摘要:

收集现有代码的糟糕实现，前事不忘后事之师，把这些烂东西定死在耻辱柱上以免再次发生

糟糕的设计:

一. DGMaterializedIterator::GetNextPackrow

函数实现:

int DimensionGroupMaterialized::DGMaterializedIterator::GetNextPackrow(int dim, int ahead) {MEASURE_FET("DGMaterializedIterator::GetNextPackrow(int dim, int ahead)");if (ahead == 0) return GetCurPackrow(dim);IndexTable *cur_t = t[dim];if (cur_t == NULL) return -1;uint64_t end_block = cur_t->EndOfCurrentBlock(cur_pos);if (next_pack[dim] >= no_obj || uint64_t(next_pack[dim]) >= end_block) return -1;uint64_t ahead_pos = 0;//	cout << "dim " << dim << ",  " << next_pack[dim] << " -> " <Get64InsideBlock(next_pack[dim]);else if (ahead == 2 && ahead1[dim] != -1)ahead_pos = t[dim]->Get64InsideBlock(ahead1[dim]);else if (ahead == 3 && ahead2[dim] != -1)ahead_pos = t[dim]->Get64InsideBlock(ahead2[dim]);else if (ahead == 4 && ahead3[dim] != -1)ahead_pos = t[dim]->Get64InsideBlock(ahead3[dim]);if (ahead_pos == 0) return -1;return int((ahead_pos - 1) >> p_power);return -1;
}

    int64_t *next_pack;  // beginning of the next pack (or no_obj if there is no// other pack)int64_t *ahead1, *ahead2,*ahead3;  // beginning of the three next packs after next_pack, or// -1 if not determined properly

问题:

1. 直接将内部对于pack的预读的实现细节暴漏出去，接口的调用方必须知道DGMaterializedIterator内部的处理
2. DGMaterializedIterator类的实现与调用方的逻辑互相耦合，调用方必须了解DGMaterializedIterator的实现才能使用该类的功能、
3. 没有更高层次的抽象，只有在iterator级别的封装使用，没有对类的功能和目的做思考，导致业务逻辑散布各个类的实现细节，对于维护和扩展是个灾难，而且出现问题难以快速定位
4. 在做大的功能模块的架构的前，最基础的就是把一个函数写好，把一个类给设计好，细节做好，才能谈宏观的，不存在连个函数写的各种逻辑漏洞，类设计的稀烂却能做好功能设计的，这个类设计的真烂的出奇

建议:

1. 先设计好每个功能的划分，类承载哪些功能和起到的作用，在恰到好处的抽象层次上解耦业务逻辑
2. 对于类，建议根据面向对象的原则，去设计类和类间的交互关系

二. AggregatePackrow使用magic number作为错误码

函数实现:

          int grouping_result = AggregatePackrow(gbw, &mit, cur_tuple);if (sender) {sender->SetAffectRows(gbw.NumOfGroups());}if (grouping_result == 2) throw common::KilledException();if (grouping_result != 5) packrows_found++;  // for statisticsif (grouping_result == 1) break;             // end of the aggregation

int AggregationAlgorithm::AggregatePackrow(GroupByWrapper &gbw, MIIterator *mit, int64_t cur_tuple) {int64_t packrow_length = mit->GetPackSizeLeft();if (!gbw.AnyTuplesLeft(cur_tuple, cur_tuple + packrow_length - 1)) {mit->NextPackrow();return 5;}

问题:

1. 使用magic number无法直接从使用角度理解其出错的语义
2. 数字无法被有意义的搜索，在定位问题时，无法根据错误码快速定位代码

建议:

1. 使用枚举定义出错误码，返回枚举值

三. 逻辑不自洽，在自身的逻辑中崩溃

函数实现:

  std::unique_ptr gbw_ptr(new GroupByWrapper(*gb_sharding));gbw_ptr->FillDimsUsed(dims);gbw_ptr->SetDistinctTuples(mit->NumOfTuples());if (!gbw_ptr->IsOnePass()) gbw_ptr->InitTupleLeft(mit->NumOfTuples());

