gStore/Database/Join.cpp

1262 lines
39 KiB
C++

/*=============================================================================
# Filename: Join.cpp
# Author: Bookug Lobert
# Mail: 1181955272@qq.com
# Last Modified: 2015-12-13 16:44
# Description: implement functions in Join.h
=============================================================================*/
#include "Join.h"
using namespace std;
Join::Join()
{
this->kvstore = NULL;
}
Join::Join(KVstore* _kvstore)
{
this->kvstore = _kvstore;
}
Join::~Join()
{
//noting to do necessarily
}
void
Join::init(BasicQuery* _basic_query)
{
this->basic_query = _basic_query;
this->var_num = this->basic_query->getVarNum();
this->id2pos = (int*)malloc(sizeof(int) * this->var_num);
memset(id2pos, -1, sizeof(int) * this->var_num);
this->pos2id = (int*)malloc(sizeof(int) * this->var_num);
memset(pos2id, -1, sizeof(int) * this->var_num);
this->id_pos = 0;
this->start_id = -1;
int triple_num = this->basic_query->getTripleNum();
this->dealed_triple = (bool*)calloc(triple_num, sizeof(bool));
}
void
Join::clear()
{
free(this->id2pos);
free(this->pos2id);
//NOTICE:maybe many BasicQuery
this->current_table.clear();
while(this->mystack.empty() == false) this->mystack.pop();
free(this->dealed_triple);
}
double
Join::score_node(unsigned _degree, unsigned _size)
{
//PARAM_DEGREE * _degree - PARAM_SIZE * _size
//BETTER?:use other cost model
return Join::PARAM_DEGREE * (double)_degree + Join::PARAM_SIZE / (double)_size;
}
int
Join::judge(int _smallest, int _biggest)
{
return 0; //DEBUG
//BETTER?:use appropiate method according to size and structure
int edge_num = this->basic_query->getTripleNum();
double dense = (double)edge_num / this->var_num;
//BETTER:how to guess the size of can_lists
double size = (_smallest + _biggest) / 2.0;
double ans = Join::PARAM_DENSE * dense - size / Join::PARAM_SIZE;
if(ans > Join::JUDGE_LIMIT)
return 0; //multi_join method
else
return 1; //index_join method
}
//select the start point, and maybe search order for index_join
//just for multi_join here, maybe diffrent for index_join later
void
Join::select()
{
//NOTICE: only consider vars in select here
double max = 0;
int maxi = -1;
//int border = this->basic_query->getVarNum();
for(int i = 0; i < this->var_num; ++i)
{
//multi-join not works better in star graph
//QUERY:worse than original in q4.sql for dbpedia2014.nt, analyse
//from which the join process start?
if(this->basic_query->isFreeLiteralVariable(i))
{
continue;
}
double tmp = this->score_node(this->basic_query->getVarDegree(i), this->basic_query->getCandidateSize(i));
if(tmp > max)
{
max = tmp;
maxi = i;
}
}
if(maxi == -1) //all vars are literal, so start from anywhere
{
this->start_id = 0;
}
else
{
this->start_id = maxi;
}
#ifdef DEBUG_JOIN
printf("the start id is: %d\n", this->start_id);
#endif
}
//join on the vector of CandidateList, available after
//retrieved from the VSTREE and store the resut in _result_set
bool
Join::join_sparql(SPARQLquery& _sparql_query)
{
int basic_query_num = _sparql_query.getBasicQueryNum();
//join each basic query
for(int i=0; i < basic_query_num; i++)
{
//printf("Basic query %d\n", i);
this->init(&(_sparql_query.getBasicQuery(i)));
long begin = Util::get_cur_time();
this->filter_before_join();
long after_filter = Util::get_cur_time();
//printf("after filter_before_join: used %ld ms\n", after_filter-begin);
this->add_literal_candidate();
long after_add_literal = Util::get_cur_time();
//printf("after add_literal_candidate: used %ld ms\n", after_add_literal-after_filter);
this->join();
long after_joinbasic = Util::get_cur_time();
//printf("after join_basic : used %ld ms\n", after_joinbasic-after_add_literal);
//this->only_pre_filter_after_join(this->basic_query);
//long after_pre_filter_after_join = Util::get_cur_time();
//printf("after only_pre_filter_after_join : used %ld ms\n", after_pre_filter_after_join-after_joinbasic);
//printf("Final result size: %lu\n", this->basic_query->getResultList().size());
this->clear();
}
return true;
}
// use the appropriate method to join candidates
bool
Join::join()
{
//TODO:compute the time cost of different parts
//the smallest candidate list size of the not-literal vars
int smallest = this->basic_query->getCandidateSize(this->basic_query->getVarID_FirstProcessWhenJoin());
if(smallest == 0)
return false; //empty result
int biggest = this->basic_query->getVarID_MaxCandidateList();
int method = this->judge(smallest, biggest);
switch(method)
{
case 0:
this->multi_join();
break;
case 1:
this->index_join();
break;
default:
break;
}
return true;
}
void
Join::add_new_to_results(TableIterator it, int id)
{
//NTC:already have one more in *it if need to push back
RecordType tmp(*it);
*(tmp.rbegin()) = id;
this->current_table.push_back(tmp);
}
int
Join::choose_next_node(bool* _dealed_triple, int id)
{
//choose a child to search deeply
int degree = this->basic_query->getVarDegree(id);
int maxi = -1;
double max = 0;
for(int i = 0; i < degree; ++i)
{
// each triple/edge need to be processed only once.
