/root/doris/be/src/pipeline/task_queue.cpp

Source (jump to first uncovered line)
// Licensed to the Apache Software Foundation (ASF) under one
// or more contributor license agreements.  See the NOTICE file
// distributed with this work for additional information
// regarding copyright ownership.  The ASF licenses this file
// to you under the Apache License, Version 2.0 (the
// "License"); you may not use this file except in compliance
// with the License.  You may obtain a copy of the License at
//
//   http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing,
// software distributed under the License is distributed on an
// "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
// KIND, either express or implied.  See the License for the
// specific language governing permissions and limitations
// under the License.

#include "task_queue.h"

// IWYU pragma: no_include <bits/chrono.h>
#include <chrono> // IWYU pragma: keep
#include <string>

#include "common/logging.h"
#include "pipeline/pipeline_task.h"

namespace doris {
namespace pipeline {

TaskQueue::~TaskQueue() = default;

PipelineTask* SubTaskQueue::try_take(bool is_steal) {
    if (_queue.empty()) {
        return nullptr;
    }
    auto task = _queue.front();
    _queue.pop();
    return task;
}

////////////////////  PriorityTaskQueue ////////////////////

PriorityTaskQueue::PriorityTaskQueue() : _closed(false) {
    double factor = 1;
    for (int i = SUB_QUEUE_LEVEL - 1; i >= 0; i--) {
        _sub_queues[i].set_level_factor(factor);
        factor *= LEVEL_QUEUE_TIME_FACTOR;
    }
}

void PriorityTaskQueue::close() {
    std::unique_lock<std::mutex> lock(_work_size_mutex);
    _closed = true;
    _wait_task.notify_all();
}

PipelineTask* PriorityTaskQueue::_try_take_unprotected(bool is_steal) {
    if (_total_task_size == 0 || _closed) {
        return nullptr;
    }

    double min_vruntime = 0;
    int level = -1;
    for (int i = 0; i < SUB_QUEUE_LEVEL; ++i) {
        double cur_queue_vruntime = _sub_queues[i].get_vruntime();
        if (!_sub_queues[i].empty()) {
            if (level == -1 || cur_queue_vruntime < min_vruntime) {
                level = i;
                min_vruntime = cur_queue_vruntime;
            }
        }
    }
    DCHECK(level != -1);
    _queue_level_min_vruntime = uint64_t(min_vruntime);

    auto task = _sub_queues[level].try_take(is_steal);
    if (task) {
        task->update_queue_level(level);
        _total_task_size--;
    }
    return task;
}

int PriorityTaskQueue::_compute_level(uint64_t runtime) {
    for (int i = 0; i < SUB_QUEUE_LEVEL - 1; ++i) {
        if (runtime <= _queue_level_limit[i]) {
            return i;
        }
    }
    return SUB_QUEUE_LEVEL - 1;
}

PipelineTask* PriorityTaskQueue::try_take(bool is_steal) {
    // TODO other efficient lock? e.g. if get lock fail, return null_ptr
    std::unique_lock<std::mutex> lock(_work_size_mutex);
    return _try_take_unprotected(is_steal);
}

PipelineTask* PriorityTaskQueue::take(uint32_t timeout_ms) {
    std::unique_lock<std::mutex> lock(_work_size_mutex);
    auto task = _try_take_unprotected(false);
    if (task) {
        return task;
    } else {
        if (timeout_ms > 0) {
            _wait_task.wait_for(lock, std::chrono::milliseconds(timeout_ms));
        } else {
            _wait_task.wait(lock);
        }
        return _try_take_unprotected(false);
    }
}

Status PriorityTaskQueue::push(PipelineTask* task) {
    if (_closed) {
        return Status::InternalError("WorkTaskQueue closed");
    }
    auto level = _compute_level(task->get_runtime_ns());
    std::unique_lock<std::mutex> lock(_work_size_mutex);

    // update empty queue's  runtime, to avoid too high priority
    if (_sub_queues[level].empty() &&
        _queue_level_min_vruntime > _sub_queues[level].get_vruntime()) {
        _sub_queues[level].adjust_runtime(_queue_level_min_vruntime);
    }

