summaryrefslogtreecommitdiffstats
path: root/include/framework/structure/MutableBuffer.h
blob: a70b86b0782e704f0d4db236338c22278e23d03e (plain)
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
/*
 * include/framework/structure/MutableBuffer.h
 *
 * Copyright (C) 2023 Douglas B. Rumbaugh <drumbaugh@psu.edu> 
 *                    Dong Xie <dongx@psu.edu>
 *
 * All rights reserved. Published under the Modified BSD License.
 *
 */
#pragma once

#include <cstdlib>
#include <atomic>
#include <condition_variable>
#include <cassert>
#include <numeric>
#include <algorithm>
#include <type_traits>

#include "psu-util/alignment.h"
#include "util/bf_config.h"
#include "psu-ds/BloomFilter.h"
#include "psu-ds/Alias.h"
#include "psu-util/timer.h"
#include "framework/interface/Record.h"

using psudb::CACHELINE_SIZE;

namespace de {

template <RecordInterface R>
class MutableBuffer {
public:
    MutableBuffer(size_t capacity, size_t max_tombstone_cap)
    : m_cap(capacity), m_tombstone_cap(max_tombstone_cap), m_reccnt(0)
    , m_tombstonecnt(0), m_weight(0), m_max_weight(0) {
        m_data = (Wrapped<R>*) psudb::sf_aligned_alloc(CACHELINE_SIZE, capacity*sizeof(Wrapped<R>));
        m_merge_data = (Wrapped<R>*) psudb::sf_aligned_alloc(CACHELINE_SIZE, capacity*sizeof(Wrapped<R>));
        m_tombstone_filter = nullptr;
        if (max_tombstone_cap > 0) {
            m_tombstone_filter = new psudb::BloomFilter<R>(BF_FPR, max_tombstone_cap, BF_HASH_FUNCS);
        }

        m_refcnt.store(0);
    }

    ~MutableBuffer() {
        assert(m_refcnt.load() == 0);

        if (m_data) free(m_data);
        if (m_tombstone_filter) delete m_tombstone_filter;
        if (m_merge_data) free(m_merge_data);
    }

    template <typename R_ = R>
    int append(const R &rec, bool tombstone=false) {
        if (tombstone && m_tombstonecnt + 1 > m_tombstone_cap) return 0;

        int32_t pos = 0;
        if ((pos = try_advance_tail()) == -1) return 0;

        Wrapped<R> wrec;
        wrec.rec = rec;
        wrec.header = 0;
        if (tombstone) wrec.set_tombstone();

        m_data[pos] = wrec;
        m_data[pos].header |= (pos << 2);

        if (tombstone) {
            m_tombstonecnt.fetch_add(1);
            if (m_tombstone_filter) m_tombstone_filter->insert(rec);
        }

        if constexpr (WeightedRecordInterface<R_>) {
            m_weight.fetch_add(rec.weight);
            double old = m_max_weight.load();
            while (old < rec.weight) {
                m_max_weight.compare_exchange_strong(old, rec.weight);
                old = m_max_weight.load();
            }
        } else {
            m_weight.fetch_add(1);
        }

        return 1;     
    }

    bool truncate() {
        m_tombstonecnt.store(0);
        m_reccnt.store(0);
        m_weight.store(0);
        m_max_weight.store(0);
        if (m_tombstone_filter) m_tombstone_filter->clear();

        return true;
    }

    size_t get_record_count() {
        return m_reccnt;
    }
    
    size_t get_capacity() {
        return m_cap;
    }

    bool is_full() {
        return m_reccnt == m_cap;
    }

    size_t get_tombstone_count() {
        return m_tombstonecnt.load();
    }

    bool delete_record(const R& rec) {
        auto offset = 0;
        while (offset < m_reccnt.load()) {
            if (m_data[offset].rec == rec) {
                m_data[offset].set_delete();
                return true;
            }
            offset++;
        }

        return false;
    }

    bool check_tombstone(const R& rec) {
        if (m_tombstone_filter && !m_tombstone_filter->lookup(rec)) return false;

        auto offset = 0;
        while (offset < m_reccnt.load()) {
            if (m_data[offset].rec == rec && m_data[offset].is_tombstone()) {
                return true;
            }
            offset++;;
        }
        return false;
    }

    size_t get_memory_usage() {
        return m_cap * sizeof(R);
    }

    size_t get_aux_memory_usage() {
        return m_tombstone_filter->get_memory_usage();
    }

    size_t get_tombstone_capacity() {
        return m_tombstone_cap;
    }

    double get_total_weight() {
        return m_weight.load();
    }

    Wrapped<R> *get_data() {
        return m_data;
    }

    double get_max_weight() {
        return m_max_weight;
    }

    /*
     * This operation assumes that no other threads have write access
     * to the buffer. This will be the case in normal operation, at
     * present, but may change (in which case this approach will need
     * to be adjusted). Other threads having read access is perfectly
     * acceptable, however.
     */
    bool start_merge() {
        memcpy(m_merge_data, m_data, sizeof(Wrapped<R>) * m_reccnt.load());
        return true;
    }

    /*
     * Concurrency-related operations
     */
    bool take_reference() {
        m_refcnt.fetch_add(1);
        return true;
    }

    bool release_reference() {
        assert(m_refcnt > 0);
        m_refcnt.fetch_add(-1);
        return true;
    }

    size_t get_reference_count() {
        return m_refcnt.load();
    }

private:
    int32_t try_advance_tail() {
        size_t new_tail = m_reccnt.fetch_add(1);

        if (new_tail < m_cap) return new_tail;
        else return -1;
    }

    size_t m_cap;
    size_t m_tombstone_cap;
    
    Wrapped<R>* m_data;
    Wrapped<R>* m_merge_data;

    psudb::BloomFilter<R>* m_tombstone_filter;

    alignas(64) std::atomic<size_t> m_tombstonecnt;
    alignas(64) std::atomic<uint32_t> m_reccnt;
    alignas(64) std::atomic<double> m_weight;
    alignas(64) std::atomic<double> m_max_weight;

    alignas(64) std::atomic<size_t> m_refcnt;
};

}