westonpace commented on a change in pull request #9095: URL: https://github.com/apache/arrow/pull/9095#discussion_r565103153
########## File path: cpp/src/arrow/util/async_iterator.h ########## @@ -0,0 +1,295 @@ +// 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. + +#pragma once +#include <queue> + +#include "arrow/util/functional.h" +#include "arrow/util/future.h" +#include "arrow/util/iterator.h" +#include "arrow/util/optional.h" +#include "arrow/util/thread_pool.h" + +namespace arrow { + +template <typename T> +using AsyncGenerator = std::function<Future<T>()>; + +/// Iterates through a generator of futures, visiting the result of each one and +/// returning a future that completes when all have been visited +template <typename T> +Future<> VisitAsyncGenerator(AsyncGenerator<T> generator, + std::function<Status(T)> visitor) { + auto loop_body = [generator, visitor] { + auto next = generator(); + return next.Then([visitor](const T& result) -> Result<ControlFlow<detail::Empty>> { + if (result == IterationTraits<T>::End()) { + return Break(detail::Empty()); + } else { + auto visited = visitor(result); + if (visited.ok()) { + return Continue(); + } else { + return visited; + } + } + }); + }; + return Loop(loop_body); +} + +template <typename T> +Future<std::vector<T>> CollectAsyncGenerator(AsyncGenerator<T> generator) { + auto vec = std::make_shared<std::vector<T>>(); + auto loop_body = [generator, vec] { + auto next = generator(); + return next.Then([vec](const T& result) -> Result<ControlFlow<std::vector<T>>> { + if (result == IterationTraits<T>::End()) { + return Break(*vec); + } else { + vec->push_back(result); + return Continue(); + } + }); + }; + return Loop(loop_body); +} + +template <typename T, typename V> +class TransformingGenerator { + public: + explicit TransformingGenerator(AsyncGenerator<T> generator, + Transformer<T, V> transformer) + : finished_(), last_value_(), generator_(generator), transformer_(transformer) {} + + // See comment on TransformingIterator::Pump + Result<util::optional<V>> Pump() { + if (!finished_ && last_value_.has_value()) { + ARROW_ASSIGN_OR_RAISE(TransformFlow<V> next, transformer_(*last_value_)); + if (next.ReadyForNext()) { + if (*last_value_ == IterationTraits<T>::End()) { + finished_ = true; + } + last_value_.reset(); + } + if (next.Finished()) { + finished_ = true; + } + if (next.HasValue()) { + return next.Value(); + } + } + if (finished_) { + return IterationTraits<V>::End(); + } + return util::nullopt; + } + + Future<V> operator()() { + while (true) { + auto maybe_next_result = Pump(); + if (!maybe_next_result.ok()) { + return Future<V>::MakeFinished(maybe_next_result.status()); + } + auto maybe_next = maybe_next_result.ValueUnsafe(); + if (maybe_next.has_value()) { + return Future<V>::MakeFinished(*maybe_next); + } + + auto next_fut = generator_(); + // If finished already, process results immediately inside the loop to avoid stack + // overflow + if (next_fut.is_finished()) { + auto next_result = next_fut.result(); + if (next_result.ok()) { + last_value_ = *next_result; + } else { + return Future<V>::MakeFinished(next_result.status()); + } + // Otherwise, if not finished immediately, add callback to process results + } else { + return next_fut.Then([this](const Result<T>& next_result) { + if (next_result.ok()) { + last_value_ = *next_result; + return (*this)(); + } else { + return Future<V>::MakeFinished(next_result.status()); + } + }); + } + } + } + + protected: + bool finished_; + util::optional<T> last_value_; + AsyncGenerator<T> generator_; + Transformer<T, V> transformer_; +}; + +template <typename T> +static std::function<void(const Result<T>&)> MakeCallback( + std::shared_ptr<bool> finished) { + return [finished](const Result<T>& next_result) { + if (!next_result.ok()) { + *finished = true; + } else { + auto next = *next_result; + *finished = (next == IterationTraits<T>::End()); + } + }; +} + +template <typename T> +AsyncGenerator<T> AddReadahead(AsyncGenerator<T> source_generator, int max_readahead) { + // Using a shared_ptr instead of a lambda capture here because it's possible that + // the inner mark_finished_if_done outlives the outer lambda + auto finished = std::make_shared<bool>(false); + auto mark_finished_if_done = [finished](const Result<T>& next_result) { + if (!next_result.ok()) { + *finished = true; + } else { + auto next = *next_result; + *finished = (next == IterationTraits<T>::End()); + } + }; + + std::queue<Future<T>> readahead_queue; + return [=]() mutable -> Future<T> { + if (readahead_queue.empty()) { + // This is the first request, let's pump the underlying queue + for (int i = 0; i < max_readahead; i++) { + auto next = source_generator(); + next.AddCallback(mark_finished_if_done); + readahead_queue.push(std::move(next)); + } + } + // Pop one and add one + auto result = readahead_queue.front(); + readahead_queue.pop(); + if (*finished) { + readahead_queue.push(Future<T>::MakeFinished(IterationTraits<T>::End())); + } else { + auto back_of_queue = source_generator(); + back_of_queue.AddCallback(mark_finished_if_done); + readahead_queue.push(std::move(back_of_queue)); + } + return result; + }; +} + +/// \brief Transforms an async generator using a transformer function returning a new +/// AsyncGenerator +/// +/// The transform function here behaves exactly the same as the transform function in +/// MakeTransformedIterator and you can safely use the same transform function to +/// transform both synchronous and asynchronous streams. +template <typename T, typename V> +AsyncGenerator<V> TransformAsyncGenerator(AsyncGenerator<T> generator, + Transformer<T, V> transformer) { + return TransformingGenerator<T, V>(generator, transformer); +} + +namespace detail { + +template <typename T> +struct BackgroundIteratorPromise : ReadaheadPromise { + ~BackgroundIteratorPromise() override {} + + explicit BackgroundIteratorPromise(Iterator<T>* it) : it_(it) {} + + bool Call() override { + auto next = it_->Next(); + auto finished = next == IterationTraits<T>::End(); + out_.MarkFinished(std::move(next)); + return finished; + } + + void End() override { out_.MarkFinished(IterationTraits<T>::End()); } + + Iterator<T>* it_; + Future<T> out_ = Future<T>::Make(); +}; + +} // namespace detail + +/// \brief Async generator that iterates on an underlying iterator in a +/// separate thread. Review comment: No, not in an executor at the moment, and probably never will. At some point, I think the filesystem interfaces will support async and then the filesystems themselves can assign executors where appropriate. Once that is done there will be no need for this "backgrounding" class. The background threads certainly don't belong on the CPU thread pool. We could create a "background" thread pool but it would have to be large enough to support the maximum parallelism we want from the filesystem (which may be greater than the # of CPUs in a slow networked I/O case). This generator (I will change that name per next comment) can indeed be called from several threads at once. Although there is no need for it and we can do away with that requirement if it helps us save on concurrency cost. I'll write a separate note on the parallelism of these generators. The underlying iterator will never be iterated on by any more than the background thread. ---------------------------------------------------------------- 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. For queries about this service, please contact Infrastructure at: [email protected]
