Parquet IntDecoder fail

[jinchengchenghh]

When I test read TPCDS parquet table, I receive coredump because bitWidth is 0
sql:

select s_store_sk from store where s_state = 'TN';

substrait plan:

{"extensions":[{"extensionFunction":{"name":"is_not_null:str"}},{"extensionFunction":{"functionAnchor":1,"name":"equal:str_str"}},{"extensionFunction":{"functionAnchor":2,"name":"and:bool_bool"}}],"relations":[{"root":{"input":{"project":{"common":{"direct":{}},"input":{"read":{"common":{"direct":{}},"baseSchema":{"names":["s_store_sk","s_state"],"struct":{"types":[{"i64":{"nullability":"NULLABILITY_NULLABLE"}},{"string":{"nullability":"NULLABILITY_NULLABLE"}}]},"partitionColumns":{"columnType":["NORMAL_COL","NORMAL_COL"]}},"filter":{"scalarFunction":{"functionReference":2,"outputType":{"bool":{"nullability":"NULLABILITY_NULLABLE"}},"arguments":[{"value":{"scalarFunction":{"outputType":{"bool":{"nullability":"NULLABILITY_REQUIRED"}},"arguments":[{"value":{"selection":{"directReference":{"structField":{"field":1}}}}}]}}},{"value":{"scalarFunction":{"functionReference":1,"outputType":{"bool":{"nullability":"NULLABILITY_NULLABLE"}},"arguments":[{"value":{"selection":{"directReference":{"structField":{"field":1}}}}},{"value":{"literal":{"string":"TN"}}}]}}}]}},"localFiles":{"items":[{"uriFile":"file:///tmp/tpcds-generated/store/part-00000-4de8c37f-0e0f-4016-bd78-421a3af40edd-c000.snappy.parquet","length":"9171","parquet":{}}]}}},"expressions":[{"selection":{"directReference":{"structField":{}}}}]}},"names":["s_store_sk#3631"]}}]}

Plan Node

Plan Node:
-- Project[expressions: (n1_0:BIGINT, ROW["n0_0"])] -> n1_0:BIGINT
  -- TableScan[table: hive_table, range filters: [(s_state, BytesRange: [TN, TN] no nulls)]] -> n0_0:BIGINT, n0_1:VARCHAR

