ClickHouse/dbms/DataStreams/ParallelInputsProcessor.h

367 lines
12 KiB
C++
Raw Normal View History

#pragma once
#include <list>
2015-04-23 00:20:41 +00:00
#include <queue>
#include <atomic>
#include <thread>
#include <mutex>
2015-09-29 19:19:54 +00:00
#include <common/logger_useful.h>
#include <DataStreams/IBlockStream_fwd.h>
#include <Common/setThreadName.h>
#include <Common/CurrentMetrics.h>
2018-05-29 18:14:31 +00:00
#include <Common/CurrentThread.h>
#include <Common/ThreadPool.h>
/** Allows to process multiple block input streams (sources) in parallel, using specified number of threads.
* Reads (pulls) blocks from any available source and passes it to specified handler.
*
2017-12-04 20:29:13 +00:00
* Before any reading, calls "readPrefix" method of sources in parallel.
*
* (As an example, "readPrefix" can prepare connections to remote servers,
* and we want this work to be executed in parallel for different sources)
*
* Implemented in following way:
* - there are multiple input sources to read blocks from;
* - there are multiple threads, that could simultaneously read blocks from different sources;
* - "available" sources (that are not read in any thread right now) are put in queue of sources;
* - when thread take a source to read from, it removes source from queue of sources,
* then read block from source and then put source back to queue of available sources.
*/
namespace CurrentMetrics
{
extern const Metric QueryThread;
}
namespace DB
{
2017-05-13 22:19:04 +00:00
/// Example of the handler.
struct ParallelInputsHandler
{
2017-05-13 22:19:04 +00:00
/// Processing the data block.
2017-12-01 19:34:51 +00:00
void onBlock(Block & /*block*/, size_t /*thread_num*/) {}
2017-05-13 22:19:04 +00:00
/// Called for each thread, when the thread has nothing else to do.
/// Due to the fact that part of the sources has run out, and now there are fewer sources left than streams.
/// Called if the `onException` method does not throw an exception; is called before the `onFinish` method.
2017-12-01 19:34:51 +00:00
void onFinishThread(size_t /*thread_num*/) {}
2015-12-01 21:20:14 +00:00
2017-05-13 22:19:04 +00:00
/// Blocks are over. Due to the fact that all sources ran out or because of the cancellation of work.
/// This method is always called exactly once, at the end of the work, if the `onException` method does not throw an exception.
void onFinish() {}
2017-05-13 22:19:04 +00:00
/// Exception handling. It is reasonable to call the ParallelInputsProcessor::cancel method in this method, and also pass the exception to the main thread.
2017-12-01 19:34:51 +00:00
void onException(std::exception_ptr & /*exception*/, size_t /*thread_num*/) {}
};
2018-11-28 14:33:40 +00:00
template <typename Handler>
class ParallelInputsProcessor
{
public:
2017-05-13 22:19:04 +00:00
/** additional_input_at_end - if not nullptr,
* then the blocks from this source will start to be processed only after all other sources are processed.
* This is done in the main thread.
*
2017-05-13 22:19:04 +00:00
* Intended for implementation of FULL and RIGHT JOIN
* - where you must first make JOIN in parallel, while noting which keys are not found,
* and only after the completion of this work, create blocks of keys that are not found.
*/
ParallelInputsProcessor(const BlockInputStreams & inputs_, const BlockInputStreamPtr & additional_input_at_end_, size_t max_threads_, Handler & handler_)
: inputs(inputs_), additional_input_at_end(additional_input_at_end_), max_threads(std::min(inputs_.size(), max_threads_)), handler(handler_)
{
for (size_t i = 0; i < inputs_.size(); ++i)
unprepared_inputs.emplace(inputs_[i], i);
}
~ParallelInputsProcessor()
{
try
{
wait();
}
catch (...)
{
tryLogCurrentException(__PRETTY_FUNCTION__);
}
}
2017-05-13 22:19:04 +00:00
/// Start background threads, start work.
void process()
{
active_threads = max_threads;
threads.reserve(max_threads);
try
{
for (size_t i = 0; i < max_threads; ++i)
2019-07-01 00:15:14 +00:00
threads.emplace_back(&ParallelInputsProcessor::thread, this, CurrentThread::getGroup(), i);
}
catch (...)
{
cancel(false);
wait();
if (active_threads)
{
active_threads = 0;
/// handler.onFinish() is supposed to be called from one of the threads when the number of
/// finished threads reaches max_threads. But since we weren't able to launch all threads,
/// we have to call onFinish() manually here.
handler.onFinish();
}
throw;
}
}
2017-05-13 22:19:04 +00:00
/// Ask all sources to stop earlier than they run out.
void cancel(bool kill)
{
finish = true;
for (auto & input : inputs)
{
try
{
input->cancel(kill);
}
catch (...)
{
/** If you can not ask one or more sources to stop.
* (for example, the connection is broken for distributed query processing)
* - then do not care.
*/
LOG_ERROR(log, "Exception while cancelling " << input->getName());
}
}
}
2017-05-13 22:19:04 +00:00
/// Wait until all threads are finished, before the destructor.
void wait()
{
if (joined_threads)
return;
for (auto & thread : threads)
thread.join();
threads.clear();
joined_threads = true;
}
size_t getNumActiveThreads() const
{
return active_threads;
}
private:
2017-05-13 22:19:04 +00:00
/// Single source data
struct InputData
{
BlockInputStreamPtr in;
2019-04-22 16:07:09 +00:00
size_t i = 0; /// The source number (for debugging).