GroupByWrapper::GroupByWrapper(const GroupByWrapper &sec): distinct_watch(sec.p_power), m_conn(sec.m_conn), gt(sec.gt) {p_power = sec.p_power;attrs_size = sec.attrs_size;just_distinct = sec.just_distinct;virt_col = new vcolumn::VirtualColumn *[attrs_size];input_mode = new GBInputMode[attrs_size];is_lookup = new bool[attrs_size];attr_mapping = new int[attrs_size];  // output attr[j] <-> gt group[attr_mapping[j]]dist_vals = new int64_t[attrs_size];for (int i = 0; i < attrs_size; i++) {attr_mapping[i] = sec.attr_mapping[i];virt_col[i] = sec.virt_col[i];input_mode[i] = sec.input_mode[i];is_lookup[i] = sec.is_lookup[i];dist_vals[i] = sec.dist_vals[i];}no_grouping_attr = sec.no_grouping_attr;no_aggregated_attr = sec.no_aggregated_attr;no_more_groups = sec.no_more_groups;no_groups = sec.no_groups;no_attr = sec.no_attr;pack_not_omitted = new bool[no_attr];packrows_omitted = 0;packrows_part_omitted = 0;for (int i = 0; i < no_attr; i++) pack_not_omitted[i] = sec.pack_not_omitted[i];tuple_left = NULL;if (sec.tuple_left) tuple_left = new Filter(*sec.tuple_left);  // a copy of filter// init distinct_watch to make copy ctor has all Initialization logicdistinct_watch.Initialize(no_attr);for (int gr_a = 0; gr_a < no_attr; gr_a++) {if (gt.AttrDistinct(gr_a)) {distinct_watch.DeclareAsDistinct(gr_a);}}
}

  if (sec.tuple_left) tuple_left = new Filter(*sec.tuple_left);  // a copy of filter

void GroupByWrapper::InitTupleLeft(int64_t n) {DEBUG_ASSERT(tuple_left == NULL);tuple_left = new Filter(n, p_power);tuple_left->Set();
}

  DEBUG_ASSERT(tuple_left == NULL);

问题:

1. GroupByWrapper的拷贝构造函数中，tuple_left存在被赋值的情况
2. 创建 GroupByWrapper后立即调用 InitTupleLeft， 此函数中如果 tuple_left 不为NULL则宕掉

四. 函数内逻辑冗余，导致被不必要的执行多次

函数实现:

问题:

1. 进入else分支后，将被执行两次IsType_JoinSimple

五. 在没有理解原有代码的设计目的前，瞎改造，将只有串行的函数强行用多线程串行

AggregationAlgorithm::MultiDimensionalGroupByScan

原有代码:

