/* GLIB - Library of useful routines for C programming * Copyright (C) 1995-1997 Peter Mattis, Spencer Kimball and Josh MacDonald * * GAsyncQueue: thread pool implementation. * Copyright (C) 2000 Sebastian Wilhelmi; University of Karlsruhe * * This library is free software; you can redistribute it and/or * modify it under the terms of the GNU Lesser General Public * License as published by the Free Software Foundation; either * version 2 of the License, or (at your option) any later version. * * This library is distributed in the hope that it will be useful, * but WITHOUT ANY WARRANTY; without even the implied warranty of * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU * Lesser General Public License for more details. * * You should have received a copy of the GNU Lesser General Public * License along with this library; if not, write to the * Free Software Foundation, Inc., 59 Temple Place - Suite 330, * Boston, MA 02111-1307, USA. */ /* * MT safe */ #include "config.h" #include "gthreadpool.h" #include "gasyncqueue.h" #include "gmain.h" #include "gtestutils.h" #include "gtimer.h" /** * SECTION: thread_pools * @title: Thread Pools * @short_description: pools of threads to execute work concurrently * @see_also: * #GThread GLib thread * system. * * * Sometimes you wish to asynchronously fork out the execution of work * and continue working in your own thread. If that will happen often, * the overhead of starting and destroying a thread each time might be * too high. In such cases reusing already started threads seems like a * good idea. And it indeed is, but implementing this can be tedious * and error-prone. * * Therefore GLib provides thread pools for your convenience. An added * advantage is, that the threads can be shared between the different * subsystems of your program, when they are using GLib. * * To create a new thread pool, you use g_thread_pool_new(). It is * destroyed by g_thread_pool_free(). * * If you want to execute a certain task within a thread pool, you call * g_thread_pool_push(). * * To get the current number of running threads you call * g_thread_pool_get_num_threads(). To get the number of still * unprocessed tasks you call g_thread_pool_unprocessed(). To control * the maximal number of threads for a thread pool, you use * g_thread_pool_get_max_threads() and g_thread_pool_set_max_threads(). * * Finally you can control the number of unused threads, that are kept * alive by GLib for future use. The current number can be fetched with * g_thread_pool_get_num_unused_threads(). The maximal number can be * controlled by g_thread_pool_get_max_unused_threads() and * g_thread_pool_set_max_unused_threads(). All currently unused threads * can be stopped by calling g_thread_pool_stop_unused_threads(). **/ #define DEBUG_MSG(x) /* #define DEBUG_MSG(args) g_printerr args ; g_printerr ("\n"); */ typedef struct _GRealThreadPool GRealThreadPool; /** * GThreadPool: * @func: the function to execute in the threads of this pool * @user_data: the user data for the threads of this pool * @exclusive: are all threads exclusive to this pool * * The #GThreadPool struct represents a thread pool. It has three * public read-only members, but the underlying struct is bigger, so * you must not copy this struct. **/ struct _GRealThreadPool { GThreadPool pool; GAsyncQueue* queue; GCond* cond; gint max_threads; gint num_threads; gboolean running; gboolean immediate; gboolean waiting; GCompareDataFunc sort_func; gpointer sort_user_data; }; /* The following is just an address to mark the wakeup order for a * thread, it could be any address (as long, as it isn't a valid * GThreadPool address) */ static const gpointer wakeup_thread_marker = (gpointer) &g_thread_pool_new; static gint wakeup_thread_serial = 0; /* Here all unused threads are waiting */ static GAsyncQueue *unused_thread_queue = NULL; static gint unused_threads = 0; static gint max_unused_threads = 0; static gint kill_unused_threads = 0; static guint max_idle_time = 0; static void g_thread_pool_queue_push_unlocked (GRealThreadPool *pool, gpointer data); static void g_thread_pool_free_internal (GRealThreadPool *pool); static gpointer g_thread_pool_thread_proxy (gpointer data); static void g_thread_pool_start_thread (GRealThreadPool *pool, GError **error); static void