int edge_id = this->basic_query->getEdgeID(id, i);
if(_dealed_triple[edge_id])
{
continue;
}
int var_id2 = this->basic_query->getEdgeNeighborID(id, i);
if(var_id2 == -1) //not in join, including constant
{
continue;
}
//NTC:not using updated degrees, other not the whole loop
double tmp = this->score_node(this->basic_query->getVarDegree(var_id2), this->basic_query->getCandidateSize(var_id2));
if(max < tmp)
{
max = tmp;
maxi = i;
}
}
return maxi;
}
void
Join::acquire_all_id_lists(IdLists& _id_lists, IdListsLen& _id_lists_len, IDList& _can_list, vector<int>& _edges, bool* _dealed_triple, int _id, int can_list_size)
{
int* tmp_id_list;
int tmp_id_list_len;
for(int i = 0; i < this->id_pos; ++i)
{
// keep empty if not valid/used
_id_lists.push_back(vector<int*>());
_id_lists_len.push_back(vector<int>());
int edge_index = _edges[i];
if(edge_index != -1)
{
int pre_id = this->basic_query->getEdgePreID(_id, edge_index);
//int edge_id = this->basic_query->getEdgeID(_id, edge_index);
int edge_type = this->basic_query->getEdgeType(_id, edge_index);
if(pre_id >= 0) // valid
{
for(int j = 0; j < can_list_size; ++j)
{
if(edge_type == BasicQuery::EDGE_IN)
{
this->kvstore->getsubIDlistByobjIDpreID(_can_list[i], \
pre_id, tmp_id_list, tmp_id_list_len);
}
else //EDGE_OUT
{
this->kvstore->getobjIDlistBysubIDpreID(_can_list[i],\
pre_id, tmp_id_list, tmp_id_list_len);
}
_id_lists.rbegin()->push_back(tmp_id_list);
_id_lists_len.rbegin()->push_back(tmp_id_list_len);
}
}
}
}
}
bool
Join::is_literal_var(int _id)
{
//if(!this->basic_query->isFreeLiteralVariable(_id) || this->basic_query->isAddedLiteralCandidate(_id))
//if(!this->basic_query->isFreeLiteralVariable(_id))
//{
//return false;
//}
//BETTER?:this is not needed because we ensure that
//all dealed nodes's literals are added!
//this->basic_query->setAddedLiteralCandidate(_id);
if(this->basic_query->isAddedLiteralCandidate(_id))
return false;
else
return true;
}
bool
Join::is_literal_ele(int _id)
{
return _id >= Util::LITERAL_FIRST_ID;
}
//TODO:add debug info and check when the var is not free
bool
Join::new_join_with_multi_vars_prepared(IdLists& _id_lists, IdListsLen& _id_lists_len, vector<int>& _edges, IDList& _can_list, int _can_list_size)
{
if(_can_list_size == 0)
{
return false; //empty result
}
bool found = false; //no record matched
bool if_new_start = false; //the first to add to end in while
//list< list<int> > temp_table;
for(TableIterator it0 = this->current_table.begin(); it0 != this->new_start;)
{
bool matched = false; //this record matched
bool added = false; //if one ele added already
for(int i = 0; i < _can_list_size; ++i)
{
int cnt = 0;
bool linked = true;
for(RecordIterator it1 = it0->begin(); it1 != it0->end(); ++it1, ++cnt)
{
int edge_index = _edges[cnt];
if(edge_index == -1)
{
continue;
}
int ele = *it1;
if(_id_lists_len[cnt][i] == 0)
{
linked = false;
break;
}
if(Util::bsearch_int_uporder(ele, _id_lists[cnt][i], _id_lists_len[cnt][i]) == -1)
{
linked = false;
break;
}
}
if(linked)
{
if(added)
{
this->add_new_to_results(it0, _can_list[i]);
if(!if_new_start)
{
if_new_start = true;
this->new_start = this->current_table.end();
this->new_start--;
}
}
else
{
added = true;
it0->push_back(_can_list[i]);
}
matched = true;
}
}
if(matched)
{
found = true;
it0++;
//it3++;
}
else
{
it0 = this->current_table.erase(it0);
//it3 = this->table_row_new.erase(it3);
}
}
return found;
}
bool
Join::new_join_with_multi_vars_not_prepared(vector<int>& _edges, IDList& _can_list, int _can_list_size, int _id, bool _is_literal)
