Mismatch column count in ScanActor (expected) vs CS (received) and out of range vector access in TPC-DS Q9

[iddqdex]

Воспроизведение:

ulimit -c unlimited
cd ~/g/ydb/tests/functional/tpc/large
ya make -tttF test_tpcds.py::TestTpcdsS1::test_tpcds[?]
ls test-results/py3test/test_tpcds/testing_out_stuff/test_tpcds.py.TestTpcdsS1.test_tpcds.1/cluster/slot_1/

стек:

warning: Unable to find libthread_db matching inferior's thread library, thread debugging will not be available.
Core was generated by `/extra_disk_1/.ya/build/build_root/k20t/000007/ydb/apps/ydbd/ydbd server --supp'.
Program terminated with signal SIGILL, Illegal instruction.
#0  std::__y1::vector<NYql::NUdf::TUnboxedValue, NKikimr::NMiniKQL::TMKQLAllocator<NYql::NUdf::TUnboxedValue, (NKikimr::NMiniKQL::EMemorySubPool)0>>::operator[][abi:fe190000](unsigned long) (this=<optimized out>, __n=<optimized out>) at /-S/contrib/libs/cxxsupp/libcxx/include/vector:1448

warning: 1448   /-S/contrib/libs/cxxsupp/libcxx/include/vector: No such file or directory
[Current thread is 1 (LWP 1566338)]
[arc] Arcadia GDB pretty-printers enabled
(gdb) bt
#0  std::__y1::vector<NYql::NUdf::TUnboxedValue, NKikimr::NMiniKQL::TMKQLAllocator<NYql::NUdf::TUnboxedValue, (NKikimr::NMiniKQL::EMemorySubPool)0>>::operator[][abi:fe190000](unsigned long) (this=<optimized out>, __n=<optimized out>) at /home/iddqd/git/ydb/contrib/libs/cxxsupp/libcxx/include/vector:1448
#1  0x000000001559f026 in NKikimr::NMiniKQL::TKqpScanComputeContext::TScanData::TBlockBatchReader::AddData (this=0x6c319b15700, dataAccessor=..., holderFactory=...)
#2  0x000000001559f5ff in NKikimr::NMiniKQL::TKqpScanComputeContext::TScanData::AddData (this=0x6c37940b488, batch=...,
    shardId=Printer initialization error: <class 'AttributeError'>, 'NoneType' object has no attribute 'dereference'
..., holderFactory=...) at /home/iddqd/git/ydb/ydb/core/kqp/runtime/kqp_scan_data.cpp:722
#3  0x000000001960eb64 in NKikimr::NKqp::NScanPrivate::TKqpScanComputeActor::Handle (this=this@entry=0x6c355fe8e80, ev=Python Exception <class 'gdb.error'>: There is no member named T_.
)
    at /home/iddqd/git/ydb/ydb/core/kqp/compute_actor/kqp_scan_compute_actor.cpp:182
#4  0x0000000019611e48 in NKikimr::NKqp::NScanPrivate::TKqpScanComputeActor::StateFunc (this=0x6c355fe8e80, ev=TAutoPtr<NActors::IEventHandle> = {...})
    at /home/iddqd/git/ydb/ydb/core/kqp/compute_actor/kqp_scan_compute_actor.h:79
#5  0x000000000aa6ce35 in NActors::IActor::Receive (this=0x6c355fe8e80, ev=TAutoPtr<NActors::IEventHandle> = {...})
    at /home/iddqd/git/ydb/ydb/library/actors/core/actor.cpp:280
#6  0x000000000aaa4711 in NActors::TExecutorThread::Execute (this=this@entry=0x6c33f40fa00, mailbox=mailbox@entry=0x6c31b1a6900, isTailExecution=<optimized out>)
    at /home/iddqd/git/ydb/ydb/library/actors/core/executor_thread.cpp:269
#7  0x000000000aaa8212 in NActors::TExecutorThread::ProcessExecutorPool()::$_0::operator()(NActors::TMailbox*, bool) const (this=this@entry=0x7f6fd9e22ea0,
    mailbox=mailbox@entry=0x6c31b1a6900, isTailExecution=false) at /home/iddqd/git/ydb/ydb/library/actors/core/executor_thread.cpp:460
#8  0x000000000aaa7e61 in NActors::TExecutorThread::ProcessExecutorPool (this=this@entry=0x6c33f40fa00) at /home/iddqd/git/ydb/ydb/library/actors/core/executor_thread.cpp:512
#9  0x000000000aaa89db in NActors::TExecutorThread::ThreadProc (this=0x6c33f40fa00) at /home/iddqd/git/ydb/ydb/library/actors/core/executor_thread.cpp:538
#10 0x0000000009d400ad in (anonymous namespace)::TPosixThread::ThreadProxy (arg=0x6c33d709480) at /home/iddqd/git/ydb/util/system/thread.cpp:244
#11 0x00007f6fecb90609 in start_thread () from /lib/x86_64-linux-gnu/libpthread.so.0
#12 0x00007f6fecab5353 in clone () from /lib/x86_64-linux-gnu/libc.so.6

#13580

[Hor911]

Single Q9 run also reproduces the problem

ya make -tttF test_tpcds.py::TestTpcdsS1::test_tpcds[9]

[maximyurchuk]

А почему на перф кластерах у нас такое не воспроизводится? Может кто-нибудь знает (@Hor911 ?)

[Hor911]

1. Там настоящий проезд в коде
2. Он остается незамеченным в build=release и почему-то стреляет только в debug
3. Я сейчас хочу вставить Y_ENSURE и посмотреть на кластере, сколько запросов попадает

[Hor911]

Problem is here https://github.com/ydb-platform/ydb/blob/main/ydb/core/kqp/runtime/kqp_scan_data.cpp#L691

CS send TEvSendData with empty DataIndexes field and 3 internal columns

_yql_plan_step
_yql_tx_id
_yql_write_id

TKqpScanComputeActor expects no columns and results out-of-range access batchValues vector