Method GetNext

be/src/exec/tuple-file-reader.cc:64–191

}

Status TupleFileReader::GetNext(RuntimeState *state,
    BufferPool::ClientHandle* bpclient, RowBatch* output_row_batch, bool* eos) {
  if (offset_ == file_size_) {
    *eos = true;
    return Status::OK();
  }
  // Only the first rowbatch starts at the zero offset of the file.
  // We use this for injecting errors for testing purposes.
  bool first_rowbatch = offset_ == 0;
  *eos = false;
  SCOPED_TIMER(read_timer_);
  // Each block starts with the sizes of the variable chunks of data:
  // 1. The header size
  // 2. The tuple data size
  // 3. The tuple offsets size
  size_t header_len;
  size_t tuple_data_len;
  size_t tuple_offsets_len;

  vector<kudu::Slice> chunk_lens_slices = {
      kudu::Slice(reinterpret_cast<const char*>(&header_len), sizeof(header_len)),
      kudu::Slice(reinterpret_cast<const char*>(&tuple_data_len),
          sizeof(tuple_data_len)),
      kudu::Slice(reinterpret_cast<const char*>(&tuple_offsets_len),
          sizeof(tuple_offsets_len))};

  KUDU_RETURN_IF_ERROR(reader_->ReadV(offset_,
      kudu::ArrayView<kudu::Slice>(chunk_lens_slices)),
      "Failed to read cache file");
  size_t chunk_lens_slices_size = sizeof(header_len) + sizeof(tuple_data_len) +
      sizeof(tuple_offsets_len);
  COUNTER_ADD(bytes_read_, chunk_lens_slices_size);

  // tuple_data_len can be zero, see IMPALA-13411.
  if (header_len == 0 || tuple_offsets_len == 0) {
    string err_msg = Substitute("Invalid data lengths at offset $0 in $1: "
        "header_len=$2, tuple_data_len=$3, tuple_offsets_len=$4", offset_, path_,
        header_len, tuple_data_len, tuple_offsets_len);
    DCHECK(false) << err_msg;
    return Status(Substitute("Invalid tuple cache file: $0", err_msg));
  }
  offset_ += chunk_lens_slices_size;

  // Now, we know the total size of the variable-length data, and we can read
  // it in a single chunk.
  size_t varlen_size = header_len + tuple_data_len + tuple_offsets_len;

  // Sanity check: The varlen_size shouldn't be larger than the rest of the file.
  // This protects us from doing very large memory allocations if any of the lengths
  // are bogus.
  if (offset_ + varlen_size > file_size_) {
    string err_msg = Substitute("Invalid data lengths at offset $0 in $1 exceed "
        "file size $2: header_len=$3, tuple_data_len=$4, tuple_offsets_len=$5",
        offset_, path_, file_size_, header_len, tuple_data_len, tuple_offsets_len);
    DCHECK(false) << err_msg;
    return Status(Substitute("Invalid tuple cache file: $0", err_msg));
  }
  std::unique_ptr<char []> varlen_data(new char[varlen_size]);