lbug 0.21.2

An in-process property graph database management system built for query speed and scalability
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230

#include "processor/operator/intersect/intersect.h"

#include <algorithm>
#include <cstring>

#include "function/hash/hash_functions.h"
#include "processor/operator/intersect/intersect_kernels.h"
#include "processor/result/factorized_table.h"

using namespace lbug::common;

namespace lbug {
namespace processor {

std::string IntersectPrintInfo::toString() const {
    std::string result = "Key: ";
    result += key->toString();
    return result;
}

void Intersect::initLocalStateInternal(ResultSet* resultSet, ExecutionContext* /*context*/) {
    outKeyVector = resultSet->getValueVector(outputDataPos);
    for (auto& dataInfo : intersectDataInfos) {
        probeKeyVectors.push_back(resultSet->getValueVector(dataInfo.keyDataPos));
        std::vector<uint32_t> columnIdxesToScanFrom;
        std::vector<ValueVector*> vectorsToReadInto;
        for (auto i = 0u; i < dataInfo.payloadsDataPos.size(); i++) {
            auto vector = resultSet->getValueVector(dataInfo.payloadsDataPos[i]);
            // Always skip the first two columns in the fTable: build key and intersect key.
            // TODO(Guodong): Remove this assumption so that keys can be stored in any order. Change
            // mapping logic too so that we don't need to maintain this order explicitly.
            columnIdxesToScanFrom.push_back(i + 2);
            vectorsToReadInto.push_back(vector.get());
        }
        payloadColumnIdxesToScanFrom.push_back(columnIdxesToScanFrom);
        payloadVectorsToScanInto.push_back(std::move(vectorsToReadInto));
    }
    for (auto& sharedHT : sharedHTs) {
        intersectSelVectors.push_back(std::make_unique<SelectionVector>(DEFAULT_VECTOR_CAPACITY));
        isIntersectListAFlatValue.push_back(
            sharedHT->getHashTable()->getTableSchema()->getColumn(1)->isFlat());
    }
}

void Intersect::probeHTs() {
    std::vector<std::vector<overflow_value_t>> flatTuples(probeKeyVectors.size());
    hash_t hashVal = 0;
    for (auto i = 0u; i < probeKeyVectors.size(); i++) {
        DASSERT(probeKeyVectors[i]->state->isFlat());
        probedFlatTuples[i].clear();
        if (sharedHTs[i]->getHashTable()->getNumEntries() == 0) {
            continue;
        }
        auto key =
            probeKeyVectors[i]->getValue<nodeID_t>(probeKeyVectors[i]->state->getSelVector()[0]);
        function::Hash::operation<nodeID_t>(key, false, hashVal);
        auto flatTuple = sharedHTs[i]->getHashTable()->getTupleForHash(hashVal);
        while (flatTuple) {
            nodeID_t tupleKey;
            memcpy(&tupleKey, flatTuple, sizeof(nodeID_t));
            if (tupleKey == key) {
                probedFlatTuples[i].push_back(flatTuple);
            }
            flatTuple = sharedHTs[i]->getHashTable()->getPrevTupleValue(flatTuple);
        }
    }
}

void Intersect::twoWayIntersect(nodeID_t* leftNodeIDs, SelectionVector& lSelVector,
    nodeID_t* rightNodeIDs, SelectionVector& rSelVector) {
    DASSERT(lSelVector.getSelSize() <= rSelVector.getSelSize());
    auto leftPositionBuffer = lSelVector.getMutableBuffer();
    auto rightPositionBuffer = rSelVector.getMutableBuffer();
    const auto outputValuePosition =
        intersectNodeIDs(leftNodeIDs, lSelVector.getSelSize(), rightNodeIDs,
            rSelVector.getSelSize(), leftPositionBuffer.data(), rightPositionBuffer.data());
    lSelVector.setToFiltered(outputValuePosition);
    rSelVector.setToFiltered(outputValuePosition);
}

static std::vector<overflow_value_t> fetchListsToIntersectFromTuples(
    const std::vector<uint8_t*>& tuples, const std::vector<bool>& isFlatValue) {
    std::vector<overflow_value_t> listsToIntersect(tuples.size());
    for (auto i = 0u; i < tuples.size(); i++) {
        if (isFlatValue[i]) {
            listsToIntersect[i] =
                overflow_value_t{1 /* numElements */, tuples[i] + sizeof(nodeID_t)};
        } else {
            // Tuples are packed without alignment padding; use memcpy.
            memcpy(&listsToIntersect[i], tuples[i] + sizeof(nodeID_t), sizeof(overflow_value_t));
        }
    }
    return listsToIntersect;
}

static std::vector<uint32_t> swapSmallestListToFront(std::vector<overflow_value_t>& lists) {
    DASSERT(lists.size() >= 2);
    std::vector<uint32_t> listIdxes(lists.size());
    iota(listIdxes.begin(), listIdxes.end(), 0);
    uint32_t smallestListIdx = 0;
    for (auto i = 1u; i < lists.size(); i++) {
        if (lists[i].numElements < lists[smallestListIdx].numElements) {
            smallestListIdx = i;
        }
    }
    if (smallestListIdx != 0) {
        std::swap(lists[smallestListIdx], lists[0]);
        std::swap(listIdxes[smallestListIdx], listIdxes[0]);
    }
    return listIdxes;
}

static void sliceSelVectors(const std::vector<SelectionVector*>& selVectorsToSlice,
    SelectionVector& slicer) {
    for (auto selVec : selVectorsToSlice) {
        for (auto i = 0u; i < slicer.getSelSize(); i++) {
            auto pos = slicer[i];
            auto buffer = selVec->getMutableBuffer();
            buffer[i] = selVec->operator[](pos);
        }
        selVec->setToFiltered(slicer.getSelSize());
    }
}

