MDEV-6936: Buffer pool list scan optimization

Merged Facebook commit 617aef9f911d825e9053f3d611d0389e02031225
authored by Inaam Rana to InnoDB storage engine (not XtraDB)
from https://github.com/facebook/mysql-5.6

WL#7047 - Optimize buffer pool list scans and related batch processing

Reduce excessive scanning of pages when doing flush list batches. The
fix is to introduce the concept of "Hazard Pointer", this reduces the
time complexity of the scan from O(n*n) to O.

The concept of hazard pointer is reversed in this work. Academically
hazard pointer is a pointer that the thread working on it will declar
such and as long as that thread is not done no other thread is allowe
do anything with it.

In this WL we declare the pointer as a hazard pointer and then if any
thread attempts to work on it, it is allowed to do so but it has to a
the hazard pointer to the next valid value. We use hazard pointer sol
reverse traversal of lists within a buffer pool instance.

Add an event to control the background flush thread. The background f
thread wait has been converted to an os event timed wait so that it c
signalled by threads that want to kick start a background flush when
buffer pool is running low on free/dirty pages.

mysql-test/suite/sys_vars/r/innodb_monitor_disable_basic.result

@@ -47,7 +47,6 @@ buffer_data_written	disabled
 buffer_flush_batch_scanned	disabled
 buffer_flush_batch_num_scan	disabled
 buffer_flush_batch_scanned_per_call	disabled
-buffer_flush_batch_rescan	disabled
 buffer_flush_batch_total_pages	disabled
 buffer_flush_batches	disabled
 buffer_flush_batch_pages	disabled
@@ -72,9 +71,12 @@ buffer_flush_background_pages	disabled
 buffer_LRU_batch_scanned	disabled
 buffer_LRU_batch_num_scan	disabled
 buffer_LRU_batch_scanned_per_call	disabled
-buffer_LRU_batch_total_pages	disabled
-buffer_LRU_batches	disabled
-buffer_LRU_batch_pages	disabled
+buffer_LRU_batch_flush_total_pages	disabled
+buffer_LRU_batches_flush	disabled
+buffer_LRU_batch_flush_pages	disabled
+buffer_LRU_batch_evict_total_pages	disabled
+buffer_LRU_batches_evict	disabled
+buffer_LRU_batch_evict_pages	disabled
 buffer_LRU_single_flush_scanned	disabled
 buffer_LRU_single_flush_num_scan	disabled
 buffer_LRU_single_flush_scanned_per_call	disabled

mysql-test/suite/sys_vars/r/innodb_monitor_enable_basic.result

@@ -47,7 +47,6 @@ buffer_data_written	disabled
 buffer_flush_batch_scanned	disabled
 buffer_flush_batch_num_scan	disabled
 buffer_flush_batch_scanned_per_call	disabled
-buffer_flush_batch_rescan	disabled
 buffer_flush_batch_total_pages	disabled
 buffer_flush_batches	disabled
 buffer_flush_batch_pages	disabled
@@ -72,9 +71,12 @@ buffer_flush_background_pages	disabled
 buffer_LRU_batch_scanned	disabled
 buffer_LRU_batch_num_scan	disabled
 buffer_LRU_batch_scanned_per_call	disabled
-buffer_LRU_batch_total_pages	disabled
-buffer_LRU_batches	disabled
-buffer_LRU_batch_pages	disabled
+buffer_LRU_batch_flush_total_pages	disabled
+buffer_LRU_batches_flush	disabled
+buffer_LRU_batch_flush_pages	disabled
+buffer_LRU_batch_evict_total_pages	disabled
+buffer_LRU_batches_evict	disabled
+buffer_LRU_batch_evict_pages	disabled
 buffer_LRU_single_flush_scanned	disabled
 buffer_LRU_single_flush_num_scan	disabled
 buffer_LRU_single_flush_scanned_per_call	disabled

mysql-test/suite/sys_vars/r/innodb_monitor_reset_all_basic.result

@@ -47,7 +47,6 @@ buffer_data_written	disabled
 buffer_flush_batch_scanned	disabled
 buffer_flush_batch_num_scan	disabled
 buffer_flush_batch_scanned_per_call	disabled
-buffer_flush_batch_rescan	disabled
 buffer_flush_batch_total_pages	disabled
 buffer_flush_batches	disabled
 buffer_flush_batch_pages	disabled
@@ -72,9 +71,12 @@ buffer_flush_background_pages	disabled
 buffer_LRU_batch_scanned	disabled
 buffer_LRU_batch_num_scan	disabled
 buffer_LRU_batch_scanned_per_call	disabled
-buffer_LRU_batch_total_pages	disabled
-buffer_LRU_batches	disabled
-buffer_LRU_batch_pages	disabled
+buffer_LRU_batch_flush_total_pages	disabled
+buffer_LRU_batches_flush	disabled
+buffer_LRU_batch_flush_pages	disabled
+buffer_LRU_batch_evict_total_pages	disabled
+buffer_LRU_batches_evict	disabled
+buffer_LRU_batch_evict_pages	disabled
 buffer_LRU_single_flush_scanned	disabled
 buffer_LRU_single_flush_num_scan	disabled
 buffer_LRU_single_flush_scanned_per_call	disabled

mysql-test/suite/sys_vars/r/innodb_monitor_reset_basic.result

@@ -47,7 +47,6 @@ buffer_data_written	disabled
 buffer_flush_batch_scanned	disabled
 buffer_flush_batch_num_scan	disabled
 buffer_flush_batch_scanned_per_call	disabled
-buffer_flush_batch_rescan	disabled
 buffer_flush_batch_total_pages	disabled
 buffer_flush_batches	disabled
 buffer_flush_batch_pages	disabled
@@ -72,9 +71,12 @@ buffer_flush_background_pages	disabled
 buffer_LRU_batch_scanned	disabled
 buffer_LRU_batch_num_scan	disabled
 buffer_LRU_batch_scanned_per_call	disabled
-buffer_LRU_batch_total_pages	disabled
-buffer_LRU_batches	disabled
-buffer_LRU_batch_pages	disabled
+buffer_LRU_batch_flush_total_pages	disabled
+buffer_LRU_batches_flush	disabled
+buffer_LRU_batch_flush_pages	disabled
+buffer_LRU_batch_evict_total_pages	disabled
+buffer_LRU_batches_evict	disabled
+buffer_LRU_batch_evict_pages	disabled
 buffer_LRU_single_flush_scanned	disabled
 buffer_LRU_single_flush_num_scan	disabled
 buffer_LRU_single_flush_scanned_per_call	disabled

storage/innobase/buf/buf0buf.cc

@@ -55,6 +55,8 @@ Created 11/5/1995 Heikki Tuuri
 #include "srv0mon.h"
 #include "buf0checksum.h"
 
+#include <new>
+
 /*
 		IMPLEMENTATION OF THE BUFFER POOL
 		=================================
@@ -1329,6 +1331,19 @@ buf_pool_init_instance(
 
 	buf_pool->try_LRU_scan = TRUE;
 
+	/* Initialize the hazard pointer for flush_list batches */
+	new(&buf_pool->flush_hp)
+		FlushHp(buf_pool, &buf_pool->flush_list_mutex);
+
+	/* Initialize the hazard pointer for LRU batches */
+	new(&buf_pool->lru_hp) LRUHp(buf_pool, &buf_pool->mutex);
+
+	/* Initialize the iterator for LRU scan search */
+	new(&buf_pool->lru_scan_itr) LRUItr(buf_pool, &buf_pool->mutex);
+
+	/* Initialize the iterator for single page scan search */
+	new(&buf_pool->single_scan_itr) LRUItr(buf_pool, &buf_pool->mutex);
+
 	buf_pool_mutex_exit(buf_pool);
 
 	return(DB_SUCCESS);
@@ -1419,6 +1434,8 @@ buf_pool_init(
 
 	btr_search_sys_create(buf_pool_get_curr_size() / sizeof(void*) / 64);
 
+	buf_flush_event = os_event_create();
+
 	return(DB_SUCCESS);
 }
 
@@ -1535,6 +1552,10 @@ buf_relocate(
 
 	memcpy(dpage, bpage, sizeof *dpage);
 
+	/* Important that we adjust the hazard pointer before
+	removing bpage from LRU list. */
+	buf_LRU_adjust_hp(buf_pool, bpage);
+
 	ut_d(bpage->in_LRU_list = FALSE);
 	ut_d(bpage->in_page_hash = FALSE);
 
@@ -1573,6 +1594,84 @@ buf_relocate(
 	HASH_INSERT(buf_page_t, hash, buf_pool->page_hash, fold, dpage);
 }
 
+/** Hazard Pointer implementation. */
+
+/** Set current value
+@param bpage	buffer block to be set as hp */
+void
+HazardPointer::set(buf_page_t* bpage)
+{
+	ut_ad(mutex_own(m_mutex));
+	ut_ad(!bpage || buf_pool_from_bpage(bpage) == m_buf_pool);
+	ut_ad(!bpage || buf_page_in_file(bpage));
+
+	m_hp = bpage;
+}
+
+/** Checks if a bpage is the hp
+@param bpage    buffer block to be compared
+@return true if it is hp */
+
+bool
+HazardPointer::is_hp(const buf_page_t* bpage)
+{
+	ut_ad(mutex_own(m_mutex));
+	ut_ad(!m_hp || buf_pool_from_bpage(m_hp) == m_buf_pool);
+	ut_ad(!bpage || buf_pool_from_bpage(bpage) == m_buf_pool);
+
+	return(bpage == m_hp);
+}
+
+/** Adjust the value of hp. This happens when some other thread working
+on the same list attempts to remove the hp from the list.
+@param bpage	buffer block to be compared */
+
+void
+FlushHp::adjust(const buf_page_t* bpage)
+{
+	ut_ad(bpage != NULL);
+
+	/** We only support reverse traversal for now. */
+	if (is_hp(bpage)) {
+		m_hp = UT_LIST_GET_PREV(list, m_hp);
+	}
+
+	ut_ad(!m_hp || m_hp->in_flush_list);
+}
+
+/** Adjust the value of hp. This happens when some other thread working
+on the same list attempts to remove the hp from the list.
+@param bpage	buffer block to be compared */
+
+void
+LRUHp::adjust(const buf_page_t* bpage)
+{
+	ut_ad(bpage);
+
+	/** We only support reverse traversal for now. */
+	if (is_hp(bpage)) {
+		m_hp = UT_LIST_GET_PREV(LRU, m_hp);
+	}
+
+	ut_ad(!m_hp || m_hp->in_LRU_list);
+}
+
+/** Selects from where to start a scan. If we have scanned too deep into
+the LRU list it resets the value to the tail of the LRU list.
+@return buf_page_t from where to start scan. */
+
+buf_page_t*
+LRUItr::start()
+{
+	ut_ad(mutex_own(m_mutex));
+
+	if (!m_hp || m_hp->old) {
+		m_hp = UT_LIST_GET_LAST(m_buf_pool->LRU);
+	}
+
+	return(m_hp);
+}
+
 /********************************************************************//**
 Determine if a block is a sentinel for a buffer pool watch.
 @return	TRUE if a sentinel for a buffer pool watch, FALSE if not */
@@ -4057,7 +4156,10 @@ UNIV_INTERN
 bool
 buf_page_io_complete(
 /*=================*/
-	buf_page_t*	bpage)	/*!< in: pointer to the block in question */
+	buf_page_t*	bpage,	/*!< in: pointer to the block in question */
+	bool		evict)	/*!< in: whether or not to evict the page
+				from LRU list. */
+
 {
 	enum buf_io_fix	io_type;
 	buf_pool_t*	buf_pool = buf_pool_from_bpage(bpage);
@@ -4238,6 +4340,7 @@ corrupt:
 	id. */
 
 	buf_page_set_io_fix(bpage, BUF_IO_NONE);
+	buf_page_monitor(bpage, io_type);
 
 	switch (io_type) {
 	case BUF_IO_READ:
@@ -4254,6 +4357,8 @@ corrupt:
 					     BUF_IO_READ);
 		}
 
