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
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
base / task / thread_pool / thread_pool_instance.h [blame]
// Copyright 2016 The Chromium Authors
// Use of this source code is governed by a BSD-style license that can be
// found in the LICENSE file.
#ifndef BASE_TASK_THREAD_POOL_THREAD_POOL_INSTANCE_H_
#define BASE_TASK_THREAD_POOL_THREAD_POOL_INSTANCE_H_
#include <memory>
#include <string_view>
#include "base/base_export.h"
#include "base/functional/callback.h"
#include "base/gtest_prod_util.h"
#include "base/task/sequenced_task_runner.h"
#include "base/task/single_thread_task_runner.h"
#include "base/task/single_thread_task_runner_thread_mode.h"
#include "base/task/task_runner.h"
#include "base/task/task_traits.h"
#include "base/time/time.h"
#include "build/build_config.h"
namespace gin {
class V8Platform;
}
namespace content {
// Can't use the FRIEND_TEST_ALL_PREFIXES macro because the test is in a
// different namespace.
class BrowserMainLoopTest_CreateThreadsInSingleProcess_Test;
} // namespace content
namespace base {
class WorkerThreadObserver;
class ThreadPoolTestHelpers;
// Interface for a thread pool and static methods to manage the instance used
// by the thread_pool.h API.
//
// The thread pool doesn't create threads until Start() is called. Tasks can
// be posted at any time but will not run until after Start() is called.
//
// The instance methods of this class are thread-safe unless otherwise noted.
//
// Note: All thread pool users should go through base/task/thread_pool.h instead
// of this interface except for the one callsite per process which manages the
// process's instance.
class BASE_EXPORT ThreadPoolInstance {
public:
struct BASE_EXPORT InitParams {
enum class CommonThreadPoolEnvironment {
// Use the default environment (no environment).
DEFAULT,
#if BUILDFLAG(IS_WIN)
// Place the pool's workers in a COM MTA.
COM_MTA,
#endif // BUILDFLAG(IS_WIN)
};
InitParams(size_t max_num_foreground_threads_in);
InitParams(size_t max_num_foreground_threads_in,
size_t max_num_utility_threads_in);
~InitParams();
// Maximum number of unblocked tasks that can run concurrently in the
// foreground thread group. This is capped at 256 (and should likely not be
// configured anywhere close to this in a browser, approaching that limit is
// most useful on compute farms running tests or compiles in parallel).
size_t max_num_foreground_threads;
// Maximum number of unblocked tasks that can run concurrently in the
// utility thread group.
size_t max_num_utility_threads;
// Whether COM is initialized when running sequenced and parallel tasks.
CommonThreadPoolEnvironment common_thread_pool_environment =
CommonThreadPoolEnvironment::DEFAULT;
// An experiment conducted in July 2019 revealed that on Android, changing
// the reclaim time from 30 seconds to 5 minutes:
// - Reduces jank by 5% at 99th percentile
// - Reduces first input delay by 5% at 99th percentile
// - Reduces input delay by 3% at 50th percentile
// - Reduces navigation to first contentful paint by 2-3% at 25-95th
// percentiles
// On Windows and Mac, we instead see no impact or small regressions.
//
// TODO(scheduler-dev): Conduct experiments to find the optimal value for
// each process type on each platform. In particular, due to regressions at
// high percentiles for *HeartbeatLatencyMicroseconds.Renderer* histograms,
// it was suggested that we might want a different reclaim time in
// renderers. Note that the regression is not present in
// *TaskLatencyMicroseconds.Renderer* histograms.
TimeDelta suggested_reclaim_time =
#if BUILDFLAG(IS_ANDROID)
Minutes(5);
#else
Seconds(30);
#endif
};
// A Scoped(BestEffort)ExecutionFence prevents new tasks of any/BEST_EFFORT
// priority from being scheduled in ThreadPoolInstance within its scope.
// Multiple fences can exist at the same time. Upon destruction of all
// Scoped(BestEffort)ExecutionFences, tasks that were preeempted are released.