void AggregationAlgorithm::MultiDimensionalGroupByScan(GroupByWrapper &gbw, _int64& limit, _int64& offset, map > &oi, ResultSender* sender, bool limit_less_than_no_groups)
{MEASURE_FET("TempTable::MultiDimensionalGroupByScan(...)");bool first_pass = true;_int64 cur_tuple = 0;				// tuples are numbered according to tuple_left filter (not used, if tuple_left is null)_int64 displayed_no_groups = 0;// Determine dimensions to be iteratedbool no_dims_found = true;DimensionVector dims(mind->NoDimensions());gbw.FillDimsUsed(dims);for(int i = 0; i < mind->NoDimensions(); i++)if(dims[i]) {no_dims_found = false;break;}if(no_dims_found)dims[0] = true;							// at least one dimension is neededvector po(mind->NoDimensions());// do not use pack orderer if there are too many expected groups// (more than 50% of tuples)if(gbw.UpperApproxOfGroups() < mind->NoTuples() / 2) {map >::iterator oi_it;bool one_group = (gbw.UpperApproxOfGroups() == 1);for(oi_it = oi.begin(); oi_it!= oi.end(); oi_it++)PackOrderer::ChoosePackOrderer(po[(*oi_it).first],(*oi_it).second, one_group);}MIIterator mit(mind, dims, po);factor = mit.Factor();if(mit.NoTuples() == NULL_VALUE_64 || mit.NoTuples() > MAX_ROW_NUMBER) {	// 2^47, a limit for filter belowthrow OutOfMemoryRCException("Aggregation is too large.");}gbw.SetDistinctTuples(mit.NoTuples());#ifndef __BH_COMMUNITY__AggregationWorkerEnt ag_worker(gbw, this);if(gbw.MayBeParallel() && ag_worker.MayBeParallel(mit) && !limit_less_than_no_groups)	// if we are going to skip groups, we cannot do it in parallelag_worker.CheckThreads(mit);	// CheckThreads() must be executed if we want to be parallel
#elseAggregationWorker ag_worker(gbw, this);
#endifif(!gbw.IsOnePass())gbw.InitTupleLeft(mit.NoTuples());bool rewind_needed = false;bool was_prefetched = false;try {do {if(rccontrol.isOn())  {if(gbw.UpperApproxOfGroups() == 1 || first_pass)rccontrol.lock(m_conn->GetThreadID()) << "Aggregating: " << mit.NoTuples() << " tuples left." << unlock;elserccontrol.lock(m_conn->GetThreadID()) << "Aggregating: " << gbw.TuplesNoOnes() << " tuples left, " << displayed_no_groups << " gr. found so far" << unlock;}cur_tuple = 0;gbw.ClearNoGroups();			// count groups locally created in this passgbw.ClearDistinctBuffers();		// reset buffers for a new contentsgbw.AddAllGroupingConstants(mit);ag_worker.Init(mit);if(rewind_needed)mit.Rewind();	// aggregated rows will be massively omitted packrow by packrowrewind_needed = true;was_prefetched = false;for(uint i = 0; i < t->NoAttrs(); i++) {		// left as uninitialized (NULL or 0)if(t->GetAttrP(i)->mode == DELAYED) {MIDummyIterator m(1);t->GetAttrP(i)->term.vc->LockSourcePacks(m);}}while(mit.IsValid()) { / First stage - some distincts may be delayedif(m_conn->killed())throw KilledRCException();/// Grouping on a packrow _int64 packrow_length = mit.GetPackSizeLeft();if(ag_worker.ThreadsUsed() == 1) {if(was_prefetched == false) {for(int i = 0; i < gbw.NoAttr(); i++)if(gbw.GetColumn(i))gbw.GetColumn(i)->InitPrefetching(mit);was_prefetched = true;}int grouping_result = AggregatePackrow(gbw, &mit, cur_tuple);if(grouping_result == 2)throw KilledRCException();if(grouping_result != 5)packrows_found++;				// for statisticsif(grouping_result == 1)break;							// end of the aggregationif(!gbw.IsFull() && gbw.MemoryBlocksLeft() == 0) {gbw.SetAsFull();}} else {if(was_prefetched) {for(int i = 0; i < gbw.NoAttr(); i++)if(gbw.GetColumn(i))gbw.GetColumn(i)->StopPrefetching();was_prefetched = false;}MIInpackIterator lmit(mit);int grouping_result = ag_worker.AggregatePackrow(lmit, cur_tuple);if(grouping_result != 5)packrows_found++;				// for statisticsif(grouping_result == 1)break;if(grouping_result == 2)throw KilledRCException();if(grouping_result == 3 || grouping_result == 4)throw NotImplementedRCException("Aggregation overflow.");if(mit.BarrierAfterPackrow()) {ag_worker.Barrier();}ag_worker.ReevaluateNumberOfThreads(mit);mit.NextPackrow();}cur_tuple += packrow_length;}MultiDimensionalDistinctScan(gbw, mit);		// if not needed, no effectag_worker.Commit();// Now it is time to prepare output valuesif(first_pass) {first_pass = false;_int64 upper_groups = gbw.NoGroups() + gbw.TuplesNoOnes();		// upper approximation: the current size + all other possible rows (if any)t->CalculatePageSize(upper_groups);if(upper_groups > gbw.UpperApproxOfGroups())upper_groups = gbw.UpperApproxOfGroups();					// another upper limitation: not more than theoretical number of combinationsMIDummyIterator m(1);for(uint i = 0; i < t->NoAttrs(); i++) {t->GetAttrP(i)->CreateBuffer(upper_groups);					// note: may be more than neededif(t->GetAttrP(i)->mode == DELAYED)t->GetAttrP(i)->term.vc->LockSourcePacks(m);}}rccontrol.lock(m_conn->GetThreadID()) << "Generating output." << unlock;gbw.RewindRows();while(gbw.RowValid()) {		// copy GroupTable into TempTable, row by rowif(t->NoObj() >= limit)break;AggregateFillOutput(gbw, gbw.GetCurrentRow(), offset);		// offset is decremented for each row, if positiveif(sender && t->NoObj() > 65535) {TempTable::RecordIterator iter = t->begin();	for(_int64 i = 0; i < t->NoObj(); i++) {sender->Send(iter);++iter;}displayed_no_groups += t->NoObj();limit -= t->NoObj();t->SetNoObj(0);}gbw.NextRow();}if(sender) {TempTable::RecordIterator iter = t->begin();	for(_int64 i = 0; i < t->NoObj(); i++) {sender->Send(iter);++iter;}displayed_no_groups += t->NoObj();limit -= t->NoObj();t->SetNoObj(0);} else displayed_no_groups = t->NoObj();if(t->NoObj() >= limit)break;if(gbw.AnyTuplesLeft())gbw.ClearUsed();				// prepare for the next pass, if needed} while(gbw.AnyTuplesLeft());			// do the next pass, if anything left} catch(...) {ag_worker.Commit(false);throw;}if(rccontrol.isOn())rccontrol.lock(m_conn->GetThreadID()) << "Aggregated (" << displayed_no_groups<< " gr). Omitted packrows: " << gbw.packrows_omitted << " + "<< gbw.packrows_part_omitted << " partially, out of " << packrows_found << " total." << unlock;
}