+		mutex_exit(buf_page_get_mutex(bpage));
+
 		break;
 
 	case BUF_IO_WRITE:
@@ -4269,14 +4374,30 @@ corrupt:
 
 		buf_pool->stat.n_pages_written++;
 
+		/* In case of flush batches i.e.: BUF_FLUSH_LIST and
+		BUF_FLUSH_LRU this function is always called from IO
+		helper thread. In this case, we decide whether or not
+		to evict the page based on flush type. The value
+		passed as evict is the default value in function
+		definition which is false.
+		We always evict in case of LRU batch and never evict
+		in case of flush list batch. For single page flush
+		the caller sets the appropriate value. */
+		if (buf_page_get_flush_type(bpage) == BUF_FLUSH_LRU) {
+			evict = true;
+		}
+
+		mutex_exit(buf_page_get_mutex(bpage));
+		if (evict) {
+			buf_LRU_free_page(bpage, true);
+		}
+
 		break;
 
 	default:
 		ut_error;
 	}
 
-	buf_page_monitor(bpage, io_type);
-
 #ifdef UNIV_DEBUG
 	if (buf_debug_prints) {
 		fprintf(stderr, "Has %s page space %lu page no %lu\n",
@@ -4286,7 +4407,6 @@ corrupt:
 	}
 #endif /* UNIV_DEBUG */
 
-	mutex_exit(buf_page_get_mutex(bpage));
 	buf_pool_mutex_exit(buf_pool);
 
 	return(true);

storage/innobase/buf/buf0flu.cc

@@ -64,14 +64,13 @@ need to protect it by a mutex. It is only ever read by the thread
 doing the shutdown */
 UNIV_INTERN ibool buf_page_cleaner_is_active = FALSE;
 
-/** LRU flush batch is further divided into this chunk size to
-reduce the wait time for the threads waiting for a clean block */
-#define PAGE_CLEANER_LRU_BATCH_CHUNK_SIZE	100
-
 #ifdef UNIV_PFS_THREAD
 UNIV_INTERN mysql_pfs_key_t buf_page_cleaner_thread_key;
 #endif /* UNIV_PFS_THREAD */
 
+/** Event to synchronise with the flushing. */
+ os_event_t	buf_flush_event;
+
 /** If LRU list of a buf_pool is less than this size then LRU eviction
 should not happen. This is because when we do LRU flushing we also put
 the blocks on free list. If LRU list is very small then we can end up
@@ -135,60 +134,6 @@ buf_flush_validate_skip(
 }
 #endif /* UNIV_DEBUG || UNIV_BUF_DEBUG */
 
-/*******************************************************************//**
-Sets hazard pointer during flush_list iteration. */
-UNIV_INLINE
-void
-buf_flush_set_hp(
-/*=============*/
-	buf_pool_t*		buf_pool,/*!< in/out: buffer pool instance */
-	const buf_page_t*	bpage)	/*!< in: buffer control block */
-{
-	ut_ad(buf_flush_list_mutex_own(buf_pool));
-	ut_ad(buf_pool->flush_list_hp == NULL || bpage == NULL);
-	ut_ad(!bpage || buf_page_in_file(bpage));
-	ut_ad(!bpage || bpage->in_flush_list);
-	ut_ad(!bpage || buf_pool_from_bpage(bpage) == buf_pool);
-
-	buf_pool->flush_list_hp = bpage;
-}
-
-/*******************************************************************//**
-Checks if the given block is a hazard pointer
-@return true if bpage is hazard pointer */
-UNIV_INLINE
-bool
-buf_flush_is_hp(
-/*============*/
-	buf_pool_t*		buf_pool,/*!< in: buffer pool instance */
-	const buf_page_t*	bpage)	/*!< in: buffer control block */
-{
-	ut_ad(buf_flush_list_mutex_own(buf_pool));
-
-	return(buf_pool->flush_list_hp == bpage);
-}
-
-/*******************************************************************//**
-Whenever we move a block in flush_list (either to remove it or to
-relocate it) we check the hazard pointer set by some other thread
-doing the flush list scan. If the hazard pointer is the same as the
-one we are about going to move then we set it to NULL to force a rescan
-in the thread doing the batch. */
-UNIV_INLINE
-void
-buf_flush_update_hp(
-/*================*/
-	buf_pool_t*	buf_pool,	/*!< in: buffer pool instance */
-	buf_page_t*	bpage)		/*!< in: buffer control block */
-{
-	ut_ad(buf_flush_list_mutex_own(buf_pool));
-
-	if (buf_flush_is_hp(buf_pool, bpage)) {
-		buf_flush_set_hp(buf_pool, NULL);
-		MONITOR_INC(MONITOR_FLUSH_HP_RESCAN);
-	}
-}
-
 /******************************************************************//**
 Insert a block in the flush_rbt and returns a pointer to its
 predecessor or NULL if no predecessor. The ordering is maintained
@@ -587,6 +532,10 @@ buf_flush_remove(
 
 	buf_flush_list_mutex_enter(buf_pool);
 
+	/* Important that we adjust the hazard pointer before removing
+	the bpage from flush list. */
+	buf_pool->flush_hp.adjust(bpage);
+
 	switch (buf_page_get_state(bpage)) {
 	case BUF_BLOCK_POOL_WATCH:
 	case BUF_BLOCK_ZIP_PAGE:
@@ -627,7 +576,6 @@ buf_flush_remove(
 	ut_a(buf_flush_validate_skip(buf_pool));
 #endif /* UNIV_DEBUG || UNIV_BUF_DEBUG */
 
-	buf_flush_update_hp(buf_pool, bpage);
 	buf_flush_list_mutex_exit(buf_pool);
 }
 
@@ -678,6 +626,10 @@ buf_flush_relocate_on_flush_list(
 		prev_b = buf_flush_insert_in_flush_rbt(dpage);
 	}
 
+	/* Important that we adjust the hazard pointer before removing
+	the bpage from the flush list. */
+	buf_pool->flush_hp.adjust(bpage);
+
 	/* Must be done after we have removed it from the flush_rbt
 	because we assert on in_flush_list in comparison function. */
 	ut_d(bpage->in_flush_list = FALSE);
@@ -706,7 +658,6 @@ buf_flush_relocate_on_flush_list(
 	ut_a(buf_flush_validate_low(buf_pool));
 #endif /* UNIV_DEBUG || UNIV_BUF_DEBUG */
 
-	buf_flush_update_hp(buf_pool, bpage);
 	buf_flush_list_mutex_exit(buf_pool);
 }
 
@@ -997,7 +948,10 @@ buf_flush_write_block_low(
 	if (sync) {
 		ut_ad(flush_type == BUF_FLUSH_SINGLE_PAGE);
 		fil_flush(buf_page_get_space(bpage));
-		buf_page_io_complete(bpage);
+
+		/* true means we want to evict this page from the
+		LRU list as well. */
+		buf_page_io_complete(bpage, true);
 	}
 
 	/* Increment the counter of I/O operations used
@@ -1091,10 +1045,10 @@ buf_flush_page(
 			rw_lock_s_lock_gen(rw_lock, BUF_IO_WRITE);
                 }
 
-                /* Even though bpage is not protected by any mutex at this
-                point, it is safe to access bpage, because it is io_fixed and
-                oldest_modification != 0.  Thus, it cannot be relocated in the
-                buffer pool or removed from flush_list or LRU_list. */
+		/* Even though bpage is not protected by any mutex at this
+		point, it is safe to access bpage, because it is io_fixed and
+		oldest_modification != 0.  Thus, it cannot be relocated in the
+		buffer pool or removed from flush_list or LRU_list. */
 
                 buf_flush_write_block_low(bpage, flush_type, sync);
         }
@@ -1458,105 +1412,70 @@ This utility flushes dirty blocks from the end of the LRU list.
 The calling thread is not allowed to own any latches on pages!
 It attempts to make 'max' blocks available in the free list. Note that
 it is a best effort attempt and it is not guaranteed that after a call
-to this function there will be 'max' blocks in the free list.
-@return number of blocks for which the write request was queued. */
-static
-ulint
+to this function there will be 'max' blocks in the free list.*/
+__attribute__((nonnull))
+void
 buf_flush_LRU_list_batch(
 /*=====================*/
 	buf_pool_t*	buf_pool,	/*!< in: buffer pool instance */
-	ulint		max)		/*!< in: desired number of
+	ulint		max,		/*!< in: desired number of
 					blocks in the free_list */
+	flush_counters_t*	n)	/*!< out: flushed/evicted page
+					counts */
 {
 	buf_page_t*	bpage;
-	ulint		count = 0;
 	ulint		scanned = 0;
 	ulint		free_len = UT_LIST_GET_LEN(buf_pool->free);
 	ulint		lru_len = UT_LIST_GET_LEN(buf_pool->LRU);
 
+	n->flushed = 0;
+	n->evicted = 0;
+	n->unzip_LRU_evicted = 0;
+
 	ut_ad(buf_pool_mutex_own(buf_pool));
 
-	bpage = UT_LIST_GET_LAST(buf_pool->LRU);
-	while (bpage != NULL && count < max
-	       && free_len < srv_LRU_scan_depth
-	       && lru_len > BUF_LRU_MIN_LEN) {
+	for (bpage = UT_LIST_GET_LAST(buf_pool->LRU);
+	     bpage != NULL && (n->evicted + n->flushed) < max
+	     && free_len < srv_LRU_scan_depth
+	     && lru_len > BUF_LRU_MIN_LEN;
+	     ++scanned,
+	     bpage = buf_pool->lru_hp.get()) {
 
-		ib_mutex_t* block_mutex = buf_page_get_mutex(bpage);
-		ibool	 evict;
+		buf_page_t* prev = UT_LIST_GET_PREV(LRU, bpage);
+		buf_pool->lru_hp.set(prev);
 
+		ib_mutex_t* block_mutex = buf_page_get_mutex(bpage);
 		mutex_enter(block_mutex);
-		evict = buf_flush_ready_for_replace(bpage);
+		bool	evict = buf_flush_ready_for_replace(bpage);
 		mutex_exit(block_mutex);
 
-		++scanned;
-
-		/* If the block is ready to be replaced we try to
-		free it i.e.: put it on the free list.
-		Otherwise we try to flush the block and its
-		neighbors. In this case we'll put it on the
-		free list in the next pass. We do this extra work
-		of putting blocks to the free list instead of
-		just flushing them because after every flush
-		we have to restart the scan from the tail of
-		the LRU list and if we don't clear the tail
-		of the flushed pages then the scan becomes
-		O(n*n). */
 		if (evict) {
+			/* block is ready for eviction i.e., it is
+			clean and is not IO-fixed or buffer fixed. */
 			if (buf_LRU_free_page(bpage, true)) {
-				/* buf_pool->mutex was potentially
-				released and reacquired. */
-				bpage = UT_LIST_GET_LAST(buf_pool->LRU);
-			} else {
-				bpage = UT_LIST_GET_PREV(LRU, bpage);
+				n->evicted++;
 			}
 		} else {
-			ulint		space;
-			ulint		offset;
-			buf_page_t*	prev_bpage;
-
-			prev_bpage = UT_LIST_GET_PREV(LRU, bpage);
-
-			/* Save the previous bpage */
-
-			if (prev_bpage != NULL) {
-				space = prev_bpage->space;
-				offset = prev_bpage->offset;
-			} else {
-				space = ULINT_UNDEFINED;
-				offset = ULINT_UNDEFINED;
-			}
-
-			if (!buf_flush_page_and_try_neighbors(
-				bpage, BUF_FLUSH_LRU, max, &count)) {
-
-				bpage = prev_bpage;
-			} else {
-				/* buf_pool->mutex was released.
-				reposition the iterator. Note: the
-				prev block could have been repositioned
-				too but that should be rare. */
-
-				if (prev_bpage != NULL) {
-
-					ut_ad(space != ULINT_UNDEFINED);
-					ut_ad(offset != ULINT_UNDEFINED);
-
-					prev_bpage = buf_page_hash_get(
-						buf_pool, space, offset);
-				}
-
-				bpage = prev_bpage;
-			}
+			/* Block is ready for flush. Dispatch an IO
+			request. The IO helper thread will put it on
+			free list in IO completion routine. */
+			buf_flush_page_and_try_neighbors(
+				bpage, BUF_FLUSH_LRU, max, &n->flushed);
 		}
 
