boost/log/sinks/unbounded_ordering_queue.hpp
/*
* Copyright Andrey Semashev 2007 - 2013.
* Distributed under the Boost Software License, Version 1.0.
* (See accompanying file LICENSE_1_0.txt or copy at
* http://www.boost.org/LICENSE_1_0.txt)
*/
/*!
* \file unbounded_ordering_queue.hpp
* \author Andrey Semashev
* \date 24.07.2011
*
* The header contains implementation of unbounded ordering record queueing strategy for
* the asynchronous sink frontend.
*/
#ifndef BOOST_LOG_SINKS_UNBOUNDED_ORDERING_QUEUE_HPP_INCLUDED_
#define BOOST_LOG_SINKS_UNBOUNDED_ORDERING_QUEUE_HPP_INCLUDED_
#include <boost/log/detail/config.hpp>
#ifdef BOOST_HAS_PRAGMA_ONCE
#pragma once
#endif
#if defined(BOOST_LOG_NO_THREADS)
#error Boost.Log: This header content is only supported in multithreaded environment
#endif
#include <queue>
#include <vector>
#include <boost/cstdint.hpp>
#include <boost/move/core.hpp>
#include <boost/move/utility.hpp>
#include <boost/thread/locks.hpp>
#include <boost/thread/mutex.hpp>
#include <boost/thread/condition_variable.hpp>
#include <boost/thread/thread_time.hpp>
#include <boost/date_time/posix_time/posix_time_types.hpp>
#include <boost/log/detail/timestamp.hpp>
#include <boost/log/keywords/order.hpp>
#include <boost/log/keywords/ordering_window.hpp>
#include <boost/log/core/record_view.hpp>
#include <boost/log/detail/header.hpp>
namespace boost {
BOOST_LOG_OPEN_NAMESPACE
namespace sinks {
/*!
* \brief Unbounded ordering log record queueing strategy
*
* The \c unbounded_ordering_queue class is intended to be used with
* the \c asynchronous_sink frontend as a log record queueing strategy.
*
* This strategy provides the following properties to the record queueing mechanism:
*
* \li The queue has no size limits.
* \li The queue has a fixed latency window. This means that each log record put
* into the queue will normally not be dequeued for a certain period of time.
* \li The queue performs stable record ordering within the latency window.
* The ordering predicate can be specified in the \c OrderT template parameter.
*
* Since this queue has no size limits, it may grow uncontrollably if sink backends
* dequeue log records not fast enough. When this is an issue, it is recommended to
* use one of the bounded strategies.
*/
template< typename OrderT >
class unbounded_ordering_queue
{
private:
typedef boost::mutex mutex_type;
//! Log record with enqueueing timestamp
class enqueued_record
{
BOOST_COPYABLE_AND_MOVABLE(enqueued_record)
public:
//! Ordering predicate
struct order :
public OrderT
{
typedef typename OrderT::result_type result_type;
order() {}
order(order const& that) : OrderT(static_cast< OrderT const& >(that)) {}
order(OrderT const& that) : OrderT(that) {}
result_type operator() (enqueued_record const& left, enqueued_record const& right) const
{
// std::priority_queue requires ordering with semantics of std::greater, so we swap arguments
return OrderT::operator() (right.m_record, left.m_record);
}
};
boost::log::aux::timestamp m_timestamp;
record_view m_record;
enqueued_record(enqueued_record const& that) : m_timestamp(that.m_timestamp), m_record(that.m_record)
{
}
enqueued_record(BOOST_RV_REF(enqueued_record) that) :
m_timestamp(that.m_timestamp),
m_record(boost::move(that.m_record))
{
}
explicit enqueued_record(record_view const& rec) :
m_timestamp(boost::log::aux::get_timestamp()),
m_record(rec)
{
}
enqueued_record& operator= (BOOST_COPY_ASSIGN_REF(enqueued_record) that)
{
m_timestamp = that.m_timestamp;
m_record = that.m_record;
return *this;
}
enqueued_record& operator= (BOOST_RV_REF(enqueued_record) that)
{
m_timestamp = that.m_timestamp;
m_record = boost::move(that.m_record);
return *this;
}
};
typedef std::priority_queue<
enqueued_record,
std::vector< enqueued_record >,
typename enqueued_record::order
> queue_type;
private:
//! Ordering window duration, in milliseconds
const uint64_t m_ordering_window;
//! Synchronization mutex
mutex_type m_mutex;
//! Condition for blocking
condition_variable m_cond;
//! Thread-safe queue
queue_type m_queue;
//! Interruption flag
bool m_interruption_requested;
public:
/*!
