/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 <memory>
#include <string>

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

namespace doris::pipeline {
#include "common/compile_check_begin.h"

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() &&
        double(_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)
        : _prio_task_queues(core_size), _closed(false), _core_size(core_size) {}

void MultiCoreTaskQueue::close() {
    if (_closed) {
        return;
    }
    _closed = true;
    // close all priority task queue
    std::ranges::for_each(_prio_task_queues,
                          [](auto& prio_task_queue) { prio_task_queue.close(); });
}

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

PipelineTask* MultiCoreTaskQueue::_steal_take(int core_id) {
    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_queues[next_id].try_take(true);
        if (task) {
            return task;
        }
    }
    return nullptr;
}

Status MultiCoreTaskQueue::push_back(PipelineTask* task) {
    int core_id = task->get_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();
    return _prio_task_queues[core_id].push(task);
}

void MultiCoreTaskQueue::update_statistics(PipelineTask* task, int64_t time_spent) {
    // if the task not execute but exception early close, core_id == -1
    // should not do update_statistics
    if (auto core_id = task->get_core_id(); core_id >= 0) {
        task->inc_runtime_ns(time_spent);
        _prio_task_queues[core_id].inc_sub_queue_runtime(task->get_queue_level(), time_spent);
    }
}

} // namespace doris::pipeline

Coverage Report

Created: 2025-03-10 18:45

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 <memory>
23		#include <string>
24
25		#include "common/logging.h"
26		#include "pipeline/pipeline_task.h"
27		#include "runtime/workload_group/workload_group.h"
28
29		namespace doris::pipeline {
30		#include "common/compile_check_begin.h"
31
32	10	PipelineTask* SubTaskQueue::try_take(bool is_steal) {
33	10	if (_queue.empty()) {
34	0	return nullptr;
35	0	}
36	10	auto task = _queue.front();
37	10	_queue.pop();
38	10	return task;
39	10	}
40
41		//////////////////// PriorityTaskQueue ////////////////////
42
43	3	PriorityTaskQueue::PriorityTaskQueue() : _closed(false) {
44	3	double factor = 1;
45	21	for (int i = SUB_QUEUE_LEVEL - 1; i >= 0; i--) {
46	18	_sub_queues[i].set_level_factor(factor);
47	18	factor *= LEVEL_QUEUE_TIME_FACTOR;
48	18	}
49	3	}
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	13	PipelineTask* PriorityTaskQueue::_try_take_unprotected(bool is_steal) {
58	13	if (_total_task_size == 0 \|\| _closed) {
59	3	return nullptr;
60	3	}
61
62	10	double min_vruntime = 0;
63	10	int level = -1;
64	70	for (int i = 0; i < SUB_QUEUE_LEVEL; ++i) {
65	60	double cur_queue_vruntime = _sub_queues[i].get_vruntime();
66	60	if (!_sub_queues[i].empty()) {
67	10	if (level == -1 \|\| cur_queue_vruntime < min_vruntime) {
68	10	level = i;
69	10	min_vruntime = cur_queue_vruntime;
70	10	}
71	10	}
72	60	}
73	10	DCHECK(level != -1);
74	10	_queue_level_min_vruntime = uint64_t(min_vruntime);
75
76	10	auto task = _sub_queues[level].try_take(is_steal);
77	10	if (task) {
78	10	task->update_queue_level(level);
79	10	_total_task_size--;
80	10	}
81	10	return task;
82	13	}
83
84	17	int PriorityTaskQueue::_compute_level(uint64_t runtime) {
85	17	for (int i = 0; i < SUB_QUEUE_LEVEL - 1; ++i) {
86	17	if (runtime <= _queue_level_limit[i]) {
87	17	return i;
88	17	}
89	17	}
90	0	return SUB_QUEUE_LEVEL - 1;
91	17	}
92
93	11	PipelineTask* PriorityTaskQueue::try_take(bool is_steal) {
94		// TODO other efficient lock? e.g. if get lock fail, return null_ptr
95	11	std::unique_lock<std::mutex> lock(_work_size_mutex);