+		ut_ad(!mutex_own(block_mutex));
+		ut_ad(buf_pool_mutex_own(buf_pool));
+
 		free_len = UT_LIST_GET_LEN(buf_pool->free);
 		lru_len = UT_LIST_GET_LEN(buf_pool->LRU);
 	}
 
+	buf_pool->lru_hp.set(NULL);
+
 	/* We keep track of all flushes happening as part of LRU
 	flush. When estimating the desired rate at which flush_list
 	should be flushed, we factor in this value. */
-	buf_lru_flush_page_count += count;
+	buf_lru_flush_page_count += n->flushed;
 
 	ut_ad(buf_pool_mutex_own(buf_pool));
 
@@ -1567,35 +1486,38 @@ buf_flush_LRU_list_batch(
 			MONITOR_LRU_BATCH_SCANNED_PER_CALL,
 			scanned);
 	}
-
-	return(count);
 }
 
 /*******************************************************************//**
 Flush and move pages from LRU or unzip_LRU list to the free list.
-Whether LRU or unzip_LRU is used depends on the state of the system.
-@return number of blocks for which either the write request was queued
-or in case of unzip_LRU the number of blocks actually moved to the
-free list */
+Whether LRU or unzip_LRU is used depends on the state of the system.*/
+__attribute__((nonnull))
 static
-ulint
+void
 buf_do_LRU_batch(
 /*=============*/
 	buf_pool_t*	buf_pool,	/*!< in: buffer pool instance */
-	ulint		max)		/*!< in: desired number of
+	ulint		max,		/*!< in: desired number of
 					blocks in the free_list */
+	flush_counters_t*	n)	/*!< out: flushed/evicted page
+					counts */
 {
-	ulint	count = 0;
-
 	if (buf_LRU_evict_from_unzip_LRU(buf_pool)) {
-		count += buf_free_from_unzip_LRU_list_batch(buf_pool, max);
+		n->unzip_LRU_evicted = buf_free_from_unzip_LRU_list_batch(buf_pool, max);
+	} else {
+		n->unzip_LRU_evicted = 0;
 	}
 
-	if (max > count) {
-		count += buf_flush_LRU_list_batch(buf_pool, max - count);
+	if (max > n->unzip_LRU_evicted) {
+		buf_flush_LRU_list_batch(buf_pool, max - n->unzip_LRU_evicted, n);
+	} else {
+		n->evicted = 0;
+		n->flushed = 0;
 	}
 
-	return(count);
+	/* Add evicted pages from unzip_LRU to the evicted pages from
+	the simple LRU. */
+	n->evicted += n->unzip_LRU_evicted;
 }
 
 /*******************************************************************//**
@@ -1637,6 +1559,7 @@ buf_do_flush_list_batch(
 	for (buf_page_t* bpage = UT_LIST_GET_LAST(buf_pool->flush_list);
 	     count < min_n && bpage != NULL && len > 0
 	     && bpage->oldest_modification < lsn_limit;
+	     bpage = buf_pool->flush_hp.get(),
 	     ++scanned) {
 
 		buf_page_t*	prev;
@@ -1645,8 +1568,7 @@ buf_do_flush_list_batch(
 		ut_ad(bpage->in_flush_list);
 
 		prev = UT_LIST_GET_PREV(list, bpage);
-		buf_flush_set_hp(buf_pool, prev);
-
+		buf_pool->flush_hp.set(prev);
 		buf_flush_list_mutex_exit(buf_pool);
 
 #ifdef UNIV_DEBUG
@@ -1657,23 +1579,12 @@ buf_do_flush_list_batch(
 
 		buf_flush_list_mutex_enter(buf_pool);
 
-		ut_ad(flushed || buf_flush_is_hp(buf_pool, prev));
+		ut_ad(flushed || buf_pool->flush_hp.is_hp(prev));
 
-		if (!buf_flush_is_hp(buf_pool, prev)) {
-			/* The hazard pointer was reset by some other
-			thread. Restart the scan. */
-			ut_ad(buf_flush_is_hp(buf_pool, NULL));
-			bpage = UT_LIST_GET_LAST(buf_pool->flush_list);
-			len = UT_LIST_GET_LEN(buf_pool->flush_list);
-		} else {
-			bpage = prev;
-			--len;
-			buf_flush_set_hp(buf_pool, NULL);
-		}
-
-		ut_ad(!bpage || bpage->in_flush_list);
+		--len;
 	}
 
+	buf_pool->flush_hp.set(NULL);
 	buf_flush_list_mutex_exit(buf_pool);
 
 	MONITOR_INC_VALUE_CUMULATIVE(MONITOR_FLUSH_BATCH_SCANNED,
@@ -1691,9 +1602,9 @@ This utility flushes dirty blocks from the end of the LRU list or flush_list.
 NOTE 1: in the case of an LRU flush the calling thread may own latches to
 pages: to avoid deadlocks, this function must be written so that it cannot
 end up waiting for these latches! NOTE 2: in the case of a flush list flush,
-the calling thread is not allowed to own any latches on pages!
-@return number of blocks for which the write request was queued */
-ulint
+the calling thread is not allowed to own any latches on pages! */
+__attribute__((nonnull))
+void
 buf_flush_batch(
 /*============*/
 	buf_pool_t*	buf_pool,	/*!< in: buffer pool instance */
@@ -1704,14 +1615,14 @@ buf_flush_batch(
 	ulint		min_n,		/*!< in: wished minimum mumber of blocks
 					flushed (it is not guaranteed that the
 					actual number is that big, though) */
-	lsn_t		lsn_limit)	/*!< in: in the case of BUF_FLUSH_LIST
+	lsn_t		lsn_limit,	/*!< in: in the case of BUF_FLUSH_LIST
 					all blocks whose oldest_modification is
 					smaller than this should be flushed
 					(if their number does not exceed
 					min_n), otherwise ignored */
+	flush_counters_t*	n)	/*!< out: flushed/evicted page
+					counts  */
 {
-	ulint		count	= 0;
-
 	ut_ad(flush_type == BUF_FLUSH_LRU || flush_type == BUF_FLUSH_LIST);
 #ifdef UNIV_SYNC_DEBUG
 	ut_ad((flush_type != BUF_FLUSH_LIST)
@@ -1724,10 +1635,11 @@ buf_flush_batch(
 	the flush functions. */
 	switch (flush_type) {
 	case BUF_FLUSH_LRU:
-		count = buf_do_LRU_batch(buf_pool, min_n);
+		buf_do_LRU_batch(buf_pool, min_n, n);
 		break;
 	case BUF_FLUSH_LIST:
-		count = buf_do_flush_list_batch(buf_pool, min_n, lsn_limit);
+		n->flushed = buf_do_flush_list_batch(buf_pool, min_n, lsn_limit);
+		n->evicted = 0;
 		break;
 	default:
 		ut_error;
@@ -1736,15 +1648,13 @@ buf_flush_batch(
 	buf_pool_mutex_exit(buf_pool);
 
 #ifdef UNIV_DEBUG
-	if (buf_debug_prints && count > 0) {
+	if (buf_debug_prints && n->flushed > 0) {
 		fprintf(stderr, flush_type == BUF_FLUSH_LRU
 			? "Flushed %lu pages in LRU flush\n"
 			: "Flushed %lu pages in flush list flush\n",
-			(ulong) count);
+			(ulong) n->flushed);
 	}
 #endif /* UNIV_DEBUG */
-
-	return(count);
 }
 
 /******************************************************************//**
@@ -1855,48 +1765,6 @@ buf_flush_wait_batch_end(
 	}
 }
 
-/*******************************************************************//**
-This utility flushes dirty blocks from the end of the LRU list and also
-puts replaceable clean pages from the end of the LRU list to the free
-list.
-NOTE: The calling thread is not allowed to own any latches on pages!
-@return true if a batch was queued successfully. false if another batch
-of same type was already running. */
-static
-bool
-buf_flush_LRU(
-/*==========*/
-	buf_pool_t*	buf_pool,	/*!< in/out: buffer pool instance */
-	ulint		min_n,		/*!< in: wished minimum mumber of blocks
-					flushed (it is not guaranteed that the
-					actual number is that big, though) */
-	ulint*		n_processed)	/*!< out: the number of pages
-					which were processed is passed
-					back to caller. Ignored if NULL */
-{
-	ulint		page_count;
-
-	if (n_processed) {
-		*n_processed = 0;
-	}
-
-	if (!buf_flush_start(buf_pool, BUF_FLUSH_LRU)) {
-		return(false);
-	}
-
-	page_count = buf_flush_batch(buf_pool, BUF_FLUSH_LRU, min_n, 0);
-
-	buf_flush_end(buf_pool, BUF_FLUSH_LRU);
-
-	buf_flush_common(BUF_FLUSH_LRU, page_count);
-
-	if (n_processed) {
-		*n_processed = page_count;
-	}
-
-	return(true);
-}
-
 /*******************************************************************//**
 This utility flushes dirty blocks from the end of the flush list of
 all buffer pool instances.
@@ -1943,8 +1811,8 @@ buf_flush_list(
 
 	/* Flush to lsn_limit in all buffer pool instances */
 	for (i = 0; i < srv_buf_pool_instances; i++) {
-		buf_pool_t*	buf_pool;
-		ulint		page_count = 0;
+		buf_pool_t*		buf_pool;
+		flush_counters_t	n;
 
 		buf_pool = buf_pool_from_array(i);
 
@@ -1964,23 +1832,23 @@ buf_flush_list(
 			continue;
 		}
 
-		page_count = buf_flush_batch(
-			buf_pool, BUF_FLUSH_LIST, min_n, lsn_limit);
+		buf_flush_batch(
+			buf_pool, BUF_FLUSH_LIST, min_n, lsn_limit, &n);
 
 		buf_flush_end(buf_pool, BUF_FLUSH_LIST);
 
-		buf_flush_common(BUF_FLUSH_LIST, page_count);
+		buf_flush_common(BUF_FLUSH_LIST, n.flushed);
 
 		if (n_processed) {
-			*n_processed += page_count;
+			*n_processed += n.flushed;
 		}
 
-		if (page_count) {
+		if (n.flushed) {
 			MONITOR_INC_VALUE_CUMULATIVE(
 				MONITOR_FLUSH_BATCH_TOTAL_PAGE,
 				MONITOR_FLUSH_BATCH_COUNT,
 				MONITOR_FLUSH_BATCH_PAGES,
-				page_count);
+				n.flushed);
 		}
 	}
 
@@ -1988,12 +1856,12 @@ buf_flush_list(
 }
 
 /******************************************************************//**
-This function picks up a single dirty page from the tail of the LRU
-list, flushes it, removes it from page_hash and LRU list and puts
-it on the free list. It is called from user threads when they are
-unable to find a replaceable page at the tail of the LRU list i.e.:
-when the background LRU flushing in the page_cleaner thread is not
-fast enough to keep pace with the workload.
+This function picks up a single page from the tail of the LRU
+list, flushes it (if it is dirty), removes it from page_hash and LRU
+list and puts it on the free list. It is called from user threads when
+they are unable to find a replaceable page at the tail of the LRU
+list i.e.: when the background LRU flushing in the page_cleaner thread
+is not fast enough to keep pace with the workload.
 @return TRUE if success. */
 UNIV_INTERN
 ibool
@@ -2003,84 +1871,67 @@ buf_flush_single_page_from_LRU(
 {
 	ulint		scanned;
 	buf_page_t*	bpage;
+	ibool		freed;
 
 	buf_pool_mutex_enter(buf_pool);
 
-	for (bpage = UT_LIST_GET_LAST(buf_pool->LRU), scanned = 1;
+	for (bpage = buf_pool->single_scan_itr.start(),
+	     scanned = 0, freed = FALSE;
 	     bpage != NULL;
-	     bpage = UT_LIST_GET_PREV(LRU, bpage), ++scanned) {
-
-		ib_mutex_t*	block_mutex = buf_page_get_mutex(bpage);
+	     ++scanned, bpage = buf_pool->single_scan_itr.get()) {
 