* Returns ordering window size specified during initialization
*/
posix_time::time_duration get_ordering_window() const
{
return posix_time::milliseconds(m_ordering_window);
}
/*!
* Returns default ordering window size.
* The default window size is specific to the operating system thread scheduling mechanism.
*/
static posix_time::time_duration get_default_ordering_window()
{
// The main idea behind this parameter is that the ordering window should be large enough
// to allow the frontend to order records from different threads on an attribute
// that contains system time. Thus this value should be:
// * No less than the minimum time resolution quant that Boost.DateTime provides on the current OS.
// For instance, on Windows it defaults to around 15-16 ms.
// * No less than thread switching quant on the current OS. For now 30 ms is large enough window size to
// switch threads on any known OS. It can be tuned for other platforms as needed.
return posix_time::milliseconds(30);
}
protected:
//! Initializing constructor
template< typename ArgsT >
explicit unbounded_ordering_queue(ArgsT const& args) :
m_ordering_window(args[keywords::ordering_window || &unbounded_ordering_queue::get_default_ordering_window].total_milliseconds()),
m_queue(args[keywords::order]),
m_interruption_requested(false)
{
}
//! Enqueues log record to the queue
void enqueue(record_view const& rec)
{
lock_guard< mutex_type > lock(m_mutex);
enqueue_unlocked(rec);
}
//! Attempts to enqueue log record to the queue
bool try_enqueue(record_view const& rec)
{
unique_lock< mutex_type > lock(m_mutex, try_to_lock);
if (lock.owns_lock())
{
enqueue_unlocked(rec);
return true;
}
else
return false;
}
//! Attempts to dequeue a log record ready for processing from the queue, does not block if no log records are ready to be processed
bool try_dequeue_ready(record_view& rec)
{
lock_guard< mutex_type > lock(m_mutex);
if (!m_queue.empty())
{
const boost::log::aux::timestamp now = boost::log::aux::get_timestamp();
enqueued_record const& elem = m_queue.top();
if (static_cast< uint64_t >((now - elem.m_timestamp).milliseconds()) >= m_ordering_window)
{
// We got a new element
rec = elem.m_record;
m_queue.pop();
return true;
}
}
return false;
}
//! Attempts to dequeue log record from the queue, does not block.
bool try_dequeue(record_view& rec)
{
lock_guard< mutex_type > lock(m_mutex);
if (!m_queue.empty())
{
enqueued_record const& elem = m_queue.top();
rec = elem.m_record;
m_queue.pop();
return true;
}
return false;
}
//! Dequeues log record from the queue, blocks if no log records are ready to be processed
bool dequeue_ready(record_view& rec)
{
unique_lock< mutex_type > lock(m_mutex);
while (!m_interruption_requested)
{
if (!m_queue.empty())
{
const boost::log::aux::timestamp now = boost::log::aux::get_timestamp();
enqueued_record const& elem = m_queue.top();
const uint64_t difference = (now - elem.m_timestamp).milliseconds();
if (difference >= m_ordering_window)
{
// We got a new element
rec = elem.m_record;
m_queue.pop();
return true;
}
else
{
// Wait until the element becomes ready to be processed
m_cond.timed_wait(lock, posix_time::milliseconds(m_ordering_window - difference));
}
}
else
{
// Wait for an element to come
m_cond.wait(lock);
}
}
m_interruption_requested = false;
return false;
}
//! Wakes a thread possibly blocked in the \c dequeue method
void interrupt_dequeue()
{
lock_guard< mutex_type > lock(m_mutex);
m_interruption_requested = true;
m_cond.notify_one();
}
private:
//! Enqueues a log record
void enqueue_unlocked(record_view const& rec)
{
const bool was_empty = m_queue.empty();
m_queue.push(enqueued_record(rec));
if (was_empty)
m_cond.notify_one();
}
};
} // namespace sinks
BOOST_LOG_CLOSE_NAMESPACE // namespace log
} // namespace boost
#include <boost/log/detail/footer.hpp>
#endif // BOOST_LOG_SINKS_UNBOUNDED_ORDERING_QUEUE_HPP_INCLUDED_