[
https://issues.apache.org/jira/browse/IMPALA-7903?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16705754#comment-16705754
]
Gabor Kaszab edited comment on IMPALA-7903 at 12/1/18 9:57 AM:
---------------------------------------------------------------
The way to reproduce this issue is to set log level to 3 and then execute an
aggregating query like this one:
{code:java}
select avg(bigint_col) from functional.alltypes;
{code}
This query above generates 2 query fragments: One will have a slot as
TYPE_FIXED_UDA_INTERMEDIATE and the other one will have a TYPE_DOUBLE slot.
This is common for both log_level=0 and log_level=3. So putting a breakpoint in
FragmentInstanceState::ExecInternal() will stop twice, once for each fragment.
+When log_level=0+
A KrpcDataStreamSender is created for the fragment with
TYPE_FIXED_UDA_INTERMEDIATE and a PlanRootSink is created for the one with
TYPE_DOUBLE. PlanRootSink::Send() is going to be the one that eventually
reaches RawValue::PrintValue() and the fragment with
TYPE_FIXED_UDA_INTERMEDIATE won't reach this function.
+When log_level=3+
This path is different because in FragmentInstanceState::ExecInternal() we
won't reach the part that calls sink_k->Send() but 2 lines before we go to the
route to call RawBatch::VLogRows().
https://github.com/apache/impala/blob/96f976534837dde6d3fb87ea02019b67d9c3ea78/be/src/runtime/fragment-instance-state.cc#L328
{code:java}
if (VLOG_ROW_IS_ON)
row_batch_->VLogRows("FragmentInstanceState::ExecInternal()");
COUNTER_ADD(rows_produced_counter_, row_batch_->num_rows());
RETURN_IF_ERROR(sink_->Send(runtime_state_, row_batch_.get()));
{code}
On this path both of the fragments will end up eventually in
RawValue::PrintValue() even if this function is not ready for taking
TYPE_FIXED_UDA_INTERMEDIATE slots.
+Solution+
As a solution I haven't decided whether to go and extend RawValue::PrintValue()
to handle TYPE_FIXED_UDA_INTERMEDIATE slots as well or to introduce a check
somewhere on the call chain to prevent PrintValue() to run on that type.
Update: I went for the second option.
was (Author: gaborkaszab):
The way to reproduce this issue is to set log level to 3 and then execute an
aggregating query like this one:
{code:java}
select avg(bigint_col) from functional.alltypes;
{code}
This query above generates 2 query fragments: One will have a slot as
TYPE_FIXED_UDA_INTERMEDIATE and the other one will have a TYPE_DOUBLE slot.
This is common for both log_level=0 and log_level=3. So putting a breakpoint in
FragmentInstanceState::ExecInternal() will stop twice, once for each fragment.
+When log_level=0+
A KrpcDataStreamSender is created for the fragment with
TYPE_FIXED_UDA_INTERMEDIATE and a PlanRootSink is created for the one with
TYPE_DOUBLE. PlanRootSink::Send() is going to be the one that eventually
reaches RawValue::PrintValue() and the fragment with
TYPE_FIXED_UDA_INTERMEDIATE won't reach this function.
+When log_level=3+
This path is different because in FragmentInstanceState::ExecInternal() we
won't reach the part that calls sink_k->Send() but 2 lines before we go to the
route to call RawBatch::VLogRows().
https://github.com/apache/impala/blob/96f976534837dde6d3fb87ea02019b67d9c3ea78/be/src/runtime/fragment-instance-state.cc#L328
{code:java}
if (VLOG_ROW_IS_ON)
row_batch_->VLogRows("FragmentInstanceState::ExecInternal()");
COUNTER_ADD(rows_produced_counter_, row_batch_->num_rows());
RETURN_IF_ERROR(sink_->Send(runtime_state_, row_batch_.get()));
{code}
On this path both of the fragments will end up eventually in
RawValue::PrintValue() even if this function is not ready for taking
TYPE_FIXED_UDA_INTERMEDIATE slots.
Solution
As a solution I haven't decided whether to go and extend RawValue::PrintValue()
to handle TYPE_FIXED_UDA_INTERMEDIATE slots as well or to introduce a check
somewhere on the call chain to prevent PrintValue() to run on that type.
Update: I went for the second option.