核心处理:

			while(mit.IsValid()) { / First stage - some distincts may be delayedif(m_conn->killed())throw KilledRCException();/// Grouping on a packrow _int64 packrow_length = mit.GetPackSizeLeft();if(ag_worker.ThreadsUsed() == 1) {if(was_prefetched == false) {for(int i = 0; i < gbw.NoAttr(); i++)if(gbw.GetColumn(i))gbw.GetColumn(i)->InitPrefetching(mit);was_prefetched = true;}int grouping_result = AggregatePackrow(gbw, &mit, cur_tuple);if(grouping_result == 2)throw KilledRCException();if(grouping_result != 5)packrows_found++;				// for statisticsif(grouping_result == 1)break;							// end of the aggregationif(!gbw.IsFull() && gbw.MemoryBlocksLeft() == 0) {gbw.SetAsFull();}} else {if(was_prefetched) {for(int i = 0; i < gbw.NoAttr(); i++)if(gbw.GetColumn(i))gbw.GetColumn(i)->StopPrefetching();was_prefetched = false;}MIInpackIterator lmit(mit);int grouping_result = ag_worker.AggregatePackrow(lmit, cur_tuple);if(grouping_result != 5)packrows_found++;				// for statisticsif(grouping_result == 1)break;if(grouping_result == 2)throw KilledRCException();if(grouping_result == 3 || grouping_result == 4)throw NotImplementedRCException("Aggregation overflow.");if(mit.BarrierAfterPackrow()) {ag_worker.Barrier();}ag_worker.ReevaluateNumberOfThreads(mit);mit.NextPackrow();}cur_tuple += packrow_length;}

被魔改的代码:

