If an inplace reorg processes a pointer / overflow pair that
reside on the same page and the processing of the records
rolls back, the page on which the records reside may be
left in a corrupted state. The undo of this operation can leave
an invalid pointer record on the page, and a random corruption
on the same page.
A variety of symptoms can be experienced including: an abnormal
termination during inplace reorg clean up, incorrect data, or an
abnormal termination while reading data from the page.
Examples of possible errors are the following db2diag.log
messages:
(1) Error detected while reading data from the affected page
(after the rollback completes) and the invalid pointer record is
read:
2014-11-12-12.59.23.430273-300 I2734809A600 LEVEL:
Severe
PID : 21168154 TID : 17031 PROC :
db2sysc 1
INSTANCE: db2user NODE : 001 DB :
SAMPLE
APPHDL : 1-5461 APPID:
9.26.96.91.46396.141112190345
AUTHID : DB2USER HOSTNAME: coralpib210
EDUID : 17031 EDUNAME: db2agent (SAMPLE) 1
FUNCTION: DB2 UDB, data management, sqldFetchOverflow,
probe:2098
MESSAGE : Followed REORGPOINTER but no OLR in progress!
DATA #1 : Hexdump, 8 bytes
0x0700000007BF71E0 : 0000 0000 0000 0008
........
followed by a "badPage" error.
(2) Error detected during the reorg process:
2014-11-05-23.39.21.271862-300 I10732900A597 LEVEL:
Severe
PID : 9240738 TID : 118481 PROC :
db2sysc 1
INSTANCE: db2user NODE : 001 DB :
TESTDB
APPHDL : 1-5303 APPID: *N1.DB2.141106050704
AUTHID : DB2USER HOSTNAME: p8svt15.torolab.ibm.com
EDUID : 118481 EDUNAME: db2reorg (TESTDB) 1
FUNCTION: DB2 UDB, data management,
sqldUndoConvertPtrOvflwRecord, probe:3577
RETCODE : ZRC=0x87040001=-2029780991=SQLD_BADPAGE "Bad Data
Page"
DIA8500C A data file error has occurred, record id is
"".
There may be other symptoms as well. The error condition is
more likely to be encountered in a shared data (pureScale)
environment. |
