DB2 FOR LUW / DB2FORLUW / B10 - DB2

Under certain timing conditions a hang/deadlatch can occur after
an agent has sent modified pages to the Group Buffer Pool (GBP).
An agent (A) will be holding one or more page latches
(SQLO_LT_SQLB_BPD__bpdLatch_SX) in shared (S) mode, and
attempting to upgrade a page latch to exclusive (X) mode in
function sqlbgbUpgradeLatchFromShareDuringWAR. This agent will
have a stack similar to the following:

getConflictComplex
getConflict
sqlbgbUpgradeLatchFromShareDuringWAR
sqlbgbWARM_PostProcessing
sqlbgbWARM

Meanwhile, another agent (B) will be holding (in S or X mode)
the page latch agent A is attempting to latch. At the same time
agent B will be attempting to latch in X mode one of the page
latches that is held in S mode by agent A.

Agent B might have a stack similar, but not limited to, the
following:

getConflictComplex
getConflict
sqlbFindPageInBPOrSim
sqlbfix

****************************************************************
* USERS AFFECTED:                                              *
* all                                                          *
****************************************************************
* PROBLEM DESCRIPTION:                                         *
* See Error Description                                        *
****************************************************************
* RECOMMENDATION:                                              *
* Upgrade to Db2 11.1.4.6 or higher                            *
****************************************************************

The hang/deadlatch is contained within a member. Killing the
member and allowing it to recover will resolve the hang.

A workaround to prevent the hang is to do the following:
db2set DB2_HIGH_CONCURRENCY_WAR=OFF. Each member would have to
have db2stop and db2start run for this to take effect. Note that
this will have a negative performance impact for heavy insert,
update, or delete workloads.

not known / see Local fix

forerunner  : 
follow-up : IT33863 

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

07.08.2020
31.03.2021
31.03.2021