void Intersect::intersectLists(const std::vector<overflow_value_t>& listsToIntersect) {
    if (listsToIntersect[0].numElements == 0) {
        outKeyVector->state->getSelVectorUnsafe().setSelSize(0);
        return;
    }
    DASSERT(listsToIntersect[0].numElements <= DEFAULT_VECTOR_CAPACITY);
    memcpy(outKeyVector->getData(), listsToIntersect[0].value,
        listsToIntersect[0].numElements * sizeof(nodeID_t));
    SelectionVector lSelVector(listsToIntersect[0].numElements);
    lSelVector.setSelSize(listsToIntersect[0].numElements);
    std::vector<SelectionVector*> selVectorsForIntersectedLists;
    intersectSelVectors[0]->setToUnfiltered(listsToIntersect[0].numElements);
    selVectorsForIntersectedLists.push_back(intersectSelVectors[0].get());
    // Reusable aligned buffer for the right-hand adjacency list. Lists live in packed
    // factorized-table/unflat storage without alignment padding, so the raw value pointer
    // may be misaligned for nodeID_t; copy it into aligned storage before intersecting.
    std::vector<nodeID_t> alignedRightList;
    for (auto i = 0u; i < listsToIntersect.size() - 1; i++) {
        intersectSelVectors[i + 1]->setToUnfiltered(listsToIntersect[i + 1].numElements);
        alignedRightList.resize(listsToIntersect[i + 1].numElements);
        if (!alignedRightList.empty()) {
            memcpy(alignedRightList.data(), listsToIntersect[i + 1].value,
                listsToIntersect[i + 1].numElements * sizeof(nodeID_t));
        }
        twoWayIntersect((nodeID_t*)outKeyVector->getData(), lSelVector, alignedRightList.data(),
            *intersectSelVectors[i + 1]);
        // Here we need to slice all selVectors that have been previously intersected, as all these
        // lists need to be selected synchronously to read payloads correctly.
        sliceSelVectors(selVectorsForIntersectedLists, lSelVector);
        lSelVector.setToUnfiltered();
        selVectorsForIntersectedLists.push_back(intersectSelVectors[i + 1].get());
    }
    outKeyVector->state->getSelVectorUnsafe().setSelSize(lSelVector.getSelSize());
}

void Intersect::populatePayloads(const std::vector<uint8_t*>& tuples,
    const std::vector<uint32_t>& listIdxes) {
    for (auto i = 0u; i < listIdxes.size(); i++) {
        auto listIdx = listIdxes[i];
        sharedHTs[listIdx]->getHashTable()->getFactorizedTable()->lookup(
            payloadVectorsToScanInto[listIdx], intersectSelVectors[i].get(),
            payloadColumnIdxesToScanFrom[listIdx], tuples[listIdx]);
    }
}

bool Intersect::hasNextTuplesToIntersect() {
    tupleIdxPerBuildSide[carryBuildSideIdx]++;
    if (tupleIdxPerBuildSide[carryBuildSideIdx] == probedFlatTuples[carryBuildSideIdx].size()) {
        if (carryBuildSideIdx == 0) {
            return false;
        }
        tupleIdxPerBuildSide[carryBuildSideIdx] = 0;
        carryBuildSideIdx--;
        if (!hasNextTuplesToIntersect()) {
            return false;
        }
        carryBuildSideIdx++;
    }
    return true;
}

bool Intersect::getNextTuplesInternal(ExecutionContext* context) {
    do {
        while (carryBuildSideIdx == -1u) {
            if (!children[0]->getNextTuple(context)) {
                return false;
            }
            // For each build side, probe its HT and return a vector of matched flat tuples.
            probeHTs();
            auto maxNumTuplesToIntersect = 1u;
            for (auto i = 0u; i < getNumBuilds(); i++) {
                maxNumTuplesToIntersect *= probedFlatTuples[i].size();
            }
            if (maxNumTuplesToIntersect == 0) {
                // Skip if any build side has no matches.
                continue;
            }
            carryBuildSideIdx = getNumBuilds() - 1;
            std::fill(tupleIdxPerBuildSide.begin(), tupleIdxPerBuildSide.end(), 0);
        }
        // Cartesian product of all flat tuples probed from all build sides.
        // Notice: when there are large adjacency lists in the build side, which means the list is
        // too large to fit in a single ValueVector, we end up chunking the list as multiple tuples
        // in FTable. Thus, when performing the intersection, we need to perform cartesian product
        // between all flat tuples probed from all build sides.
        std::vector<uint8_t*> flatTuplesToIntersect(getNumBuilds());
        for (auto i = 0u; i < getNumBuilds(); i++) {
            flatTuplesToIntersect[i] = probedFlatTuples[i][tupleIdxPerBuildSide[i]];
        }
        auto listsToIntersect =
            fetchListsToIntersectFromTuples(flatTuplesToIntersect, isIntersectListAFlatValue);
        auto listIdxes = swapSmallestListToFront(listsToIntersect);
        intersectLists(listsToIntersect);
        if (outKeyVector->state->getSelVector().getSelSize() != 0) {
            populatePayloads(flatTuplesToIntersect, listIdxes);
        }
        if (!hasNextTuplesToIntersect()) {
            carryBuildSideIdx = -1u;
        }
    } while (outKeyVector->state->getSelVector().getSelSize() == 0);
    metrics->numOutputTuple.increase(outKeyVector->state->getSelVector().getSelSize());
    return true;
}

} // namespace processor
} // namespace lbug