#7  0x00007fd9aed703e7 in facebook::velox::dwio::common::IntDecoder<false>::decodeBitsLE<int> (bits=0x7fdbe801a178, bitOffset=0,
    rows=..., rowBias=0, bitWidth=0 '\000',
    bufferEnd=0x7fdbe801a178 '\241' <repeats 24 times>, "\255\336\335ں\335ںP\246\001\350\333\177", result=0x7fdbe802a2c0)
    at /mnt/DP_disk1/code/gluten/tools/build/velox_ep/velox/dwio/common/IntDecoder.cpp:2602
#8  0x00007fd9ad7b56df in facebook::velox::parquet::RleDecoder<false>::processRun<true, false, true, facebook::velox::dwio::common::StringDictionaryColumnVisitor<facebook::velox::common::BytesRange, facebook::velox::dwio::common::ExtractToReader<facebook::velox::parquet::StringColumnReader>, true> > (this=0x7fdbe802a440, rows=0x7fdbe802af00, rowIndex=0, currentRow=0, numRows=2, scatterRows=0x7fdbe802a700,
    filterHits=0x7fdbe802a1c0, values=0x7fdbe802a2c0, numValues=@0x7fdc8aef1bac: 0, visitor=...)
    at /mnt/DP_disk1/code/gluten/tools/build/velox_ep/./velox/dwio/parquet/reader/RleDecoder.h:259
#9  0x00007fd9ad7a873e in facebook::velox::parquet::RleDecoder<false>::bulkScan<true, false, true, facebook::velox::dwio::common::StringDictionaryColumnVisitor<facebook::velox::common::BytesRange, facebook::velox::dwio::common::ExtractToReader<facebook::velox::parquet::StringColumnReader>, true> > (this=0x7fdbe802a440, nonNullRows=..., scatterRows=0x7fdbe802a700, visitor=...)
    at /mnt/DP_disk1/code/gluten/tools/build/velox_ep/./velox/dwio/parquet/reader/RleDecoder.h:338
#10 0x00007fd9ad79fca3 in facebook::velox::parquet::RleDecoder<false>::fastPath<true, facebook::velox::dwio::common::StringDictionaryColumnVisitor<facebook::velox::common::BytesRange, facebook::velox::dwio::common::ExtractToReader<facebook::velox::parquet::StringColumnReader>, true> > (this=0x7fdbe802a440, nulls=0x7fdbe802a5c0, visitor=...)
    at /mnt/DP_disk1/code/gluten/tools/build/velox_ep/./velox/dwio/parquet/reader/RleDecoder.h:209
#11 0x00007fd9ad77b69f in facebook::velox::parquet::RleDecoder<false>::readWithVisitor<true, facebook::velox::dwio::common::StringDictionaryColumnVisitor<facebook::velox::common::BytesRange, facebook::velox::dwio::common::ExtractToReader<facebook::velox::parquet::StringColumnReader>, true> > (this=0x7fdbe802a440, nulls=0x7fdbe802a5c0, visitor=...)
    at /mnt/DP_disk1/code/gluten/tools/build/velox_ep/./velox/dwio/parquet/reader/RleDecoder.h:84
#12 0x00007fd9ad77409c in facebook::velox::parquet::PageReader::callDecoder<facebook::velox::dwio::common::ColumnVisitor<folly::Range<char const*>, facebook::velox::common::BytesRange, facebook::velox::dwio::common::ExtractToReader<facebook::velox::parquet::StringColumnReader>, true>, 0> (this=0x7fdbe802ac70, nulls=0x7fdbe802a5c0, nullsFromFastPath=@0x7fdc8aef1f48: true, visitor=...)
    at /mnt/DP_disk1/code/gluten/tools/build/velox_ep/./velox/dwio/parquet/reader/PageReader.h:196
#13 0x00007fd9ad76bf1f in facebook::velox::parquet::PageReader::readWithVisitor<facebook::velox::dwio::common::ColumnVisitor<folly::Range<char const*>, facebook::velox::common::BytesRange, facebook::velox::dwio::common::ExtractToReader<facebook::velox::parquet::StringColumnReader>, true> > (this=0x7fdbe802ac70, visitor=...)
    at /mnt/DP_disk1/code/gluten/tools/build/velox_ep/./velox/dwio/parquet/reader/PageReader.h:325
#14 0x00007fd9ad76a47e in facebook::velox::parquet::ParquetData::readWithVisitor<facebook::velox::dwio::common::ColumnVisitor<folly::Range<char const*>, facebook::velox::common::BytesRange, facebook::velox::dwio::common::ExtractToReader<facebook::velox::parquet::StringColumnReader>, true> > (this=0x7fdbe8029bd0, visitor=...)
    at /mnt/DP_disk1/code/gluten/tools/build/velox_ep/./velox/dwio/parquet/reader/ParquetData.h:118
--Type <RET> for more, q to quit, c to continue without paging--
#15 0x00007fd9ad768861 in facebook::velox::parquet::StringColumnReader::readHelper<facebook::velox::common::BytesRange, true, facebook::velox::dwio::common::ExtractToReader<facebook::velox::parquet::StringColumnReader> > (this=0x7fdbe8029a30, filter=0x7fdbe8014c50, rows=...,
    extractValues=...) at /mnt/DP_disk1/code/gluten/tools/build/velox_ep/velox/dwio/parquet/reader/StringColumnReader.cpp:38
#16 0x00007fd9ad767fed in facebook::velox::parquet::StringColumnReader::processFilter<true, facebook::velox::dwio::common::ExtractToReader<facebook::velox::parquet::StringColumnReader> > (this=0x7fdbe8029a30, filter=0x7fdbe8014c50, rows=..., extractValues=...)
    at /mnt/DP_disk1/code/gluten/tools/build/velox_ep/velox/dwio/parquet/reader/StringColumnReader.cpp:70
#17 0x00007fd9ad767175 in facebook::velox::parquet::StringColumnReader::read (this=0x7fdbe8029a30, offset=0, rows=..., incomingNulls=0x0)
    at /mnt/DP_disk1/code/gluten/tools/build/velox_ep/velox/dwio/parquet/reader/StringColumnReader.cpp:111
#18 0x00007fd9aeda54b9 in facebook::velox::dwio::common::SelectiveStructColumnReader::read (this=0x7fdbe80147c0, offset=0, rows=...,
    incomingNulls=0x0) at /mnt/DP_disk1/code/gluten/tools/build/velox_ep/velox/dwio/common/SelectiveStructColumnReader.cpp:139
#19 0x00007fd9aeda5111 in facebook::velox::dwio::common::SelectiveStructColumnReader::next (this=0x7fdbe80147c0, numValues=2, result=
    std::shared_ptr<class facebook::velox::BaseVector> (use count 1, weak count 0) = {...}, incomingNulls=0x0)
    at /mnt/DP_disk1/code/gluten/tools/build/velox_ep/velox/dwio/common/SelectiveStructColumnReader.cpp:104
#20 0x00007fd9ad4faff1 in facebook::velox::parquet::ParquetRowReader::next (this=0x7fdbe801ed20, size=32768,