// Note: the constructor of Scoped(BestEffort)ExecutionFence will not wait for
// currently running tasks (as they were posted before entering this scope and
// do not violate the contract; some of them could be CONTINUE_ON_SHUTDOWN and
// waiting for them to complete is ill-advised).
class BASE_EXPORT ScopedExecutionFence {
public:
ScopedExecutionFence();
ScopedExecutionFence(const ScopedExecutionFence&) = delete;
ScopedExecutionFence& operator=(const ScopedExecutionFence&) = delete;
~ScopedExecutionFence();
};
class BASE_EXPORT ScopedBestEffortExecutionFence {
public:
ScopedBestEffortExecutionFence();
ScopedBestEffortExecutionFence(const ScopedBestEffortExecutionFence&) =
delete;
ScopedBestEffortExecutionFence& operator=(
const ScopedBestEffortExecutionFence&) = delete;
~ScopedBestEffortExecutionFence();
};
// Used to restrict the maximum number of concurrent tasks that can run in a
// scope.
class BASE_EXPORT ScopedRestrictedTasks {
public:
ScopedRestrictedTasks();
ScopedRestrictedTasks(const ScopedRestrictedTasks&) = delete;
ScopedRestrictedTasks& operator=(const ScopedRestrictedTasks&) = delete;
~ScopedRestrictedTasks();
};
// Used to allow posting `BLOCK_SHUTDOWN` tasks after shutdown in a scope. The
// tasks will fizzle (not run) but not trigger any checks that aim to catch
// this class of ordering bugs.
class BASE_EXPORT ScopedFizzleBlockShutdownTasks {
public:
ScopedFizzleBlockShutdownTasks();
ScopedFizzleBlockShutdownTasks(const ScopedFizzleBlockShutdownTasks&) =
delete;
ScopedFizzleBlockShutdownTasks& operator=(
const ScopedFizzleBlockShutdownTasks&) = delete;
~ScopedFizzleBlockShutdownTasks();
};
// Destroying a ThreadPoolInstance is not allowed in production; it is always
// leaked. In tests, it should only be destroyed after JoinForTesting() has
// returned.
virtual ~ThreadPoolInstance() = default;
// Allows the thread pool to create threads and run tasks following the
// |init_params| specification.
//
// If specified, |worker_thread_observer| will be notified when a worker
// enters and exits its main function. It must not be destroyed before
// JoinForTesting() has returned (must never be destroyed in production).
//
// CHECKs on failure.
virtual void Start(
const InitParams& init_params,
WorkerThreadObserver* worker_thread_observer = nullptr) = 0;
// Returns true if Start() was called. This will continue returning true even
// after Shutdown() is called. Must be called on the same sequence as Start().
virtual bool WasStarted() const = 0;
// Same as WasStarted(), but can be called from any sequence. The caller must
// make sure this call is properly synchronized with Start(), to avoid
// undefined behavior.
virtual bool WasStartedUnsafe() const = 0;
// Synchronously shuts down the thread pool. Once this is called, only tasks
// posted with the BLOCK_SHUTDOWN behavior will be run. When this returns:
// - All SKIP_ON_SHUTDOWN tasks that were already running have completed their
// execution.
// - All posted BLOCK_SHUTDOWN tasks have completed their execution.
// - CONTINUE_ON_SHUTDOWN tasks might still be running.
// Note that an implementation can keep threads and other resources alive to
// support running CONTINUE_ON_SHUTDOWN after this returns. This can only be
// called once. Must be called on the same sequence as Start().
virtual void Shutdown() = 0;
// Waits until there are no pending undelayed tasks. May be called in tests
// to validate that a condition is met after all undelayed tasks have run.
//
// Does not wait for delayed tasks. Waits for undelayed tasks posted from
// other threads during the call. Returns immediately when shutdown completes.
virtual void FlushForTesting() = 0;
// Returns and calls |flush_callback| when there are no incomplete undelayed
// tasks. |flush_callback| may be called back on any thread and should not
// perform a lot of work. May be used when additional work on the current
// thread needs to be performed during a flush. Only one
// FlushAsyncForTesting() may be pending at any given time.
virtual void FlushAsyncForTesting(OnceClosure flush_callback) = 0;
// Joins all threads. Tasks that are already running are allowed to complete
// their execution. This can only be called once. Using this thread pool
// instance to create task runners or post tasks is not permitted during or
// after this call.
virtual void JoinForTesting() = 0;
virtual void BeginFizzlingBlockShutdownTasks() = 0;
virtual void EndFizzlingBlockShutdownTasks() = 0;
// CreateAndStartWithDefaultParams(), Create(), and SetInstance() register a
// ThreadPoolInstance to handle tasks posted through the thread_pool.h API for
// this process.