-		mutex_enter(block_mutex);
+		ut_ad(buf_pool_mutex_own(buf_pool));
 
-		if (buf_flush_ready_for_flush(bpage, BUF_FLUSH_SINGLE_PAGE)) {
+		buf_page_t* prev = UT_LIST_GET_PREV(LRU, bpage);
+		buf_pool->single_scan_itr.set(prev);
 
-			/* The following call will release the buffer pool
-			and block mutex. */
-
-			ibool	flushed = buf_flush_page(
-				buf_pool, bpage, BUF_FLUSH_SINGLE_PAGE, true);
+		ib_mutex_t* block_mutex = buf_page_get_mutex(bpage);
+		mutex_enter(block_mutex);
 
-			if (flushed) {
-				/* buf_flush_page() will release the
-				block mutex */
+		if (buf_flush_ready_for_replace(bpage)) {
+			/* block is ready for eviction i.e., it is
+			clean and is not IO-fixed or buffer fixed. */
+			mutex_exit(block_mutex);
+			if (buf_LRU_free_page(bpage, true)) {
+				buf_pool_mutex_exit(buf_pool);
+				freed = TRUE;
 				break;
 			}
+		} else if (buf_flush_ready_for_flush(
+				bpage, BUF_FLUSH_SINGLE_PAGE)) {
+			/* Block is ready for flush. Dispatch an IO
+			request. We'll put it on free list in IO
+			completion routine. The following call, if
+			successful, will release the buffer pool and
+			block mutex. */
+			freed = buf_flush_page(buf_pool, bpage,
+					       BUF_FLUSH_SINGLE_PAGE, true);
+			if (freed) {
+				/* block and buffer pool mutex have
+				already been reelased. */
+				break;
+			}
+			mutex_exit(block_mutex);
+		} else {
+			mutex_exit(block_mutex);
 		}
-
-		mutex_exit(block_mutex);
 	}
 
-	MONITOR_INC_VALUE_CUMULATIVE(
-		MONITOR_LRU_SINGLE_FLUSH_SCANNED,
-		MONITOR_LRU_SINGLE_FLUSH_SCANNED_NUM_CALL,
-		MONITOR_LRU_SINGLE_FLUSH_SCANNED_PER_CALL,
-		scanned);
-
-	if (bpage == NULL) {
+	if (!freed) {
 		/* Can't find a single flushable page. */
+		ut_ad(!bpage);
 		buf_pool_mutex_exit(buf_pool);
-		return(FALSE);
 	}
 
-
-	ibool	freed = FALSE;
-
-	/* At this point the page has been written to the disk.
-	As we are not holding buffer pool or block mutex therefore
-	we cannot use the bpage safely. It may have been plucked out
-	of the LRU list by some other thread or it may even have
-	relocated in case of a compressed page. We need to start
-	the scan of LRU list again to remove the block from the LRU
-	list and put it on the free list. */
-	buf_pool_mutex_enter(buf_pool);
-
-	for (bpage = UT_LIST_GET_LAST(buf_pool->LRU);
-	     bpage != NULL;
-	     bpage = UT_LIST_GET_PREV(LRU, bpage)) {
-
-		ib_mutex_t*	block_mutex = buf_page_get_mutex(bpage);
-
-		mutex_enter(block_mutex);
-
-		ibool	ready = buf_flush_ready_for_replace(bpage);
-
-		mutex_exit(block_mutex);
-
-		if (ready) {
-			bool	evict_zip;
-
-			evict_zip = !buf_LRU_evict_from_unzip_LRU(buf_pool);;
-
-			freed = buf_LRU_free_page(bpage, evict_zip);
-
-			break;
-		}
+	if (scanned) {
+		MONITOR_INC_VALUE_CUMULATIVE(
+			MONITOR_LRU_SINGLE_FLUSH_SCANNED,
+			MONITOR_LRU_SINGLE_FLUSH_SCANNED_NUM_CALL,
+			MONITOR_LRU_SINGLE_FLUSH_SCANNED_PER_CALL,
+			scanned);
 	}
 
-	buf_pool_mutex_exit(buf_pool);
-
+	ut_ad(!buf_pool_mutex_own(buf_pool));
 	return(freed);
 }
 
@@ -2107,6 +1958,7 @@ buf_flush_LRU_tail(void)
 
 		buf_pool_t*	buf_pool = buf_pool_from_array(i);
 		ulint		scan_depth;
+		flush_counters_t	n;
 
 		/* srv_LRU_scan_depth can be arbitrarily large value.
 		We cap it with current LRU size. */
@@ -2116,44 +1968,37 @@ buf_flush_LRU_tail(void)
 
 		scan_depth = ut_min(srv_LRU_scan_depth, scan_depth);
 
-		/* We divide LRU flush into smaller chunks because
-		there may be user threads waiting for the flush to
-		end in buf_LRU_get_free_block(). */
-		for (ulint j = 0;
-		     j < scan_depth;
-		     j += PAGE_CLEANER_LRU_BATCH_CHUNK_SIZE) {
-
-			ulint	n_flushed = 0;
-
-			/* Currently page_cleaner is the only thread
-			that can trigger an LRU flush. It is possible
-			that a batch triggered during last iteration is
-			still running, */
-			if (buf_flush_LRU(buf_pool,
-					  PAGE_CLEANER_LRU_BATCH_CHUNK_SIZE,
-					  &n_flushed)) {
-
-				/* Allowed only one batch per
-				buffer pool instance. */
-				buf_flush_wait_batch_end(
-					buf_pool, BUF_FLUSH_LRU);
-			}
+		/* Currently page_cleaner is the only thread
+		that can trigger an LRU flush. It is possible
+		that a batch triggered during last iteration is
+		still running, */
+		if (!buf_flush_start(buf_pool, BUF_FLUSH_LRU)) {
+			continue;
+		}
 
-			if (n_flushed) {
-				total_flushed += n_flushed;
-			} else {
-				/* Nothing to flush */
-				break;
-			}
+		buf_flush_batch(buf_pool, BUF_FLUSH_LRU, scan_depth, 0, &n);
+
+		buf_flush_end(buf_pool, BUF_FLUSH_LRU);
+
+		buf_flush_common(BUF_FLUSH_LRU, n.flushed);
+
+		if (n.flushed) {
+			MONITOR_INC_VALUE_CUMULATIVE(
+				MONITOR_LRU_BATCH_FLUSH_TOTAL_PAGE,
+				MONITOR_LRU_BATCH_FLUSH_COUNT,
+				MONITOR_LRU_BATCH_FLUSH_PAGES,
+				n.flushed);
 		}
-	}
 
-	if (total_flushed) {
-		MONITOR_INC_VALUE_CUMULATIVE(
-			MONITOR_LRU_BATCH_TOTAL_PAGE,
-			MONITOR_LRU_BATCH_COUNT,
-			MONITOR_LRU_BATCH_PAGES,
-			total_flushed);
+		if (n.evicted) {
+			MONITOR_INC_VALUE_CUMULATIVE(
+				MONITOR_LRU_BATCH_EVICT_TOTAL_PAGE,
+				MONITOR_LRU_BATCH_EVICT_COUNT,
+				MONITOR_LRU_BATCH_EVICT_PAGES,
+				n.evicted);
+		}
+
+		total_flushed += (n.flushed + n.evicted);
 	}
 
 	return(total_flushed);
@@ -2404,11 +2249,14 @@ page_cleaner_sleep_if_needed(
 
 	if (next_loop_time > cur_time) {
 		/* Get sleep interval in micro seconds. We use
-		ut_min() to avoid long sleep in case of
-		wrap around. */
-		os_thread_sleep(ut_min(1000000,
-				(next_loop_time - cur_time)
-				 * 1000));
+		ut_min() to avoid long sleep in case of wrap around. */
+		ulint	sleep_us;
+
+		sleep_us = ut_min(1000000, (next_loop_time - cur_time) * 1000);
+
+		ib_int64_t	sig_count = os_event_reset(buf_flush_event);
+
+		os_event_wait_time_low(buf_flush_event, sleep_us, sig_count);
 	}
 }
 
@@ -2538,6 +2386,8 @@ DECLARE_THREAD(buf_flush_page_cleaner_thread)(
 thread_exit:
 	buf_page_cleaner_is_active = FALSE;
 
+	os_event_free(buf_flush_event);
+
 	/* We count the number of threads in os_thread_exit(). A created
 	thread should always use that to exit and not use return() to exit. */
 	os_thread_exit(NULL);

storage/innobase/buf/buf0lru.cc

@@ -81,6 +81,10 @@ are not blocked for extended period of time when using very large
 buffer pools. */
 #define BUF_LRU_DROP_SEARCH_SIZE	1024
 
+/** We scan these many blocks when looking for a clean page to evict
+during LRU eviction. */
+#define BUF_LRU_SEARCH_SCAN_THRESHOLD	100
+
 /** If we switch on the InnoDB monitor because there are too few available
 frames in the buffer pool, we set this to TRUE */
 static ibool	buf_lru_switched_on_innodb_mon	= FALSE;
@@ -961,7 +965,7 @@ buf_LRU_free_from_unzip_LRU_list(
 	}
 
 	for (block = UT_LIST_GET_LAST(buf_pool->unzip_LRU),
-	     scanned = 1, freed = FALSE;
+	     scanned = 0, freed = FALSE;
 	     block != NULL && !freed
 	     && (scan_all || scanned < srv_LRU_scan_depth);
 	     ++scanned) {
@@ -978,11 +982,13 @@ buf_LRU_free_from_unzip_LRU_list(
 		block = prev_block;
 	}
 
-	MONITOR_INC_VALUE_CUMULATIVE(
-		MONITOR_LRU_UNZIP_SEARCH_SCANNED,
-		MONITOR_LRU_UNZIP_SEARCH_SCANNED_NUM_CALL,
-		MONITOR_LRU_UNZIP_SEARCH_SCANNED_PER_CALL,
-		scanned);
+	if (scanned) {
+		MONITOR_INC_VALUE_CUMULATIVE(
+			MONITOR_LRU_UNZIP_SEARCH_SCANNED,
+			MONITOR_LRU_UNZIP_SEARCH_SCANNED_NUM_CALL,
+			MONITOR_LRU_UNZIP_SEARCH_SCANNED_PER_CALL,
+			scanned);
+	}
 	return(freed);
 }
 
@@ -1004,21 +1010,30 @@ buf_LRU_free_from_common_LRU_list(
 
 	ut_ad(buf_pool_mutex_own(buf_pool));
 
-	for (bpage = UT_LIST_GET_LAST(buf_pool->LRU),
-	     scanned = 1, freed = FALSE;
+	for (bpage = buf_pool->lru_scan_itr.start(),
+	     scanned = 0, freed = false;
 	     bpage != NULL && !freed
-	     && (scan_all || scanned < srv_LRU_scan_depth);
-	     ++scanned) {
+	     && (scan_all || scanned < BUF_LRU_SEARCH_SCAN_THRESHOLD);
+	     ++scanned, bpage = buf_pool->lru_scan_itr.get()) {
 
-		unsigned	accessed;
-		buf_page_t*	prev_bpage = UT_LIST_GET_PREV(LRU,
-						bpage);
+		buf_page_t* prev = UT_LIST_GET_PREV(LRU, bpage);
+		buf_pool->lru_scan_itr.set(prev);
+
+		ib_mutex_t* mutex = buf_page_get_mutex(bpage);
+		mutex_enter(mutex);
 
 		ut_ad(buf_page_in_file(bpage));
 		ut_ad(bpage->in_LRU_list);
 