(gdb) f 8

   #8  0x00007fd9ad7b56df in facebook::velox::parquet::RleDecoder<false>::processRun<true, false, true, facebook::velox::dwio::common::StringDictionaryColumnVisitor<facebook::velox::common::BytesRange, facebook::velox::dwio::common::ExtractToReader<facebook::velox::parquet::StringColumnReader>, true> > (this=0x7fdbe802a440,
    rows=0x7fdbe802af00, rowIndex=0, currentRow=0, numRows=2, scatterRows=0x7fdbe802a700, filterHits=0x7fdbe802a1c0,
    values=0x7fdbe802a2c0, numValues=@0x7fdc8aef1bac: 0, visitor=...)
    at /mnt/DP_disk1/code/gluten/tools/build/velox_ep/./velox/dwio/parquet/reader/RleDecoder.h:259
warning: Source file is more recent than executable.
259         super::decodeBitsLE(
(gdb) p this
$1 = (facebook::velox::parquet::RleDecoder<false> * const) 0x7fdbe802a440
(gdb) p *this
$2 = {<facebook::velox::dwio::common::IntDecoder<false>> = {
    _vptr.IntDecoder = 0x7fd9b478b608 <vtable for facebook::velox::parquet::RleDecoder<false>+16>,
    static kMinDenseBatch = <error reading variable: Missing ELF symbol "facebook::velox::dwio::common::IntDecoder<false>::kMinDenseBatch".>, inputStream = std::unique_ptr<facebook::velox::dwio::common::SeekableInputStream> = {get() = 0x0},
    bufferStart = 0x7fdbe801a178 '\241' <repeats 24 times>, "\255\336\335ں\335ںP\246\001\350\333\177",
    bufferEnd = 0x7fdbe801a178 '\241' <repeats 24 times>, "\255\336\335ں\335ںP\246\001\350\333\177", useVInts = false, numBytes = 0},
  bitWidth_ = 0 '\000', byteWidth_ = 0 '\000', bitMask_ = 0, lastSafeWord_ = 0x7fdbe801a170 "\002", remainingValues_ = 8, value_ = 14,
  bitOffset_ = 0 '\000', repeating_ = false}

relevant code is

    if (redZone > 0) {
      anyUnsafe = true;
      auto numRed = (redZone + 1) * 8 / bitWidth;
      int32_t lastSafeIndex = rows.back() - numRed;
      --numSafeRows;
      for (; numSafeRows >= 1; --numSafeRows) {
        if (rows[numSafeRows - 1] < lastSafeIndex) {
          break;
        }
      }
    }

