be/src/exec/pipeline/task_scheduler.cpp
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_scheduler.h" |
19 | | |
20 | | #include <fmt/format.h> |
21 | | #include <gen_cpp/types.pb.h> |
22 | | #include <glog/logging.h> |
23 | | #include <sched.h> |
24 | | |
25 | | // IWYU pragma: no_include <bits/chrono.h> |
26 | | #include <algorithm> |
27 | | #include <chrono> // IWYU pragma: keep |
28 | | #include <cstddef> |
29 | | #include <exception> |
30 | | #include <functional> |
31 | | #include <memory> |
32 | | #include <mutex> |
33 | | #include <ostream> |
34 | | #include <string> |
35 | | #include <utility> |
36 | | |
37 | | #include "common/logging.h" |
38 | | #include "common/status.h" |
39 | | #include "core/value/vdatetime_value.h" |
40 | | #include "exec/pipeline/pipeline_fragment_context.h" |
41 | | #include "exec/pipeline/pipeline_task.h" |
42 | | #include "runtime/query_context.h" |
43 | | #include "runtime/thread_context.h" |
44 | | #include "util/thread.h" |
45 | | #include "util/threadpool.h" |
46 | | #include "util/time.h" |
47 | | #include "util/uid_util.h" |
48 | | |
49 | | namespace doris { |
50 | 48 | TaskScheduler::~TaskScheduler() { |
51 | 48 | stop(); |
52 | 48 | LOG(INFO) << "Task scheduler " << _name << " shutdown"; |
53 | 48 | } |
54 | | |
55 | 2 | Status TaskScheduler::start() { |
56 | 2 | RETURN_IF_ERROR(ThreadPoolBuilder(_name) |
57 | 2 | .set_min_threads(_num_threads) |
58 | 2 | .set_max_threads(_num_threads) |
59 | 2 | .set_max_queue_size(0) |
60 | 2 | .set_cgroup_cpu_ctl(_cgroup_cpu_ctl) |
61 | 2 | .build(&_fix_thread_pool)); |
62 | 2 | LOG_INFO("TaskScheduler set cores").tag("size", _num_threads); |
63 | 4 | for (int32_t i = 0; i < _num_threads; ++i) { |
64 | 2 | RETURN_IF_ERROR(_fix_thread_pool->submit_func([this, i] { _do_work(i); })); |
65 | 2 | } |
66 | 2 | return Status::OK(); |
67 | 2 | } |
68 | | |
69 | 1 | Status TaskScheduler::submit(PipelineTaskSPtr task) { |
70 | 1 | return _task_queue.push_back(task); |
71 | 1 | } |
72 | | |
73 | | // after close_task, task maybe destructed. |
74 | 1 | void close_task(PipelineTask* task, Status exec_status, PipelineFragmentContext* ctx) { |
75 | | // Has to attach memory tracker here, because the close task will also release some memory. |
76 | | // Should count the memory to the query or the query's memory will not decrease when part of |
77 | | // task finished. |
78 | 1 | SCOPED_ATTACH_TASK(task->runtime_state()); |
79 | 1 | if (!exec_status.ok()) { |
80 | 1 | ctx->cancel(exec_status); |
81 | 1 | LOG(WARNING) << fmt::format("Pipeline task failed. query_id: {} reason: {}", |
82 | 1 | print_id(ctx->get_query_id()), exec_status.to_string()); |
83 | 1 | } |
84 | 1 | Status status = task->close(exec_status); |
85 | 1 | if (!status.ok()) { |
86 | 0 | ctx->cancel(status); |
87 | 0 | } |
88 | 1 | status = task->finalize(); |
89 | 1 | if (!status.ok()) { |
90 | 0 | ctx->cancel(status); |
91 | 0 | } |
92 | 1 | } |
93 | | |
94 | 2 | void TaskScheduler::_do_work(int index) { |
95 | 5 | while (!_need_to_stop) { |
96 | 3 | auto task = _task_queue.take(index); |
97 | 3 | if (!task) { |
98 | 2 | continue; |
99 | 2 | } |
100 | | |
101 | | // The task is already running, maybe block in now dependency wake up by other thread |
102 | | // but the block thread still hold the task, so put it back to the queue, until the hold |
103 | | // thread set task->set_running(false) |
104 | | // set_running return the old value |
105 | 1 | if (task->set_running(true)) { |