void AggregationAlgorithm::MultiDimensionalGroupByScan(GroupByWrapper &gbw, int64_t &limit, int64_t &offset,ResultSender *sender, bool limit_less_than_no_groups) {MEASURE_FET("TempTable::MultiDimensionalGroupByScan(...)");bool first_pass = true;// tuples are numbered according to tuple_left filter (not used, if tuple_left// is null)int64_t cur_tuple = 0;int64_t displayed_no_groups = 0;// Determine dimensions to be iteratedbool no_dims_found = true;DimensionVector dims(mind->NumOfDimensions());gbw.FillDimsUsed(dims);for (int i = 0; i < mind->NumOfDimensions(); i++)if (dims[i]) {no_dims_found = false;break;}if (no_dims_found) dims[0] = true;  // at least one dimension is neededstd::vector po(mind->NumOfDimensions());MIIterator mit(mind, dims, po);factor = mit.Factor();if (mit.NumOfTuples() == common::NULL_VALUE_64 ||mit.NumOfTuples() > common::MAX_ROW_NUMBER) {  // 2^47, a limit for filter belowthrow common::OutOfMemoryException("Aggregation is too large.");}gbw.SetDistinctTuples(mit.NumOfTuples());int thd_cnt = 1;if (ParallelAllowed(gbw) && !limit_less_than_no_groups) {thd_cnt = std::thread::hardware_concurrency() / 4;  // For concurrence reason, don't swallow all cores once.}AggregationWorkerEnt ag_worker(gbw, mind, thd_cnt, this);if (!gbw.IsOnePass()) gbw.InitTupleLeft(mit.NumOfTuples());bool rewind_needed = false;try {do {if (rccontrol.isOn()) {if (gbw.UpperApproxOfGroups() == 1 || first_pass)rccontrol.lock(m_conn->GetThreadID())<< "Aggregating: " << mit.NumOfTuples() << " tuples left." << system::unlock;elserccontrol.lock(m_conn->GetThreadID()) << "Aggregating: " << gbw.TuplesNoOnes() << " tuples left, "<< displayed_no_groups << " gr. found so far" << system::unlock;}cur_tuple = 0;gbw.ClearNoGroups();         // count groups locally created in this passgbw.ClearDistinctBuffers();  // reset buffers for a new contentsgbw.AddAllGroupingConstants(mit);ag_worker.Init(mit);if (rewind_needed)mit.Rewind();  // aggregated rows will be massively omitted packrow by// packrowrewind_needed = true;for (uint i = 0; i < t->NumOfAttrs(); i++) {  // left as uninitialized (NULL or 0)if (t->GetAttrP(i)->mode == common::ColOperation::DELAYED) {MIDummyIterator m(1);t->GetAttrP(i)->term.vc->LockSourcePacks(m);}}if (ag_worker.ThreadsUsed() > 1) {ag_worker.DistributeAggreTaskAverage(mit);} else {while (mit.IsValid()) {  // need muti thread// First stage -//  some distincts may be delayedif (m_conn->Killed()) throw common::KilledException();// Grouping on a packrowint64_t packrow_length = mit.GetPackSizeLeft();int grouping_result = AggregatePackrow(gbw, &mit, cur_tuple);if (sender) {sender->SetAffectRows(gbw.NumOfGroups());}if (grouping_result == 2) throw common::KilledException();if (grouping_result != 5) packrows_found++;  // for statisticsif (grouping_result == 1) break;             // end of the aggregationif (!gbw.IsFull() && gbw.MemoryBlocksLeft() == 0) {gbw.SetAsFull();}cur_tuple += packrow_length;}}gbw.ClearDistinctBuffers();              // reset buffers for a new contentsMultiDimensionalDistinctScan(gbw, mit);  // if not needed, no effectag_worker.Commit();// Now it is time to prepare output valuesif (first_pass) {first_pass = false;int64_t upper_groups = gbw.NumOfGroups() + gbw.TuplesNoOnes();  // upper approximation: the current size +// all other possible rows (if any)t->CalculatePageSize(upper_groups);if (upper_groups > gbw.UpperApproxOfGroups())upper_groups = gbw.UpperApproxOfGroups();  // another upper limitation: not more// than theoretical number of// combinationsMIDummyIterator m(1);for (uint i = 0; i < t->NumOfAttrs(); i++) {if (t->GetAttrP(i)->mode == common::ColOperation::GROUP_CONCAT) {t->GetAttrP(i)->SetTypeName(common::CT::VARCHAR);t->GetAttrP(i)->OverrideStringSize(tianmu_group_concat_max_len);}t->GetAttrP(i)->CreateBuffer(upper_groups);  // note: may be more than neededif (t->GetAttrP(i)->mode == common::ColOperation::DELAYED) t->GetAttrP(i)->term.vc->LockSourcePacks(m);}}rccontrol.lock(m_conn->GetThreadID()) << "Group/Aggregate end. Begin generating output." << system::unlock;rccontrol.lock(m_conn->GetThreadID()) << "Output rows: " << gbw.NumOfGroups() + gbw.TuplesNoOnes()<< ", output table row limit: " << t->GetPageSize() << system::unlock;int64_t output_size = (gbw.NumOfGroups() + gbw.TuplesNoOnes()) * t->GetOneOutputRecordSize();gbw.RewindRows();if (t->GetPageSize() >= (gbw.NumOfGroups() + gbw.TuplesNoOnes()) && output_size > (1L << 29) &&!t->HasHavingConditions() && tianmu_sysvar_parallel_filloutput) {// Turn on parallel output when:// 1. output page is large enough to hold all output rows// 2. output result is larger than 512MB// 3. no have conditionrccontrol.lock(m_conn->GetThreadID()) << "Start parallel output" << system::unlock;ParallelFillOutputWrapper(gbw, offset, limit, mit);} else {while (gbw.RowValid()) {// copy GroupTable into TempTable, row by rowif (t->NumOfObj() >= limit) break;AggregateFillOutput(gbw, gbw.GetCurrentRow(),offset);  // offset is decremented for each row, if positiveif (sender && t->NumOfObj() > (1 << mind->ValueOfPower()) - 1) {TempTable::RecordIterator iter = t->begin();for (int64_t i = 0; i < t->NumOfObj(); i++) {sender->Send(iter);++iter;}displayed_no_groups += t->NumOfObj();limit -= t->NumOfObj();t->SetNumOfObj(0);}gbw.NextRow();}}if (sender) {TempTable::RecordIterator iter = t->begin();for (int64_t i = 0; i < t->NumOfObj(); i++) {sender->Send(iter);++iter;}displayed_no_groups += t->NumOfObj();limit -= t->NumOfObj();t->SetNumOfObj(0);} elsedisplayed_no_groups = t->NumOfObj();if (t->NumOfObj() >= limit) break;if (gbw.AnyTuplesLeft()) gbw.ClearUsed();  // prepare for the next pass, if needed} while (gbw.AnyTuplesLeft());               // do the next pass, if anything left} catch (...) {ag_worker.Commit(false);throw;}if (rccontrol.isOn())rccontrol.lock(m_conn->GetThreadID())<< "Generating output end. "<< "Aggregated (" << displayed_no_groups << " group). Omitted packrows: " << gbw.packrows_omitted << " + "<< gbw.packrows_part_omitted << " partially, out of " << packrows_found << " total." << system::unlock;
}