{
if(_can_list_size == 0 && !_is_literal)
{
return false; //empty result
}
bool found = false;
bool if_new_start = false; //the first to add to end in while
for(TableIterator it0 = this->current_table.begin(); it0 != this->new_start;)
{
#ifdef DEBUG_JOIN
if(this->new_start != this->current_table.end())
{
printf("now the new_start is:");
for(RecordIterator it1 = this->new_start->begin(); it1 != this->new_start->end(); ++it1)
{
printf(" %d", *it1);
}
printf("\n");
}
else
printf("new_start still in end?!\n");
printf("now the record is:");
for(RecordIterator it1 = it0->begin(); it1 != it0->end(); ++it1)
{
printf(" %d", *it1);
}
printf("\n");
#endif
int cnt = 0;
//update the valid id num according to restrictions by multi vars
//also ordered while id_list and can_list are ordered
list<int> valid_ans_list;
bool matched = true;
for(RecordIterator it1 = it0->begin(); it1 != it0->end(); ++it1, ++cnt)
{
#ifdef DEBUG_JOIN
printf("cnt is: %d\n", cnt);
#endif
int edge_index = _edges[cnt];
if(edge_index == -1)
{
continue;
}
#ifdef DEBUG_JOIN
printf("edge exists!\n");
#endif
int ele = *it1;
int edge_type = this->basic_query->getEdgeType(_id, edge_index);
int pre_id = this->basic_query->getEdgePreID(_id, edge_index);
int* id_list;
int id_list_len;
if (edge_type == BasicQuery::EDGE_IN)
{
#ifdef DEBUG_JOIN
printf("this is an edge to our id to join!\n");
#endif
this->kvstore->getobjIDlistBysubIDpreID(ele,\
pre_id, id_list, id_list_len);
}
else
{
#ifdef DEBUG_JOIN
printf("this is an edge from our id to join!\n");
#endif
this->kvstore->getsubIDlistByobjIDpreID(ele,\
pre_id, id_list, id_list_len);
}
if(id_list_len == 0)
{
//id_list == NULL in this case, no need to free
matched = false;
#ifdef DEBUG_JOIN
printf("this id_list is empty!\n");
#endif
break;
}
//BETTER?:to union/intersect two ordered lists directly?
//The cost to join two ordered lists are the basic operation
//of the whole join process!
//only can occur the first time, means cnt == 0
if(valid_ans_list.empty())
{
for(int i = 0; i < id_list_len; ++i)
{
//if we found this element(entity/literal) in
//var1's candidate list, or this is a literal
//element and var2 is a free literal variable,
//we should add this one to result.
bool flag = false;
//NOTICE:this var is free, but it can also contain
//entities. Candidates after retrieved from vstree will
//contain all possible entities, but no literals.
if(this->is_literal_ele(id_list[i]))
{
if(_is_literal)
{
flag = true;
#ifdef DEBUG_JOIN
printf("to add literal for free variable!\n");
#endif
}
}
else
{
flag = _can_list.bsearch_uporder(id_list[i]) >= 0;
}
if(!flag) continue;
//printf("add the ele to list!\n");
valid_ans_list.push_back(id_list[i]);
}
}
else
{
for(list<int>::iterator it2 = valid_ans_list.begin(); it2 != valid_ans_list.end();)
{
int tmp = *it2;
if(Util::bsearch_int_uporder(tmp, id_list, id_list_len) == -1)
{
it2 = valid_ans_list.erase(it2);
}
else
{
it2++;
}
}
}
delete[] id_list;
if(valid_ans_list.empty())
{
matched = false;
break;
}
}
if(matched)
{
#ifdef DEBUG_JOIN
printf("this record is matched!!\n");
#endif
found = true;
bool added = false;
//add new var results to table from valid_ans_list
for(list<int>::iterator it2 = valid_ans_list.begin(); it2 != valid_ans_list.end(); ++it2)
{
if(added)
{
this->add_new_to_results(it0, *it2);
if(!if_new_start)
{
if_new_start = true;
//DEBUG:we can acquire ele by *it in this way,
//but it still equals end()!
