kaka11chen commented on code in PR #49671: URL: https://github.com/apache/doris/pull/49671#discussion_r2048809115
########## be/src/vec/exec/executor/time_sharing/time_sharing_task_executor.cpp: ########## @@ -0,0 +1,393 @@ +// 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 "vec/exec/executor/time_sharing/time_sharing_task_executor.h" + +#include <functional> +#include <iostream> + +#include "util/defer_op.h" +#include "util/thread.h" +#include "util/threadpool.h" +#include "vec/exec/executor/time_sharing/time_sharing_task_handle.h" + +namespace doris { +namespace vectorized { + +TimeSharingTaskExecutor::TimeSharingTaskExecutor( + ThreadConfig thread_config, int min_concurrency, int guaranteed_concurrency_per_task, + int max_concurrency_per_task, std::shared_ptr<Ticker> ticker, + std::chrono::milliseconds stuck_split_warning_threshold, + std::shared_ptr<MultilevelSplitQueue> split_queue) + : _thread_config(thread_config), + _min_concurrency(min_concurrency), + _guaranteed_concurrency_per_task(guaranteed_concurrency_per_task), + _max_concurrency_per_task(max_concurrency_per_task), + _ticker(ticker != nullptr ? ticker : std::make_shared<SystemTicker>()), + _stuck_split_warning_threshold(stuck_split_warning_threshold), + _waiting_splits(split_queue != nullptr ? std::move(split_queue) + : std::make_shared<MultilevelSplitQueue>(2)) {} + +Status TimeSharingTaskExecutor::init() { + static_cast<void>(_stuck_split_warning_threshold); + + ThreadPoolBuilder builder(_thread_config.thread_name); + builder.set_min_threads(_thread_config.min_thread_num) + .set_max_threads(_thread_config.max_thread_num) + .set_max_queue_size(_thread_config.max_queue_size) + .set_cgroup_cpu_ctl(_thread_config.cgroup_cpu_ctl); + RETURN_IF_ERROR(builder.build(&_thread_pool)); + return Status::OK(); +} + +TimeSharingTaskExecutor::~TimeSharingTaskExecutor() { + if (!_stopped.exchange(true)) { + stop(); + } + + std::vector<std::shared_ptr<PrioritizedSplitRunner>> splits_to_destroy; + { + std::lock_guard<std::mutex> guard(_mutex); + for (auto& [task_id, handle] : _tasks) { + if (handle->is_closed()) { + LOG(WARNING) << "task is already destroyed, task_id: " << task_id.to_string(); + continue; + } + auto task_splits = handle->close(); + splits_to_destroy.insert(splits_to_destroy.end(), + std::make_move_iterator(task_splits.begin()), + std::make_move_iterator(task_splits.end())); + _record_leaf_splits_size(guard); + } + //_tasks.clear(); + //_all_splits.clear(); + //_intermediate_splits.clear(); + //_blocked_splits.clear(); + _waiting_splits->remove_all(splits_to_destroy); + } + + if (splits_to_destroy.empty()) { + return; + } + + for (auto& split : splits_to_destroy) { + split->close(Status::OK()); + } +} + +Status TimeSharingTaskExecutor::start() { + std::lock_guard<std::mutex> guard(_mutex); + for (int i = 0; i < _thread_config.min_thread_num; ++i) { + RETURN_IF_ERROR(_add_runner_thread()); + } + return Status::OK(); +} + +void TimeSharingTaskExecutor::stop() { + _waiting_splits->interrupt(); + { + std::lock_guard<std::mutex> guard(_mutex); + _stopped = true; + } + if (_thread_pool) { + _thread_pool->shutdown(); + _thread_pool->wait(); + } +} + +Status TimeSharingTaskExecutor::_add_runner_thread() { + return _thread_pool->submit_func([this]() { + Thread::set_self_name("SplitRunner"); + + while (!_stopped) { + std::shared_ptr<PrioritizedSplitRunner> split; + split = _waiting_splits->take(); + if (!split) { + return; + } + + { + std::lock_guard<std::mutex> guard(_mutex); + _running_splits.insert(split); + } + Defer defer {[&]() { + std::lock_guard<std::mutex> guard(_mutex); + _running_splits.erase(split); + }}; + Result<SharedListenableFuture<Void>> blocked_future_result = split->process(); + if (!blocked_future_result.has_value()) { + return; + } + auto blocked_future = blocked_future_result.value(); + + if (split->is_finished()) { + _split_finished(split, split->finished_status()); + } else { + std::lock_guard<std::mutex> guard(_mutex); + if (blocked_future.is_done()) { + _waiting_splits->offer(split); + } else { + _blocked_splits[split] = blocked_future; + + _blocked_splits[split].add_callback( + [this, split](const Void& value, const Status& status) { + if (status.ok()) { + std::lock_guard<std::mutex> guard(_mutex); + _blocked_splits.erase(split); + split->reset_level_priority(); + _waiting_splits->offer(split); + } else { + LOG(WARNING) << "blocked split is failed, split_id: " + << split->split_id() << ", status: " << status; + _split_finished(split, status); + } + }); + } + } + } + }); +} + +Result<std::shared_ptr<TaskHandle>> TimeSharingTaskExecutor::create_task( + const TaskId& task_id, std::function<double()> utilization_supplier, + int initial_split_concurrency, std::chrono::nanoseconds split_concurrency_adjust_frequency, + std::optional<int> max_concurrency_per_task) { + auto task_handle = std::make_shared<TimeSharingTaskHandle>( + task_id, _waiting_splits, utilization_supplier, initial_split_concurrency, + split_concurrency_adjust_frequency, max_concurrency_per_task); + RETURN_IF_ERROR_RESULT(task_handle->init()); + + std::lock_guard<std::mutex> lock(_mutex); + + _tasks[task_id] = task_handle; + + return task_handle; +} + +Status TimeSharingTaskExecutor::add_task(const TaskId& task_id, + std::shared_ptr<TaskHandle> task_handle) { + std::lock_guard<std::mutex> lock(_mutex); + _tasks[task_id] = + std::dynamic_pointer_cast<doris::vectorized::TimeSharingTaskHandle>(task_handle); + return Status::OK(); +} + +Status TimeSharingTaskExecutor::remove_task(std::shared_ptr<TaskHandle> task_handle) { + auto handle = std::dynamic_pointer_cast<TimeSharingTaskHandle>(task_handle); + std::vector<std::shared_ptr<PrioritizedSplitRunner>> splits_to_destroy; + + { + std::lock_guard<std::mutex> guard(_mutex); + auto it = _tasks.find(handle->task_id()); + if (it == _tasks.end() || handle->is_closed()) { + return Status::OK(); + } + _tasks.erase(it); + + // Task is already closed + if (task_handle->is_closed()) { + return Status::OK(); + } + + splits_to_destroy = handle->close(); + + for (const auto& split : splits_to_destroy) { + _all_splits.erase(split); + _intermediate_splits.erase(split); + _blocked_splits.erase(split); + } + _waiting_splits->remove_all(splits_to_destroy); + _record_leaf_splits_size(guard); + } + + // call destroy outside of synchronized block as it is expensive and doesn't need a lock on the task executor + for (auto& split : splits_to_destroy) { + split->close(Status::OK()); + } + + // record completed stats + int64_t thread_usage_nanos = handle->scheduled_nanos(); + int level = MultilevelSplitQueue::compute_level(thread_usage_nanos); + _completed_tasks_per_level[level]++; + + if (splits_to_destroy.empty()) { + return Status::OK(); + } + + // replace blocked splits that were terminated + { + std::lock_guard<std::mutex> guard(_mutex); + _add_new_entrants(guard); + _record_leaf_splits_size(guard); + } + return Status::OK(); +} + +Result<std::vector<SharedListenableFuture<Void>>> TimeSharingTaskExecutor::enqueue_splits( + std::shared_ptr<TaskHandle> task_handle, bool intermediate, + const std::vector<std::shared_ptr<SplitRunner>>& splits) { + std::vector<std::shared_ptr<PrioritizedSplitRunner>> splits_to_destroy; + Defer defer {[&]() { + for (auto& split : splits_to_destroy) { + split->close(Status::OK()); + } + }}; + std::vector<SharedListenableFuture<Void>> finished_futures; + auto handle = std::dynamic_pointer_cast<TimeSharingTaskHandle>(task_handle); + { + std::lock_guard<std::mutex> guard(_mutex); + for (const auto& task_split : splits) { + TaskId task_id = handle->task_id(); + int split_id = handle->next_split_id(); + + auto prioritized_split = + PrioritizedSplitRunner::create_shared(handle, split_id, task_split, _ticker); + RETURN_IF_ERROR_RESULT(prioritized_split->init()); + if (intermediate) { + if (handle->record_intermediate_split(prioritized_split)) { + _start_intermediate_split(prioritized_split, guard); + } else { + splits_to_destroy.push_back(prioritized_split); + } + } else { + if (handle->enqueue_split(prioritized_split)) { + _schedule_task_if_necessary(handle, guard); + _add_new_entrants(guard); + } else { + splits_to_destroy.push_back(prioritized_split); + } + } + finished_futures.push_back(prioritized_split->finished_future()); + } + _record_leaf_splits_size(guard); + } + return finished_futures; +} + +void TimeSharingTaskExecutor::_split_finished(std::shared_ptr<PrioritizedSplitRunner> split, + const Status& status) { + _completed_splits_per_level[split->priority().level()]++; + { + std::lock_guard<std::mutex> guard(_mutex); + _all_splits.erase(split); + + auto task_handle = split->task_handle(); + task_handle->split_finished(split); + + _schedule_task_if_necessary(task_handle, guard); + + _add_new_entrants(guard); + _record_leaf_splits_size(guard); + } + // call close outside of synchronized block as it is expensive and doesn't need a lock on the task executor + split->close(status); +} + +void TimeSharingTaskExecutor::_schedule_task_if_necessary( + std::shared_ptr<TimeSharingTaskHandle> task_handle, std::lock_guard<std::mutex>& guard) { + int guaranteed_concurrency = std::min( + _guaranteed_concurrency_per_task, + task_handle->max_concurrency_per_task().value_or(std::numeric_limits<int>::max())); + int splits_to_schedule = guaranteed_concurrency - task_handle->running_leaf_splits(); + + for (int i = 0; i < splits_to_schedule; ++i) { + auto split = task_handle->poll_next_split(); + if (!split) return; + + _start_split(split, guard); + auto elapsed_nanos = std::chrono::nanoseconds( + std::chrono::steady_clock::now().time_since_epoch().count() - + split->created_nanos()); + _split_queued_time + << std::chrono::duration_cast<std::chrono::microseconds>(elapsed_nanos).count(); + } + _record_leaf_splits_size(guard); +} + +void TimeSharingTaskExecutor::_add_new_entrants(std::lock_guard<std::mutex>& guard) { + int running = _all_splits.size() - _intermediate_splits.size(); + for (int i = 0; i < _min_concurrency - running; i++) { + auto split = _poll_next_split_worker(guard); + if (!split) { + break; + } + + auto elapsed_nanos = std::chrono::nanoseconds( + std::chrono::steady_clock::now().time_since_epoch().count() - + split->created_nanos()); + _split_queued_time + << std::chrono::duration_cast<std::chrono::microseconds>(elapsed_nanos).count(); + _start_split(split, guard); + } +} + +void TimeSharingTaskExecutor::_start_intermediate_split( + std::shared_ptr<PrioritizedSplitRunner> split, std::lock_guard<std::mutex>& guard) { + _start_split(split, guard); + _intermediate_splits.insert(split); +} + +void TimeSharingTaskExecutor::_start_split(std::shared_ptr<PrioritizedSplitRunner> split, + std::lock_guard<std::mutex>& guard) { + _all_splits.insert(split); + _waiting_splits->offer(split); +} + +std::shared_ptr<PrioritizedSplitRunner> TimeSharingTaskExecutor::_poll_next_split_worker( + std::lock_guard<std::mutex>& guard) { + for (auto it = _tasks.begin(); it != _tasks.end();) { + auto task = it->second; + if (task->running_leaf_splits() >= + task->max_concurrency_per_task().value_or(_max_concurrency_per_task)) { + ++it; + continue; + } + + auto split = task->poll_next_split(); + if (split) { + auto task_copy = task; + auto task_id = it->first; + it = _tasks.erase(it); + _tasks[task_id] = task_copy; + return split; + } + ++it; + } + return nullptr; +} + +void TimeSharingTaskExecutor::_record_leaf_splits_size(std::lock_guard<std::mutex>& guard) {} Review Comment: can remove it. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: commits-unsubscr...@doris.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org --------------------------------------------------------------------- To unsubscribe, e-mail: commits-unsubscr...@doris.apache.org For additional commands, e-mail: commits-h...@doris.apache.org