核心处理:

      if (ag_worker.ThreadsUsed() > 1) {ag_worker.DistributeAggreTaskAverage(mit);} else {while (mit.IsValid()) {  // need muti thread// First stage -//  some distincts may be delayedif (m_conn->Killed()) throw common::KilledException();// Grouping on a packrowint64_t packrow_length = mit.GetPackSizeLeft();int grouping_result = AggregatePackrow(gbw, &mit, cur_tuple);if (sender) {sender->SetAffectRows(gbw.NumOfGroups());}if (grouping_result == 2) throw common::KilledException();if (grouping_result != 5) packrows_found++;  // for statisticsif (grouping_result == 1) break;             // end of the aggregationif (!gbw.IsFull() && gbw.MemoryBlocksLeft() == 0) {gbw.SetAsFull();}cur_tuple += packrow_length;}}

单线程中遍历迭代器给GroupByWrapper的filter的blocks赋值，用来在下一次遍历中使用:

(gdb) bt
#0  Tianmu::core::AggregationAlgorithm::AggregatePackrow (this=0x7ff52da4b580, gbw=..., mit=0x7ff52da4aee0, cur_tuple=41953)at /home/jenkins/workspace/stonedb5.7-zsl-centos7.9-30-119-20220805/storage/tianmu/core/aggregation_algorithm.cpp:580
#1  0x0000000003005b74 in Tianmu::core::AggregationAlgorithm::MultiDimensionalGroupByScan (this=0x7ff52da4b580, gbw=..., limit=@0x7ff52da4b208: 7422784, offset=@0x7ff52da4b608: 0, sender=0x0, limit_less_than_no_groups=false) at /home/jenkins/workspace/stonedb5.7-zsl-centos7.9-30-119-20220805/storage/tianmu/core/aggregation_algorithm.cpp:280
#2  0x00000000030053ca in Tianmu::core::AggregationAlgorithm::Aggregate (this=0x7ff52da4b580, just_distinct=false, limit=@0x7ff52da4b600: -1, offset=@0x7ff52da4b608: 0, sender=0x0)at /home/jenkins/workspace/stonedb5.7-zsl-centos7.9-30-119-20220805/storage/tianmu/core/aggregation_algorithm.cpp:196
#3  0x0000000002df1e3e in Tianmu::core::TempTable::Materialize (this=0x7fd1f8001e10, in_subq=false, sender=0x7fd1f8931ed0, lazy=false)at /home/jenkins/workspace/stonedb5.7-zsl-centos7.9-30-119-20220805/storage/tianmu/core/temp_table.cpp:1972
#4  0x0000000002d3a414 in Tianmu::core::Engine::Execute (this=0x7cef220, thd=0x7fd1f80125f0, lex=0x7fd1f8014918, result_output=0x7fd1f8ab1c20, unit_for_union=0x0)at /home/jenkins/workspace/stonedb5.7-zsl-centos7.9-30-119-20220805/storage/tianmu/core/engine_execute.cpp:426
#5  0x0000000002d395b6 in Tianmu::core::Engine::HandleSelect (this=0x7cef220, thd=0x7fd1f80125f0, lex=0x7fd1f8014918, result=@0x7ff52da4bd18: 0x7fd1f8ab1c20, setup_tables_done_option=0, res=@0x7ff52da4bd14: 0, optimize_after_tianmu=@0x7ff52da4bd0c: 1, tianmu_free_join=@0x7ff52da4bd10: 1, with_insert=0)at /home/jenkins/workspace/stonedb5.7-zsl-centos7.9-30-119-20220805/storage/tianmu/core/engine_execute.cpp:232
#6  0x0000000002e21e47 in Tianmu::dbhandler::TIANMU_HandleSelect (thd=0x7fd1f80125f0, lex=0x7fd1f8014918, result=@0x7ff52da4bd18: 0x7fd1f8ab1c20, setup_tables_done_option=0, res=@0x7ff52da4bd14: 0, optimize_after_tianmu=@0x7ff52da4bd0c: 1, tianmu_free_join=@0x7ff52da4bd10: 1, with_insert=0)at /home/jenkins/workspace/stonedb5.7-zsl-centos7.9-30-119-20220805/storage/tianmu/handler/ha_rcengine.cpp:82
#7  0x0000000002462f6a in execute_sqlcom_select (thd=0x7fd1f80125f0, all_tables=0x7fd1f8b3ec88) at /home/jenkins/workspace/stonedb5.7-zsl-centos7.9-30-119-20220805/sql/sql_parse.cc:5182
#8  0x000000000245c2ee in mysql_execute_command (thd=0x7fd1f80125f0, first_level=true) at /home/jenkins/workspace/stonedb5.7-zsl-centos7.9-30-119-20220805/sql/sql_parse.cc:2831
#9  0x0000000002463f33 in mysql_parse (thd=0x7fd1f80125f0, parser_state=0x7ff52da4ceb0) at /home/jenkins/workspace/stonedb5.7-zsl-centos7.9-30-119-20220805/sql/sql_parse.cc:5621
#10 0x00000000024591cb in dispatch_command (thd=0x7fd1f80125f0, com_data=0x7ff52da4d650, command=COM_QUERY) at /home/jenkins/workspace/stonedb5.7-zsl-centos7.9-30-119-20220805/sql/sql_parse.cc:1495
#11 0x00000000024580f7 in do_command (thd=0x7fd1f80125f0) at /home/jenkins/workspace/stonedb5.7-zsl-centos7.9-30-119-20220805/sql/sql_parse.cc:1034
#12 0x000000000258accd in handle_connection (arg=0x9c1ef60) at /home/jenkins/workspace/stonedb5.7-zsl-centos7.9-30-119-20220805/sql/conn_handler/connection_handler_per_thread.cc:313
#13 0x0000000002c71102 in pfs_spawn_thread (arg=0x8140030) at /home/jenkins/workspace/stonedb5.7-zsl-centos7.9-30-119-20220805/storage/perfschema/pfs.cc:2197
#14 0x00007ff57ca91ea5 in start_thread () from /lib64/libpthread.so.0
#15 0x00007ff57acc6b0d in clone () from /lib64/libc.so.6

