database/mariadb-columnstore-engine
1
0
Fork 0
mirror of https://github.com/mariadb-corporation/mariadb-columnstore-engine.git synced 2025-08-05 16:15:50 +03:00
Code Activity

feat(optimizer,rules): use EI statistics for range filters

Browse Source
This commit is contained in:
drrtuy
2025-07-16 22:18:09 +00:00
parent 3f9ce7779e
commit 67295b4320
2 changed files with 30 additions and 12 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
dbcon/mysql/ha_mcs_execplan.cpp
View File

@@ -6321,7 +6321,7 @@ void extractColumnStatistics(Item_field* ifp, gp_walk_info& gwi)
if (histogram) if (histogram)
{ {
std::cout << "Type of histogram object: " << typeid(*histogram).name() << std::endl; std::cout << "Type of histogram object: " << typeid(*histogram).name() << std::endl;
std::vector<Histogram_bucket> histogramBuckets = histogram->get_histogram(); std::vector<Histogram_bucket> histogramBuckets = histogram->get_json_histogram();
std::cout << "gwi.columnStatisticsMap[ifp->field->field_name.str].size() " << histogramBuckets.size() << std::endl; std::cout << "gwi.columnStatisticsMap[ifp->field->field_name.str].size() " << histogramBuckets.size() << std::endl;
gwi.columnStatisticsMap[ifp->field->field_name.str] = histogramBuckets; gwi.columnStatisticsMap[ifp->field->field_name.str] = histogramBuckets;
} }

40
dbcon/mysql/rulebased_optimizer.cpp
View File

@@ -202,10 +202,11 @@ execplan::CalpontSelectExecutionPlan::SelectList makeUnionFromTable(
execplan::CalpontSelectExecutionPlan::SelectList unionVec; execplan::CalpontSelectExecutionPlan::SelectList unionVec;
// unionVec.reserve(numberOfLegs); // unionVec.reserve(numberOfLegs);
execplan::SimpleColumn* keyColumn = findSuitableKeyColumn(csep); execplan::SimpleColumn* keyColumn = findSuitableKeyColumn(csep);
std::cout << "looking for " << keyColumn->columnName() << " in ctx.gwi.columnStatisticsMap " << " with size " << ctx.gwi.columnStatisticsMap.size() << std::endl; std::cout << "looking for " << keyColumn->columnName() << " in ctx.gwi.columnStatisticsMap "
<< " with size " << ctx.gwi.columnStatisticsMap.size() << std::endl;
for (auto& [k, v] : ctx.gwi.columnStatisticsMap) for (auto& [k, v] : ctx.gwi.columnStatisticsMap)
{ {
std::cout << "key " << k << std::endl; std::cout << "key " << k << " vector size " << v.size() << std::endl;
} }
if (!keyColumn || if (!keyColumn ||
ctx.gwi.columnStatisticsMap.find(keyColumn->columnName()) == ctx.gwi.columnStatisticsMap.end()) ctx.gwi.columnStatisticsMap.find(keyColumn->columnName()) == ctx.gwi.columnStatisticsMap.end())
@@ -216,16 +217,33 @@ execplan::CalpontSelectExecutionPlan::SelectList makeUnionFromTable(
auto columnStatistics = ctx.gwi.columnStatisticsMap[keyColumn->columnName()]; auto columnStatistics = ctx.gwi.columnStatisticsMap[keyColumn->columnName()];
std::cout << "columnStatistics.size() " << columnStatistics.size() << std::endl; std::cout << "columnStatistics.size() " << columnStatistics.size() << std::endl;
// TODO char and other numerical types support // TODO char and other numerical types support
size_t numberOfUnionUnits = 2;
size_t numberOfBucketsPerUnionUnit = columnStatistics.size() / numberOfUnionUnits;
std::vector<std::pair<uint64_t, uint64_t>> bounds; std::vector<std::pair<uint64_t, uint64_t>> bounds;
std::transform(columnStatistics.begin(), columnStatistics.end(), std::back_inserter(bounds),
[](const auto& bucket) // TODO need to process tail if number of buckets is not divisible by number of union units
{ // TODO non-overlapping buckets if it is a problem at all
uint64_t lowerBound = std::stoul(bucket.start_value); for (size_t i = 0; i < numberOfUnionUnits; ++i)
uint64_t upperBound = lowerBound + bucket.ndv; {
return std::make_pair(lowerBound, upperBound); auto bucket = columnStatistics.begin() + i * numberOfBucketsPerUnionUnit;
}); auto endBucket = columnStatistics.begin() + (i + 1) * numberOfBucketsPerUnionUnit;
// std::vector<std::pair<uint64_t, uint64_t>> bounds({{0, 3000961}, // TODO find a median b/w the current bucket start and the previous bucket end
// // {3000961, std::numeric_limits<uint64_t>::max()}}); uint64_t currentLowerBound =
(bounds.empty() ? *(uint32_t*)bucket->start_value.data()
: std::min((uint64_t)*(uint32_t*)bucket->start_value.data(), bounds.back().second));
uint64_t currentUpperBound = currentLowerBound;
for (; bucket != endBucket; ++bucket)
{
uint64_t bucketLowerBound = *(uint32_t*)bucket->start_value.data();
std::cout << "bucket.start_value " << bucketLowerBound << std::endl;
currentUpperBound = bucketLowerBound + bucket->ndv;
}
std::cout << "currentLowerBound " << currentLowerBound << " currentUpperBound " << currentUpperBound
<< std::endl;
bounds.push_back(std::make_pair(currentLowerBound, currentUpperBound));
}
for (auto& bound : bounds) for (auto& bound : bounds)
{ {
auto clonedCSEP = csep.cloneWORecursiveSelects(); auto clonedCSEP = csep.cloneWORecursiveSelects();