PK90219: DFR GETS INFINITE LOOP IN DFRCMRCX (X=9, A OR B)

Obtain the fix for this APAR.

APAR status

• Closed as program error.

Error description

• If a failed IMS was in the following situation, DFR generates
  an External Subsystem UOR Status list:
    - The IMS is the sync point coordinator.
    - The IMS application program makes one or more External
      Subsystem (ESS) calls.
    - The external subsystem contains any in-doubt UORs.
  Then, DFR analyzes the condition of the external subsystem with
  IMS logs and gets an infinite loop in DFRCMRCx (x=9, A or B).
  The loop is due to inaccurate resources that are left by
  the previous analysis for the condition of the external
  subsystem.
  
  Keywords: DFRCMRC9 DFRCMRCA DFRCMRCB looping loops
  

Local fix

Problem summary

• ****************************************************************
  * USERS AFFECTED: All IMS DEDB FAST RECOVERY Version 2         *
  *                 Release 2 (FMID=H1J2220) users who use an    *
  *                 IMS application program makes one or more    *
  *                 External Subsystem (ESS) calls in IMS V9,    *
  *                 IMS V10 or IMS V11.                          *
  ****************************************************************
  * PROBLEM DESCRIPTION: DFR gets infinite loop in DFRCMRCx      *
  *                      (x=9, A or B) by inaccurate resources   *
  *                      (RCIEs) that are left by previous       *
  *                      analysis for the situation of the       *
  *                      external subsystem.                     *
  ****************************************************************
  * RECOMMENDATION: Apply the maintenance for this APAR.         *
  ****************************************************************
  If a failed IMS was in the following situation, DFR generates
  an External Subsystem UOR Status list.
  - IMS is the sync point coordinator.
  - IMS application program makes one or more External Subsystem
    (ESS) calls.
  - The external subsystem contains any in-doubt UORs.
  Then, DFR analyzes the situation of the external subsystem with
  IMS logs.
  Usually, when a log record for a sync-abort-point is processed,
  DFR deletes a RPST and related resources, before processing
  X'07' or X'5612' log record for this RPST. However, when the
  external subsystem is being used, DFR saves the information of
  the external subsystem into an External Subsystem Block (ESBK)
  and even if DFR processes the log record for the sync-abort-
  point, DFR does not delete the RPST in order to process the
  log records of the external subsystem.
  Moreover, DFR processes only one of the log records about the
  RPST for the sync-commit-point or for the sync-abort-point.
  Therefore, DFR does not process the log record for the sync-
  commit-point in this case.
  When IMS application program is being run on an IFP region, or
  it issues a ROLB call, IMS writes the log records for the sync-
  abort-point. And IMS increases a recovery token of a UOR only
  after the sync-commit-point. After the ROLB call or an internal
  ROLB call for the IFP region, the recovery token is not
  increased. So, if DB updates and the sync-commit-point were
  performed after the log records for the sync-abort-point were
  written, these log records have the same recovery-token as the
  log records for the sync-abort-point.
  Even if the log record for two or more sync-point exists in one
  UOR, DFR recognizes the log record for the first sync-point to
  be the last situation of the UOR, and continues processing.
  Although DFR registers a RCIE by the log record for DB update
  after the log record for the sync-abort-point, the log record
  for the sync-commit-point is not processed.
  And even if the target RPST is deleted with the log record for
  the external subsystem before X'5612' or X'07' log record for
  the target RPST, the RCIE remains as inaccurate information.
  *
  The following two problems may occur depending on the situation
  of the failed IMS.
  . When other UOR updated the same CI of DEDB that is registered
    with the RCIE left as inaccurate information, DFR gets
    infinite loop in DFRCMRCx. (x=9, A or B)
  . When IMS failed before writing X'5612' or X'07' log record for
    the target UOR, DFR loses CI updates of DEDB which should be
    recovered.
  

Problem conclusion

• The following macro and modules have been modified.
  (x = 9, A and B)
  DFRRPSTx - Add a new flag RPSTCOMT (X'01') which indicates that
             X'37', X'5937' or X'5637' log record was processed.
  DFRCMRCx - This module was changed so that it ends with abend
             code U3999 in a subroutine RPUTONE when a previous
             RPST chained from a RCIE does not have a chain to the
             RCIE (RPSTRCIE=0).
  DFRLUORx - This module was changed as follows.
             . A RPSTWTDE flag is not turned ON when X'38',
               X'5938' or X'5638' log record is processed.
               (It turns ON only when X'5612' or X'07' log record
                is processed.)
             . A RPSTABOT flag is reset and a new RPSTCOMT flag is
               turned ON when X'37', X'5937' or X'5637' log record
               is processed.
  DFRL593x - This module was changed so that a new RPSTCOMT flag
             might be used for a judgement of whether a UOR was
             committed instead of a RPSTSYNC flag.
             Multiple log records for one sync-abort-point came to
             be processed by this change.
             For a mixed mode environment, the following processes
             were added so that multiple registration of the same
             information to a Sync-Abort UOR list might not be
             carried out by multiple log records for one sync-
             abort-point.
             . Judge whether X'5950' or X'5951' log record was
               processed by this sync-abort-point.
             . After an entry is registered, information required
               for registration (RPSTLSNF, RPSTLSNL and RPSTBSN)
               is cleared.
  DFRNLG0x - This module was changed as follows.
             . A Sync-Abort UOR entry has two log sequence
               numbers.
               One is the first update log record sequence
               number of a UOR, the other is the last update log
               record sequence number of the UOR aborted.
               This module was changed as follows in consideration
               of the log sequence numbers overlapping case.
               . When checking whether X'5950' or X'5951' log
                 record is registered into a Sync-Abort UOR list,
                 not only the log sequence number but also a
                 recovery token is checked.
               . When the entry is searched, it is searched from
                 the first Sync-Abort UOR list at every log
                 record.
               . When the log sequence number of the target log is
                 smaller than the first log sequence number
                 (ABTELSNF) of the Sync-Abort UOR entry, it
                 changed so that the search process might be
                 ended.
             . When a value of ABTLENXT field was 0 (when only one
               ABTL is used and 100 entries of the ABTL are used
               up), it changed so that the search process might be
               ended.
             . The error of the timing which clears a RSYNSTAT
               flag is corrected.
  *
  

Temporary fix

Comments

APAR Information

• APAR is sysrouted FROM one or more of the following:

• APAR is sysrouted TO one or more of the following:

  UK48423

Modules/Macros

•    DFRCMRCA DFRCMRCB DFRCMRC9 DFRLUORA DFRLUORB
  DFRLUOR9 DFRL593A DFRL593B DFRL5939 DFRNLG0A DFRNLG0B DFRNLG09
  DFRRPSTA DFRRPSTB DFRRPST9
  

Fix information

• Fixed component name

  DEDB FAST RECOV

• Fixed component ID

  5655E3200

Applicable component levels

• R220 PSY UK48423

     UP09/07/18 P F907

Fix is available

• Select the PTF appropriate for your component level. You will be required to sign in. Distribution on physical media is not available in all countries.