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
|
// Copyright Vespa.ai. Licensed under the terms of the Apache 2.0 license. See LICENSE in the project root.
#pragma once
#include "isequencedtaskexecutor.h"
#include "thread.h"
#include <vespa/vespalib/util/executor_idle_tracking.h>
#include <vespa/vespalib/util/arrayqueue.hpp>
#include <vespa/vespalib/util/gate.h>
#include <vespa/vespalib/util/eventbarrier.hpp>
#include <mutex>
#include <condition_variable>
#include <optional>
#include <cassert>
namespace vespalib {
/**
* Sequenced executor that balances the number of active threads in
* order to optimize for throughput over latency by minimizing the
* number of critical-path wakeups.
**/
class AdaptiveSequencedExecutor : public ISequencedTaskExecutor
{
private:
using Task = Executor::Task;
struct TaggedTask {
Task::UP task;
uint32_t token;
TaggedTask() : task(nullptr), token(0) {}
TaggedTask(Task::UP task_in, uint32_t token_in)
: task(std::move(task_in)), token(token_in) {}
TaggedTask(TaggedTask &&rhs) = default;
TaggedTask(const TaggedTask &rhs) = delete;
TaggedTask &operator=(const TaggedTask &rhs) = delete;
TaggedTask &operator=(TaggedTask &&rhs) {
assert(task.get() == nullptr); // no overwrites
task = std::move(rhs.task);
token = rhs.token;
return *this;
}
operator bool() const { return bool(task); }
};
/**
* Values used to configure the executor.
**/
struct Config {
size_t num_threads;
size_t max_waiting;
size_t max_pending;
size_t wakeup_limit;
bool is_max_pending_hard;
void set_max_pending(size_t max_pending_in) {
max_pending = std::max(1uL, max_pending_in);
wakeup_limit = std::max(1uL, size_t(max_pending * 0.9));
assert(wakeup_limit > 0);
assert(wakeup_limit <= max_pending);
}
bool is_above_max_pending(size_t pending) {
return (pending >= max_pending) && is_max_pending_hard;
}
Config(size_t num_threads_in, size_t max_waiting_in, size_t max_pending_in, bool is_max_pending_hard_in)
: num_threads(num_threads_in),
max_waiting(max_waiting_in),
max_pending(1000),
wakeup_limit(900),
is_max_pending_hard(is_max_pending_hard_in)
{
assert(num_threads > 0);
set_max_pending(max_pending_in);
}
};
/**
* Tasks that need to be sequenced are handled by a single strand.
**/
struct Strand {
enum class State { IDLE, WAITING, ACTIVE };
State state;
ArrayQueue<TaggedTask> queue;
Strand() noexcept;
~Strand();
};
/**
* The state of a single worker thread.
**/
struct Worker {
enum class State { RUNNING, BLOCKED, DONE };
std::condition_variable cond;
ThreadIdleTracker idleTracker;
State state;
Strand *strand;
Worker();
~Worker();
};
/**
* State related to the executor itself.
**/
struct Self {
enum class State { OPEN, BLOCKED, CLOSED };
std::condition_variable cond;
State state;
size_t waiting_tasks;
size_t pending_tasks;
Self();
~Self();
};
/**
* Stuff related to worker thread startup and shutdown.
**/
struct ThreadTools {
AdaptiveSequencedExecutor &parent;
ThreadPool pool;
Gate allow_worker_exit;
ThreadTools(AdaptiveSequencedExecutor &parent_in);
~ThreadTools();
void start(size_t num_threads);
void close();
};
struct BarrierCompletion {
Gate gate;
void completeBarrier() { gate.countDown(); }
};
std::unique_ptr<ThreadTools> _thread_tools;
mutable std::mutex _mutex;
std::vector<Strand> _strands;
ArrayQueue<Strand*> _wait_queue;
ArrayQueue<Worker*> _worker_stack;
EventBarrier<BarrierCompletion> _barrier;
Self _self;
ExecutorStats _stats;
ExecutorIdleTracker _idleTracker;
Config _cfg;
void maybe_block_self(std::unique_lock<std::mutex> &lock);
void maybe_unblock_self(const std::unique_lock<std::mutex> &lock);
void maybe_wake_worker(const std::unique_lock<std::mutex> &lock);
bool obtain_strand(Worker &worker, std::unique_lock<std::mutex> &lock);
bool exchange_strand(Worker &worker, std::unique_lock<std::mutex> &lock);
TaggedTask next_task(Worker &worker, std::optional<uint32_t> prev_token);
void worker_main();
public:
AdaptiveSequencedExecutor(size_t num_strands, size_t num_threads,
size_t max_waiting, size_t max_pending,
bool is_max_pending_hard);
~AdaptiveSequencedExecutor() override;
ExecutorId getExecutorId(uint64_t component) const override;
void executeTask(ExecutorId id, Task::UP task) override;
void sync_all() override;
void setTaskLimit(uint32_t task_limit) override;
ExecutorStats getStats() override;
Config get_config() const;
};
}
|
