Details
-
Bug
-
Status: Resolved
-
Blocker
-
Resolution: Fixed
-
Kudu_Impala
Description
After loading a few more functional tables for Kudu, a simple agg on alltypes crashed.
[localhost:21000] > use functional_kudu; Query: use functional_kudu [localhost:21000] > select count(*) from alltypes where year = 2009.0; Error communicating with impalad: TSocket read 0 bytes
F0712 14:23:59.508661 5016 kudu-scan-node.cc:204] Check failed: predicate.nodes.size() == 3 (4 vs. 3) *** Check failure stack trace: *** @ 0x7fb405dec19d google::LogMessage::Fail() @ 0x7fb405deeac6 google::LogMessage::SendToLog() @ 0x7fb405debcbd google::LogMessage::Flush() @ 0x7fb405def56e google::LogMessageFatal::~LogMessageFatal() @ 0x7fb40a13b832 impala::KuduScanNode::TransformPushableConjunctsToRangePredicates() @ 0x7fb40a13ad6f impala::KuduScanNode::Open() @ 0x7fb40a0eb7a7 impala::PartitionedAggregationNode::Open() @ 0x7fb4081c3ec8 impala::PlanFragmentExecutor::OpenInternal() @ 0x7fb4081c3bff impala::PlanFragmentExecutor::Open() @ 0x7fb407880f6a impala::FragmentMgr::FragmentExecState::Exec() @ 0x7fb40788e213 impala::FragmentMgr::FragmentThread() @ 0x7fb4078a8d64 boost::_mfi::mf1<>::operator()() @ 0x7fb4078a768d boost::_bi::list2<>::operator()<>() @ 0x7fb4078a6103 boost::_bi::bind_t<>::operator()() @ 0x7fb4078a39a6 boost::detail::function::void_function_obj_invoker0<>::invoke() @ 0x7d61be boost::function0<>::operator()() @ 0x7fb40675c9f9 impala::Thread::SuperviseThread() @ 0x7fb406764480 boost::_bi::list4<>::operator()<>() @ 0x7fb4067643c3 boost::_bi::bind_t<>::operator()() @ 0x7fb40676431e boost::detail::thread_data<>::run() @ 0x8acd2a thread_proxy @ 0x7fb406270184 start_thread @ 0x7fb402d7537d (unknown) Picked up JAVA_TOOL_OPTIONS: -agentlib:jdwp=transport=dt_socket,address=30000,server=y,suspend=n Wrote minidump to /home/mj/dev/Impala/logs/cluster/minidumps/impalad/1511eaea-952d-efae-5ea04079-1217af5c.dmp
It looks like year gets cast to a double, but then we fail the DCHECK since the predicate isn't as simple as we expect.
For now we might need to avoid pushing such predicates down, but this kind of thing could lead to real perf issues.
Attachments
Issue Links
- relates to
-
IMPALA-3871 Casting literals to TIMESTAMP throw when pushed to KuduScanNode
- Resolved