    _sub_queues[level].push_back(task);
    _total_task_size++;
    _wait_task.notify_one();
    return Status::OK();
}

MultiCoreTaskQueue::~MultiCoreTaskQueue() = default;

MultiCoreTaskQueue::MultiCoreTaskQueue(int core_size) : TaskQueue(core_size), _closed(false) {
    _prio_task_queue_list =
            std::make_shared<std::vector<std::unique_ptr<PriorityTaskQueue>>>(core_size);
    for (int i = 0; i < core_size; i++) {
        (*_prio_task_queue_list)[i] = std::make_unique<PriorityTaskQueue>();
    }
}

void MultiCoreTaskQueue::close() {
    if (_closed) {
        return;
    }
    _closed = true;
    for (int i = 0; i < _core_size; ++i) {
        (*_prio_task_queue_list)[i]->close();
    }
    std::atomic_store(&_prio_task_queue_list,
                      std::shared_ptr<std::vector<std::unique_ptr<PriorityTaskQueue>>>(nullptr));
}

PipelineTask* MultiCoreTaskQueue::take(int core_id) {
    PipelineTask* task = nullptr;
    auto prio_task_queue_list =
            std::atomic_load_explicit(&_prio_task_queue_list, std::memory_order_relaxed);
    while (!_closed) {
        DCHECK(prio_task_queue_list->size() > core_id)
                << " list size: " << prio_task_queue_list->size() << " core_id: " << core_id
                << " _core_size: " << _core_size << " _next_core: " << _next_core.load();
        task = (*prio_task_queue_list)[core_id]->try_take(false);
        if (task) {
            task->set_core_id(core_id);
            break;
        }
        task = _steal_take(core_id, *prio_task_queue_list);
        if (task) {
            break;
        }
        task = (*prio_task_queue_list)[core_id]->take(WAIT_CORE_TASK_TIMEOUT_MS /* timeout_ms */);
        if (task) {
            task->set_core_id(core_id);
            break;
        }
    }
    if (task) {
        task->pop_out_runnable_queue();
    }
    return task;
}

PipelineTask* MultiCoreTaskQueue::_steal_take(
        int core_id, std::vector<std::unique_ptr<PriorityTaskQueue>>& prio_task_queue_list) {
    DCHECK(core_id < _core_size);
    int next_id = core_id;
    for (int i = 1; i < _core_size; ++i) {
        ++next_id;
        if (next_id == _core_size) {
            next_id = 0;
        }
        DCHECK(next_id < _core_size);
        auto task = prio_task_queue_list[next_id]->try_take(true);
        if (task) {
            task->set_core_id(next_id);
            return task;
        }
    }
    return nullptr;
}

Status MultiCoreTaskQueue::push_back(PipelineTask* task) {
    int core_id = task->get_previous_core_id();
    if (core_id < 0) {
        core_id = _next_core.fetch_add(1) % _core_size;
    }
    return push_back(task, core_id);
}

Status MultiCoreTaskQueue::push_back(PipelineTask* task, int core_id) {
    DCHECK(core_id < _core_size);
    task->put_in_runnable_queue();
    auto prio_task_queue_list =
            std::atomic_load_explicit(&_prio_task_queue_list, std::memory_order_relaxed);
    return (*prio_task_queue_list)[core_id]->push(task);
}