-		accessed = buf_page_is_accessed(bpage);
-		freed = buf_LRU_free_page(bpage, true);
+		unsigned accessed = buf_page_is_accessed(bpage);
+
+		if (buf_flush_ready_for_replace(bpage)) {
+			mutex_exit(mutex);
+			freed = buf_LRU_free_page(bpage, true);
+		} else {
+			mutex_exit(mutex);
+		}
+
 		if (freed && !accessed) {
 			/* Keep track of pages that are evicted without
 			ever being accessed. This gives us a measure of
@@ -1026,14 +1041,17 @@ buf_LRU_free_from_common_LRU_list(
 			++buf_pool->stat.n_ra_pages_evicted;
 		}
 
-		bpage = prev_bpage;
+		ut_ad(buf_pool_mutex_own(buf_pool));
+		ut_ad(!mutex_own(mutex));
 	}
 
-	MONITOR_INC_VALUE_CUMULATIVE(
-		MONITOR_LRU_SEARCH_SCANNED,
-		MONITOR_LRU_SEARCH_SCANNED_NUM_CALL,
-		MONITOR_LRU_SEARCH_SCANNED_PER_CALL,
-		scanned);
+	if (scanned) {
+		MONITOR_INC_VALUE_CUMULATIVE(
+			MONITOR_LRU_SEARCH_SCANNED,
+			MONITOR_LRU_SEARCH_SCANNED_NUM_CALL,
+			MONITOR_LRU_SEARCH_SCANNED_PER_CALL,
+			scanned);
+	}
 
 	return(freed);
 }
@@ -1217,8 +1235,6 @@ the free list. Even when we flush a page or find a page in LRU scan
 we put it to free list to be used.
 * iteration 0:
   * get a block from free list, success:done
-  * if there is an LRU flush batch in progress:
-    * wait for batch to end: retry free list
   * if buf_pool->try_LRU_scan is set
     * scan LRU up to srv_LRU_scan_depth to find a clean block
     * the above will put the block on free list
@@ -1231,7 +1247,7 @@ we put it to free list to be used.
     * scan whole LRU list
     * scan LRU list even if buf_pool->try_LRU_scan is not set
 * iteration > 1:
-  * same as iteration 1 but sleep 100ms
+  * same as iteration 1 but sleep 10ms
 @return	the free control block, in state BUF_BLOCK_READY_FOR_USE */
 UNIV_INTERN
 buf_block_t*
@@ -1269,20 +1285,6 @@ loop:
 		return(block);
 	}
 
-	if (buf_pool->init_flush[BUF_FLUSH_LRU]
-	    && srv_use_doublewrite_buf
-	    && buf_dblwr != NULL) {
-
-		/* If there is an LRU flush happening in the background
-		then we wait for it to end instead of trying a single
-		page flush. If, however, we are not using doublewrite
-		buffer then it is better to do our own single page
-		flush instead of waiting for LRU flush to end. */
-		buf_pool_mutex_exit(buf_pool);
-		buf_flush_wait_batch_end(buf_pool, BUF_FLUSH_LRU);
-		goto loop;
-	}
-
 	freed = FALSE;
 	if (buf_pool->try_LRU_scan || n_iterations > 0) {
 		/* If no block was in the free list, search from the
@@ -1299,6 +1301,10 @@ loop:
 			TRUE again when we flush a batch from this
 			buffer pool. */
 			buf_pool->try_LRU_scan = FALSE;
+
+			/* Also tell the page_cleaner thread that
+			there is work for it to do. */
+			os_event_set(buf_flush_event);
 		}
 	}
 
@@ -1347,12 +1353,10 @@ loop:
 
 	/* If we have scanned the whole LRU and still are unable to
 	find a free block then we should sleep here to let the
-	page_cleaner do an LRU batch for us.
-	TODO: It'd be better if we can signal the page_cleaner. Perhaps
-	we should use timed wait for page_cleaner. */
-	if (n_iterations > 1) {
+	page_cleaner do an LRU batch for us. */
 
-		os_thread_sleep(100000);
+	if (n_iterations > 1) {
+		os_thread_sleep(10000);
 	}
 
 	/* No free block was found: try to flush the LRU list.
@@ -1502,6 +1506,20 @@ buf_unzip_LRU_remove_block_if_needed(
 	}
 }
 
+/******************************************************************//**
+Adjust LRU hazard pointers if needed. */
+
+void
+buf_LRU_adjust_hp(
+/*==============*/
+	buf_pool_t*		buf_pool,/*!< in: buffer pool instance */
+	const buf_page_t*	bpage)	/*!< in: control block */
+{
+	buf_pool->lru_hp.adjust(bpage);
+	buf_pool->lru_scan_itr.adjust(bpage);
+	buf_pool->single_scan_itr.adjust(bpage);
+}
+
 /******************************************************************//**
 Removes a block from the LRU list. */
 UNIV_INLINE
@@ -1521,6 +1539,10 @@ buf_LRU_remove_block(
 
 	ut_ad(bpage->in_LRU_list);
 
+	/* Important that we adjust the hazard pointers before removing
+	bpage from the LRU list. */
+	buf_LRU_adjust_hp(buf_pool, bpage);
+
 	/* If the LRU_old pointer is defined and points to just this block,
 	move it backward one step */
 

storage/innobase/buf/buf0mtflu.cc

@@ -110,7 +110,8 @@ typedef struct wrk_itm
 					will be used */
 	wr_tsk_t	wr;		/*!< Flush page list */
 	rd_tsk_t	rd;		/*!< Decompress page list */
-        ulint		n_flushed; 	/*!< Flushed pages count  */
+        ulint		n_flushed; 	/*!< Number of flushed pages */
+	ulint		n_evicted;	/*!< Number of evicted pages */
  	os_thread_id_t	id_usr;		/*!< Thread-id currently working */
     	wrk_status_t    wi_status;	/*!< Work item status */
 	mem_heap_t      *wheap;         /*!< Heap were to allocate memory
@@ -180,6 +181,7 @@ buf_mtflu_flush_pool_instance(
 /*==========================*/
 	wrk_t	*work_item)	/*!< inout: work item to be flushed */
 {
+	flush_counters_t	n;
 	ut_a(work_item != NULL);
 	ut_a(work_item->wr.buf_pool != NULL);
 
@@ -211,14 +213,16 @@ buf_mtflu_flush_pool_instance(
         	work_item->wr.min = ut_min(srv_LRU_scan_depth,work_item->wr.min);
     	}
 
-	work_item->n_flushed = buf_flush_batch(work_item->wr.buf_pool,
-                                    		work_item->wr.flush_type,
-                                    		work_item->wr.min,
-						work_item->wr.lsn_limit);
-
+	buf_flush_batch(work_item->wr.buf_pool,
+		work_item->wr.flush_type,
+		work_item->wr.min,
+		work_item->wr.lsn_limit,
+		&n);
 
 	buf_flush_end(work_item->wr.buf_pool, work_item->wr.flush_type);
-	buf_flush_common(work_item->wr.flush_type, work_item->n_flushed);
+	buf_flush_common(work_item->wr.flush_type, n.flushed);
+	work_item->n_flushed = n.flushed;
+	work_item->n_evicted = n.evicted;
 
 	return work_item->n_flushed;
 }
@@ -359,7 +363,7 @@ void
 buf_mtflu_io_thread_exit(void)
 /*==========================*/
 {
-	long i;
+	ulint i;
 	thread_sync_t* mtflush_io = mtflush_ctx;
 	wrk_t* work_item = NULL;
 
@@ -393,7 +397,7 @@ buf_mtflu_io_thread_exit(void)
 	ut_a(ib_wqueue_is_empty(mtflush_io->wq));
 
 	/* Send one exit work item/thread */
-	for (i=0; i < srv_mtflush_threads; i++) {
+	for (i=0; i < (ulint)srv_mtflush_threads; i++) {
 		work_item[i].tsk = MT_WRK_NONE;
 		work_item[i].wi_status = WRK_ITEM_EXIT;
 		work_item[i].wheap = mtflush_io->wheap;
@@ -413,11 +417,8 @@ buf_mtflu_io_thread_exit(void)
 
 	ut_a(ib_wqueue_is_empty(mtflush_io->wq));
 
-	/* Requests sent */
-	os_fast_mutex_unlock(&mtflush_mtx);
-
 	/* Collect all work done items */
-	for (i=0; i < srv_mtflush_threads;) {
+	for (i=0; i < (ulint)srv_mtflush_threads;) {
 		wrk_t* work_item = NULL;
 
 		work_item = (wrk_t *)ib_wqueue_timedwait(mtflush_io->wr_cq, MT_WAIT_IN_USECS);
@@ -432,6 +433,11 @@ buf_mtflu_io_thread_exit(void)
 	/* Wait about 1/2 sec to allow threads really exit */
 	os_thread_sleep(MT_WAIT_IN_USECS);
 
+	while(!ib_wqueue_is_empty(mtflush_io->wq))
+	{
+		ib_wqueue_nowait(mtflush_io->wq);
+	}
+
 	ut_a(ib_wqueue_is_empty(mtflush_io->wq));
 	ut_a(ib_wqueue_is_empty(mtflush_io->wr_cq));
 	ut_a(ib_wqueue_is_empty(mtflush_io->rd_cq));
@@ -441,6 +447,8 @@ buf_mtflu_io_thread_exit(void)
 	ib_wqueue_free(mtflush_io->wr_cq);
 	ib_wqueue_free(mtflush_io->rd_cq);
 
+	/* Requests sent */
+	os_fast_mutex_unlock(&mtflush_mtx);
 	os_fast_mutex_free(&mtflush_mtx);
 	os_fast_mutex_free(&mtflush_io->thread_global_mtx);
 
@@ -518,8 +526,8 @@ ulint
 buf_mtflu_flush_work_items(
 /*=======================*/
 	ulint buf_pool_inst,		/*!< in: Number of buffer pool instances */
-	ulint *per_pool_pages_flushed,	/*!< out: Number of pages
-					flushed/instance */
+	flush_counters_t *per_pool_cnt,	/*!< out: Number of pages
+					flushed or evicted /instance */
 	buf_flush_t flush_type,		/*!< in: Type of flush */
 	ulint min_n,			/*!< in: Wished minimum number of
 					blocks to be flushed */
@@ -549,6 +557,7 @@ buf_mtflu_flush_work_items(
 		work_item[i].wheap = work_heap;
 		work_item[i].rheap = reply_heap;
 		work_item[i].n_flushed = 0;
+		work_item[i].n_evicted = 0;
 		work_item[i].id_usr = 0;
 
 		ib_wqueue_add(mtflush_ctx->wq,
@@ -562,7 +571,8 @@ buf_mtflu_flush_work_items(
 		done_wi = (wrk_t *)ib_wqueue_wait(mtflush_ctx->wr_cq);
 
 		if (done_wi != NULL) {
-			per_pool_pages_flushed[i] = done_wi->n_flushed;
+			per_pool_cnt[i].flushed = done_wi->n_flushed;
+			per_pool_cnt[i].evicted = done_wi->n_evicted;
 
 #ifdef UNIV_MTFLUSH_DEBUG
 			if((int)done_wi->id_usr == 0 &&
@@ -576,7 +586,7 @@ buf_mtflu_flush_work_items(
 			}
 #endif
 
-			n_flushed+= done_wi->n_flushed;
+			n_flushed+= done_wi->n_flushed+done_wi->n_evicted;
 			i++;
 		}
 	}
@@ -607,9 +617,9 @@ buf_mtflu_flush_list(
 					back to caller. Ignored if NULL */
 
