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
|
/*
* include/framework/Scheduler.h
*
* Copyright (C) 2023 Douglas Rumbaugh <drumbaugh@psu.edu>
* Dong Xie <dongx@psu.edu>
*
* All rights reserved. Published under the Modified BSD License.
*
*/
#pragma once
#include <vector>
#include <memory>
#include <queue>
#include <thread>
#include <condition_variable>
#include "util/types.h"
#include "framework/ShardInterface.h"
#include "framework/QueryInterface.h"
#include "framework/RecordInterface.h"
#include "framework/MutableBuffer.h"
#include "framework/Configuration.h"
#include "framework/ExtensionStructure.h"
namespace de {
template <RecordInterface R, ShardInterface S, QueryInterface Q, LayoutPolicy L>
class Scheduler {
typedef ExtensionStructure<R, S, Q, L> Structure;
typedef MutableBuffer<R> Buffer;
public:
/*
* Memory budget stated in bytes, with 0 meaning unlimited. Likewise, 0 threads means
* unlimited.
*/
Scheduler(size_t memory_budget, size_t thread_cnt)
: m_memory_budget((memory_budget) ? memory_budget : UINT64_MAX)
, m_thread_cnt((thread_cnt) ? thread_cnt : UINT64_MAX)
, m_used_memory(0)
, m_used_threads(0)
, m_shutdown(false)
{
m_sched_thrd = std::thread(&Scheduler::run_scheduler, this);
}
~Scheduler() {
m_shutdown = true;
m_cv.notify_all();
m_sched_thrd.join();
}
bool schedule_merge(Structure *version, MutableBuffer<R> *buffer) {
/*
* temporary hack
*/
pending_version = version;
pending_buffer = buffer;
/*
* Get list of individual level reconstructions that are necessary
* for completing the overall merge
*/
std::vector<MergeTask> merges = version->get_merge_tasks(buffer->get_record_count());
/*
* Schedule the merge tasks (FIXME: currently this just
* executes them sequentially in a blocking fashion)
*/
for (ssize_t i=0; i<merges.size(); i++) {
merges[i].m_timestamp = m_timestamp.fetch_add(1);
m_merge_queue_lock.lock();
m_merge_queue.push(merges[i]);
m_merge_queue_lock.unlock();
}
MergeTask buffer_merge;
buffer_merge.m_source_level = -1;
buffer_merge.m_target_level = 0;
buffer_merge.m_size = buffer->get_record_count() * sizeof(R) * 2;
buffer_merge.m_timestamp = m_timestamp.fetch_add(1);
m_merge_queue_lock.lock();
m_merge_queue.push(buffer_merge);
m_merge_queue_lock.unlock();
m_cv.notify_all();
do {
std::unique_lock<std::mutex> merge_cv_lock(m_merge_cv_lock);
m_merge_cv.wait(merge_cv_lock);
} while (m_merge_queue.size() > 0);
assert(version->get_levels()[version->get_levels().size() - 1]->get_shard(0)->get_tombstone_count() == 0);
return true;
}
private:
size_t get_timestamp() {
auto ts = m_timestamp.fetch_add(1);
return ts;
}
void schedule_next_task() {
m_merge_queue_lock.lock();
auto task = m_merge_queue.top();
m_merge_queue.pop();
m_merge_queue_lock.unlock();
if (task.m_source_level == -1 && task.m_target_level == 0) {
run_buffer_merge(pending_buffer, pending_version);
} else {
run_merge(task, pending_version);
}
if (m_merge_queue.size() == 0) {
m_merge_cv.notify_all();
}
}
void run_merge(MergeTask task, Structure *version) {
version->merge_levels(task.m_target_level, task.m_source_level);
if (!version->validate_tombstone_proportion(task.m_target_level)) {
auto tasks = version->get_merge_tasks(task.m_target_level);
/*
* Schedule the merge tasks (FIXME: currently this just
* executes them sequentially in a blocking fashion)
*/
for (ssize_t i=tasks.size()-1; i>=0; i--) {
tasks[i].m_timestamp = m_timestamp.fetch_add(1);
m_merge_queue_lock.lock();
m_merge_queue.push(tasks[i]);
m_merge_queue_lock.unlock();
}
}
}
void run_buffer_merge(Buffer *buffer, Structure *version) {
version->merge_buffer(buffer);
if (!version->validate_tombstone_proportion(0)) {
auto tasks = version->get_merge_tasks_from_level(0);
/*
* Schedule the merge tasks (FIXME: currently this just
* executes them sequentially in a blocking fashion)
*/
for (ssize_t i=tasks.size()-1; i>=0; i--) {
tasks[i].m_timestamp = m_timestamp.fetch_add(1);
m_merge_queue_lock.lock();
m_merge_queue.push(tasks[i]);
m_merge_queue_lock.unlock();
}
}
}
void run_scheduler() {
do {
std::unique_lock<std::mutex> cv_lock(m_cv_lock);
m_cv.wait(cv_lock);
while (m_merge_queue.size() > 0 && m_used_threads < m_thread_cnt) {
schedule_next_task();
}
cv_lock.unlock();
} while(!m_shutdown);
}
size_t m_memory_budget;
size_t m_thread_cnt;
Buffer *pending_buffer;
Structure *pending_version;
alignas(64) std::atomic<size_t> m_used_memory;
alignas(64) std::atomic<size_t> m_used_threads;
alignas(64) std::atomic<size_t> m_timestamp;
std::priority_queue<MergeTask, std::vector<MergeTask>, std::greater<MergeTask>> m_merge_queue;
std::mutex m_merge_queue_lock;
std::mutex m_cv_lock;
std::condition_variable m_cv;
std::mutex m_merge_cv_lock;
std::condition_variable m_merge_cv;
std::thread m_sched_thrd;
bool m_shutdown;
};
}
|