//this->new_start = this->current_table.rbegin().base();
this->new_start = this->current_table.end();
this->new_start--; //-1 is not allowed
#ifdef DEBUG_JOIN
if(this->new_start == this->current_table.end())
printf("error to set new_start!\n");
else
printf("now the new_start is set again!\n");
#endif
}
}
else
{
added = true;
it0->push_back(*it2);
}
}
it0++;
}
else
{
it0 = this->current_table.erase(it0);
#ifdef DEBUG_JOIN
printf("this record is not matched!\n");
#endif
}
}
return found;
}
bool
Join::if_prepare_idlist(int _can_list_size, bool _is_literal)
{
if(!_is_literal && _can_list_size < Join::LIMIT_CANDIDATE_LIST_SIZE)
return true;
else
return false;
}
void
Join::add_id_pos_mapping(int _id)
{
this->pos2id[this->id_pos] = _id;
this->id2pos[_id] = this->id_pos;
this->id_pos++;
}
void
Join::multi_join()
{
this->select();
//bool* dealed_id_list = (bool*)malloc(sizeof(bool) * var_num);
//memset(dealed_id_list, 0, sizeof(bool) * var_num);
//bool* dealed_triple = (bool*)malloc(sizeof(bool) * triple_num);
//memset(dealed_triple, 0, sizeof(bool) * triple_num);
//keep an increasing vector for temp results, not in id order
//vals num generally < 10, so just enum them and check if conncted
//finally, copy in order to result_list in BasicQuery
TableIterator it0;
list<int>::iterator it1;
vector<int>::iterator it2;
//list<bool>::iterator it3;
IDList& start_table = this->basic_query->getCandidateList(this->start_id);
int start_size = this->basic_query->getCandidateSize(this->start_id);
for(int i = 0; i < start_size; ++i)
{
RecordType record(1, start_table.getID(i));
this->current_table.push_back(record);
//this->table_row_new.push_back(false);
}
this->add_id_pos_mapping(this->start_id);
this->new_start = this->current_table.end();
//BETTER?:we can use nodes in stack to consider links instead of
//nodes in current_table, but this needs the stack to be visited
//below top, requiring us to implement on our own(array/vector)
//DEBUG: var_num > 100, maybe using vector, increasing dynamicly
//int mystack[100];
//int top = -1;
//mystack[++top] = this->start_id;
//
//if using nodes in current_table to consider links, no []
//can be used(except changing to vector, but wasteful)
//and then visit eles below top in stack is not ok,
//so choose STL stack
this->mystack.push(this->start_id);
while(!this->mystack.empty())
{
int id = this->mystack.top();
//int id = mystack[top];
int maxi = this->choose_next_node(dealed_triple, id);
if(maxi == -1) //all edges of this node are dealed
{
//top--;
this->mystack.pop();
continue;
}
int id2 = this->basic_query->getEdgeNeighborID(id, maxi);
//pre_id == -1 means we cannot find such predicate in rdf file, so the result set of this sparql should be empty.
//note that we cannot support to query sparqls with predicate variables ?p.
//TODO: if all missed?!
//preid < 0 !
//if(id_list[cnt].empty())
//{
// ifEmpty = true;
// break;
//}
vector<int> edges; //the edge index for table column in id2
// the outer is node-loop, inner is canlist-loop
vector< vector<int*> > id_lists;
vector< vector<int> > id_lists_len;
//int* tmp_id_list;
//int tmp_id_list_len;
IDList& can_list = this->basic_query->getCandidateList(id2);
int can_list_size = can_list.size();
for(int i = 0; i < this->id_pos; ++i)
{
int edge_index = this->basic_query->getEdgeIndex(id2, this->pos2id[i]);
edges.push_back(edge_index);
}
//NOTICE: there are several ways to join two tables
//h is the cost to search kvstore, m is the returned list size
//n is the normal can_list_size, k is the vars num to
//consider now, r is the record num
//0. expand and intersect with another table: not ok!
