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
|
// -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*-
// vim: ts=8 sw=2 smarttab
/*
* Ceph distributed storage system
*
* Copyright (C) 2013 Cloudwatt <libre.licensing@cloudwatt.com>
*
* Author: Loic Dachary <loic@dachary.org>
*
* This library is free software; you can redistribute it and/or
* modify it under the terms of the GNU Lesser General Public
* License as published by the Free Software Foundation; either
* version 2.1 of the License, or (at your option) any later version.
*
*/
#ifndef CEPH_ERASURE_CODE_EXAMPLE_H
#define CEPH_ERASURE_CODE_EXAMPLE_H
#include <unistd.h>
#include <errno.h>
#include <algorithm>
#include <sstream>
#include "crush/CrushWrapper.h"
#include "osd/osd_types.h"
#include "erasure-code/ErasureCode.h"
#define FIRST_DATA_CHUNK 0
#define SECOND_DATA_CHUNK 1
#define DATA_CHUNKS 2u
#define CODING_CHUNK 2
#define CODING_CHUNKS 1u
#define MINIMUM_TO_RECOVER 2u
class ErasureCodeExample final : public ErasureCode {
public:
~ErasureCodeExample() override {}
int create_rule(const string &name,
CrushWrapper &crush,
ostream *ss) const override {
return crush.add_simple_rule(name, "default", "host", "",
"indep", pg_pool_t::TYPE_ERASURE, ss);
}
int minimum_to_decode_with_cost(const set<int> &want_to_read,
const map<int, int> &available,
set<int> *minimum) override {
//
// If one chunk is more expensive to fetch than the others,
// recover it instead. For instance, if the cost reflects the
// time it takes for a chunk to be retrieved from a remote
// OSD and if CPU is cheap, it could make sense to recover
// instead of fetching the chunk.
//
map<int, int> c2c(available);
if (c2c.size() > DATA_CHUNKS) {
if (c2c[FIRST_DATA_CHUNK] > c2c[SECOND_DATA_CHUNK] &&
c2c[FIRST_DATA_CHUNK] > c2c[CODING_CHUNK])
c2c.erase(FIRST_DATA_CHUNK);
else if(c2c[SECOND_DATA_CHUNK] > c2c[FIRST_DATA_CHUNK] &&
c2c[SECOND_DATA_CHUNK] > c2c[CODING_CHUNK])
c2c.erase(SECOND_DATA_CHUNK);
else if(c2c[CODING_CHUNK] > c2c[FIRST_DATA_CHUNK] &&
c2c[CODING_CHUNK] > c2c[SECOND_DATA_CHUNK])
c2c.erase(CODING_CHUNK);
}
set <int> available_chunks;
for (map<int, int>::const_iterator i = c2c.begin();
i != c2c.end();
++i)
available_chunks.insert(i->first);
return _minimum_to_decode(want_to_read, available_chunks, minimum);
}
unsigned int get_chunk_count() const override {
return DATA_CHUNKS + CODING_CHUNKS;
}
unsigned int get_data_chunk_count() const override {
return DATA_CHUNKS;
}
unsigned int get_chunk_size(unsigned int object_size) const override {
return ( object_size / DATA_CHUNKS ) + 1;
}
int encode(const set<int> &want_to_encode,
const bufferlist &in,
map<int, bufferlist> *encoded) override {
//
// make sure all data chunks have the same length, allocating
// padding if necessary.
//
unsigned int chunk_length = get_chunk_size(in.length());
bufferlist out(in);
unsigned int width = get_chunk_count() * get_chunk_size(in.length());
bufferptr pad(width - in.length());
pad.zero(0, get_data_chunk_count());
out.push_back(pad);
//
// compute the coding chunk with first chunk ^ second chunk
//
char *p = out.c_str();
for (unsigned i = 0; i < chunk_length; i++)
p[i + CODING_CHUNK * chunk_length] =
p[i + FIRST_DATA_CHUNK * chunk_length] ^
p[i + SECOND_DATA_CHUNK * chunk_length];
//
// populate the bufferlist with bufferptr pointing
// to chunk boundaries
//
const bufferptr &ptr = out.front();
for (set<int>::iterator j = want_to_encode.begin();
j != want_to_encode.end();
++j) {
bufferlist tmp;
bufferptr chunk(ptr, (*j) * chunk_length, chunk_length);
tmp.push_back(chunk);
tmp.claim_append((*encoded)[*j]);
(*encoded)[*j].swap(tmp);
}
return 0;
}
int encode_chunks(const set<int> &want_to_encode,
map<int, bufferlist> *encoded) override {
ceph_abort();
return 0;
}
int _decode(const set<int> &want_to_read,
const map<int, bufferlist> &chunks,
map<int, bufferlist> *decoded) {
//
// All chunks have the same size
//
unsigned chunk_length = (*chunks.begin()).second.length();
for (set<int>::iterator i = want_to_read.begin();
i != want_to_read.end();
++i) {
if (chunks.find(*i) != chunks.end()) {
//
// If the chunk is available, just copy the bufferptr pointer
// to the decoded argument.
//
(*decoded)[*i] = chunks.find(*i)->second;
} else if(chunks.size() != 2) {
//
// If a chunk is missing and there are not enough chunks
// to recover, abort.
//
return -ERANGE;
} else {
//
// No matter what the missing chunk is, XOR of the other
// two recovers it.
//
map<int, bufferlist>::const_iterator k = chunks.begin();
const char *a = k->second.front().c_str();
++k;
const char *b = k->second.front().c_str();
bufferptr chunk(chunk_length);
char *c = chunk.c_str();
for (unsigned j = 0; j < chunk_length; j++) {
c[j] = a[j] ^ b[j];
}
bufferlist tmp;
tmp.append(chunk);
tmp.claim_append((*decoded)[*i]);
(*decoded)[*i].swap(tmp);
}
}
return 0;
}
int decode_chunks(const set<int> &want_to_read,
const map<int, bufferlist> &chunks,
map<int, bufferlist> *decoded) override {
ceph_abort();
return 0;
}
const vector<int> &get_chunk_mapping() const override {
static vector<int> mapping;
return mapping;
}
};
#endif
|