(gdb) bt
#0  Tianmu::core::Filter::Set (this=0x7fd1f8b8f990, b=0, n=41953) at /home/jenkins/workspace/stonedb5.7-zsl-centos7.9-30-119-20220805/storage/tianmu/core/filter.cpp:243
#1  0x0000000002d5c465 in Tianmu::core::Filter::Set (this=0x7fd1f8b8f990, n=41953) at /home/jenkins/workspace/stonedb5.7-zsl-centos7.9-30-119-20220805/storage/tianmu/core/filter.h:74
#2  0x0000000003074929 in Tianmu::core::DistinctWrapper::SetAsOmitted (this=0x7ff52da4b220, attr=3, obj=41953)at /home/jenkins/workspace/stonedb5.7-zsl-centos7.9-30-119-20220805/storage/tianmu/core/groupby_wrapper.h:44
#3  0x0000000003073071 in Tianmu::core::GroupByWrapper::DistinctlyOmitted (this=0x7ff52da4b220, attr=3, obj=41953)at /home/jenkins/workspace/stonedb5.7-zsl-centos7.9-30-119-20220805/storage/tianmu/core/groupby_wrapper.cpp:534
#4  0x0000000003007427 in Tianmu::core::AggregationAlgorithm::AggregatePackrow (this=0x7ff52da4b580, gbw=..., mit=0x7ff52da4aee0, cur_tuple=41953)at /home/jenkins/workspace/stonedb5.7-zsl-centos7.9-30-119-20220805/storage/tianmu/core/aggregation_algorithm.cpp:580
#5  0x0000000003005b74 in Tianmu::core::AggregationAlgorithm::MultiDimensionalGroupByScan (this=0x7ff52da4b580, gbw=..., limit=@0x7ff52da4b208: 7422784, offset=@0x7ff52da4b608: 0, sender=0x0, limit_less_than_no_groups=false) at /home/jenkins/workspace/stonedb5.7-zsl-centos7.9-30-119-20220805/storage/tianmu/core/aggregation_algorithm.cpp:280
#6  0x00000000030053ca in Tianmu::core::AggregationAlgorithm::Aggregate (this=0x7ff52da4b580, just_distinct=false, limit=@0x7ff52da4b600: -1, offset=@0x7ff52da4b608: 0, sender=0x0)at /home/jenkins/workspace/stonedb5.7-zsl-centos7.9-30-119-20220805/storage/tianmu/core/aggregation_algorithm.cpp:196
#7  0x0000000002df1e3e in Tianmu::core::TempTable::Materialize (this=0x7fd1f8001e10, in_subq=false, sender=0x7fd1f8931ed0, lazy=false)at /home/jenkins/workspace/stonedb5.7-zsl-centos7.9-30-119-20220805/storage/tianmu/core/temp_table.cpp:1972
#8  0x0000000002d3a414 in Tianmu::core::Engine::Execute (this=0x7cef220, thd=0x7fd1f80125f0, lex=0x7fd1f8014918, result_output=0x7fd1f8ab1c20, unit_for_union=0x0)at /home/jenkins/workspace/stonedb5.7-zsl-centos7.9-30-119-20220805/storage/tianmu/core/engine_execute.cpp:426
#9  0x0000000002d395b6 in Tianmu::core::Engine::HandleSelect (this=0x7cef220, thd=0x7fd1f80125f0, lex=0x7fd1f8014918, result=@0x7ff52da4bd18: 0x7fd1f8ab1c20, setup_tables_done_option=0, res=@0x7ff52da4bd14: 0, optimize_after_tianmu=@0x7ff52da4bd0c: 1, tianmu_free_join=@0x7ff52da4bd10: 1, with_insert=0)at /home/jenkins/workspace/stonedb5.7-zsl-centos7.9-30-119-20220805/storage/tianmu/core/engine_execute.cpp:232
#10 0x0000000002e21e47 in Tianmu::dbhandler::TIANMU_HandleSelect (thd=0x7fd1f80125f0, lex=0x7fd1f8014918, result=@0x7ff52da4bd18: 0x7fd1f8ab1c20, setup_tables_done_option=0, res=@0x7ff52da4bd14: 0, optimize_after_tianmu=@0x7ff52da4bd0c: 1, tianmu_free_join=@0x7ff52da4bd10: 1, with_insert=0)at /home/jenkins/workspace/stonedb5.7-zsl-centos7.9-30-119-20220805/storage/tianmu/handler/ha_rcengine.cpp:82
#11 0x0000000002462f6a in execute_sqlcom_select (thd=0x7fd1f80125f0, all_tables=0x7fd1f8b3ec88) at /home/jenkins/workspace/stonedb5.7-zsl-centos7.9-30-119-20220805/sql/sql_parse.cc:5182
#12 0x000000000245c2ee in mysql_execute_command (thd=0x7fd1f80125f0, first_level=true) at /home/jenkins/workspace/stonedb5.7-zsl-centos7.9-30-119-20220805/sql/sql_parse.cc:2831
#13 0x0000000002463f33 in mysql_parse (thd=0x7fd1f80125f0, parser_state=0x7ff52da4ceb0) at /home/jenkins/workspace/stonedb5.7-zsl-centos7.9-30-119-20220805/sql/sql_parse.cc:5621
#14 0x00000000024591cb in dispatch_command (thd=0x7fd1f80125f0, com_data=0x7ff52da4d650, command=COM_QUERY) at /home/jenkins/workspace/stonedb5.7-zsl-centos7.9-30-119-20220805/sql/sql_parse.cc:1495
#15 0x00000000024580f7 in do_command (thd=0x7fd1f80125f0) at /home/jenkins/workspace/stonedb5.7-zsl-centos7.9-30-119-20220805/sql/sql_parse.cc:1034
#16 0x000000000258accd in handle_connection (arg=0x9c1ef60) at /home/jenkins/workspace/stonedb5.7-zsl-centos7.9-30-119-20220805/sql/conn_handler/connection_handler_per_thread.cc:313
#17 0x0000000002c71102 in pfs_spawn_thread (arg=0x8140030) at /home/jenkins/workspace/stonedb5.7-zsl-centos7.9-30-119-20220805/storage/perfschema/pfs.cc:2197
#18 0x00007ff57ca91ea5 in start_thread () from /lib64/libpthread.so.0
#19 0x00007ff57acc6b0d in clone () from /lib64/libc.so.6

问题：

1. AggregationAlgorithm::AggregatePackrow在遍历迭代器过程中给GroupByWrapper的filter的blocks赋值
2. 不考虑此前代码的实现，强行用多线程来跑只有单线程遍历的逻辑，不顾底层block数据实现的逻辑

建议:

1. 没看明白原有代码的业务逻辑的因果关系前，不要随便按照臆想瞎几把搞
2. 写不出来就空着说写不出来，也别为了糊弄写一坨南辕北辙的垃圾上去