//1. given two node to find if exist right pre:
//O(1) space, O(rhknlogn) time,
//2. bsearch in can_list: O(mk+n) space, O(rmkhlogn) time
//3. bsearch in id_list: O(nkm) space, O(rnklogm+knh)
//
//most queries will contain many constants(entity/literal)
//var's can_list with one constant neighbor will be small,
//otherwise will be big compared with id_list
//the can_list of var representing literals is not valid,
//must use kvstore->get...() to join
bool is_literal = this->is_literal_var(id2);
bool flag = false;
bool if_prepare = this->if_prepare_idlist(can_list_size, is_literal);
//#ifdef DEBUG_JOIN
if_prepare = false;
//#endif
//needed if place can_list in the outer loop to join
if(if_prepare)
{
this->acquire_all_id_lists(id_lists, id_lists_len, can_list, edges, dealed_triple, id2, can_list_size);
flag = this->new_join_with_multi_vars_prepared(id_lists, id_lists_len, edges, can_list, can_list_size);
//need to release id_lists if using acquire_all_id_lists() firstly
for(vector< vector<int*> >::iterator p1 = id_lists.begin(); p1 != id_lists.end(); ++p1)
{
for(vector<int*>::iterator p2 = p1->begin(); p2 != p1->end(); ++p2)
{
delete[] *p2;
}
}
}
else
{
flag = this->new_join_with_multi_vars_not_prepared(edges, can_list, can_list_size, id2, is_literal);
}
//if current_table is empty, ends directly
if(!flag)
{
//break;
return; //to avoid later invalid copy
}
for(int i = 0; i < this->id_pos; ++i)
{
int edge_index = edges[i];
if(edge_index != -1)
{
int edge_id = this->basic_query->getEdgeID(id2, edge_index);
dealed_triple[edge_id] = true;
}
}
this->new_start = this->current_table.end();
this->add_id_pos_mapping(id2);
this->mystack.push(id2);
}
this->only_pre_filter_after_join();
//copy to result list, adjust the vars order
vector<int*>& result_list = this->basic_query->getResultList();
result_list.clear();
int select_var_num = this->basic_query->getSelectVarNum();
for(it0 = this->current_table.begin(); it0 != this->current_table.end(); ++it0)
{
//int* record = (int*)malloc(sizeof(int) * select_var_num);
int* record = new int[select_var_num];
for(int i = 0; i < this->id_pos; ++i)
{
if(this->pos2id[i] < select_var_num)
record[this->pos2id[i]] = (*it0)[i];
}
result_list.push_back(record);
}
//QUERY:the var order in output seems to be different
//on screen compared with in file(the right one)
//BETTER?:though the whole current_table is ordered here, the
//selected columns are not definitely ordered, needing to be
//sorted at the end. We can join based on the selected var's
//candidate to ensure the order, but this may be complicated.
//If we want to ensure the order here, new table is a must!
//and the duplicates cannot be checked unless the last step!
//The result list will not be too large generally, and the sort
//is not in any loop.(but if the size is too large?)
}
//BETTER?:place multi_join and index_join in separated files
//TODO:add debug info and check
void
Join::index_join()
{
//NOTICE:can combine multi-join if do index_join for each edge
//firstly(two-sided), then search deeply(check return edge),
//each time a path,if all is linked, then it is a result
//However, the performance is hard to guarantee, because the
//pre index-join is costly(always all candidates)
//index-join saves the cost to compute links every time, while
//other temporal-table based method may get a smaller table size
//to continue(no need to compute all candidates).
//
//BETTER?:there is two ways to do better, notice that the question to
//add literals or not can be solved by isAdded...
//The index join method saves the memory cost because 2m+2mn < 3mn,
//and time may be reduced if the pre-process is not too costly
//because we can reuse the links other than recompute in temporal table
//New struct is needed for node, i.e. list<int, list<iterator> >,
//because we may have to delete, but how can we know if an iterator
//is valid if the one it points to is removed?
//1. based on edges: process each time only in valid area(already
//macthed with others, invalid is removed), and finally it must be
//all ok, just copy to result_list. We should select the edge order
//to better the efficiency, but how can we keep only a neighbor links
//set if we want to save memory?(ensure all can be linked later)
//2. based on points: search deeply while multi-index-join, only a
//neighbor links set is kept for a node(not every edge), so memory
//cost is low. Finally, travel around along valid iterator, copy...
}
//sort the candidate lists and deal with all constant neigbors
void
Join::filter_before_join()
{
//printf("*****IIIIIIN filter_before_join\n");
for(int i = 0; i < this->var_num; i++)
{
//printf("\tVar%d %s\n", i, this->basic_query->getVarName(i).c_str());
IDList &can_list = this->basic_query->getCandidateList(i);
//printf("\t\tsize of canlist before filter: %d\n", can_list.size());
//NOTICE:must sort before using binary search.
can_list.sort();
long begin = Util::get_cur_time();
this->literal_edge_filter(i);
long after_literal_edge_filter = Util::get_cur_time();
//printf("\t\tliteral_edge_filter: used %ld ms\n", after_literal_edge_filter-begin);
// this->preid_filter(this->basic_query, i);
// long after_preid_filter = Util::get_cur_time();
// cout << "\t\tafter_preid_filter: used " << (after_preid_filter-after_literal_edge_filter) << " ms" << endl;
//printf("\t\t[%d] after filter, candidate size = %d\n\n\n", i, can_list.size());
//debug
// {
// stringstream _ss;
// for(int i = 0; i < can_list.size(); i ++)
// {
// string _can = this->kvstore->getEntityByID(can_list[i]);
// _ss << "[" << _can << ", " << can_list[i] << "]\t";
// }
// _ss << endl;
// Util::logging(_ss.str());
// cout << can_list.to_str() << endl;
// }
}
//printf("OOOOOOUT filter_before_join\n");
}
//decrease the candidates of _var_i using its constant neighbors
void
Join::literal_edge_filter(int _var_i)
{
//Util::logging("IN literal_edge_filter"); //debug
int var_degree = this->basic_query->getVarDegree(_var_i);
for(int j = 0; j < var_degree; j ++)
{
int neighbor_id = this->basic_query->getEdgeNeighborID(_var_i, j);
if(neighbor_id != -1) //variables in join not considered here
{
continue;
}
//QUERY:not only constants considered?