96	11	return _try_take_unprotected(is_steal);
97	11	}
98
99	1	PipelineTask* PriorityTaskQueue::take(uint32_t timeout_ms) {
100	1	std::unique_lock<std::mutex> lock(_work_size_mutex);
101	1	auto task = _try_take_unprotected(false);
102	1	if (task) {
103	0	return task;
104	1	} else {
105	1	if (timeout_ms > 0) {
106	1	_wait_task.wait_for(lock, std::chrono::milliseconds(timeout_ms));
107	1	} else {
108	0	_wait_task.wait(lock);
109	0	}
110	1	return _try_take_unprotected(false);
111	1	}
112	1	}
113
114	17	Status PriorityTaskQueue::push(PipelineTask* task) {
115	17	if (_closed) {
116	0	return Status::InternalError("WorkTaskQueue closed");
117	0	}
118	17	auto level = _compute_level(task->get_runtime_ns());
119	17	std::unique_lock<std::mutex> lock(_work_size_mutex);
120
121		// update empty queue's runtime, to avoid too high priority
122	17	if (_sub_queues[level].empty() &&
123	17	double(_queue_level_min_vruntime) > _sub_queues[level].get_vruntime()) {
124	0	_sub_queues[level].adjust_runtime(_queue_level_min_vruntime);
125	0	}
126
127	17	_sub_queues[level].push_back(task);
128	17	_total_task_size++;
129	17	_wait_task.notify_one();
130	17	return Status::OK();
131	17	}
132
133	3	MultiCoreTaskQueue::~MultiCoreTaskQueue() = default;
134
135		MultiCoreTaskQueue::MultiCoreTaskQueue(int core_size)
136	3	: _prio_task_queues(core_size), _closed(false), _core_size(core_size) {}
137
138	0	void MultiCoreTaskQueue::close() {
139	0	if (_closed) {
140	0	return;
141	0	}
142	0	_closed = true;
143		// close all priority task queue
144	0	std::ranges::for_each(_prio_task_queues,
145	0	[](auto& prio_task_queue) { prio_task_queue.close(); });
146	0	}
147
148	0	PipelineTask* MultiCoreTaskQueue::take(int core_id) {
149	0	PipelineTask* task = nullptr;
150	0	while (!_closed) {
151	0	DCHECK(_prio_task_queues.size() > core_id)
152	0	<< " list size: " << _prio_task_queues.size() << " core_id: " << core_id
153	0	<< " _core_size: " << _core_size << " _next_core: " << _next_core.load();
154	0	task = _prio_task_queues[core_id].try_take(false);
155	0	if (task) {
156	0	break;
157	0	}
158	0	task = _steal_take(core_id);
159	0	if (task) {
160	0	break;
161	0	}
162	0	task = _prio_task_queues[core_id].take(WAIT_CORE_TASK_TIMEOUT_MS /* timeout_ms */);
163	0	if (task) {
164	0	break;
165	0	}
166	0	}
167	0	if (task) {
168	0	task->pop_out_runnable_queue();
169	0	}
170	0	return task;
171	0	}
172
173	1	PipelineTask* MultiCoreTaskQueue::_steal_take(int core_id) {
174	1	DCHECK(core_id < _core_size);
175	1	int next_id = core_id;
176	1	for (int i = 1; i < _core_size; ++i) {
177	0	++next_id;
178	0	if (next_id == _core_size) {
179	0	next_id = 0;
180	0	}
181	0	DCHECK(next_id < _core_size);
182	0	auto task = _prio_task_queues[next_id].try_take(true);
183	0	if (task) {
184	0	return task;
185	0	}
186	0	}
187	1	return nullptr;
188	1	}
189
190	17	Status MultiCoreTaskQueue::push_back(PipelineTask* task) {
191	17	int core_id = task->get_core_id();
192	17	if (core_id < 0) {
193	17	core_id = _next_core.fetch_add(1) % _core_size;
194	17	}
195	17	return push_back(task, core_id);
196	17	}
197
198	17	Status MultiCoreTaskQueue::push_back(PipelineTask* task, int core_id) {
199	17	DCHECK(core_id < _core_size);
200	17	task->put_in_runnable_queue();
201	17	return _prio_task_queues[core_id].push(task);
202	17	}
203
204	27	void MultiCoreTaskQueue::update_statistics(PipelineTask* task, int64_t time_spent) {
205		// if the task not execute but exception early close, core_id == -1
206		// should not do update_statistics
207	27	if (auto core_id = task->get_core_id(); core_id >= 0) {
208	0	task->inc_runtime_ns(time_spent);
209	0	_prio_task_queues[core_id].inc_sub_queue_runtime(task->get_queue_level(), time_spent);
210	0	}
211	27	}
212
213		} // namespace doris::pipeline