106 | 0 | static_cast<void>(_task_queue.push_back(task, index)); |
107 | 0 | continue; |
108 | 0 | } |
109 | | |
110 | 1 | if (task->is_finalized()) { |
111 | 0 | task->set_running(false); |
112 | 0 | continue; |
113 | 0 | } |
114 | | |
115 | 1 | auto fragment_context = task->fragment_context().lock(); |
116 | 1 | if (!fragment_context) { |
117 | | // Fragment already finished |
118 | 0 | task->set_running(false); |
119 | 0 | continue; |
120 | 0 | } |
121 | | |
122 | 1 | task->set_thread_id(index); |
123 | | |
124 | 1 | bool done = false; |
125 | 1 | auto status = Status::OK(); |
126 | 1 | int64_t exec_ns = 0; |
127 | 1 | SCOPED_RAW_TIMER(&exec_ns); |
128 | 1 | Defer task_running_defer {[&]() { |
129 | | // If fragment is finished, fragment context will be de-constructed with all tasks in it. |
130 | 1 | if (done || !status.ok()) { |
131 | 1 | auto id = task->pipeline_id(); |
132 | 1 | close_task(task.get(), status, fragment_context.get()); |
133 | 1 | task->set_running(false); |
134 | 1 | fragment_context->decrement_running_task(id); |
135 | 1 | } else { |
136 | 0 | task->set_running(false); |
137 | 0 | } |
138 | 1 | _task_queue.update_statistics(task.get(), exec_ns); |
139 | 1 | }}; |
140 | 1 | bool canceled = fragment_context->is_canceled(); |
141 | | |
142 | | // Close task if canceled |
143 | 1 | if (canceled) { |
144 | 0 | status = fragment_context->get_query_ctx()->exec_status(); |
145 | 0 | DCHECK(!status.ok()); |
146 | 0 | continue; |
147 | 0 | } |
148 | | |
149 | | // Main logics of execution |
150 | 1 | try { |
151 | 1 | ASSIGN_STATUS_IF_CATCH_EXCEPTION(status = task->execute(&done), status); |
152 | 1 | } catch (const std::exception& e) { |
153 | 1 | status = Status::InternalError("Catch std::exception: {}", e.what()); |
154 | 1 | } |
155 | 1 | fragment_context->trigger_report_if_necessary(); |
156 | 1 | } |
157 | 2 | } |
158 | | |
159 | 54 | void TaskScheduler::stop() { |
160 | 54 | if (!_shutdown) { |
161 | 48 | _task_queue.close(); |
162 | 48 | if (_fix_thread_pool) { |
163 | 2 | _need_to_stop = true; |
164 | 2 | _fix_thread_pool->shutdown(); |
165 | 2 | _fix_thread_pool->wait(); |
166 | 2 | } |
167 | | // Should set at the ending of the stop to ensure that the |
168 | | // pool is stopped. For example, if there are 2 threads call stop |
169 | | // then if one thread set shutdown = false, then another thread will |
170 | | // not check it and will free task scheduler. |
171 | 48 | _shutdown = true; |
172 | 48 | } |
173 | 54 | } |
174 | | |
175 | 3 | Status HybridTaskScheduler::submit(PipelineTaskSPtr task) { |
176 | 3 | bool blockable = false; |
177 | 3 | { |
178 | 3 | std::unique_lock<std::mutex> blockable_check_lock(task->_blockable_check_lock); |
179 | 3 | if (!task->_accept_submit) { |
180 | 2 | return Status::OK(); |
181 | 2 | } |
182 | 1 | blockable = task->is_blockable(); |
183 | 1 | } |
184 | 1 | if (blockable) { |
185 | 0 | return _blocking_scheduler.submit(task); |
186 | 1 | } else { |
187 | 1 | return _simple_scheduler.submit(task); |
188 | 1 | } |
189 | 1 | } |
190 | | |
191 | 1 | Status HybridTaskScheduler::start() { |
192 | 1 | RETURN_IF_ERROR(_blocking_scheduler.start()); |
193 | 1 | RETURN_IF_ERROR(_simple_scheduler.start()); |
194 | 1 | return Status::OK(); |
195 | 1 | } |
196 | | |
197 | 3 | void HybridTaskScheduler::stop() { |
198 | 3 | _blocking_scheduler.stop(); |
199 | 3 | _simple_scheduler.stop(); |
200 | 3 | } |
201 | | |
202 | | } // namespace doris |