> DCHECK in RawValue::PrintValue() hit with VLOG level 3
> ------------------------------------------------------
>
> Key: IMPALA-7903
> URL: https://issues.apache.org/jira/browse/IMPALA-7903
> Project: IMPALA
> Issue Type: Improvement
> Components: Backend
> Reporter: Tim Armstrong
> Assignee: Gabor Kaszab
> Priority: Critical
> Labels: broken-build, crash
>
> We hit this DCHECK here on an exhaustive test run:
> {noformat}
> #0 0x00007fda93d441f7 in raise () from /lib64/libc.so.6
> #1 0x00007fda93d458e8 in abort () from /lib64/libc.so.6
> #2 0x00000000046679f4 in google::DumpStackTraceAndExit() ()
> #3 0x000000000465e44d in google::LogMessage::Fail() ()
> #4 0x000000000465fcf2 in google::LogMessage::SendToLog() ()
> #5 0x000000000465de27 in google::LogMessage::Flush() ()
> #6 0x00000000046613ee in google::LogMessageFatal::~LogMessageFatal() ()
> #7 0x0000000001f2a9b5 in impala::RawValue::PrintValue (value=0x3a21f93e,
> type=..., scale=-1, stream=0x7fd9e92a9bd0) at
> /data/jenkins/workspace/impala-asf-master-exhaustive/repos/Impala/be/src/runtime/raw-value.cc:279
> #8 0x0000000001f29b42 in impala::RawValue::PrintValue (value=0x3a21f93e,
> type=..., scale=-1, str=0x7fd9e92a9f30) at
> /data/jenkins/workspace/impala-asf-master-exhaustive/repos/Impala/be/src/runtime/raw-value.cc:111
> #9 0x0000000002121f9e in impala::PrintTuple (t=0x3a21f93e, d=...) at
> /data/jenkins/workspace/impala-asf-master-exhaustive/repos/Impala/be/src/util/debug-util.cc:168
> #10 0x000000000212217c in impala::PrintRow (row=0xf1a6000, d=...) at
> /data/jenkins/workspace/impala-asf-master-exhaustive/repos/Impala/be/src/util/debug-util.cc:181
> #11 0x0000000001ec0829 in impala::RowBatch::VLogRows (this=0x5cda7740,
> context=...) at
> /data/jenkins/workspace/impala-asf-master-exhaustive/repos/Impala/be/src/runtime/row-batch.cc:579
> #12 0x0000000001f11d82 in impala::FragmentInstanceState::ExecInternal
> (this=0x1c1b8000) at
> /data/jenkins/workspace/impala-asf-master-exhaustive/repos/Impala/be/src/runtime/fragment-instance-state.cc:328
> #13 0x0000000001f0ea39 in impala::FragmentInstanceState::Exec
> (this=0x1c1b8000) at
> /data/jenkins/workspace/impala-asf-master-exhaustive/repos/Impala/be/src/runtime/fragment-instance-state.cc:93
> #14 0x0000000001f1f01d in impala::QueryState::ExecFInstance (this=0x3f565000,
> fis=0x1c1b8000) at
> /data/jenkins/workspace/impala-asf-master-exhaustive/repos/Impala/be/src/runtime/query-state.cc:571
> #15 0x0000000001f1d326 in impala::QueryState::<lambda()>::operator()(void)
> const (__closure=0x7fd9e92aaba8) at
> /data/jenkins/workspace/impala-asf-master-exhaustive/repos/Impala/be/src/runtime/query-state.cc:480
> #16 0x0000000001f1fdc5 in
> boost::detail::function::void_function_obj_invoker0<impala::QueryState::StartFInstances()::<lambda()>,
> void>::invoke(boost::detail::function::function_buffer &)
> (function_obj_ptr=...) at
> /data/jenkins/workspace/impala-asf-master-exhaustive/Impala-Toolchain/boost-1.57.0-p3/include/boost/function/function_template.hpp:153
> #17 0x0000000001d38206 in boost::function0<void>::operator()
> (this=0x7fd9e92aaba0) at
> /data/jenkins/workspace/impala-asf-master-exhaustive/Impala-Toolchain/boost-1.57.0-p3/include/boost/function/function_template.hpp:767
> #18 0x00000000021baa5f in impala::Thread::SuperviseThread(std::string const&,
> std::string const&, boost::function<void ()>, impala::ThreadDebugInfo const*,
> impala::Promise<long, (impala::PromiseMode)0>*) (name=..., category=...,
> functor=..., parent_thread_info=0x7fd9d4d87850,
> thread_started=0x7fd9d4d867f0) at
> /data/jenkins/workspace/impala-asf-master-exhaustive/repos/Impala/be/src/util/thread.cc:359
