in the following cases:



• The session uses non-pass-through partitioning. All partition points must be pass-through partitioning unless the partition pointpasses data to an Aggregator or Rank transformation. If Enable Recovery is enabled in the session properties, you can onlyselect the default partition types for each partition point.



• The session uses database partitioning. PowerCenter does not support recovery for a session that uses database partitioning.



• You want to debug the mapping. You cannot perform recovery from the Debugger.



• You want to perform a test load. You cannot perform a test load if the session is enabled for recovery. If you enable recovery,the Workflow Manager disables the Test Load option.

In general, it is safe to assume a changed(edited and saved) mapping or session will result in an unrecoverable session or anindeterministic data load, depending on the extent of the changes. To perform consistent data recovery, the session properties for therecovery session must be the same as the session properties for the normal failed session. You might get inconsistent data if youperform recovery under the following circumstances:

• The Integration Service runs in Unicode mode and you change the session sort order.



• The Integration Service code page or source and target code pages change after the initial session failure.



• Data movement mode changes after the initial session failure.



• The session performs incremental aggregation and the Integration Service stops unexpectedly.



• The mapping uses a Sequence Generator transformation.



• The mapping uses a Normalizer transformation.



• The sources or targets change after the initial session failure.



• The mapping uses a lookup transformation that performs a lookup on the target table, or the data in the lookup table changesbetween session runs.



• An XML/VSAM source generates keys that the Integration Service uses to populate the target.



• You edited the mapping in a way that causes the Integration Service to distribute data differently.