|
Problem IT01180
|
Status: Closed |
POTENTIAL DEAD LATCH OR HANG WITH A LOAD AND TWO INDEX REBUILDS |
| product: |
DB2 FOR LUW / DB2FORLUW / A50 - DB2 |
| Problem description: |
A timing window exists between three applications that could
result in a dead latch. In order for the this to materialize
the
following things must happen:
- A range-partitioned table exists with multiple ranges, where
each range has at least one partitioned index.
- At least two of the partitioned indexes in the table have
indexes that were marked invalid and require index recreation to
rebuild them.
- Both partitioned indexes that are marked invalid must be
defined in the same table space.
- Two separate applications must drive index recreation for each
invalid index above.
- A third application has to be running the LOAD utility. The
LOAD utility can be for another table, as long as it shares the
same table space as the table space used for the partitioned
indexes above.
- The LOAD utility must be using the 'COPY NO' option
- The connection for the LOAD utility must be using LOCK TIMEOUT
set to
infinite
This dead latch can be detected by analyzing a combination of
the
lock monitoring, db2pd -latches and db2pd -stack output.
The three applications will be holding and requesting the
resources in the following states:
1) First index recreate
Holding latch type identified by
SQLO_LT_SQLD_TCB_LOTCH__tcbIndexValidityLotch
Waiting for the table space lock. Requesting the lock in IX
mode.
2) LOAD
Waiting for the table space lock. Requesting the lock in U
mode.
A snippet of the stack while it is waiting, which can be
generated with db2pd -stack:
_ZN8sqeAgent14IntrptWaitLock
_Z18sqleIntrptWaitLock
_Z21sqlplWaitForLockGrantP9sqeBsuEdu
_Z13sqlplWaitOnWP
_Z21sqlplMakeNewRequestSD
_Z19sqlplMakeNewRequest
_Z7sqlplrqP9sqeBsuEduP14SQLP_LOCK_INFO
_Z13sqluLockPoolsP8sqeAgentP10sqluCTablej
_Z23sqluRegisterLoadRestart
_Z29sqluAcquirePartitionResources
_Z24sqlu_register_table_load
3) Second index recreate
Holding the table space lock in IX mode.
Waiting for the latch type identified by
SQLO_LT_SQLD_TCB_LOTCH__tcbIndexValidityLotch |
| Problem Summary: |
****************************************************************
* USERS AFFECTED: *
* All users *
****************************************************************
* PROBLEM DESCRIPTION: *
* See Error Description *
****************************************************************
* RECOMMENDATION: *
* Upgrade to the latest fix pack. *
**************************************************************** |
| Local Fix: |
To break the dead latch the LOAD application must be
interrupted. This can be done with the FORCE APPLICATION
command. |
| available fix packs: |
DB2 Cancun Release 10.5.0.4 (also known as Fix Pack 4) for Linux, UNIX, and Windows
DB2 Version 10.5 Fix Pack 9 for Linux, UNIX, and Windows
|
| Solution |
Problem is first fixed in DB2 UDB version 10.5 fix pack 4. |
| Workaround |
not known / see Local fix |
| Timestamps |
Date - problem reported :
Date - problem closed :
Date - last modified :
| 22.04.2014
08.09.2014
08.09.2014 |
| Problem solved at the following versions (IBM BugInfos) |
|
| Problem solved according to the fixlist(s) of the following version(s) |
| 10.5.0.4
|
 |