> #19 0x00000000021c2d7f in 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)
> (this=0x119ff8fc0, f=@0x119ff8fb8: 0x21ba6f8
> <impala::Thread::SuperviseThread(std::string const&, std::string const&,
> boost::function<void ()>, impala::ThreadDebugInfo const*,
> impala::Promise<long, (impala::PromiseMode)0>*)>, a=...) at
> /data/jenkins/workspace/impala-asf-master-exhaustive/Impala-Toolchain/boost-1.57.0-p3/include/boost/bind/bind.hpp:525
> #20 0x00000000021c2ca3 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()() (this=0x119ff8fb8) at
> /data/jenkins/workspace/impala-asf-master-exhaustive/Impala-Toolchain/boost-1.57.0-p3/include/boost/bind/bind_template.hpp:20
> #21 0x00000000021c2c66 in boost::detail::thread_data<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>*> > >
> >::run() (this=0x119ff8e00) at
> /data/jenkins/workspace/impala-asf-master-exhaustive/Impala-Toolchain/boost-1.57.0-p3/include/boost/thread/detail/thread.hpp:116
> #22 0x0000000003589b7a in thread_proxy ()
> #23 0x00007fda940d9e25 in start_thread () from /lib64/libpthread.so.0
> #24 0x00007fda93e0734d in clone () from /lib64/libc.so.6
> {noformat}
> I think this started happening because test_webpages toggles the log level
> and was enable recently IMPALA-7691. Clearly we shouldn't be hitting this
> DCHECK anyway.
> It looks like this is a FIXED_UDA_INTERMEDIATE, so I think the batches being
> logged were being shuffled as part of the exchange for an aggregate:
> {noformat}
> (gdb) frame 7
> #7 0x0000000001f2a9b5 in impala::RawValue::PrintValue (value=0x3a21f93e,
> type=..., scale=-1,
> stream=0x7fd9e92a9bd0)
> at
> /data/jenkins/workspace/impala-asf-master-exhaustive/repos/Impala/be/src/runtime/raw-value.cc:279
> 279 in
> /data/jenkins/workspace/impala-asf-master-exhaustive/repos/Impala/be/src/runtime/raw-value.cc
> (gdb) info args
> value = 0x3a21f93e
> type = @0xee87688: {
> type = impala::TYPE_FIXED_UDA_INTERMEDIATE,
> len = 24,
> static MAX_VARCHAR_LENGTH = 65535,
> static MAX_CHAR_LENGTH = 255,
> precision = -1,
> scale = -1,
> static MAX_PRECISION = 38,
> static MAX_SCALE = 38,
> static MIN_ADJUSTED_SCALE = 6,
> static MAX_DECIMAL4_PRECISION = 9,
> static MAX_DECIMAL8_PRECISION = 18,
> children = {
> <std::_Vector_base<impala::ColumnType, std::allocator<impala::ColumnType>
> >> = {
> _M_impl = {
> <std::allocator<impala::ColumnType>> = {
> <__gnu_cxx::new_allocator<impala::ColumnType>> = {<No data
> fields>}, <No data fields>},
> members of std::_Vector_base<impala::ColumnType,
> std::allocator<impala::ColumnType> >::_Vector_impl:
> _M_start = 0x0,
> _M_finish = 0x0,
> _M_end_of_storage = 0x0
> }
> }, <No data fields>},
> field_names = {
> <std::_Vector_base<std::basic_string<char, std::char_traits<char>,
> std::allocator<char> >, std::allocator<std::basic_string<char,
> std::char_traits<char>, std::allocator<char> > > >> = {
> _M_impl = {
> <std::allocator<std::basic_string<char, std::char_traits<char>,
> std::allocator<char> > >> = {
> <__gnu_cxx::new_allocator<std::basic_string<char,
> std::char_traits<char>, std::allocator<char> > >> = {<No data fields>}, <No
> data fields>},
> members of std::_Vector_base<std::basic_string<char,
> std::char_traits<char>, std::allocator<char> >,
> std::allocator<std::basic_string<char, std::char_traits<char>,
> std::allocator<char> > > >::_Vector_impl:
> _M_start = 0x0,
> _M_finish = 0x0,
> _M_end_of_storage = 0x0
> }
> }, <No data fields>},
> static LLVM_CLASS_NAME = 0x4a57fb0 "struct.impala::ColumnType"
> }
> scale = -1
> {noformat}
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