 {
-	ulint		i;
-	bool		success = true;
-	ulint		cnt_flush[MTFLUSH_MAX_WORKER];
+	ulint				i;
+	bool				success = true;
+	flush_counters_t		cnt[MTFLUSH_MAX_WORKER];
 
 	if (n_processed) {
 		*n_processed = 0;
@@ -627,20 +637,29 @@ buf_mtflu_flush_list(
 	/* This lock is to safequard against re-entry if any. */
 	os_fast_mutex_lock(&mtflush_mtx);
 	buf_mtflu_flush_work_items(srv_buf_pool_instances,
-                cnt_flush, BUF_FLUSH_LIST,
+                cnt, BUF_FLUSH_LIST,
                 min_n, lsn_limit);
 	os_fast_mutex_unlock(&mtflush_mtx);
 
 	for (i = 0; i < srv_buf_pool_instances; i++) {
 		if (n_processed) {
-			*n_processed += cnt_flush[i];
+			*n_processed += cnt[i].flushed+cnt[i].evicted;
 		}
-		if (cnt_flush[i]) {
+
+		if (cnt[i].flushed) {
 			MONITOR_INC_VALUE_CUMULATIVE(
 				MONITOR_FLUSH_BATCH_TOTAL_PAGE,
 				MONITOR_FLUSH_BATCH_COUNT,
 				MONITOR_FLUSH_BATCH_PAGES,
-				cnt_flush[i]);
+				cnt[i].flushed);
+		}
+
+		if(cnt[i].evicted) {
+				MONITOR_INC_VALUE_CUMULATIVE(
+				MONITOR_LRU_BATCH_EVICT_TOTAL_PAGE,
+				MONITOR_LRU_BATCH_EVICT_COUNT,
+				MONITOR_LRU_BATCH_EVICT_PAGES,
+				cnt[i].evicted);
 		}
 	}
 #ifdef UNIV_MTFLUSH_DEBUG
@@ -663,25 +682,38 @@ buf_mtflu_flush_LRU_tail(void)
 /*==========================*/
 {
 	ulint	total_flushed=0, i;
-	ulint	cnt_flush[MTFLUSH_MAX_WORKER];
+	flush_counters_t	cnt[MTFLUSH_MAX_WORKER];
 
 	ut_a(buf_mtflu_init_done());
 
+	/* At shutdown do not send requests anymore */
+	if (!mtflush_ctx || mtflush_ctx->gwt_status == WTHR_KILL_IT) {
+		return (total_flushed);
+	}
+
 	/* This lock is to safeguard against re-entry if any */
 	os_fast_mutex_lock(&mtflush_mtx);
 	buf_mtflu_flush_work_items(srv_buf_pool_instances,
-		cnt_flush, BUF_FLUSH_LRU, srv_LRU_scan_depth, 0);
+		cnt, BUF_FLUSH_LRU, srv_LRU_scan_depth, 0);
 	os_fast_mutex_unlock(&mtflush_mtx);
 
 	for (i = 0; i < srv_buf_pool_instances; i++) {
-		if (cnt_flush[i]) {
-			total_flushed += cnt_flush[i];
+		total_flushed += cnt[i].flushed+cnt[i].evicted;
 
+		if (cnt[i].flushed) {
 			MONITOR_INC_VALUE_CUMULATIVE(
-			        MONITOR_LRU_BATCH_TOTAL_PAGE,
-			        MONITOR_LRU_BATCH_COUNT,
-			        MONITOR_LRU_BATCH_PAGES,
-			        cnt_flush[i]);
+			        MONITOR_LRU_BATCH_FLUSH_TOTAL_PAGE,
+			        MONITOR_LRU_BATCH_FLUSH_COUNT,
+			        MONITOR_LRU_BATCH_FLUSH_PAGES,
+			        cnt[i].flushed);
+		}
+
+		if(cnt[i].evicted) {
+				MONITOR_INC_VALUE_CUMULATIVE(
+				MONITOR_LRU_BATCH_EVICT_TOTAL_PAGE,
+				MONITOR_LRU_BATCH_EVICT_COUNT,
+				MONITOR_LRU_BATCH_EVICT_PAGES,
+				cnt[i].evicted);
 		}
 	}
 

storage/innobase/include/buf0buf.h

@@ -1199,7 +1199,9 @@ UNIV_INTERN
 bool
 buf_page_io_complete(
 /*=================*/
-	buf_page_t*	bpage);	/*!< in: pointer to the block in question */
+	buf_page_t*	bpage,	/*!< in: pointer to the block in question */
+	bool		evict = false);/*!< in: whether or not to evict
+				the page from LRU list. */
 /********************************************************************//**
 Calculates a folded value of a file page address to use in the page hash
 table.
@@ -1762,6 +1764,133 @@ Compute the hash fold value for blocks in buf_pool->zip_hash. */
 #define BUF_POOL_ZIP_FOLD_BPAGE(b) BUF_POOL_ZIP_FOLD((buf_block_t*) (b))
 /* @} */
 
+/** A "Hazard Pointer" class used to iterate over page lists
+inside the buffer pool. A hazard pointer is a buf_page_t pointer
+which we intend to iterate over next and we want it remain valid
+even after we release the buffer pool mutex. */
+class HazardPointer {
+
+public:
+	/** Constructor
+	@param buf_pool buffer pool instance
+	@param mutex	mutex that is protecting the hp. */
+	HazardPointer(const buf_pool_t* buf_pool, const ib_mutex_t* mutex)
+		:
+		m_buf_pool(buf_pool)
+#ifdef UNIV_DEBUG
+		, m_mutex(mutex)
+#endif /* UNIV_DEBUG */
+		, m_hp() {}
+
+	/** Destructor */
+	virtual ~HazardPointer() {}
+
+	/** Get current value */
+	buf_page_t* get()
+	{
+		ut_ad(mutex_own(m_mutex));
+		return(m_hp);
+	}
+
+	/** Set current value
+	@param bpage	buffer block to be set as hp */
+	void set(buf_page_t* bpage);
+
+	/** Checks if a bpage is the hp
+	@param bpage	buffer block to be compared
+	@return true if it is hp */
+	bool is_hp(const buf_page_t* bpage);
+
+	/** Adjust the value of hp. This happens when some
+	other thread working on the same list attempts to
+	remove the hp from the list. Must be implemented
+	by the derived classes.
+	@param bpage	buffer block to be compared */
+	virtual void adjust(const buf_page_t*) = 0;
+
+protected:
+	/** Disable copying */
+	HazardPointer(const HazardPointer&);
+	HazardPointer& operator=(const HazardPointer&);
+
+	/** Buffer pool instance */
+	const buf_pool_t*	m_buf_pool;
+
+#if UNIV_DEBUG
+	/** mutex that protects access to the m_hp. */
+	const ib_mutex_t*	m_mutex;
+#endif /* UNIV_DEBUG */
+
+	/** hazard pointer. */
+	buf_page_t*		m_hp;
+};
+
+/** Class implementing buf_pool->flush_list hazard pointer */
+class FlushHp: public HazardPointer {
+
+public:
+	/** Constructor
+	@param buf_pool buffer pool instance
+	@param mutex	mutex that is protecting the hp. */
+	FlushHp(const buf_pool_t* buf_pool, const ib_mutex_t* mutex)
+		:
+		HazardPointer(buf_pool, mutex) {}
+
+	/** Destructor */
+	virtual ~FlushHp() {}
+
+	/** Adjust the value of hp. This happens when some
+	other thread working on the same list attempts to
+	remove the hp from the list.
+	@param bpage	buffer block to be compared */
+	void adjust(const buf_page_t* bpage);
+};
+
+/** Class implementing buf_pool->LRU hazard pointer */
+class LRUHp: public HazardPointer {
+
+public:
+	/** Constructor
+	@param buf_pool buffer pool instance
+	@param mutex	mutex that is protecting the hp. */
+	LRUHp(const buf_pool_t* buf_pool, const ib_mutex_t* mutex)
+		:
+		HazardPointer(buf_pool, mutex) {}
+
+	/** Destructor */
+	virtual ~LRUHp() {}
+
+	/** Adjust the value of hp. This happens when some
+	other thread working on the same list attempts to
+	remove the hp from the list.
+	@param bpage	buffer block to be compared */
+	void adjust(const buf_page_t* bpage);
+};
+
+/** Special purpose iterators to be used when scanning the LRU list.
+The idea is that when one thread finishes the scan it leaves the
+itr in that position and the other thread can start scan from
+there */
+class LRUItr: public LRUHp {
+
+public:
+	/** Constructor
+	@param buf_pool buffer pool instance
+	@param mutex	mutex that is protecting the hp. */
+	LRUItr(const buf_pool_t* buf_pool, const ib_mutex_t* mutex)
+		:
+		LRUHp(buf_pool, mutex) {}
+
+	/** Destructor */
+	virtual ~LRUItr() {}
+
+	/** Selects from where to start a scan. If we have scanned
+	too deep into the LRU list it resets the value to the tail
+	of the LRU list.
+	@return buf_page_t from where to start scan. */
+	buf_page_t* start();
+};
+
 /** Struct that is embedded in the free zip blocks */
 struct buf_buddy_free_t {
 	union {
@@ -1894,7 +2023,7 @@ struct buf_pool_t{
 					also protects writes to
 					bpage::oldest_modification and
 					flush_list_hp */
-	const buf_page_t*	flush_list_hp;/*!< "hazard pointer"
+	FlushHp			flush_hp;/*!< "hazard pointer"
 					used during scan of flush_list
 					while doing flush list batch.
 					Protected by flush_list_mutex */
@@ -1952,6 +2081,19 @@ struct buf_pool_t{
 	UT_LIST_BASE_NODE_T(buf_page_t) free;
 					/*!< base node of the free
 					block list */
+
+	/** "hazard pointer" used during scan of LRU while doing
+	LRU list batch.  Protected by buf_pool::mutex */
+	LRUHp		lru_hp;
+
+	/** Iterator used to scan the LRU list when searching for
+	replacable victim. Protected by buf_pool::mutex. */
+	LRUItr		lru_scan_itr;
+
+	/** Iterator used to scan the LRU list when searching for
+	single page flushing victim.  Protected by buf_pool::mutex. */
+	LRUItr		single_scan_itr;
+
 	UT_LIST_BASE_NODE_T(buf_page_t) LRU;
 					/*!< base node of the LRU list */
 	buf_page_t*	LRU_old;	/*!< pointer to the about

storage/innobase/include/buf0flu.h

@@ -37,6 +37,17 @@ Created 11/5/1995 Heikki Tuuri
 /** Flag indicating if the page_cleaner is in active state. */
 extern ibool buf_page_cleaner_is_active;
 
+/** Event to synchronise with the flushing. */
+extern os_event_t	buf_flush_event;
+
+/** Handled page counters for a single flush */
+struct flush_counters_t {
+	ulint	flushed;	/*!< number of dirty pages flushed */
+	ulint	evicted;	/*!< number of clean pages evicted */
+	ulint	unzip_LRU_evicted;/*!< number of uncompressed page images
+				evicted */
+};
+
 /********************************************************************//**
 Remove a block from the flush list of modified blocks. */
 UNIV_INTERN
@@ -111,12 +122,12 @@ buf_flush_list(
 					which were processed is passed
 					back to caller. Ignored if NULL */
 /******************************************************************//**
-This function picks up a single dirty page from the tail of the LRU
-list, flushes it, removes it from page_hash and LRU list and puts
-it on the free list. It is called from user threads when they are
-unable to find a replacable page at the tail of the LRU list i.e.:
-when the background LRU flushing in the page_cleaner thread is not
-fast enough to keep pace with the workload.
+This function picks up a single page from the tail of the LRU
+list, flushes it (if it is dirty), removes it from page_hash and LRU
+list and puts it on the free list. It is called from user threads when
+they are unable to find a replaceable page at the tail of the LRU
+list i.e.: when the background LRU flushing in the page_cleaner thread
+is not fast enough to keep pace with the workload.
 @return TRUE if success. */
 UNIV_INTERN
 ibool
@@ -309,9 +320,9 @@ This utility flushes dirty blocks from the end of the LRU list or flush_list.
 NOTE 1: in the case of an LRU flush the calling thread may own latches to
 pages: to avoid deadlocks, this function must be written so that it cannot
 end up waiting for these latches! NOTE 2: in the case of a flush list flush,
-the calling thread is not allowed to own any latches on pages!
-@return number of blocks for which the write request was queued */
-ulint
+the calling thread is not allowed to own any latches on pages! */
+__attribute__((nonnull))
+void
 buf_flush_batch(
 /*============*/
 	buf_pool_t*	buf_pool,	/*!< in: buffer pool instance */
@@ -322,11 +333,13 @@ buf_flush_batch(
 	ulint		min_n,		/*!< in: wished minimum mumber of blocks
 					flushed (it is not guaranteed that the
 					actual number is that big, though) */
-	lsn_t		lsn_limit);	/*!< in: in the case of BUF_FLUSH_LIST
+	lsn_t		lsn_limit,	/*!< in: in the case of BUF_FLUSH_LIST
 					all blocks whose oldest_modification is
 					smaller than this should be flushed
 					(if their number does not exceed
 					min_n), otherwise ignored */
+	flush_counters_t*	n);	/*!< out: flushed/evicted page
+					counts  */
 