InputData() {}
InputData(const BlockInputStreamPtr & in_, size_t i_) : in(in_), i(i_) {}
};
2018-11-28 14:33:40 +00:00
void publishPayload(Block & block, size_t thread_num)
{
2018-11-28 14:33:40 +00:00
handler.onBlock(block, thread_num);
}
void thread(ThreadGroupStatusPtr thread_group, size_t thread_num)
{
std::exception_ptr exception;
CurrentMetrics::Increment metric_increment{CurrentMetrics::QueryThread};
try
{
setThreadName("ParalInputsProc");
if (thread_group)
CurrentThread::attachTo(thread_group);
2017-12-04 20:29:13 +00:00
while (!finish)
{
InputData unprepared_input;
{
2019-01-02 06:44:36 +00:00
std::lock_guard lock(unprepared_inputs_mutex);
if (unprepared_inputs.empty())
break;
unprepared_input = unprepared_inputs.front();
unprepared_inputs.pop();
}
unprepared_input.in->readPrefix();
{
2019-01-02 06:44:36 +00:00
std::lock_guard lock(available_inputs_mutex);
available_inputs.push(unprepared_input);
}
}
2017-12-04 20:29:13 +00:00
loop(thread_num);
handler.onFinishThread(thread_num);
}
catch (...)
{
exception = std::current_exception();
}
if (exception)
{
handler.onException(exception, thread_num);
}
2017-05-13 22:19:04 +00:00
/// The last thread on the output indicates that there is no more data.
if (0 == --active_threads)
{
2017-05-13 22:19:04 +00:00
/// And then it processes an additional source, if there is one.
if (additional_input_at_end)
{
try
{
additional_input_at_end->readPrefix();
while (Block block = additional_input_at_end->read())
2018-11-28 14:33:40 +00:00
publishPayload(block, thread_num);
}
catch (...)
{
exception = std::current_exception();
}
if (exception)
{
handler.onException(exception, thread_num);
}
}
try
{
handler.onFinish();
}
catch (...)
{
exception = std::current_exception();
}
if (exception)
{
handler.onException(exception, thread_num);
}
}
}
void loop(size_t thread_num)
{
2017-05-13 22:19:04 +00:00
while (!finish) /// You may need to stop work earlier than all sources run out.
{
InputData input;
2017-05-13 22:19:04 +00:00
/// Select the next source.
{
2019-01-02 06:44:36 +00:00
std::lock_guard lock(available_inputs_mutex);
2017-05-13 22:19:04 +00:00
/// If there are no free sources, then this thread is no longer needed. (But other threads can work with their sources.)
if (available_inputs.empty())
break;
input = available_inputs.front();
2017-05-13 22:19:04 +00:00
/// We remove the source from the queue of available sources.
available_inputs.pop();
}
2017-05-13 22:19:04 +00:00
/// The main work.
Block block = input.in->read();
{
if (finish)
break;
2017-05-13 22:19:04 +00:00
/// If this source is not run out yet, then put the resulting block in the ready queue.
{
2019-01-02 06:44:36 +00:00
std::lock_guard lock(available_inputs_mutex);
if (block)
{
available_inputs.push(input);
}
else
{
if (available_inputs.empty())
break;
}
}
if (finish)
break;
if (block)
2018-11-28 14:33:40 +00:00
publishPayload(block, thread_num);
}
}
}
BlockInputStreams inputs;
BlockInputStreamPtr additional_input_at_end;
unsigned max_threads;
Handler & handler;
/// Threads.
using ThreadsData = std::vector<ThreadFromGlobalPool>;
ThreadsData threads;
2017-05-13 22:19:04 +00:00
/** A set of available sources that are not currently processed by any thread.
* Each thread takes one source from this set, takes a block out of the source (at this moment the source does the calculations)
* and (if the source is not run out), puts it back into the set of available sources.
*
2017-05-13 22:19:04 +00:00
* The question arises what is better to use:
* - the queue (just processed source will be processed the next time later than the rest)
* - stack (just processed source will be processed as soon as possible).
*
2017-05-13 22:19:04 +00:00
* The stack is better than the queue when you need to do work on reading one source more consequentially,
* and theoretically, this allows you to achieve more consequent/consistent reads from the disk.
*
2017-05-13 22:19:04 +00:00
* But when using the stack, there is a problem with distributed query processing:
* data is read only from a part of the servers, and on the other servers
* a timeout occurs during send, and the request processing ends with an exception.
*
2017-05-13 22:19:04 +00:00
* Therefore, a queue is used. This can be improved in the future.
*/
using AvailableInputs = std::queue<InputData>;
AvailableInputs available_inputs;
2017-12-04 20:29:13 +00:00
/** For parallel preparing (readPrefix) child streams.
* First, streams are located here.
* After a stream was prepared, it is moved to "available_inputs" for reading.
*/
using UnpreparedInputs = std::queue<InputData>;
UnpreparedInputs unprepared_inputs;
2017-05-13 22:19:04 +00:00
/// For operations with available_inputs.
std::mutex available_inputs_mutex;
/// For operations with unprepared_inputs.
std::mutex unprepared_inputs_mutex;
2017-05-13 22:19:04 +00:00
/// How many sources ran out.
std::atomic<size_t> active_threads { 0 };
2017-05-13 22:19:04 +00:00
/// Finish the threads work (before the sources run out).
std::atomic<bool> finish { false };
2017-05-13 22:19:04 +00:00
/// Wait for the completion of all threads.
std::atomic<bool> joined_threads { false };
Logger * log = &Logger::get("ParallelInputsProcessor");
};
}