[IMPALA-8140] Grouping aggregation with limit breaks asan build

• Type: Bug
• Status: Resolved
• Priority: Blocker
• Resolution: Fixed
• Affects Version/s: Impala 3.1.0, Impala 3.2.0
• Fix Version/s: Impala 3.2.0
• Component/s: Backend
• Labels: asan crash
• Target Version: Impala 3.2.0
• Epic Color: ghx-label-8
• Assignee: Lars Volker
• Reporter: Lars Volker
• Created: 29/Jan/19 21:22
• Updated: 02/Feb/19 07:01
• Resolved: 01/Feb/19 11:39

Description

Commit 4af3a7853e9 for IMPALA-7333 breaks the following query on ASAN:

select count(*) from tpch_parquet.orders o group by o.o_clerk limit 10;

==30219==ERROR: AddressSanitizer: use-after-poison on address 0x631000c4569c at pc 0x0000020163cc bp 0x7f73a12a5700 sp 0x7f73a12a56f8
READ of size 1 at 0x631000c4569c thread T276
    #0 0x20163cb in impala::Tuple::IsNull(impala::NullIndicatorOffset const&) const /tmp/be/src/runtime/tuple.h:241:13
    #1 0x280c3d1 in impala::AggFnEvaluator::SerializeOrFinalize(impala::Tuple*, impala::SlotDescriptor const&, impala::Tuple*, void*) /tmp/be/src/exprs/agg-fn-evaluator.cc:393:29
    #2 0x2777bc8 in impala::AggFnEvaluator::Finalize(std::vector<impala::AggFnEvaluator*, std::allocator<impala::AggFnEvaluator*> > const&, impala::Tuple*, impala::Tuple*) /tmp/be/src/exprs/agg-fn-evaluator.h:307:15
    #3 0x27add96 in impala::GroupingAggregator::CleanupHashTbl(std::vector<impala::AggFnEvaluator*, std::allocator<impala::AggFnEvaluator*> > const&, impala::HashTable::Iterator) /tmp/be/src/exec/grouping-aggregator.cc:351:7
    #4 0x27ae2b2 in impala::GroupingAggregator::ClosePartitions() /tmp/be/src/exec/grouping-aggregator.cc:930:5
    #5 0x27ae5f4 in impala::GroupingAggregator::Close(impala::RuntimeState*) /tmp/be/src/exec/grouping-aggregator.cc:383:3
    #6 0x27637f7 in impala::AggregationNode::Close(impala::RuntimeState*) /tmp/be/src/exec/aggregation-node.cc:139:32
    #7 0x206b7e9 in impala::FragmentInstanceState::Close() /tmp/be/src/runtime/fragment-instance-state.cc:368:42
    #8 0x2066b1a in impala::FragmentInstanceState::Exec() /tmp/be/src/runtime/fragment-instance-state.cc:99:3
    #9 0x2080e12 in impala::QueryState::ExecFInstance(impala::FragmentInstanceState*) /tmp/be/src/runtime/query-state.cc:584:24
    #10 0x1d79036 in boost::function0<void>::operator()() const /opt/Impala-Toolchain/boost-1.57.0-p3/include/boost/function/function_template.hpp:766:14
    #11 0x24bbe06 in impala::Thread::SuperviseThread(std::string const&, std::string const&, boost::function<void ()>, impala::ThreadDebugInfo const*, impala::Promise<long, (impala::PromiseMode)0>*) /tmp/be/src/util/thread.cc:359:3
    #12 0x24c72f8 in void boost::_bi::list5<boost::_bi::value<std::string>, boost::_bi::value<std::string>, boost::_bi::value<boost::function<void ()> >, boost::_bi::value<impala::ThreadDebugInfo*>, boost::_bi::value<impala::Promise<long, (impala::PromiseMode)0>*> >::operator()<void (*)(std::string const&, std::string const&, boost::function<void ()>, impala::ThreadDebugInfo const*, impala::Promise<long, (impala::PromiseMode)0>*), boost::_bi::list0>(boost::_bi::type<void>, void (*&)(std::string const&, std::string const&, boost::function<void ()>, impala::ThreadDebugInfo const*, impala::Promise<long, (impala::PromiseMode)0>*), boost::_bi::list0&, int) /opt/Impala-Toolchain/boost-1.57.0-p3/include/boost/bind/bind.hpp:525:9
    #13 0x24c714b in boost::_bi::bind_t<void, void (*)(std::string const&, std::string const&, boost::function<void ()>, impala::ThreadDebugInfo const*, impala::Promise<long, (impala::PromiseMode)0>*), boost::_bi::list5<boost::_bi::value<std::string>, boost::_bi::value<std::string>, boost::_bi::value<boost::function<void ()> >, boost::_bi::value<impala::ThreadDebugInfo*>, boost::_bi::value<impala::Promise<long, (impala::PromiseMode)0>*> > >::operator()() /opt/Impala-Toolchain/boost-1.57.0-p3/include/boost/bind/bind_template.hpp:20:16
    #14 0x3c83949 in thread_proxy (/home/lv/i4/be/build/debug/service/impalad+0x3c83949)
    #15 0x7f768ce73183 in start_thread /build/eglibc-ripdx6/eglibc-2.19/nptl/pthread_create.c:312
    #16 0x7f768c98a03c in clone /build/eglibc-ripdx6/eglibc-2.19/misc/../sysdeps/unix/sysv/linux/x86_64/clone.S:111

The problem seems to be that we call output_partition_>aggregated_row_stream>Close() in be/src/exec/grouping-aggregator.cc:284 when hitting the limit, and then later the tuple creation in CleanupHashTbl() in be/src/exec/grouping-aggregator.cc:341 reads from poisoned memory.

A similar query does not show the crash:

select count(*) from functional_parquet.alltypes a group by a.string_col limit 2;

tarmstrong - Do you have an idea why the query on a much smaller dataset wouldn't crash?

Issue Links

• is duplicated by: IMPALA-8118 ASAN build failure: query_test/test_scanners.py
• is duplicated by: IMPALA-8141 ASAN build failure in query_test/test_mem_usage_scaling.py
• is duplicated by: IMPALA-8142 ASAN build failure in query_test/test_nested_types.py