 
 #ifndef UNIV_NONINL

storage/innobase/include/buf0lru.h

@@ -117,7 +117,7 @@ buf_LRU_get_free_only(
 	buf_pool_t*	buf_pool);	/*!< buffer pool instance */
 /******************************************************************//**
 Returns a free block from the buf_pool. The block is taken off the
-free list. If it is empty, blocks are moved from the end of the
+free list. If free list is empty, blocks are moved from the end of the
 LRU list to the free list.
 This function is called from a user thread when it needs a clean
 block to read in a page. Note that we only ever get a block from
@@ -125,8 +125,6 @@ the free list. Even when we flush a page or find a page in LRU scan
 we put it to free list to be used.
 * iteration 0:
   * get a block from free list, success:done
-  * if there is an LRU flush batch in progress:
-    * wait for batch to end: retry free list
   * if buf_pool->try_LRU_scan is set
     * scan LRU up to srv_LRU_scan_depth to find a clean block
     * the above will put the block on free list
@@ -139,7 +137,7 @@ we put it to free list to be used.
     * scan whole LRU list
     * scan LRU list even if buf_pool->try_LRU_scan is not set
 * iteration > 1:
-  * same as iteration 1 but sleep 100ms
+  * same as iteration 1 but sleep 10ms
 @return	the free control block, in state BUF_BLOCK_READY_FOR_USE */
 UNIV_INTERN
 buf_block_t*
@@ -231,6 +229,15 @@ buf_LRU_free_one_page(
 				may or may not be a hash index to the page */
 	__attribute__((nonnull));
 
+/******************************************************************//**
+Adjust LRU hazard pointers if needed. */
+
+void
+buf_LRU_adjust_hp(
+/*==============*/
+	buf_pool_t*		buf_pool,/*!< in: buffer pool instance */
+	const buf_page_t*	bpage);	/*!< in: control block */
+
 #if defined UNIV_DEBUG || defined UNIV_BUF_DEBUG
 /**********************************************************************//**
 Validates the LRU list.

storage/innobase/include/srv0mon.h

@@ -174,7 +174,6 @@ enum monitor_id_t {
 	MONITOR_FLUSH_BATCH_SCANNED,
 	MONITOR_FLUSH_BATCH_SCANNED_NUM_CALL,
 	MONITOR_FLUSH_BATCH_SCANNED_PER_CALL,
-	MONITOR_FLUSH_HP_RESCAN,
 	MONITOR_FLUSH_BATCH_TOTAL_PAGE,
 	MONITOR_FLUSH_BATCH_COUNT,
 	MONITOR_FLUSH_BATCH_PAGES,
@@ -199,9 +198,12 @@ enum monitor_id_t {
 	MONITOR_LRU_BATCH_SCANNED,
 	MONITOR_LRU_BATCH_SCANNED_NUM_CALL,
 	MONITOR_LRU_BATCH_SCANNED_PER_CALL,
-	MONITOR_LRU_BATCH_TOTAL_PAGE,
-	MONITOR_LRU_BATCH_COUNT,
-	MONITOR_LRU_BATCH_PAGES,
+	MONITOR_LRU_BATCH_FLUSH_TOTAL_PAGE,
+	MONITOR_LRU_BATCH_FLUSH_COUNT,
+	MONITOR_LRU_BATCH_FLUSH_PAGES,
+	MONITOR_LRU_BATCH_EVICT_TOTAL_PAGE,
+	MONITOR_LRU_BATCH_EVICT_COUNT,
+	MONITOR_LRU_BATCH_EVICT_PAGES,
 	MONITOR_LRU_SINGLE_FLUSH_SCANNED,
 	MONITOR_LRU_SINGLE_FLUSH_SCANNED_NUM_CALL,
 	MONITOR_LRU_SINGLE_FLUSH_SCANNED_PER_CALL,

storage/innobase/log/log0recv.cc

@@ -348,7 +348,10 @@ DECLARE_THREAD(recv_writer_thread)(
 
 	while (srv_shutdown_state == SRV_SHUTDOWN_NONE) {
 
-		os_thread_sleep(100000);
+		/* Wait till we get a signal to clean the LRU list.
+		Bounded by max wait time of 100ms. */
+		ib_int64_t      sig_count = os_event_reset(buf_flush_event);
+		os_event_wait_time_low(buf_flush_event, 100000, sig_count);
 
 		mutex_enter(&recv_sys->writer_mutex);
 

storage/innobase/srv/srv0mon.cc

@@ -350,11 +350,6 @@ static monitor_info_t	innodb_counter_info[] =
 	 MONITOR_SET_MEMBER, MONITOR_FLUSH_BATCH_SCANNED,
 	 MONITOR_FLUSH_BATCH_SCANNED_PER_CALL},
 
-	{"buffer_flush_batch_rescan", "buffer",
-	 "Number of times rescan of flush list forced",
-	 MONITOR_NONE,
-	 MONITOR_DEFAULT_START, MONITOR_FLUSH_HP_RESCAN},
-
 	/* Cumulative counter for pages flushed in flush batches */
 	{"buffer_flush_batch_total_pages", "buffer",
 	 "Total pages flushed as part of flush batch",
@@ -482,20 +477,36 @@ static monitor_info_t	innodb_counter_info[] =
 	 MONITOR_LRU_BATCH_SCANNED_PER_CALL},
 
 	/* Cumulative counter for LRU batch pages flushed */
-	{"buffer_LRU_batch_total_pages", "buffer",
+	{"buffer_LRU_batch_flush_total_pages", "buffer",
 	 "Total pages flushed as part of LRU batches",
-	 MONITOR_SET_OWNER, MONITOR_LRU_BATCH_COUNT,
-	 MONITOR_LRU_BATCH_TOTAL_PAGE},
+	 MONITOR_SET_OWNER, MONITOR_LRU_BATCH_FLUSH_COUNT,
+	 MONITOR_LRU_BATCH_FLUSH_TOTAL_PAGE},
+
+	{"buffer_LRU_batches_flush", "buffer",
+	 "Number of LRU batches",
+	 MONITOR_SET_MEMBER, MONITOR_LRU_BATCH_FLUSH_TOTAL_PAGE,
+	 MONITOR_LRU_BATCH_FLUSH_COUNT},
+
+	{"buffer_LRU_batch_flush_pages", "buffer",
+	 "Pages queued as an LRU batch",
+	 MONITOR_SET_MEMBER, MONITOR_LRU_BATCH_FLUSH_TOTAL_PAGE,
+	 MONITOR_LRU_BATCH_FLUSH_PAGES},
+
+	/* Cumulative counter for LRU batch pages flushed */
+	{"buffer_LRU_batch_evict_total_pages", "buffer",
+	 "Total pages evicted as part of LRU batches",
+	 MONITOR_SET_OWNER, MONITOR_LRU_BATCH_EVICT_COUNT,
+	 MONITOR_LRU_BATCH_EVICT_TOTAL_PAGE},
 
-	{"buffer_LRU_batches", "buffer",
+	{"buffer_LRU_batches_evict", "buffer",
 	 "Number of LRU batches",
-	 MONITOR_SET_MEMBER, MONITOR_LRU_BATCH_TOTAL_PAGE,
-	 MONITOR_LRU_BATCH_COUNT},
+	 MONITOR_SET_MEMBER, MONITOR_LRU_BATCH_EVICT_TOTAL_PAGE,
+	 MONITOR_LRU_BATCH_EVICT_COUNT},
 
-	{"buffer_LRU_batch_pages", "buffer",
+	{"buffer_LRU_batch_evict_pages", "buffer",
 	 "Pages queued as an LRU batch",
-	 MONITOR_SET_MEMBER, MONITOR_LRU_BATCH_TOTAL_PAGE,
-	 MONITOR_LRU_BATCH_PAGES},
+	 MONITOR_SET_MEMBER, MONITOR_LRU_BATCH_EVICT_TOTAL_PAGE,
+	 MONITOR_LRU_BATCH_EVICT_PAGES},
 
 	/* Cumulative counter for single page LRU scans */
 	{"buffer_LRU_single_flush_scanned", "buffer",

storage/xtradb/buf/buf0flu.cc

@@ -2338,19 +2338,26 @@ buf_flush_LRU_tail(void)
 					free_len = UT_LIST_GET_LEN(
 						buf_pool->free);
 				}
+				if (n.flushed) {
+					MONITOR_INC_VALUE_CUMULATIVE(
+						MONITOR_LRU_BATCH_FLUSH_TOTAL_PAGE,
+						MONITOR_LRU_BATCH_FLUSH_COUNT,
+						MONITOR_LRU_BATCH_FLUSH_PAGES,
+						n.flushed);
+				}
+
+				if (n.evicted) {
+					MONITOR_INC_VALUE_CUMULATIVE(
+						MONITOR_LRU_BATCH_EVICT_TOTAL_PAGE,
+						MONITOR_LRU_BATCH_EVICT_COUNT,
+						MONITOR_LRU_BATCH_EVICT_PAGES,
+						n.evicted);
+				}
 			} while (active_instance[i]
 				 && free_len <= free_list_lwm);
 		}
 	}
 
-	if (total_flushed) {
-		MONITOR_INC_VALUE_CUMULATIVE(
-			MONITOR_LRU_BATCH_TOTAL_PAGE,
-			MONITOR_LRU_BATCH_COUNT,
-			MONITOR_LRU_BATCH_PAGES,
-			total_flushed);
-	}
-
 	return(total_flushed);
 }
 

storage/xtradb/buf/buf0mtflu.cc

@@ -110,7 +110,8 @@ typedef struct wrk_itm
 					will be used */
 	wr_tsk_t	wr;		/*!< Flush page list */
 	rd_tsk_t	rd;		/*!< Decompress page list */
-        ulint		n_flushed; 	/*!< Flushed pages count  */
+        ulint		n_flushed; 	/*!< Number of flushed pages */
+	ulint		n_evicted;	/*!< Number of evicted pages */
  	os_thread_id_t	id_usr;		/*!< Thread-id currently working */
     	wrk_status_t    wi_status;	/*!< Work item status */
 	mem_heap_t      *wheap;         /*!< Heap were to allocate memory
@@ -366,7 +367,7 @@ void
 buf_mtflu_io_thread_exit(void)
 /*==========================*/
 {
-	long i;
+	ulint i;
 	thread_sync_t* mtflush_io = mtflush_ctx;
 	wrk_t* work_item = NULL;
 
@@ -397,7 +398,7 @@ buf_mtflu_io_thread_exit(void)
 	os_fast_mutex_lock(&mtflush_mtx);
 
 	/* Send one exit work item/thread */
-	for (i=0; i < srv_mtflush_threads; i++) {
+	for (i=0; i < (ulint)srv_mtflush_threads; i++) {
 		work_item[i].tsk = MT_WRK_NONE;
 		work_item[i].wi_status = WRK_ITEM_EXIT;
 		work_item[i].wheap = mtflush_io->wheap;
@@ -417,11 +418,8 @@ buf_mtflu_io_thread_exit(void)
 
 	ut_a(ib_wqueue_is_empty(mtflush_io->wq));
 
-	/* Requests sent */
-	os_fast_mutex_unlock(&mtflush_mtx);
-
 	/* Collect all work done items */
-	for (i=0; i < srv_mtflush_threads;) {
+	for (i=0; i < (ulint)srv_mtflush_threads;) {
 		wrk_t* work_item = NULL;
 
 		work_item = (wrk_t *)ib_wqueue_timedwait(mtflush_io->wr_cq, MT_WAIT_IN_USECS);
@@ -436,6 +434,16 @@ buf_mtflu_io_thread_exit(void)
 	/* Wait about 1/2 sec to allow threads really exit */
 	os_thread_sleep(MT_WAIT_IN_USECS);
 