} // namespace pipeline
} // namespace doris

Coverage Report

Created: 2024-11-20 12:30

Line	Count	Source (jump to first uncovered line)
1		// Licensed to the Apache Software Foundation (ASF) under one
2		// or more contributor license agreements. See the NOTICE file
3		// distributed with this work for additional information
4		// regarding copyright ownership. The ASF licenses this file
5		// to you under the Apache License, Version 2.0 (the
6		// "License"); you may not use this file except in compliance
7		// with the License. You may obtain a copy of the License at
8		//
9		// http://www.apache.org/licenses/LICENSE-2.0
10		//
11		// Unless required by applicable law or agreed to in writing,
12		// software distributed under the License is distributed on an
13		// "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
14		// KIND, either express or implied. See the License for the
15		// specific language governing permissions and limitations
16		// under the License.
17
18		#include "task_queue.h"
19
20		// IWYU pragma: no_include <bits/chrono.h>
21		#include <chrono> // IWYU pragma: keep
22		#include <string>
23
24		#include "common/logging.h"
25		#include "pipeline/pipeline_task.h"
26
27		namespace doris {
28		namespace pipeline {
29
30	0	TaskQueue::~TaskQueue() = default;
31
32	0	PipelineTask* SubTaskQueue::try_take(bool is_steal) {
33	0	if (_queue.empty()) {
34	0	return nullptr;
35	0	}
36	0	auto task = _queue.front();
37	0	_queue.pop();
38	0	return task;
39	0	}
40
41		//////////////////// PriorityTaskQueue ////////////////////
42
43	0	PriorityTaskQueue::PriorityTaskQueue() : _closed(false) {
44	0	double factor = 1;
45	0	for (int i = SUB_QUEUE_LEVEL - 1; i >= 0; i--) {
46	0	_sub_queues[i].set_level_factor(factor);
47	0	factor *= LEVEL_QUEUE_TIME_FACTOR;
48	0	}
49	0	}
50
51	0	void PriorityTaskQueue::close() {
52	0	std::unique_lock<std::mutex> lock(_work_size_mutex);
53	0	_closed = true;
54	0	_wait_task.notify_all();
55	0	}
56
57	0	PipelineTask* PriorityTaskQueue::_try_take_unprotected(bool is_steal) {
58	0	if (_total_task_size == 0 \|\| _closed) {
59	0	return nullptr;
60	0	}
61
62	0	double min_vruntime = 0;
63	0	int level = -1;
64	0	for (int i = 0; i < SUB_QUEUE_LEVEL; ++i) {
65	0	double cur_queue_vruntime = _sub_queues[i].get_vruntime();
66	0	if (!_sub_queues[i].empty()) {
67	0	if (level == -1 \|\| cur_queue_vruntime < min_vruntime) {
68	0	level = i;
69	0	min_vruntime = cur_queue_vruntime;
70	0	}
71	0	}
72	0	}
73	0	DCHECK(level != -1);
74	0	_queue_level_min_vruntime = uint64_t(min_vruntime);
75
76	0	auto task = _sub_queues[level].try_take(is_steal);
77	0	if (task) {
78	0	task->update_queue_level(level);
79	0	_total_task_size--;
80	0	}
81	0	return task;
82	0	}
83
84	0	int PriorityTaskQueue::_compute_level(uint64_t runtime) {
85	0	for (int i = 0; i < SUB_QUEUE_LEVEL - 1; ++i) {
86	0	if (runtime <= _queue_level_limit[i]) {
87	0	return i;
88	0	}
89	0	}
90	0	return SUB_QUEUE_LEVEL - 1;
91	0	}
92
93	0	PipelineTask* PriorityTaskQueue::try_take(bool is_steal) {
94		// TODO other efficient lock? e.g. if get lock fail, return null_ptr
95	0	std::unique_lock<std::mutex> lock(_work_size_mutex);
96	0	return _try_take_unprotected(is_steal);
97	0	}
98
99	0	PipelineTask* PriorityTaskQueue::take(uint32_t timeout_ms) {
100	0	std::unique_lock<std::mutex> lock(_work_size_mutex);
101	0	auto task = _try_take_unprotected(false);
102	0	if (task) {
103	0	return task;
104	0	} else {
105	0	if (timeout_ms > 0) {
106	0	_wait_task.wait_for(lock, std::chrono::milliseconds(timeout_ms));
107	0	} else {
108	0	_wait_task.wait(lock);
109	0	}
110	0	return _try_take_unprotected(false);
