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
|
// -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*-
// vim: ts=8 sw=2 smarttab ft=cpp
/*
* Ceph - scalable distributed file system
*
* Copyright (C) 2018 Red Hat, Inc.
*
* 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.
*
*/
#ifndef RGW_DMCLOCK_ASYNC_SCHEDULER_H
#define RGW_DMCLOCK_ASYNC_SCHEDULER_H
#include "common/async/completion.h"
#include <boost/asio.hpp>
#include "rgw_dmclock_scheduler.h"
#include "rgw_dmclock_scheduler_ctx.h"
namespace rgw::dmclock {
namespace async = ceph::async;
/*
* A dmclock request scheduling service for use with boost::asio.
*
* An asynchronous dmclock priority queue, where scheduled requests complete
* on a boost::asio executor.
*/
class AsyncScheduler : public md_config_obs_t, public Scheduler {
public:
template <typename ...Args> // args forwarded to PullPriorityQueue ctor
AsyncScheduler(CephContext *cct, boost::asio::io_context& context,
GetClientCounters&& counters, md_config_obs_t *observer,
Args&& ...args);
~AsyncScheduler();
using executor_type = boost::asio::io_context::executor_type;
/// return the default executor for async_request() callbacks
executor_type get_executor() noexcept {
return timer.get_executor();
}
/// submit an async request for dmclock scheduling. the given completion
/// handler will be invoked with (error_code, PhaseType) when the request
/// is ready or canceled. on success, this grants a throttle unit that must
/// be returned with a call to request_complete()
template <typename CompletionToken>
auto async_request(const client_id& client, const ReqParams& params,
const Time& time, Cost cost, CompletionToken&& token);
/// returns a throttle unit granted by async_request()
void request_complete() override;
/// cancel all queued requests, invoking their completion handlers with an
/// operation_aborted error and default-constructed result
void cancel();
/// cancel all queued requests for a given client, invoking their completion
/// handler with an operation_aborted error and default-constructed result
void cancel(const client_id& client);
const char** get_tracked_conf_keys() const override;
void handle_conf_change(const ConfigProxy& conf,
const std::set<std::string>& changed) override;
private:
int schedule_request_impl(const client_id& client, const ReqParams& params,
const Time& time, const Cost& cost,
optional_yield yield_ctx) override;
static constexpr bool IsDelayed = false;
using Queue = crimson::dmclock::PullPriorityQueue<client_id, Request, IsDelayed>;
using RequestRef = typename Queue::RequestRef;
Queue queue; //< dmclock priority queue
using Signature = void(boost::system::error_code, PhaseType);
using Completion = async::Completion<Signature, async::AsBase<Request>>;
using Clock = ceph::coarse_real_clock;
using Timer = boost::asio::basic_waitable_timer<Clock,
boost::asio::wait_traits<Clock>, executor_type>;
Timer timer; //< timer for the next scheduled request
CephContext *const cct;
md_config_obs_t *const observer; //< observer to update ClientInfoFunc
GetClientCounters counters; //< provides per-client perf counters
/// max request throttle
std::atomic<int64_t> max_requests;
std::atomic<int64_t> outstanding_requests = 0;
/// set a timer to process the next request
void schedule(const Time& time);
/// process ready requests, then schedule the next pending request
void process(const Time& now);
};
template <typename ...Args>
AsyncScheduler::AsyncScheduler(CephContext *cct, boost::asio::io_context& context,
GetClientCounters&& counters,
md_config_obs_t *observer, Args&& ...args)
: queue(std::forward<Args>(args)...),
timer(context), cct(cct), observer(observer),
counters(std::move(counters)),
max_requests(cct->_conf.get_val<int64_t>("rgw_max_concurrent_requests"))
{
if (max_requests <= 0) {
max_requests = std::numeric_limits<int64_t>::max();
}
if (observer) {
cct->_conf.add_observer(this);
}
}
template <typename CompletionToken>
auto AsyncScheduler::async_request(const client_id& client,
const ReqParams& params,
const Time& time, Cost cost,
CompletionToken&& token)
{
boost::asio::async_completion<CompletionToken, Signature> init(token);
auto ex1 = get_executor();
auto& handler = init.completion_handler;
// allocate the Request and add it to the queue
auto completion = Completion::create(ex1, std::move(handler),
Request{client, time, cost});
// cast to unique_ptr<Request>
auto req = RequestRef{std::move(completion)};
int r = queue.add_request(std::move(req), client, params, time, cost);
if (r == 0) {
// schedule an immediate call to process() on the executor
schedule(crimson::dmclock::TimeZero);
if (auto c = counters(client)) {
c->inc(queue_counters::l_qlen);
c->inc(queue_counters::l_cost, cost);
}
} else {
// post the error code
boost::system::error_code ec(r, boost::system::system_category());
// cast back to Completion
auto completion = static_cast<Completion*>(req.release());
async::post(std::unique_ptr<Completion>{completion},
ec, PhaseType::priority);
if (auto c = counters(client)) {
c->inc(queue_counters::l_limit);
c->inc(queue_counters::l_limit_cost, cost);
}
}
return init.result.get();
}
class SimpleThrottler : public md_config_obs_t, public dmclock::Scheduler {
public:
SimpleThrottler(CephContext *cct) :
max_requests(cct->_conf.get_val<int64_t>("rgw_max_concurrent_requests")),
counters(cct, "simple-throttler")
{
if (max_requests <= 0) {
max_requests = std::numeric_limits<int64_t>::max();
}
cct->_conf.add_observer(this);
}
const char** get_tracked_conf_keys() const override {
static const char* keys[] = { "rgw_max_concurrent_requests", nullptr };
return keys;
}
void handle_conf_change(const ConfigProxy& conf,
const std::set<std::string>& changed) override
{
if (changed.count("rgw_max_concurrent_requests")) {
auto new_max = conf.get_val<int64_t>("rgw_max_concurrent_requests");
max_requests = new_max > 0 ? new_max : std::numeric_limits<int64_t>::max();
}
}
void request_complete() override {
--outstanding_requests;
if (auto c = counters();
c != nullptr) {
c->inc(throttle_counters::l_outstanding, -1);
}
}
private:
int schedule_request_impl(const client_id&, const ReqParams&,
const Time&, const Cost&,
optional_yield) override {
if (outstanding_requests++ >= max_requests) {
if (auto c = counters();
c != nullptr) {
c->inc(throttle_counters::l_outstanding);
c->inc(throttle_counters::l_throttle);
}
return -EAGAIN;
}
return 0 ;
}
std::atomic<int64_t> max_requests;
std::atomic<int64_t> outstanding_requests = 0;
ThrottleCounters counters;
};
} // namespace rgw::dmclock
#endif /* RGW_DMCLOCK_ASYNC_SCHEDULER_H */
|