g_thread_pool_wakeup_and_stop_all (GRealThreadPool *pool); static GRealThreadPool* g_thread_pool_wait_for_new_pool (void); static gpointer g_thread_pool_wait_for_new_task (GRealThreadPool *pool); static void g_thread_pool_queue_push_unlocked (GRealThreadPool *pool, gpointer data) { if (pool->sort_func) g_async_queue_push_sorted_unlocked (pool->queue, data, pool->sort_func, pool->sort_user_data); else g_async_queue_push_unlocked (pool->queue, data); } static GRealThreadPool* g_thread_pool_wait_for_new_pool (void) { GRealThreadPool *pool; gint local_wakeup_thread_serial; guint local_max_unused_threads; gint local_max_idle_time; gint last_wakeup_thread_serial; gboolean have_relayed_thread_marker = FALSE; local_max_unused_threads = g_atomic_int_get (&max_unused_threads); local_max_idle_time = g_atomic_int_get (&max_idle_time); last_wakeup_thread_serial = g_atomic_int_get (&wakeup_thread_serial); g_atomic_int_inc (&unused_threads); do { if (g_atomic_int_get (&unused_threads) >= local_max_unused_threads) { /* If this is a superfluous thread, stop it. */ pool = NULL; } else if (local_max_idle_time > 0) { /* If a maximal idle time is given, wait for the given time. */ GTimeVal end_time; g_get_current_time (&end_time); g_time_val_add (&end_time, local_max_idle_time * 1000); DEBUG_MSG (("thread %p waiting in global pool for %f seconds.", g_thread_self (), local_max_idle_time / 1000.0)); pool = g_async_queue_timed_pop (unused_thread_queue, &end_time); } else { /* If no maximal idle time is given, wait indefinitely. */ DEBUG_MSG (("thread %p waiting in global pool.", g_thread_self ())); pool = g_async_queue_pop (unused_thread_queue); } if (pool == wakeup_thread_marker) { local_wakeup_thread_serial = g_atomic_int_get (&wakeup_thread_serial); if (last_wakeup_thread_serial == local_wakeup_thread_serial) { if (!have_relayed_thread_marker) { /* If this wakeup marker has been received for * the second time, relay it. */ DEBUG_MSG (("thread %p relaying wakeup message to " "waiting thread with lower serial.", g_thread_self ())); g_async_queue_push (unused_thread_queue, wakeup_thread_marker); have_relayed_thread_marker = TRUE; /* If a wakeup marker has been relayed, this thread * will get out of the way for 100 microseconds to * avoid receiving this marker again. */ g_usleep (100); } } else { if (g_atomic_int_exchange_and_add (&kill_unused_threads, -1) > 0) { pool = NULL; break; } DEBUG_MSG (("thread %p updating to new limits.", g_thread_self ())); local_max_unused_threads = g_atomic_int_get (&max_unused_threads); local_max_idle_time = g_atomic_int_get (&max_idle_time); last_wakeup_thread_serial = local_wakeup_thread_serial; have_relayed_thread_marker = FALSE; } } } while (pool == wakeup_thread_marker); g_atomic_int_add (&unused_threads, -1); return pool; } static gpointer g_thread_pool_wait_for_new_task (GRealThreadPool *pool) { gpointer task = NULL; if (pool->running || (!pool->immediate && g_async_queue_length_unlocked (pool->queue) > 0)) { /* This thread pool is still active. */ if (pool->num_threads > pool->max_threads && pool->max_threads != -1) { /* This is a superfluous thread, so it goes to the global pool. */ DEBUG_MSG (("superfluous thread %p in pool %p.", g_thread_self (), pool)); } else if (pool->pool.exclusive) { /* Exclusive threads stay attached to the pool. */ task = g_async_queue_pop_unlocked (pool->queue); DEBUG_MSG (("thread %p in exclusive pool %p waits for task " "(%d running, %d unprocessed).", g_thread_self (), pool, pool->num_threads, g_async_queue_length_unlocked (pool->queue))); } else { /* A thread will wait for new tasks for at most 1/2 * second before going to the global pool. */ GTimeVal end_time; g_get_current_time (&end_time); g_time_val_add (&end_time, G_USEC_PER_SEC / 2); /* 1/2 second */ DEBUG_MSG (("thread %p in pool %p waits for up to a 1/2 second for task " "(%d running, %d unprocessed).", g_thread_self (), pool, pool->num_threads, g_async_queue_length_unlocked (pool->queue))); task = g_async_queue_timed_pop_unlocked (pool->queue, &end_time); } } else { /* This thread pool is inactive, it will no longer process tasks. */ DEBUG_MSG (("pool %p not active, thread %p will go to global pool " "(running: %s, immediate: %s, len: %d).", pool, g_thread_self (), pool->running ? "true" : "false", pool->immediate ? "true" : "false", g_async_queue_length_unlocked (pool->queue))); } return task; } static gpointer g_thread_pool_thread_proxy (gpointer data) { GRealThreadPool *pool; pool = data; DEBUG_MSG (("thread %p started for pool %p.", g_thread_self (), pool)); g_async_queue_lock (pool->queue); while (TRUE) { gpointer task; task = g_thread_pool_wait_for_new_task (pool); if (task) { if (pool->running || !pool->immediate) { /* A task was received and the thread pool is active, so * execute the function. */ g_async_queue_unlock (pool->queue); DEBUG_MSG (("thread %p in pool %p calling func.", g_thread_self (), pool)); pool->pool.func (task, pool->pool.user_data); g_async_queue_lock (pool->queue); } } else { /* No task was received, so this thread goes to the global * pool. */ gboolean free_pool = FALSE; DEBUG_MSG (("thread %p leaving pool %p for global pool.", g_thread_self (), pool)); pool->num_threads--; if (!pool->running) { if (!pool->waiting) { if (pool->num_threads == 0) { /* If the pool is not running and no other * thread is waiting for this thread pool to * finish and this is the last thread of this * pool, free the pool. */ free_pool = TRUE; } else { /* If the pool is not running and no other * thread is waiting for this thread pool to * finish and this is not the last thread of * this pool and there are no tasks left in the * queue, wakeup the remaining threads. */ if (g_async_queue_length_unlocked (pool->queue) == - pool->num_threads) g_thread_pool_wakeup_and_stop_all (pool); } } else if (pool->immediate || g_async_queue_length_unlocked (pool->queue) <= 0) { /* If the pool is not running and another thread is * waiting for this thread pool to finish and there * are either no tasks left or the pool shall stop * immediatly, inform the waiting thread of a change * of the thread pool state. */ g_cond_broadcast (pool->cond); } } g_async_queue_unlock (pool->queue); if (free_pool) g_thread_pool_free_internal (pool); if ((pool = g_thread_pool_wait_for_new_pool ()) == NULL) break; g_async_queue_lock (pool->queue); DEBUG_MSG (("thread %p entering pool %p from global pool.", g_thread_self (), pool)); /* pool->num_threads++ is not done here, but in * g_thread_pool_start_thread to make the new started thread * known to the pool, before itself can do it. */ } } return NULL; } static void g_thread_pool_start_thread (GRealThreadPool *pool, GError **error) { gboolean success = FALSE; if (pool->num_threads >= pool->max_threads && pool->max_threads != -1) /* Enough threads are already running */ return; g_async_queue_lock (unused_thread_queue); if (g_async_queue_length_unlocked (unused_thread_queue) < 0) { g_async_queue_push_unlocked (unused_thread_queue, pool); success = TRUE; } g_async_queue_unlock (unused_thread_queue); if (!success) { GError *local_error = NULL; /* No thread was found, we have to start a new one */ g_thread_create (g_thread_pool_thread_proxy, pool, FALSE, &local_error); if (local_error) { g_propagate_error (error, local_error); return; } } /* See comment in g_thread_pool_thread_proxy as to why this is done * here and not there */ pool->num_threads++; } /** * g_thread_pool_new: * @func: a function to execute in the threads of the new thread pool * @user_data: user data that is handed over to @func every time it * is called * @max_threads: the maximal number of threads to execute concurrently in * the new thread pool, -1 means no limit * @exclusive: should this thread pool be exclusive? * @error: return location for error * * This function creates a new thread pool. * * Whenever you call g_thread_pool_push(), either a new thread is * created or an unused one is reused. At most @max_threads threads * are running concurrently for this thread pool. @max_threads = -1 * allows unlimited threads to be created for this thread pool. The * newly created or reused thread now executes the function @func with * the two arguments. The first one is the parameter to * g_thread_pool_push() and the second one is @user_data. * * The parameter @exclusive determines, whether the thread pool owns * all threads exclusive or whether the threads are shared * globally. If @exclusive is %TRUE, @max_threads threads are started * immediately and they will run exclusively for this thread pool until * it is destroyed by g_thread_pool_free(). If @exclusive is %FALSE, * threads are created, when