1
0
mirror of https://github.com/mariadb-corporation/mariadb-columnstore-engine.git synced 2025-07-02 17:22:27 +03:00

MCOL-1829 Subquery with limited order by could potentially return onordered set.

There were two code mistakes: Eq::operator() always returned true for
	any pair and Hasher::operator() always returned 0 as a key.
This commit is contained in:
Roman Nozdrin
2018-12-25 12:50:18 +03:00
parent 0584b114da
commit 35a17a87c4
4 changed files with 16 additions and 17 deletions

View File

@ -113,7 +113,6 @@ void LimitedOrderBy::processRow(const rowgroup::Row& row)
if (fOrderByQueue.size() < fStart + fCount) if (fOrderByQueue.size() < fStart + fCount)
{ {
copyRow(row, &fRow0); copyRow(row, &fRow0);
//memcpy(fRow0.getData(), row.getData(), row.getSize());
OrderByRow newRow(fRow0, fRule); OrderByRow newRow(fRow0, fRule);
fOrderByQueue.push(newRow); fOrderByQueue.push(newRow);
@ -121,8 +120,6 @@ void LimitedOrderBy::processRow(const rowgroup::Row& row)
if (fDistinct) if (fDistinct)
fDistinctMap->insert(fRow0.getPointer()); fDistinctMap->insert(fRow0.getPointer());
//fDistinctMap->insert(make_pair((fRow0.getData()+2), fRow0.getData()));
fRowGroup.incRowCount(); fRowGroup.incRowCount();
fRow0.nextRow(); fRow0.nextRow();
@ -150,23 +147,16 @@ void LimitedOrderBy::processRow(const rowgroup::Row& row)
{ {
OrderByRow swapRow = fOrderByQueue.top(); OrderByRow swapRow = fOrderByQueue.top();
row1.setData(swapRow.fData); row1.setData(swapRow.fData);
fOrderByQueue.pop();
if (!fDistinct) copyRow(row, &row1);
if (fDistinct)
{ {
copyRow(row, &row1); fDistinctMap->erase(fOrderByQueue.top().fData);
//memcpy(swapRow.fData, row.getData(), row.getSize());
}
else
{
fDistinctMap->erase(row.getPointer());
copyRow(row, &row1);
fDistinctMap->insert(row1.getPointer()); fDistinctMap->insert(row1.getPointer());
//fDistinctMap->erase(fDistinctMap->find(row.getData() + 2));
//memcpy(swapRow.fData, row.getData(), row.getSize());
//fDistinctMap->insert(make_pair((swapRow.fData+2), swapRow.fData));
} }
fOrderByQueue.pop();
fOrderByQueue.push(swapRow); fOrderByQueue.push(swapRow);
} }
} }
@ -228,6 +218,7 @@ void LimitedOrderBy::finalize()
const OrderByRow& topRow = fOrderByQueue.top(); const OrderByRow& topRow = fOrderByQueue.top();
row1.setData(topRow.fData); row1.setData(topRow.fData);
copyRow(row1, &fRow0); copyRow(row1, &fRow0);
//cerr << "LimitedOrderBy::finalize fRow0 " << fRow0.toString() << endl;
fRowGroup.incRowCount(); fRowGroup.incRowCount();
offset--; offset--;
fRow0.prevRow(rSize); fRow0.prevRow(rSize);

View File

@ -5554,6 +5554,8 @@ void TupleAggregateStep::threadedAggregateRowGroups(uint32_t threadID)
rowIn.copyField(distRow[j], k, multiDist->subAggregators()[j]->getGroupByCols()[k].get()->fInputColumnIndex); rowIn.copyField(distRow[j], k, multiDist->subAggregators()[j]->getGroupByCols()[k].get()->fInputColumnIndex);
} }
// TBD This approach could potentiall
// put all values in on bucket.
bucketID = distRow[j].hash(hashLens[j] - 1) % fNumOfBuckets; bucketID = distRow[j].hash(hashLens[j] - 1) % fNumOfBuckets;
rowBucketVecs[bucketID][j].push_back(rowIn.getPointer()); rowBucketVecs[bucketID][j].push_back(rowIn.getPointer());
rowIn.nextRow(); rowIn.nextRow();
@ -5572,6 +5574,8 @@ void TupleAggregateStep::threadedAggregateRowGroups(uint32_t threadID)
for (uint64_t i = 0; i < fRowGroupIns[threadID].getRowCount(); ++i) for (uint64_t i = 0; i < fRowGroupIns[threadID].getRowCount(); ++i)
{ {
// The key is the groupby columns, which are the leading columns. // The key is the groupby columns, which are the leading columns.
// TBD This approach could potentiall
// put all values in on bucket.
int bucketID = rowIn.hash(hashLens[0] - 1) % fNumOfBuckets; int bucketID = rowIn.hash(hashLens[0] - 1) % fNumOfBuckets;
rowBucketVecs[bucketID][0].push_back(rowIn.getPointer()); rowBucketVecs[bucketID][0].push_back(rowIn.getPointer());
rowIn.nextRow(); rowIn.nextRow();

View File

@ -1183,6 +1183,7 @@ inline bool Row::equals(const Row& r2, const std::vector<uint32_t>& keyCols) con
inline bool Row::equals(const Row& r2, uint32_t lastCol) const inline bool Row::equals(const Row& r2, uint32_t lastCol) const
{ {
// This check fires with empty r2 only.
if (lastCol >= columnCount) if (lastCol >= columnCount)
return true; return true;

View File

@ -461,7 +461,8 @@ uint64_t IdbOrderBy::Hasher::operator()(const Row::Pointer& p) const
{ {
Row& row = ts->row1; Row& row = ts->row1;
row.setPointer(p); row.setPointer(p);
uint64_t ret = row.hash(colCount); // MCOL-1829 Row::h uses colcount as an array idx down a callstack.
uint64_t ret = row.hash();//(colCount - 1);
//cout << "hash(): returning " << ret << " for row: " << row.toString() << endl; //cout << "hash(): returning " << ret << " for row: " << row.toString() << endl;
return ret; return ret;
} }
@ -471,7 +472,9 @@ bool IdbOrderBy::Eq::operator()(const Row::Pointer& d1, const Row::Pointer& d2)
Row& r1 = ts->row1, &r2 = ts->row2; Row& r1 = ts->row1, &r2 = ts->row2;
r1.setPointer(d1); r1.setPointer(d1);
r2.setPointer(d2); r2.setPointer(d2);
bool ret = r1.equals(r2, colCount); // MCOL-1829 Row::equals uses 2nd argument as container size boundary
// so it must be column count - 1.
bool ret = r1.equals(r2, colCount - 1);
//cout << "equals(): returning " << (int) ret << " for r1: " << r1.toString() << " r2: " << r2.toString() //cout << "equals(): returning " << (int) ret << " for r1: " << r1.toString() << " r2: " << r2.toString()
// << endl; // << endl;