Performance issue on HashProbe::getBuildSideOutput

[SeaRise]

HashProbe build side output handles only single concurrency

velox/velox/exec/HashProbe.cpp

Lines 908 to 914 in b4ea92c

	if (!hasMoreInput()) {
	if (needLastProbe() && lastProber_) {
	auto output = getBuildSideOutput();
	if (output != nullptr) {
	return output;
	}
	}

Is it possible for multiple operators to call listRows on different tables for parallel execution?

velox/velox/exec/HashTable.cpp

Lines 1865 to 1892 in b4ea92c

	template <bool ignoreNullKeys>
	template <RowContainer::ProbeType probeType>
	int32_t HashTable<ignoreNullKeys>::listRows(
	RowsIterator* iter,
	int32_t maxRows,
	uint64_t maxBytes,
	char** rows) {
	if (iter->hashTableIndex_ == -1) {
	const auto numRows = rows_->listRows<probeType>(
	&iter->rowContainerIterator_, maxRows, maxBytes, rows);
	if (numRows > 0) {
	return numRows;
	}
	iter->hashTableIndex_ = 0;
	iter->rowContainerIterator_.reset();
	}
	while (iter->hashTableIndex_ < otherTables_.size()) {
	const auto numRows =
	otherTables_[iter->hashTableIndex_]
	->rows()
	->template listRows<probeType>(
	&iter->rowContainerIterator_, maxRows, maxBytes, rows);
	if (numRows > 0) {
	return numRows;
	}
	++iter->hashTableIndex_;
	iter->rowContainerIterator_.reset();
	}

[kevincmchen]

@rui-mo please take a look