//
// Processes that need to initialize ThreadPoolInstance with custom params or
// that need to allow tasks to be posted before the ThreadPoolInstance creates
// its threads should use Create() followed by Start(). Other processes can
// use CreateAndStartWithDefaultParams().
//
// A registered ThreadPoolInstance is only deleted when a new
// ThreadPoolInstance is registered. The last registered ThreadPoolInstance is
// leaked on shutdown. The methods below must not be called when TaskRunners
// created by a previous ThreadPoolInstance are still alive. The methods are
// not thread-safe; proper synchronization is required to use the
// thread_pool.h API after registering a new ThreadPoolInstance.
#if !BUILDFLAG(IS_NACL)
// Creates and starts a thread pool using default params. |name| is used to
// label histograms, it must not be empty. It should identify the component
// that calls this. Start() is called by this method; it is invalid to call it
// again afterwards. CHECKs on failure. For tests, prefer
// base::test::TaskEnvironment (ensures isolation).
static void CreateAndStartWithDefaultParams(std::string_view name);
// Same as CreateAndStartWithDefaultParams() but allows callers to split the
// Create() and StartWithDefaultParams() calls. Start() is called by this
// method; it is invalid to call it again afterwards.
void StartWithDefaultParams();
#endif // !BUILDFLAG(IS_NACL)
// Creates a ready to start thread pool. |name| is used to label histograms,
// it must not be empty. It should identify the component that creates the
// ThreadPoolInstance. The thread pool doesn't create threads until Start() is
// called. Tasks can be posted at any time but will not run until after
// Start() is called. For tests, prefer base::test::TaskEnvironment
// (ensures isolation).
static void Create(std::string_view name);
// Registers |thread_pool| to handle tasks posted through the thread_pool.h
// API for this process. For tests, prefer base::test::TaskEnvironment
// (ensures isolation).
static void Set(std::unique_ptr<ThreadPoolInstance> thread_pool);
// Retrieve the ThreadPoolInstance set via SetInstance() or Create(). This
// should be used very rarely; most users of the thread pool should use the
// thread_pool.h API. In particular, refrain from doing
// if (!ThreadPoolInstance::Get()) {
// ThreadPoolInstance::Set(...);
// base::ThreadPool::PostTask(...);
// }
// instead make sure to SetInstance() early in one determinstic place in the
// process' initialization phase.
// In doubt, consult with //base/task/thread_pool/OWNERS.
static ThreadPoolInstance* Get();
private:
friend class ThreadPoolTestHelpers;
friend class gin::V8Platform;
friend class content::BrowserMainLoopTest_CreateThreadsInSingleProcess_Test;
// Returns the maximum number of non-single-threaded non-blocked tasks posted
// with |traits| that can run concurrently in this thread pool. |traits|
// can't contain TaskPriority::BEST_EFFORT.
//
// Do not use this method. To process n items, post n tasks that each process
// 1 item rather than GetMaxConcurrentNonBlockedTasksWithTraitsDeprecated()
// tasks that each process
// n/GetMaxConcurrentNonBlockedTasksWithTraitsDeprecated() items.
//
// TODO(fdoray): Remove this method. https://crbug.com/687264
virtual size_t GetMaxConcurrentNonBlockedTasksWithTraitsDeprecated(
const TaskTraits& traits) const = 0;
// Starts/stops a fence that prevents scheduling of tasks of any / BEST_EFFORT
// priority. Ongoing tasks will still be allowed to complete and not be
// waited upon. This is useful for use cases where a second component
// (e.g. content) needs a "single-threaded" startup phase where tasks it
// posts do not run before it "enables the ThreadPool"
// (via ThreadPoolInstance::EndFence instead of the typical
// ThreadPoolInstance::Start). For example, because a lightweight service
// manager was already running prior to launching full chrome. BeginFence
// does not wait for ongoing tasks as those pertain to the previous phase and
// cannot interfere with the upcoming "single-threaded" initialization
// phase. These methods must be called from the same sequence as Start().
virtual void BeginFence() = 0;
virtual void EndFence() = 0;
virtual void BeginBestEffortFence() = 0;
virtual void EndBestEffortFence() = 0;
virtual void BeginRestrictedTasks() = 0;
virtual void EndRestrictedTasks() = 0;
};
} // namespace base
#endif // BASE_TASK_THREAD_POOL_THREAD_POOL_INSTANCE_H_