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
|
// -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*-
// vim: ts=8 sw=2 smarttab
/*
* Ceph - scalable distributed file system
*
* Copyright (C) 2016 XSky <haomai@xsky.com>
*
* Author: Haomai Wang <haomaiwang@gmail.com>
*
* This is free software; you can redistribute it and/or
* modify it under the terms of the GNU Lesser General Public
* License version 2.1, as published by the Free Software
* Foundation. See file COPYING.
*
*/
#include <mutex>
#include "include/compat.h"
#include "common/Cond.h"
#include "common/errno.h"
#include "PosixStack.h"
#ifdef HAVE_RDMA
#include "rdma/RDMAStack.h"
#endif
#ifdef HAVE_DPDK
#include "dpdk/DPDKStack.h"
#endif
#include "common/dout.h"
#include "include/ceph_assert.h"
#define dout_subsys ceph_subsys_ms
#undef dout_prefix
#define dout_prefix *_dout << "stack "
std::function<void ()> NetworkStack::add_thread(unsigned i)
{
Worker *w = workers[i];
return [this, w]() {
char tp_name[16];
sprintf(tp_name, "msgr-worker-%u", w->id);
ceph_pthread_setname(pthread_self(), tp_name);
const unsigned EventMaxWaitUs = 30000000;
w->center.set_owner();
ldout(cct, 10) << __func__ << " starting" << dendl;
w->initialize();
w->init_done();
while (!w->done) {
ldout(cct, 30) << __func__ << " calling event process" << dendl;
ceph::timespan dur;
int r = w->center.process_events(EventMaxWaitUs, &dur);
if (r < 0) {
ldout(cct, 20) << __func__ << " process events failed: "
<< cpp_strerror(errno) << dendl;
// TODO do something?
}
w->perf_logger->tinc(l_msgr_running_total_time, dur);
}
w->reset();
w->destroy();
};
}
std::shared_ptr<NetworkStack> NetworkStack::create(CephContext *c, const string &t)
{
if (t == "posix")
return std::make_shared<PosixNetworkStack>(c, t);
#ifdef HAVE_RDMA
else if (t == "rdma")
return std::make_shared<RDMAStack>(c, t);
#endif
#ifdef HAVE_DPDK
else if (t == "dpdk")
return std::make_shared<DPDKStack>(c, t);
#endif
lderr(c) << __func__ << " ms_async_transport_type " << t <<
" is not supported! " << dendl;
ceph_abort();
return nullptr;
}
Worker* NetworkStack::create_worker(CephContext *c, const string &type, unsigned i)
{
if (type == "posix")
return new PosixWorker(c, i);
#ifdef HAVE_RDMA
else if (type == "rdma")
return new RDMAWorker(c, i);
#endif
#ifdef HAVE_DPDK
else if (type == "dpdk")
return new DPDKWorker(c, i);
#endif
lderr(c) << __func__ << " ms_async_transport_type " << type <<
" is not supported! " << dendl;
ceph_abort();
return nullptr;
}
NetworkStack::NetworkStack(CephContext *c, const string &t): type(t), started(false), cct(c)
{
ceph_assert(cct->_conf->ms_async_op_threads > 0);
const int InitEventNumber = 5000;
num_workers = cct->_conf->ms_async_op_threads;
if (num_workers >= EventCenter::MAX_EVENTCENTER) {
ldout(cct, 0) << __func__ << " max thread limit is "
<< EventCenter::MAX_EVENTCENTER << ", switching to this now. "
<< "Higher thread values are unnecessary and currently unsupported."
<< dendl;
num_workers = EventCenter::MAX_EVENTCENTER;
}
for (unsigned i = 0; i < num_workers; ++i) {
Worker *w = create_worker(cct, type, i);
w->center.init(InitEventNumber, i, type);
workers.push_back(w);
}
}
void NetworkStack::start()
{
std::unique_lock<decltype(pool_spin)> lk(pool_spin);
if (started) {
return ;
}
for (unsigned i = 0; i < num_workers; ++i) {
if (workers[i]->is_init())
continue;
std::function<void ()> thread = add_thread(i);
spawn_worker(i, std::move(thread));
}
started = true;
lk.unlock();
for (unsigned i = 0; i < num_workers; ++i)
workers[i]->wait_for_init();
}
Worker* NetworkStack::get_worker()
{
ldout(cct, 30) << __func__ << dendl;
// start with some reasonably large number
unsigned min_load = std::numeric_limits<int>::max();
Worker* current_best = nullptr;
pool_spin.lock();
// find worker with least references
// tempting case is returning on references == 0, but in reality
// this will happen so rarely that there's no need for special case.
for (unsigned i = 0; i < num_workers; ++i) {
unsigned worker_load = workers[i]->references.load();
if (worker_load < min_load) {
current_best = workers[i];
min_load = worker_load;
}
}
pool_spin.unlock();
ceph_assert(current_best);
++current_best->references;
return current_best;
}
void NetworkStack::stop()
{
std::lock_guard<decltype(pool_spin)> lk(pool_spin);
for (unsigned i = 0; i < num_workers; ++i) {
workers[i]->done = true;
workers[i]->center.wakeup();
join_worker(i);
}
started = false;
}
class C_drain : public EventCallback {
Mutex drain_lock;
Cond drain_cond;
unsigned drain_count;
public:
explicit C_drain(size_t c)
: drain_lock("C_drain::drain_lock"),
drain_count(c) {}
void do_request(uint64_t id) override {
Mutex::Locker l(drain_lock);
drain_count--;
if (drain_count == 0) drain_cond.Signal();
}
void wait() {
Mutex::Locker l(drain_lock);
while (drain_count)
drain_cond.Wait(drain_lock);
}
};
void NetworkStack::drain()
{
ldout(cct, 30) << __func__ << " started." << dendl;
pthread_t cur = pthread_self();
pool_spin.lock();
C_drain drain(num_workers);
for (unsigned i = 0; i < num_workers; ++i) {
ceph_assert(cur != workers[i]->center.get_owner());
workers[i]->center.dispatch_event_external(EventCallbackRef(&drain));
}
pool_spin.unlock();
drain.wait();
ldout(cct, 30) << __func__ << " end." << dendl;
}
|
