2003
Standart Edition.
There were series of similar instance crashes. At the begining, some
errors appeared in the db2diag.log like:
=================
2004-06-06-17.21.48.671000 Instance:DB2 Node:000
PID:2348(dbm.exe) TID:1696 Appid:none
oper system services sqloSSemClose Probe:20
Unexpected system error 0x6 has occurred.
This has been mapped to ZRC 0x83000006.
PID:2348 TID:1696 Node:000 Title: SYSTEM ERROR DESCRIPTION
The handle is invalid.
=================
and after a while the instance crashes without additional logging into
the
db2diag.log.
Crash recovery went fine after instance restart.
I opened PMR, and received an answer
The one message that is appearing over and over again is the first
one listed above. That ZRC code indicatest that this is a "Resource
Capacity Error" in buffer pool services. Specifically, this is an
"SQLZ_RC_BPFULL" which means that there are no available buffer pool
pages. The associated SQLCODE is:
SQL1218N There are no pages currently available in bufferpool "".
Could you please tune your bufferpool size and then please let me
know if these messages persist? Thank you.
It sounds to me as a formal answer, which doesn't solve the problem at
all.
Actually, we have 2 separate bufferpools (each about 500 Mb) for
indexes
and data. The bufferpool for indexes is large enough to cache them
all, and
the bufferpool for data can cashe more than 25% of data. All update
transactions are pretty small (not more than 10-20 rows) and only one
application processes them, all heavy selects are issued by
applications
with UR isolation level. We are monitoring
SQL_ELM_POOL_ASINC_DATA/INDEXES_WRITES and
SQL_ELM_POOL_DATA/INDEXES_WRITES
on a regular basis and even before crashes they were still almost
equal, so
as far as I understand, it means that there are a lot of pages that
could be removed from a bufferpool at any given moment.
Occasionally, I changed db2ntworkset from DB2NTWORKSET=1024,2560 to
DB2NTWORKSET=1024,3072, and since that changes everything works fine,
although I did not change any bufferpools parameters.
I am not sure if it is a coincedence or it does cure the problem.
Does enyone have an idea or advice?