DB2 FOR LUW / DB2FORLUW / A50 - DB2

If smp mode is enabled (database manager configuration setting 
for INTRA_PARALLEL is set to ON), and a query is run that 
contains a hash join, then the hash join logic will be run in 
parallel using multiple db2 agents. 
 
  There exists a timing issue in this setup where if the very 
first agent to start initializating the hash join is currently 
working, and then if one of the other agents working on the hash 
join is interrupted, then there is a logic problem in the error 
handling and cleanup processing that results in freeing memory 
while the first agent is still doing the initialization work. 
This leads to the agents making changes to memory incorrectly. 
 
  And this issue is a regression of IC92447, IC92731, IC95345 on 
the following db2 level. 
  - v97 fp9 
  - v10.1 fp3 and fp4 
  - v10.5 fp3 and fp4 
 
  Since the problem is a memory corruption, a variety of 
possible crash symptoms may result.  The following are the 
symptoms that were seen when the problem was discovered: 
 
  Sample 1: 
    sqloCrashOnCriticalMemoryValidationFailure 
    diagnoseMemoryCorruptionAndCrash 
    diagnoseMemoryCorruptionAndCrash 
    .MemTreePut 
    sqlofmblkEx 
    sqlofmblkEx 
    sqleIntrptWaitPostTerm 
    sqleIntrptWaitPostTerm 
    sqlri_hsjnClose 
    sqlrihsjn 
    sqlriSectInvoke 
    sqlrr_smp_route 
    sqlrr_subagent_router 
    sqleSubRequestRouter 
    sqleProcessSubRequest 
    RunEDU 
    EDUDriver 
    sqloEDUEntry 
 
  Sample 2: 
    sqloxltc_track 
    sqlri_hsjnNewTuple 
    sqlrihsjnpd 
    . 
    sqlifnxt 
    .sqlirdk 
    .sqldIndexFetch 
    sqldRowFetch 
    sqlritaSimplePerm 
    sqlriExecThread 
    sqlrihsjn 
    sqlriExecThread 
    sqlrihsjn 
    sqlriExecThread 
    sqlrihsjn 
    sqlriSectInvoke 
    sqlrr_smp_route 
    sqlrr_subagent_router 
    sqleSubRequestRouter 
    sqleProcessSubRequest 
    RunEDU 
    EDUDrive 
    sqloEDUEntry

**************************************************************** 
* USERS AFFECTED:                                              * 
* db2 v10.5 fp3, fp4 user                                      * 
**************************************************************** 
* PROBLEM DESCRIPTION:                                         * 
* See Error Description                                        * 
**************************************************************** 
* RECOMMENDATION:                                              * 
* Upgrade to db2 Version 10.5 FixPack 5 or higher              * 
****************************************************************

Any of the following changes can avoid the problem: 
  1) Turn off intra_parallel or set the query degree to 1 
  2) diable hash join (query optimization level 3, or db2set 
DB2_HASH_JOIN=OFF)

Fixed in Version 10.5 FixPack 5

not known / see Local fix

Date  - problem reported    :
Date  - problem closed      :
Date  - last modified       :

24.06.2014
15.04.2016
15.04.2016