(gdb) f 12
#12 0x00007fd9ad77409c in facebook::velox::parquet::PageReader::callDecoder<facebook::velox::dwio::common::ColumnVisitor<folly::Range<char const*>, facebook::velox::common::BytesRange, facebook::velox::dwio::common::ExtractToReader<facebook::velox::parquet::StringColumnReader>, true>, 0> (this=0x7fdbe802ac70, nulls=0x7fdbe802a5c0, nullsFromFastPath=@0x7fdc8aef1f48: true, visitor=...)
    at /mnt/DP_disk1/code/gluten/tools/build/velox_ep/./velox/dwio/parquet/reader/PageReader.h:196
196             rleDecoder_->readWithVisitor<true>(nulls, dictVisitor);
(gdb) p *this
$5 = {pool_ = @0x7fdbe801adb0, inputStream_ = std::unique_ptr<facebook::velox::dwio::common::SeekableInputStream> = {
    get() = 0x7fdbe802bfb0}, type_ = std::shared_ptr<const facebook::velox::parquet::ParquetTypeWithId> (use count 4, weak count 0) = {
    get() = 0x7fdbe8023d90}, maxRepeat_ = 0, maxDefine_ = 1, codec_ = facebook::velox::parquet::thrift::CompressionCodec::SNAPPY,
  chunkSize_ = 60, bufferStart_ = 0x7fdbe802b991 "", bufferEnd_ = 0x7fdbe802b991 "", tempNulls_ = {px = 0x0}, nullsInReadRange_ = {
    px = 0x0}, multiPageNulls_ = {px = 0x0}, repeatDecoder_ = std::unique_ptr<facebook::velox::parquet::RleDecoder<false>> = {
    get() = 0x0}, defineDecoder_ = std::unique_ptr<facebook::velox::parquet::RleDecoder<false>> = {get() = 0x7fdbe802a3e0},
  encoding_ = facebook::velox::parquet::thrift::Encoding::PLAIN_DICTIONARY, rowOfPage_ = 0, numRowsInPage_ = 2, pageBuffer_ = {px = 0x0},
  uncompressedData_ = {px = 0x7fdbe801a130}, pageData_ = 0x7fdbe801a176 "", dictionary_ = {values = {px = 0x7fdbe8017220}, strings = {
      px = 0x7fdbe802a350}, numValues = 1, sorted = false},
  dictionaryEncoding_ = facebook::velox::parquet::thrift::Encoding::PLAIN_DICTIONARY, pageStart_ = 60, pageDataStart_ = 50,
  encodedDataSize_ = 2, visitorRows_ = 0x7fdbe802af00, numVisitorRows_ = 2, initialRowOfPage_ = 0, currentVisitorRow_ = 2,
  firstUnvisited_ = 2, visitBase_ = 0, rowsCopy_ = 0x7fdbe8029b70, rowNumberBias_ = 0, nullConcatenation_ = {pool_ = @0x7fdbe801adb0,
    buffer_ = 0x0, numBits_ = 0, hasZeros_ = false}, dictionaryValues_ = std::shared_ptr<facebook::velox::BaseVector> (empty) = {
    get() = 0x0}, directDecoder_ = std::unique_ptr<facebook::velox::dwio::common::DirectDecoder<true>> = {get() = 0x0},
  rleDecoder_ = std::unique_ptr<facebook::velox::parquet::RleDecoder<false>> = {get() = 0x7fdbe802a440},
  stringDecoder_ = std::unique_ptr<facebook::velox::parquet::StringDecoder> = {get() = 0x0}}
(gdb) p this->pageData_
$6 = 0x7fdbe801a176 ""
(gdb) p *this->pageData_
$7 = 0 '\000'

[mbasmanova]

@oerling Orri, would you check it out?

CC: @yingsu00 @majetideepak

[mbasmanova]

@jinchengchenghh Thank you for reporting a crash in the Parquet reader. Next time, please, use 3 backticks to format blocks of code for readability. Looks like you have done some debugging. Would you like to send a PR with the fix?

[mbasmanova]

According to @oerling , this might be fixed in #2807

[jinchengchenghh]

I verified locally, #2807 cannot fix this issue, I will try to reproduce by unit test and continue debugging