char edge_type = this->basic_query->getEdgeType(_var_i, j);
int triple_id = this->basic_query->getEdgeID(_var_i, j);
Triple triple = this->basic_query->getTriple(triple_id);
string neighbor_name;
if (edge_type == BasicQuery::EDGE_OUT)
{
neighbor_name = triple.object;
}
else
{
neighbor_name = triple.subject;
}
bool only_preid_filter = (this->basic_query->isOneDegreeNotSelectVar(neighbor_name));
if(only_preid_filter)
{
continue;
}
int pre_id = this->basic_query->getEdgePreID(_var_i, j);
IDList &_list = this->basic_query->getCandidateList(_var_i);
int lit_id = (this->kvstore)->getIDByEntity(neighbor_name);
if(lit_id == -1)
{
lit_id = (this->kvstore)->getIDByLiteral(neighbor_name);
}
// cout << "\t\tedge[" << j << "] "<< lit_string << " has id " << lit_id << "";
// cout << " preid:" << pre_id << " type:" << edge_type
// << endl;
// {
// stringstream _ss;
// _ss << "\t\tedge[" << j << "] "<< lit_string << " has id " << lit_id << "";
// _ss << " preid:" << pre_id << " type:" << edge_type
// << endl;
// Util::logging(_ss.str());
// }
int id_list_len = 0;
int* id_list = NULL;
if (pre_id >= 0)
{
if (edge_type == BasicQuery::EDGE_OUT)
{
(this->kvstore)->getsubIDlistByobjIDpreID(lit_id, pre_id, id_list, id_list_len);
}
else
{
(this->kvstore)->getobjIDlistBysubIDpreID(lit_id, pre_id, id_list, id_list_len);
}
}
else
// pre_id == -1 means we cannot find such predicate in rdf file, so the result set of this sparql should be empty.
// note that we cannot support to query sparqls with predicate variables ?p.
{
id_list_len = 0;
// if (edge_type == BasicQuery::EDGE_OUT)
// {
// (this->kvstore)->getsubIDlistByobjID(lit_id, id_list, id_list_len);
// }
// else
// {
// (this->kvstore)->getobjIDlistBysubID(lit_id, id_list, id_list_len);
// }
}
//debug
// {
// stringstream _ss;
// _ss << "id_list: ";
// for (int i=0;i<id_list_len;i++)
// {
// _ss << "[" << id_list[i] << "]\t";
// }
// _ss<<endl;
// Util::logging(_ss.str());
// }
if(id_list_len == 0)
{
_list.clear();
delete []id_list;
return;
}
// cout << "\t\t can:" << can_list.to_str() << endl;
// cout << "\t\t idlist has :";
// for(int i_ = 0; i_ < id_list_len; i_ ++)
// {
// cout << "[" << id_list[i_] << "]\t";
// }
// cout << endl;
_list.intersectList(id_list, id_list_len);
delete []id_list;
}
//Util::logging("OUT literal_edge_filter"); //debug
}
// this part can be omited or improved if the encode way of predicate
// is good enough
//also, we can decide whether we need run this part (if there are predicates encode overlap) by var_i's edge in queryGraph,
//for each edge e of var_i,
//if neightbor on e is an var, but not in select
//then, if the var's degree is 1, it has none contribution to filter
//only its sole edge property(predicate) makes sense
//we should make sure that var_i has an edge matching the predicate
//so this function will do the filtering
//TBD:
//if pre_id = -1,
//it means the entity id must has at least one edge
//NOTICE:not used now!
