be/src/exec/pipeline/task_queue.cpp

Source
// 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 "exec/pipeline/task_queue.h"

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

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

namespace doris {

PipelineTaskSPtr 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();
    DorisMetrics::instance()->pipeline_task_queue_size->increment(-_total_task_size);
}

PipelineTaskSPtr 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--;
        DorisMetrics::instance()->pipeline_task_queue_size->increment(-1);
    }
    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;
}

PipelineTaskSPtr 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);
}

PipelineTaskSPtr 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(PipelineTaskSPtr 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++;
    DorisMetrics::instance()->pipeline_task_queue_size->increment(1);
    _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(); });
}

PipelineTaskSPtr MultiCoreTaskQueue::take(int core_id) {
    PipelineTaskSPtr 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;
}

PipelineTaskSPtr 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(PipelineTaskSPtr task) {
    int thread_id = task->get_thread_id(_core_size);
    if (thread_id < 0) {
        thread_id = _next_core.fetch_add(1) % _core_size;
    }
    return push_back(task, thread_id);
}

Status MultiCoreTaskQueue::push_back(PipelineTaskSPtr 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_thread_id(_core_size); 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

Coverage Report

Created: 2026-04-15 12:36

Line	Count	Source
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 "exec/pipeline/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 "exec/pipeline/pipeline_task.h"
27		#include "runtime/workload_group/workload_group.h"
28
29		namespace doris {
30
31	11	PipelineTaskSPtr SubTaskQueue::try_take(bool is_steal) {
32	11	if (_queue.empty()) {
33	0	return nullptr;
34	0	}
35	11	auto task = _queue.front();
36	11	_queue.pop();
37	11	return task;
38	11	}
39
40		//////////////////// PriorityTaskQueue ////////////////////
41
42	26	PriorityTaskQueue::PriorityTaskQueue() : _closed(false) {
43	26	double factor = 1;
44	182	for (int i = SUB_QUEUE_LEVEL - 1; i >= 0; i--) {
45	156	_sub_queues[i].set_level_factor(factor);
46	156	factor *= LEVEL_QUEUE_TIME_FACTOR;
47	156	}
48	26	}
49
50	0	void PriorityTaskQueue::close() {
51	0	std::unique_lock<std::mutex> lock(_work_size_mutex);
52	0	_closed = true;
53	0	_wait_task.notify_all();
54	0	DorisMetrics::instance()->pipeline_task_queue_size->increment(-_total_task_size);
55	0	}
56
57	14	PipelineTaskSPtr PriorityTaskQueue::_try_take_unprotected(bool is_steal) {
58	14	if (_total_task_size == 0 \|\| _closed) {
59	3	return nullptr;
60	3	}
61
62	11	double min_vruntime = 0;
63	11	int level = -1;
64	77	for (int i = 0; i < SUB_QUEUE_LEVEL; ++i) {
65	66	double cur_queue_vruntime = _sub_queues[i].get_vruntime();
66	66	if (!_sub_queues[i].empty()) {
67	11	if (level == -1 \|\| cur_queue_vruntime < min_vruntime) {
68	11	level = i;
69	11	min_vruntime = cur_queue_vruntime;
70	11	}
71	11	}
72	66	}
73	11	DCHECK(level != -1);
74	11	_queue_level_min_vruntime = uint64_t(min_vruntime);
75
76	11	auto task = _sub_queues[level].try_take(is_steal);
77	11	if (task) {
78	11	task->update_queue_level(level);
79	11	_total_task_size--;
80	11	DorisMetrics::instance()->pipeline_task_queue_size->increment(-1);
81	11	}
82	11	return task;
83	14	}
84
85	31	int PriorityTaskQueue::_compute_level(uint64_t runtime) {
86	31	for (int i = 0; i < SUB_QUEUE_LEVEL - 1; ++i) {
87	31	if (runtime <= _queue_level_limit[i]) {
88	31	return i;
89	31	}
90	31	}
91	0	return SUB_QUEUE_LEVEL - 1;
92	31	}
93
94	12	PipelineTaskSPtr PriorityTaskQueue::try_take(bool is_steal) {
95		// TODO other efficient lock? e.g. if get lock fail, return null_ptr