needed and shared between all * non-exclusive thread pools. This implies that @max_threads may not * be -1 for exclusive thread pools. * * @error can be %NULL to ignore errors, or non-%NULL to report * errors. An error can only occur when @exclusive is set to %TRUE and * not all @max_threads threads could be created. * * Return value: the new #GThreadPool **/ GThreadPool* g_thread_pool_new (GFunc func, gpointer user_data, gint max_threads, gboolean exclusive, GError **error) { GRealThreadPool *retval; G_LOCK_DEFINE_STATIC (init); g_return_val_if_fail (func, NULL); g_return_val_if_fail (!exclusive || max_threads != -1, NULL); g_return_val_if_fail (max_threads >= -1, NULL); g_return_val_if_fail (g_thread_supported (), NULL); retval = g_new (GRealThreadPool, 1); retval->pool.func = func; retval->pool.user_data = user_data; retval->pool.exclusive = exclusive; retval->queue = g_async_queue_new (); retval->cond = NULL; retval->max_threads = max_threads; retval->num_threads = 0; retval->running = TRUE; retval->sort_func = NULL; retval->sort_user_data = NULL; G_LOCK (init); if (!unused_thread_queue) unused_thread_queue = g_async_queue_new (); G_UNLOCK (init); if (retval->pool.exclusive) { g_async_queue_lock (retval->queue); while (retval->num_threads < retval->max_threads) { GError *local_error = NULL; g_thread_pool_start_thread (retval, &local_error); if (local_error) { g_propagate_error (error, local_error); break; } } g_async_queue_unlock (retval->queue); } return (GThreadPool*) retval; } /** * g_thread_pool_push: * @pool: a #GThreadPool * @data: a new task for @pool * @error: return location for error * * Inserts @data into the list of tasks to be executed by @pool. When * the number of currently running threads is lower than the maximal * allowed number of threads, a new thread is started (or reused) with * the properties given to g_thread_pool_new (). Otherwise @data stays * in the queue until a thread in this pool finishes its previous task * and processes @data. * * @error can be %NULL to ignore errors, or non-%NULL to report * errors. An error can only occur when a new thread couldn't be * created. In that case @data is simply appended to the queue of work * to do. **/ void g_thread_pool_push (GThreadPool *pool, gpointer data, GError **error) { GRealThreadPool *real; real = (GRealThreadPool*) pool; g_return_if_fail (real); g_return_if_fail (real->running); g_async_queue_lock (real->queue); if (g_async_queue_length_unlocked (real->queue) >= 0) /* No thread is waiting in the queue */ g_thread_pool_start_thread (real, error); g_thread_pool_queue_push_unlocked (real, data); g_async_queue_unlock (real->queue); } /** * g_thread_pool_set_max_threads: * @pool: a #GThreadPool * @max_threads: a new maximal number of threads for @pool * @error: return location for error * * Sets the maximal allowed number of threads for @pool. A value of -1 * means, that the maximal number of threads is unlimited. * * Setting @max_threads to 0 means stopping all work for @pool. It is * effectively frozen until @max_threads is set to a non-zero value * again. * * A thread is never terminated while calling @func, as supplied by * g_thread_pool_new (). Instead the maximal number of threads only * has effect for the allocation of new threads in g_thread_pool_push(). * A new thread is allocated, whenever the number of currently * running threads in @pool is smaller than the maximal number. * * @error can be %NULL to ignore errors, or non-%NULL to report * errors. An error can only occur when a new thread couldn't be * created. **/ void g_thread_pool_set_max_threads (GThreadPool *pool, gint max_threads, GError **error) { GRealThreadPool *real; gint to_start; real = (GRealThreadPool*) pool; g_return_if_fail (real); g_return_if_fail (real->running); g_return_if_fail (!real->pool.exclusive || max_threads != -1); g_return_if_fail (max_threads >= -1); g_async_queue_lock (real->queue); real->max_threads = max_threads; if (pool->exclusive) to_start = real->max_threads - real->num_threads; else to_start = g_async_queue_length_unlocked (real->queue); for ( ; to_start > 0; to_start--) { GError *local_error = NULL; g_thread_pool_start_thread (real, &local_error); if (local_error) { g_propagate_error (error, local_error); break; } } g_async_queue_unlock (real->queue); } /** * g_thread_pool_get_max_threads: * @pool: a #GThreadPool * * Returns the maximal number of threads for @pool. * * Return value: the maximal number of threads **/ gint g_thread_pool_get_max_threads (GThreadPool *pool) { GRealThreadPool *real; gint retval; real = (GRealThreadPool*) pool; g_return_val_if_fail (real, 0); g_return_val_if_fail (real->running, 0); g_async_queue_lock (real->queue); retval = real->max_threads; g_async_queue_unlock (real->queue); return retval; } /** * g_thread_pool_get_num_threads: * @pool: a #GThreadPool * * Returns the number of threads currently running in @pool. * * Return value: the number of threads currently running **/ guint g_thread_pool_get_num_threads (GThreadPool *pool) { GRealThreadPool *real; guint retval; real = (GRealThreadPool*) pool; g_return_val_if_fail (real, 0); g_return_val_if_fail (real->running, 0); g_async_queue_lock (real->queue); retval = real->num_threads; g_async_queue_unlock (real->queue); return retval; } /** * g_thread_pool_unprocessed: * @pool: a #GThreadPool * * Returns the number of tasks still unprocessed in @pool. * * Return value: the number of unprocessed tasks **/ guint g_thread_pool_unprocessed (GThreadPool *pool) { GRealThreadPool *real; gint unprocessed; real = (GRealThreadPool*) pool; g_return_val_if_fail (real, 0); g_return_val_if_fail (real->running, 0); unprocessed = g_async_queue_length (real->queue); return MAX (unprocessed, 0); } /** * g_thread_pool_free: * @pool: a #GThreadPool * @immediate: should @pool shut down immediately? * @wait_: should the function wait for all tasks to be finished? * * Frees all resources allocated for @pool. * * If @immediate is %TRUE, no new task is processed for * @pool. Otherwise @pool is not freed before the last task is * processed. Note however, that no thread of this pool is * interrupted, while processing a task. Instead at least all still * running threads can finish their tasks before the @pool is freed. * * If @wait_ is %TRUE, the functions does not return before all tasks * to be processed (dependent on @immediate, whether all or only the * currently running) are ready. Otherwise the function returns immediately. * * After calling this function @pool must not be used anymore. **/ void g_thread_pool_free (GThreadPool *pool, gboolean immediate, gboolean wait_) { GRealThreadPool *real; real = (GRealThreadPool*) pool; g_return_if_fail (real); g_return_if_fail (real->running); /* If there's no thread allowed here, there is not much sense in * not stopping this pool immediately, when it's not empty */ g_return_if_fail (immediate || real->max_threads != 0 || g_async_queue_length (real->queue) == 0); g_async_queue_lock (real->queue); real->running = FALSE; real->immediate = immediate; real->waiting = wait_; if (wait_) { real->cond = g_cond_new (); while (g_async_queue_length_unlocked (real->queue) != -real->num_threads && !(immediate && real->num_threads == 0)) g_cond_wait (real->cond, _g_async_queue_get_mutex (real->queue)); } if (immediate || g_async_queue_length_unlocked (real->queue) == -real->num_threads) { /* No thread is currently doing something (and nothing is left * to process in the queue) */ if (real->num_threads == 0) { /* No threads left, we clean up */ g_async_queue_unlock (real->queue); g_thread_pool_free_internal (real); return; } g_thread_pool_wakeup_and_stop_all (real); } /* The last thread should cleanup the pool */ real->waiting = FALSE; g_async_queue_unlock (real->queue); } static void g_thread_pool_free_internal (GRealThreadPool* pool) { g_return_if_fail (pool); g_return_if_fail (pool->running == FALSE); g_return_if_fail (pool->num_threads == 0); g_async_queue_unref (pool->queue); if (pool->cond) g_cond_free (pool->cond); g_free (pool); } static void g_thread_pool_wakeup_and_stop_all (GRealThreadPool* pool) { guint i; g_return_if_fail (pool); g_return_if_fail (pool->running == FALSE); g_return_if_fail (pool->num_threads != 0); pool->immediate = TRUE; for (i = 0; i < pool->num_threads; i++) g_thread_pool_queue_push_unlocked (pool, GUINT_TO_POINTER (1)); } /** * g_thread_pool_set_max_unused_threads: * @max_threads: maximal number of unused threads * * Sets the maximal number of unused threads to @max_threads. If * @max_threads is -1, no limit is imposed on the number of unused * threads. **/ void