void
Join::preid_filter(int _var_i)
{
//IDList & _list, int _pre_id, char _edge_type
for (int j = 0; j < this->basic_query->getVarDegree(_var_i); j++)
{
int neighbor_id = this->basic_query->getEdgeNeighborID(_var_i, j);
// continue;
if (neighbor_id != -1)
{
continue;
}
char edge_type = this->basic_query->getEdgeType(_var_i, j);
int triple_id = this->basic_query->getEdgeID(_var_i, j);
Triple triple = this->basic_query->getTriple(triple_id);
string neighbor_name;
if (edge_type == BasicQuery::EDGE_OUT)
{
neighbor_name = triple.object;
}
else
{
neighbor_name = triple.subject;
}
// if neightbor is an var, but not in select
// then, if its degree is 1, it has none contribution to filter
// only its sole edge property(predicate) makes sense
// we should make sure that current candidateVar has an edge matching the predicate
bool only_preid_filter = (this->basic_query->isOneDegreeNotSelectVar(neighbor_name));
if (!only_preid_filter)
{
continue;
}
int pre_id = this->basic_query->getEdgePreID(_var_i, j);
IDList& _list = this->basic_query->getCandidateList(_var_i);
int* remain_list = new int[_list.size()];
int remain_len = 0;
int _entity_id = -1;
int* pair_list = NULL;
int pair_len = 0;
for (int i = 0; i < _list.size(); i++)
{
_entity_id = _list[i];
if (edge_type == BasicQuery::EDGE_IN)
{
(this->kvstore)->getpreIDsubIDlistByobjID
(_entity_id, pair_list, pair_len);
}
else
{
(this->kvstore)->getpreIDobjIDlistBysubID
(_entity_id, pair_list, pair_len);
}
bool exist_preid = Util::bsearch_preid_uporder
(pre_id, pair_list, pair_len);
if (exist_preid)
{
remain_list[remain_len] = _entity_id;
remain_len++;
}
delete[] pair_list;
pair_len = 0;
}// end for i 0 to _list.size
_list.intersectList(remain_list, remain_len);
//can be imported
delete[] remain_list;
} //end for j : varDegree
}
//if neighbor is an var, but not in select
//then, if its degree is 1, it has none contribution to filter
//only its sole edge property(predicate) makes sense
//we should make sure that current candidateVar has an edge matching the predicate
void
Join::only_pre_filter_after_join()
{
for(int var_id = 0; var_id < this->var_num; var_id++)
{
int var_degree = this->basic_query->getVarDegree(var_id);
//get all the only predicate filter edges for this variable.
vector<int> in_edge_pre_id;
vector<int> out_edge_pre_id;
for(int i = 0; i < var_degree; i++)
{
//WARN:one degree not in select var's id is also -1 !!
//constant neighbors already be dealed in literal_edge_filter
//if(this->basic_query->getEdgeNeighborID(var_id, i) == -1)
//continue;
char edge_type = this->basic_query->getEdgeType(var_id, i);
int triple_id = this->basic_query->getEdgeID(var_id, i);
Triple triple = this->basic_query->getTriple(triple_id);
string neighbor_name;
if (edge_type == BasicQuery::EDGE_OUT)
{
neighbor_name = triple.object;
}
else
{
neighbor_name = triple.subject;
}
bool only_preid_filter = (this->basic_query->isOneDegreeNotSelectVar(neighbor_name));
if (!only_preid_filter)
{
continue;
}
int pre_id = this->basic_query->getEdgePreID(var_id, i);
if (edge_type == BasicQuery::EDGE_OUT)
{
out_edge_pre_id.push_back(pre_id);
}
else
{
in_edge_pre_id.push_back(pre_id);
}
}
if (in_edge_pre_id.empty() && out_edge_pre_id.empty())
{
continue;
}
for(TableIterator it = this->current_table.begin(); it != this->current_table.end();)
{
int entity_id = (*it)[this->id2pos[var_id]];
int* pair_list = NULL;
int pair_len = 0;
bool exist_preid = true;
//BETTER?:use getsubIDlistByobjID or getobjIDlistBysubID to
//replace getpreIDlist...
if(exist_preid && !in_edge_pre_id.empty())
{
(this->kvstore)->getpreIDsubIDlistByobjID(entity_id, pair_list, pair_len);
for(vector<int>::iterator itr_pre = in_edge_pre_id.begin(); itr_pre != in_edge_pre_id.end(); itr_pre++)
{
int pre_id = (*itr_pre);
exist_preid = Util::bsearch_preid_uporder(pre_id, pair_list, pair_len);
if(!exist_preid)
{
break;
}
}
delete[] pair_list;
}
if(exist_preid && !out_edge_pre_id.empty())
{
(this->kvstore)->getpreIDobjIDlistBysubID(entity_id, pair_list, pair_len);
for(vector<int>::iterator itr_pre = out_edge_pre_id.begin(); itr_pre != out_edge_pre_id.end(); itr_pre++)
{
int pre_id = (*itr_pre);
exist_preid = Util::bsearch_preid_uporder(pre_id, pair_list, pair_len);
if(!exist_preid)
{
break;
}
}
delete[] pair_list;
}
//result sequence is illegal when there exists any missing filter predicate id.
if(!exist_preid)
{
it = this->current_table.erase(it);
}
else
{
it++;
}
}
}
}
//BETTER?:merge with literal_edge_filter?
//this only consider subject constant neighbors, while the latter also
//consider constant object neighbors(literal), as well as entities
//neighbors.