96	12	std::unique_lock<std::mutex> lock(_work_size_mutex);
97	12	return _try_take_unprotected(is_steal);
98	12	}
99
100	1	PipelineTaskSPtr PriorityTaskQueue::take(uint32_t timeout_ms) {
101	1	std::unique_lock<std::mutex> lock(_work_size_mutex);
102	1	auto task = _try_take_unprotected(false);
103	1	if (task) {
104	0	return task;
105	1	} else {
106	1	if (timeout_ms > 0) {
107	1	_wait_task.wait_for(lock, std::chrono::milliseconds(timeout_ms));
108	1	} else {
109	0	_wait_task.wait(lock);
110	0	}
111	1	return _try_take_unprotected(false);
112	1	}
113	1	}
114
115	31	Status PriorityTaskQueue::push(PipelineTaskSPtr task) {
116	31	if (_closed) {
117	0	return Status::InternalError("WorkTaskQueue closed");
118	0	}
119	31	auto level = _compute_level(task->get_runtime_ns());
120	31	std::unique_lock<std::mutex> lock(_work_size_mutex);
121
122		// update empty queue's runtime, to avoid too high priority
123	31	if (_sub_queues[level].empty() &&
124	31	double(_queue_level_min_vruntime) > _sub_queues[level].get_vruntime()) {
125	0	_sub_queues[level].adjust_runtime(_queue_level_min_vruntime);
126	0	}
127
128	31	_sub_queues[level].push_back(task);
129	31	_total_task_size++;
130	31	DorisMetrics::instance()->pipeline_task_queue_size->increment(1);
131	31	_wait_task.notify_one();
132	31	return Status::OK();
133	31	}
134
135	52	MultiCoreTaskQueue::~MultiCoreTaskQueue() = default;
136
137		MultiCoreTaskQueue::MultiCoreTaskQueue(int core_size)
138	52	: _prio_task_queues(core_size), _closed(false), _core_size(core_size) {}
139
140	26	void MultiCoreTaskQueue::close() {
141	26	if (_closed) {
142	0	return;
143	0	}
144	26	_closed = true;
145		// close all priority task queue
146	26	std::ranges::for_each(_prio_task_queues,
147	26	[](auto& prio_task_queue) { prio_task_queue.close(); });
148	26	}
149
150	0	PipelineTaskSPtr MultiCoreTaskQueue::take(int core_id) {
151	0	PipelineTaskSPtr task = nullptr;
152	0	while (!_closed) {
153	0	DCHECK(_prio_task_queues.size() > core_id)
154	0	<< " list size: " << _prio_task_queues.size() << " core_id: " << core_id
155	0	<< " _core_size: " << _core_size << " _next_core: " << _next_core.load();
156	0	task = _prio_task_queues[core_id].try_take(false);
157	0	if (task) {
158	0	break;
159	0	}
160	0	task = _steal_take(core_id);
161	0	if (task) {
162	0	break;
163	0	}
164	0	task = _prio_task_queues[core_id].take(WAIT_CORE_TASK_TIMEOUT_MS /* timeout_ms */);
165	0	if (task) {
166	0	break;
167	0	}
168	0	}
169	0	if (task) {
170	0	task->pop_out_runnable_queue();
171	0	}
172	0	return task;
173	0	}
174
175	1	PipelineTaskSPtr MultiCoreTaskQueue::_steal_take(int core_id) {
176	1	DCHECK(core_id < _core_size);
177	1	int next_id = core_id;
178	1	for (int i = 1; i < _core_size; ++i) {
179	0	++next_id;
180	0	if (next_id == _core_size) {
181	0	next_id = 0;
182	0	}
183	0	DCHECK(next_id < _core_size);
184	0	auto task = _prio_task_queues[next_id].try_take(true);
185	0	if (task) {
186	0	return task;
187	0	}
188	0	}
189	1	return nullptr;
190	1	}
191
192	31	Status MultiCoreTaskQueue::push_back(PipelineTaskSPtr task) {
193	31	int thread_id = task->get_thread_id(_core_size);
194	31	if (thread_id < 0) {
195	31	thread_id = _next_core.fetch_add(1) % _core_size;
196	31	}
197	31	return push_back(task, thread_id);
198	31	}
199
200	31	Status MultiCoreTaskQueue::push_back(PipelineTaskSPtr task, int core_id) {
201	31	DCHECK(core_id < _core_size);
202	31	task->put_in_runnable_queue();
203	31	return _prio_task_queues[core_id].push(task);
204	31	}
205
206	0	void MultiCoreTaskQueue::update_statistics(PipelineTask* task, int64_t time_spent) {
207		// if the task not execute but exception early close, core_id == -1
208		// should not do update_statistics
209	0	if (auto core_id = task->get_thread_id(_core_size); core_id >= 0) {
210	0	task->inc_runtime_ns(time_spent);
211	0	_prio_task_queues[core_id].inc_sub_queue_runtime(task->get_queue_level(), time_spent);
212	0	}
213	0	}
214
215		} // namespace doris