111	0	}
112	0	}
113
114	0	Status PriorityTaskQueue::push(PipelineTask* task) {
115	0	if (_closed) {
116	0	return Status::InternalError("WorkTaskQueue closed");
117	0	}
118	0	auto level = _compute_level(task->get_runtime_ns());
119	0	std::unique_lock<std::mutex> lock(_work_size_mutex);
120
121		// update empty queue's runtime, to avoid too high priority
122	0	if (_sub_queues[level].empty() &&
123	0	_queue_level_min_vruntime > _sub_queues[level].get_vruntime()) {
124	0	_sub_queues[level].adjust_runtime(_queue_level_min_vruntime);
125	0	}
126
127	0	_sub_queues[level].push_back(task);
128	0	_total_task_size++;
129	0	_wait_task.notify_one();
130	0	return Status::OK();
131	0	}
132
133	0	MultiCoreTaskQueue::~MultiCoreTaskQueue() = default;
134
135	0	MultiCoreTaskQueue::MultiCoreTaskQueue(int core_size) : TaskQueue(core_size), _closed(false) {
136	0	_prio_task_queue_list =
137	0	std::make_shared<std::vector<std::unique_ptr<PriorityTaskQueue>>>(core_size);
138	0	for (int i = 0; i < core_size; i++) {
139	0	(*_prio_task_queue_list)[i] = std::make_unique<PriorityTaskQueue>();
140	0	}
141	0	}
142
143	0	void MultiCoreTaskQueue::close() {
144	0	if (_closed) {
145	0	return;
146	0	}
147	0	_closed = true;
148	0	for (int i = 0; i < _core_size; ++i) {
149	0	(*_prio_task_queue_list)[i]->close();
150	0	}
151	0	std::atomic_store(&_prio_task_queue_list,
152	0	std::shared_ptr<std::vector<std::unique_ptr<PriorityTaskQueue>>>(nullptr));
153	0	}
154
155	0	PipelineTask* MultiCoreTaskQueue::take(int core_id) {
156	0	PipelineTask* task = nullptr;
157	0	auto prio_task_queue_list =
158	0	std::atomic_load_explicit(&_prio_task_queue_list, std::memory_order_relaxed);
159	0	while (!_closed) {
160	0	DCHECK(prio_task_queue_list->size() > core_id)
161	0	<< " list size: " << prio_task_queue_list->size() << " core_id: " << core_id
162	0	<< " _core_size: " << _core_size << " _next_core: " << _next_core.load();
163	0	task = (*prio_task_queue_list)[core_id]->try_take(false);
164	0	if (task) {
165	0	task->set_core_id(core_id);
166	0	break;
167	0	}
168	0	task = _steal_take(core_id, *prio_task_queue_list);
169	0	if (task) {
170	0	break;
171	0	}
172	0	task = (prio_task_queue_list)[core_id]->take(WAIT_CORE_TASK_TIMEOUT_MS / timeout_ms */);
173	0	if (task) {
174	0	task->set_core_id(core_id);
175	0	break;
176	0	}
177	0	}
178	0	if (task) {
179	0	task->pop_out_runnable_queue();
180	0	}
181	0	return task;
182	0	}
183
184		PipelineTask* MultiCoreTaskQueue::_steal_take(
185	0	int core_id, std::vector<std::unique_ptr<PriorityTaskQueue>>& prio_task_queue_list) {
186	0	DCHECK(core_id < _core_size);
187	0	int next_id = core_id;
188	0	for (int i = 1; i < _core_size; ++i) {
189	0	++next_id;
190	0	if (next_id == _core_size) {
191	0	next_id = 0;
192	0	}
193	0	DCHECK(next_id < _core_size);
194	0	auto task = prio_task_queue_list[next_id]->try_take(true);
195	0	if (task) {
196	0	task->set_core_id(next_id);
197	0	return task;
198	0	}
199	0	}
200	0	return nullptr;
201	0	}
202
203	0	Status MultiCoreTaskQueue::push_back(PipelineTask* task) {
204	0	int core_id = task->get_previous_core_id();
205	0	if (core_id < 0) {
206	0	core_id = _next_core.fetch_add(1) % _core_size;
207	0	}
208	0	return push_back(task, core_id);
209	0	}
210
211	0	Status MultiCoreTaskQueue::push_back(PipelineTask* task, int core_id) {
212	0	DCHECK(core_id < _core_size);
213	0	task->put_in_runnable_queue();
214	0	auto prio_task_queue_list =
215	0	std::atomic_load_explicit(&_prio_task_queue_list, std::memory_order_relaxed);
216	0	return (*prio_task_queue_list)[core_id]->push(task);
217	0	}
218
219		} // namespace pipeline
220		} // namespace doris