//(only in objects, no constant neighbors are called free, dealed in join)
//
//NOTICE:not only literals, but also entities may be added here!!!
//(candidates already contain all possible entities, and entities
//produced here may not be ok!)
//add literal candidates to these variables' candidate list
//which may include literal results.
void
Join::add_literal_candidate()
{
//Util::logging("IN add_literal_candidate");
//
// deal with literal variable candidate list.
// because we do not insert any literal elements into VSTree, we can not retrieve them from VSTree.
// for these variable which may include some literal results, we should add all possible literal candidates to the candidate list.
for(int i = 0; i < this->var_num; i++)
{
//debug
//{
// stringstream _ss;
// _ss << "var[" << i << "]\t";
// if (this->basic_query->isLiteralVariable(i))
// {
// _ss << "may have literal result.";
// }
// else
// {
// _ss << "do not have literal result.";
// }
// _ss << endl;
// //Util::logging(_ss.str());
//}
if(!this->basic_query->isLiteralVariable(i))
{
// if this variable is not literal variable, we can assume that its literal candidates have been added.
this->basic_query->setAddedLiteralCandidate(i);
continue;
}
// for these literal variable without any linking entities(we call free literal variable),
// we will add their literal candidates when join-step.
if(this->basic_query->isFreeLiteralVariable(i))
{
continue;
}
int var_id = i;
int var_degree = this->basic_query->getVarDegree(var_id);
IDList literal_candidate_list;
bool flag = false;
// intersect each edge's literal candidate.
for(int j = 0; j < var_degree; j ++)
{
int neighbor_id = this->basic_query->getEdgeNeighborID(var_id, j);
int predicate_id = this->basic_query->getEdgePreID(var_id, j);
int triple_id = this->basic_query->getEdgeID(var_id, j);
Triple triple = this->basic_query->getTriple(triple_id);
string neighbor_name = triple.subject;
IDList this_edge_literal_list;
// if the neighbor of this edge is an entity, we can add all literals which has an exact predicate edge linking to this entity.
if(neighbor_id == -1)
{
int subject_id = (this->kvstore)->getIDByEntity(neighbor_name);
int* object_list = NULL;
int object_list_len = 0;
(this->kvstore)->getobjIDlistBysubIDpreID(subject_id, predicate_id, object_list, object_list_len);
this_edge_literal_list.unionList(object_list, object_list_len);
delete []object_list;
}
// if the neighbor of this edge is variable, then the neighbor variable can not have any literal results,
// we should add literals when join these two variables, see the Database::join function for details.
// deprecated...
// if the neighbor of this edge is variable, we should add all this neighbor variable's candidate entities' neighbor literal,
// which has one corresponding predicate edge linking to this variable.
else
{
continue;
/*
IDList& neighbor_candidate_list = this->basic_query->getCandidateList(neighbor_id);
int neighbor_candidate_list_size = neighbor_candidate_list.size();
for (int k = 0;k < neighbor_candidate_list_size; k ++)
{
int subject_id = neighbor_candidate_list.getID(k);
int* object_list = NULL;
int object_list_len = 0;
(this->kvstore)->getobjIDlistBysubIDpreID(subject_id, predicate_id, object_list, object_list_len);
this_edge_literal_list.unionList(object_list, object_list_len);
delete []object_list;
}
*/
}
if(!flag)
{
flag = true;
literal_candidate_list.unionList(this_edge_literal_list);
}
else
{
literal_candidate_list.intersectList(this_edge_literal_list);
}
}
// add the literal_candidate_list to the original candidate list.
IDList& origin_candidate_list = this->basic_query->getCandidateList(var_id);
int origin_candidate_list_len = origin_candidate_list.size();
origin_candidate_list.unionList(literal_candidate_list);
int after_add_literal_candidate_list_len = origin_candidate_list.size();
// this variable's literal candidates have been added.
this->basic_query->setAddedLiteralCandidate(var_id);
//debug
//{
//stringstream _ss;
//_ss << "var[" << var_id << "] candidate list after add literal:\t"
//<< origin_candidate_list_len << "-->" << after_add_literal_candidate_list_len << endl;
/*
for (int i = 0; i < after_add_literal_candidate_list_len; i ++)
{
int candidate_id = origin_candidate_list.getID(i);
string candidate_name;
if (i < origin_candidate_list_len)
{
candidate_name = (this->kvstore)->getEntityByID(origin_candidate_list.getID(i));
}
else
{
candidate_name = (this->kvstore)->getLiteralByID(origin_candidate_list.getID(i));
}
_ss << candidate_name << "(" << candidate_id << ")\t";
}
*/
//Util::logging(_ss.str());
//}
}
//Util::logging("OUT add_literal_candidate");
}