A deadlatch can happen between a VectorizedADC thread and normal
apphandle thread which is rolling back. This will cause the
apphandle to hang and not be forced. The only way out is to
recycle the Db2 instance.
The stacks are:
VectorizedADC thread:
semop
SQLO_SLATCH_CAS64::getConflict(unsigned long)
sqlbfix(SQLB_FIX_CB*)
sqlbSFVSFixPage(SQLB_FIX_CB*, int, unsigned int, int, bool*)
ibm_cde::data::DatabaseTableAccessor::fixExistingPage(ibm_cde::d
ata::PageRequest&, int) const
ibm_cde::data::DatabaseTableAccessor::writeTableMetadata(unsigne
d long, unsigned long, bool)
ibm_cde::data::VectorizedADC::linkWriteAndEnableDictLocally(ibm_
cde::data::TableAccessor&)
ibm_cde::data::VectorizedADC::buildDictionary(ibm_cde::data::Tab
leAccessor&, bool, ADC_DICT_BUILD_SUMMARY*)
ibm_cde::data::VectorizedADC::mergeHistogramsFromRPC(ibm_cde::da
ta::TableAccessor&, ibm_cde::data::DataReader const&, unsigned
long, short, bool, bool&, bool&)
ibm_cde::data::VectADCPushHistogramsRPC::unmarshalReq(ibm_cde::d
ata::RPCDataReader const&)
ibm_cde::data::BaseRPC::receive(ibm_cde::data::RPCMessageHeader
const&, ibm_cde::data::RPCDataReader const&,
ibm_cde::data::RPCDataWriter&, ibm_cde::data::RPCReplyInfo&)
ibm_cde::data::RPCReceiver::routeRPC(ibm_cde::data::RPCMessageHe
ader const&, ibm_cde::data::RPCDataReader const&,
ibm_cde::data::RPCDataWriter&, ibm_cde::data::RPCReplyInfo&)
ibm_cde::data::RPCHandler::receiveReq(sqlrr_cb*,
SQLKD_RQST_REPLY_FORMAT*, unsigned long, SQLR_RPCMESSAGE**,
ibm_cde::data::RPCReplyInfo&)
ibm_cde::data::cdeRPCHandlerInterface::receiveReq(sqlrr_cb*,
SQLKD_RQST_REPLY_FORMAT*, unsigned long, SQLR_RPCMESSAGE**,
ibm_cde::data::RPCReplyInfo&)
sqlrr_rpc_router(sqlrr_cb*)
This thread is holding: SAVEPOINTS latch and waiting on BPD
latch
The apphandle thread which has got the rollback request:
__nanosleep
sqloSpinLockConflict
sqlpGetSPLatch(sqeAgent*, SQLP_SAVEPOINTS*, bool*, SQLP_SPTYPE)
sqlpSpReq(sqeAgent*, SQLP_SPTYPE, unsigned long*)
ibm_cde::data::Recovery::ScopedSavepoint::startSavepoint()
ibm_cde::data::ExtendedInsertRangeList::createNewExtendedInsertR
ange(ibm_cde::data::TableAccessor const&,
ibm_cde::data::TableCache const&,
ibm_cde::data::ExtendedInsertRangeRecord const&,
ibm_cde::data::ExclusivePageRequest&)
ibm_cde::data::DatabaseTableAccessor::writeTableStatistics(ibm_c
de::data::TableStatisticsWriter&)
ibm_cde::data::TableStatisticsWriter::writeTableStatistics(bool)
ibm_cde::data::InsertRangeList::createExtendedInsertRangeIfNeces
sary(unsigned long, ibm_cde::data::TableAccessor&,
ibm_cde::data::TableAccessor*)
ibm_cde::data::InsertRangeDistributor::reserveInsertRange(ibm_cd
e::data::CDEApplication*, ibm_cde::data::TableAccessor&,
ibm_cde::data::TableAccessor*, unsigned long)
ibm_cde::data::CDEApplication::findOrReserveInsertRange(ibm_cde:
:data::TableAccessor&, ibm_cde::data::TableAccessor*, unsigned
long, bool&)
ibm_cde::data::TableAccessor::associateWithInsertRange(ibm_cde::
data::TableAccessor*, unsigned long, bool&)
ibm_cde::data::TableAccessor::setInsertRangeForCurrentApplicatio
n(ibm_cde::data::TableAccessor*, unsigned long)
ibm_cde::data::TableAccessor::setInsertRangeForCurrentApplicatio
n(ibm_cde::data::TableAccessorAllocator*, unsigned long)
ibm_cde::query::Scheduler::prepInsertUpdateSTQ(ibm_cde::query::S
yntaxTree*)
This thread is holding the BPD latch and waiting on SAVEPOINTS
latch. |