+	while(!ib_wqueue_is_empty(mtflush_io->wq))
+	{
+		ib_wqueue_nowait(mtflush_io->wq);
+	}
+
+	while(!ib_wqueue_is_empty(mtflush_io->wq))
+	{
+		ib_wqueue_nowait(mtflush_io->wq);
+	}
+
 	ut_a(ib_wqueue_is_empty(mtflush_io->wq));
 	ut_a(ib_wqueue_is_empty(mtflush_io->wr_cq));
 	ut_a(ib_wqueue_is_empty(mtflush_io->rd_cq));
@@ -445,6 +453,8 @@ buf_mtflu_io_thread_exit(void)
 	ib_wqueue_free(mtflush_io->wr_cq);
 	ib_wqueue_free(mtflush_io->rd_cq);
 
+	/* Requests sent */
+	os_fast_mutex_unlock(&mtflush_mtx);
 	os_fast_mutex_free(&mtflush_mtx);
 	os_fast_mutex_free(&mtflush_io->thread_global_mtx);
 
@@ -522,8 +532,8 @@ ulint
 buf_mtflu_flush_work_items(
 /*=======================*/
 	ulint buf_pool_inst,		/*!< in: Number of buffer pool instances */
-	ulint *per_pool_pages_flushed,	/*!< out: Number of pages
-					flushed/instance */
+	flush_counters_t *per_pool_cnt,	/*!< out: Number of pages
+					flushed or evicted /instance */
 	buf_flush_t flush_type,		/*!< in: Type of flush */
 	ulint min_n,			/*!< in: Wished minimum number of
 					blocks to be flushed */
@@ -553,6 +563,7 @@ buf_mtflu_flush_work_items(
 		work_item[i].wheap = work_heap;
 		work_item[i].rheap = reply_heap;
 		work_item[i].n_flushed = 0;
+		work_item[i].n_evicted = 0;
 		work_item[i].id_usr = 0;
 
 		ib_wqueue_add(mtflush_ctx->wq,
@@ -566,7 +577,8 @@ buf_mtflu_flush_work_items(
 		done_wi = (wrk_t *)ib_wqueue_wait(mtflush_ctx->wr_cq);
 
 		if (done_wi != NULL) {
-			per_pool_pages_flushed[i] = done_wi->n_flushed;
+			per_pool_cnt[i].flushed = done_wi->n_flushed;
+			per_pool_cnt[i].evicted = done_wi->n_evicted;
 
 #ifdef UNIV_MTFLUSH_DEBUG
 			if((int)done_wi->id_usr == 0 &&
@@ -580,7 +592,7 @@ buf_mtflu_flush_work_items(
 			}
 #endif
 
-			n_flushed+= done_wi->n_flushed;
+			n_flushed+= done_wi->n_flushed+done_wi->n_evicted;
 			i++;
 		}
 	}
@@ -614,9 +626,9 @@ buf_mtflu_flush_list(
 					back to caller. Ignored if NULL */
 
 {
-	ulint		i;
-	bool		success = true;
-	ulint		cnt_flush[MTFLUSH_MAX_WORKER];
+	ulint				i;
+	bool				success = true;
+	flush_counters_t		cnt[MTFLUSH_MAX_WORKER];
 
 	if (n_processed) {
 		*n_processed = 0;
@@ -634,20 +646,29 @@ buf_mtflu_flush_list(
 	/* This lock is to safequard against re-entry if any. */
 	os_fast_mutex_lock(&mtflush_mtx);
 	buf_mtflu_flush_work_items(srv_buf_pool_instances,
-                cnt_flush, BUF_FLUSH_LIST,
+                cnt, BUF_FLUSH_LIST,
                 min_n, lsn_limit);
 	os_fast_mutex_unlock(&mtflush_mtx);
 
 	for (i = 0; i < srv_buf_pool_instances; i++) {
 		if (n_processed) {
-			*n_processed += cnt_flush[i];
+			*n_processed += cnt[i].flushed+cnt[i].evicted;
 		}
-		if (cnt_flush[i]) {
+
+		if (cnt[i].flushed) {
 			MONITOR_INC_VALUE_CUMULATIVE(
 				MONITOR_FLUSH_BATCH_TOTAL_PAGE,
 				MONITOR_FLUSH_BATCH_COUNT,
 				MONITOR_FLUSH_BATCH_PAGES,
-				cnt_flush[i]);
+				cnt[i].flushed);
+		}
+
+		if(cnt[i].evicted) {
+				MONITOR_INC_VALUE_CUMULATIVE(
+				MONITOR_LRU_BATCH_EVICT_TOTAL_PAGE,
+				MONITOR_LRU_BATCH_EVICT_COUNT,
+				MONITOR_LRU_BATCH_EVICT_PAGES,
+				cnt[i].evicted);
 		}
 	}
 #ifdef UNIV_MTFLUSH_DEBUG
@@ -670,25 +691,37 @@ buf_mtflu_flush_LRU_tail(void)
 /*==========================*/
 {
 	ulint	total_flushed=0, i;
-	ulint	cnt_flush[MTFLUSH_MAX_WORKER];
+	flush_counters_t	cnt[MTFLUSH_MAX_WORKER];
 
 	ut_a(buf_mtflu_init_done());
 
+	/* At shutdown do not send requests anymore */
+	if (!mtflush_ctx || mtflush_ctx->gwt_status == WTHR_KILL_IT) {
+		return (total_flushed);
+	}
+
 	/* This lock is to safeguard against re-entry if any */
 	os_fast_mutex_lock(&mtflush_mtx);
 	buf_mtflu_flush_work_items(srv_buf_pool_instances,
-		cnt_flush, BUF_FLUSH_LRU, srv_LRU_scan_depth, 0);
+		cnt, BUF_FLUSH_LRU, srv_LRU_scan_depth, 0);
 	os_fast_mutex_unlock(&mtflush_mtx);
 
 	for (i = 0; i < srv_buf_pool_instances; i++) {
-		if (cnt_flush[i]) {
-			total_flushed += cnt_flush[i];
+		total_flushed += cnt[i].flushed+cnt[i].evicted;
 
+		if (cnt[i].flushed) {
 			MONITOR_INC_VALUE_CUMULATIVE(
-			        MONITOR_LRU_BATCH_TOTAL_PAGE,
-			        MONITOR_LRU_BATCH_COUNT,
-			        MONITOR_LRU_BATCH_PAGES,
-			        cnt_flush[i]);
+			        MONITOR_LRU_BATCH_FLUSH_TOTAL_PAGE,
+			        MONITOR_LRU_BATCH_FLUSH_COUNT,
+			        MONITOR_LRU_BATCH_FLUSH_PAGES,
+			        cnt[i].flushed);
+		}
+		if(cnt[i].evicted) {
+				MONITOR_INC_VALUE_CUMULATIVE(
+				MONITOR_LRU_BATCH_EVICT_TOTAL_PAGE,
+				MONITOR_LRU_BATCH_EVICT_COUNT,
+				MONITOR_LRU_BATCH_EVICT_PAGES,
+				cnt[i].evicted);
 		}
 	}
 

storage/xtradb/include/srv0mon.h

@@ -199,9 +199,12 @@ enum monitor_id_t {
 	MONITOR_LRU_BATCH_SCANNED,
 	MONITOR_LRU_BATCH_SCANNED_NUM_CALL,
 	MONITOR_LRU_BATCH_SCANNED_PER_CALL,
-	MONITOR_LRU_BATCH_TOTAL_PAGE,
-	MONITOR_LRU_BATCH_COUNT,
-	MONITOR_LRU_BATCH_PAGES,
+	MONITOR_LRU_BATCH_FLUSH_TOTAL_PAGE,
+	MONITOR_LRU_BATCH_FLUSH_COUNT,
+	MONITOR_LRU_BATCH_FLUSH_PAGES,
+	MONITOR_LRU_BATCH_EVICT_TOTAL_PAGE,
+	MONITOR_LRU_BATCH_EVICT_COUNT,
+	MONITOR_LRU_BATCH_EVICT_PAGES,
 	MONITOR_LRU_SINGLE_FLUSH_SCANNED,
 	MONITOR_LRU_SINGLE_FLUSH_SCANNED_NUM_CALL,
 	MONITOR_LRU_SINGLE_FLUSH_SCANNED_PER_CALL,

storage/xtradb/srv/srv0mon.cc

@@ -350,11 +350,6 @@ static monitor_info_t	innodb_counter_info[] =
 	 MONITOR_SET_MEMBER, MONITOR_FLUSH_BATCH_SCANNED,
 	 MONITOR_FLUSH_BATCH_SCANNED_PER_CALL},
 
-	{"buffer_flush_batch_rescan", "buffer",
-	 "Number of times rescan of flush list forced",
-	 MONITOR_NONE,
-	 MONITOR_DEFAULT_START, MONITOR_FLUSH_HP_RESCAN},
-
 	/* Cumulative counter for pages flushed in flush batches */
 	{"buffer_flush_batch_total_pages", "buffer",
 	 "Total pages flushed as part of flush batch",
@@ -482,20 +477,36 @@ static monitor_info_t	innodb_counter_info[] =
 	 MONITOR_LRU_BATCH_SCANNED_PER_CALL},
 
 	/* Cumulative counter for LRU batch pages flushed */
-	{"buffer_LRU_batch_total_pages", "buffer",
+	{"buffer_LRU_batch_flush_total_pages", "buffer",
 	 "Total pages flushed as part of LRU batches",
-	 MONITOR_SET_OWNER, MONITOR_LRU_BATCH_COUNT,
-	 MONITOR_LRU_BATCH_TOTAL_PAGE},
+	 MONITOR_SET_OWNER, MONITOR_LRU_BATCH_FLUSH_COUNT,
+	 MONITOR_LRU_BATCH_FLUSH_TOTAL_PAGE},
+
+	{"buffer_LRU_batches_flush", "buffer",
+	 "Number of LRU batches",
+	 MONITOR_SET_MEMBER, MONITOR_LRU_BATCH_FLUSH_TOTAL_PAGE,
+	 MONITOR_LRU_BATCH_FLUSH_COUNT},
+
+	{"buffer_LRU_batch_flush_pages", "buffer",
+	 "Pages queued as an LRU batch",
+	 MONITOR_SET_MEMBER, MONITOR_LRU_BATCH_FLUSH_TOTAL_PAGE,
+	 MONITOR_LRU_BATCH_FLUSH_PAGES},
+
+	/* Cumulative counter for LRU batch pages flushed */
+	{"buffer_LRU_batch_evict_total_pages", "buffer",
+	 "Total pages evicted as part of LRU batches",
+	 MONITOR_SET_OWNER, MONITOR_LRU_BATCH_EVICT_COUNT,
+	 MONITOR_LRU_BATCH_EVICT_TOTAL_PAGE},
 
-	{"buffer_LRU_batches", "buffer",
+	{"buffer_LRU_batches_evict", "buffer",
 	 "Number of LRU batches",
-	 MONITOR_SET_MEMBER, MONITOR_LRU_BATCH_TOTAL_PAGE,
-	 MONITOR_LRU_BATCH_COUNT},
+	 MONITOR_SET_MEMBER, MONITOR_LRU_BATCH_EVICT_TOTAL_PAGE,
+	 MONITOR_LRU_BATCH_EVICT_COUNT},
 
-	{"buffer_LRU_batch_pages", "buffer",
+	{"buffer_LRU_batch_evict_pages", "buffer",
 	 "Pages queued as an LRU batch",
-	 MONITOR_SET_MEMBER, MONITOR_LRU_BATCH_TOTAL_PAGE,
-	 MONITOR_LRU_BATCH_PAGES},
+	 MONITOR_SET_MEMBER, MONITOR_LRU_BATCH_EVICT_TOTAL_PAGE,
+	 MONITOR_LRU_BATCH_EVICT_PAGES},
 
 	/* Cumulative counter for single page LRU scans */
 	{"buffer_LRU_single_flush_scanned", "buffer",