g_thread_pool_set_max_unused_threads (gint max_threads) { g_return_if_fail (max_threads >= -1); g_atomic_int_set (&max_unused_threads, max_threads); if (max_threads != -1) { max_threads -= g_atomic_int_get (&unused_threads); if (max_threads < 0) { g_atomic_int_set (&kill_unused_threads, -max_threads); g_atomic_int_inc (&wakeup_thread_serial); g_async_queue_lock (unused_thread_queue); do { g_async_queue_push_unlocked (unused_thread_queue, wakeup_thread_marker); } while (++max_threads); g_async_queue_unlock (unused_thread_queue); } } } /** * g_thread_pool_get_max_unused_threads: * * Returns the maximal allowed number of unused threads. * * Return value: the maximal number of unused threads **/ gint g_thread_pool_get_max_unused_threads (void) { return g_atomic_int_get (&max_unused_threads); } /** * g_thread_pool_get_num_unused_threads: * * Returns the number of currently unused threads. * * Return value: the number of currently unused threads **/ guint g_thread_pool_get_num_unused_threads (void) { return g_atomic_int_get (&unused_threads); } /** * g_thread_pool_stop_unused_threads: * * Stops all currently unused threads. This does not change the * maximal number of unused threads. This function can be used to * regularly stop all unused threads e.g. from g_timeout_add(). **/ void g_thread_pool_stop_unused_threads (void) { guint oldval; oldval = g_thread_pool_get_max_unused_threads (); g_thread_pool_set_max_unused_threads (0); g_thread_pool_set_max_unused_threads (oldval); } /** * g_thread_pool_set_sort_function: * @pool: a #GThreadPool * @func: the #GCompareDataFunc used to sort the list of tasks. * This function is passed two tasks. It should return * 0 if the order in which they are handled does not matter, * a negative value if the first task should be processed before * the second or a positive value if the second task should be * processed first. * @user_data: user data passed to @func. * * Sets the function used to sort the list of tasks. This allows the * tasks to be processed by a priority determined by @func, and not * just in the order in which they were added to the pool. * * Note, if the maximum number of threads is more than 1, the order * that threads are executed can not be guranteed 100%. Threads are * scheduled by the operating system and are executed at random. It * cannot be assumed that threads are executed in the order they are * created. * * Since: 2.10 **/ void g_thread_pool_set_sort_function (GThreadPool *pool, GCompareDataFunc func, gpointer user_data) { GRealThreadPool *real; real = (GRealThreadPool*) pool; g_return_if_fail (real); g_return_if_fail (real->running); g_async_queue_lock (real->queue); real->sort_func = func; real->sort_user_data = user_data; if (func) g_async_queue_sort_unlocked (real->queue, real->sort_func, real->sort_user_data); g_async_queue_unlock (real->queue); } /** * g_thread_pool_set_max_idle_time: * @interval: the maximum @interval (1/1000ths of a second) a thread * can be idle. * * This function will set the maximum @interval that a thread waiting * in the pool for new tasks can be idle for before being * stopped. This function is similar to calling * g_thread_pool_stop_unused_threads() on a regular timeout, except, * this is done on a per thread basis. * * By setting @interval to 0, idle threads will not be stopped. * * This function makes use of g_async_queue_timed_pop () using * @interval. * * Since: 2.10 **/ void g_thread_pool_set_max_idle_time (guint interval) { guint i; g_atomic_int_set (&max_idle_time, interval); i = g_atomic_int_get (&unused_threads); if (i > 0) { g_atomic_int_inc (&wakeup_thread_serial); g_async_queue_lock (unused_thread_queue); do { g_async_queue_push_unlocked (unused_thread_queue, wakeup_thread_marker); } while (--i); g_async_queue_unlock (unused_thread_queue); } } /** * g_thread_pool_get_max_idle_time: * * This function will return the maximum @interval that a thread will * wait in the thread pool for new tasks before being stopped. * * If this function returns 0, threads waiting in the thread pool for * new work are not stopped. * * Return value: the maximum @interval to wait for new tasks in the * thread pool before stopping the thread (1/1000ths of a second). * * Since: 2.10 **/ guint g_thread_pool_get_max_idle_time (void) { return g_atomic_int_get (&max_idle_time); }