Applying InnoDB Plugin 1.0.5 snapshot, part 4

From revision r5703 to r5716

Detailed revision comments:

r5703 | marko | 2009-08-27 02:25:00 -0500 (Thu, 27 Aug 2009) | 41 lines
branches/zip: Replace the constant 3/8 ratio that controls the LRU_old
size with the settable global variable innodb_old_blocks_pct. The
minimum and maximum values are 5 and 95 per cent, respectively. The
default is 100*3/8, in line with the old behavior.

ut_time_ms(): New utility function, to return the current time in
milliseconds. TODO: Is there a more efficient timestamp function, such
as rdtsc divided by a power of two?

buf_LRU_old_threshold_ms: New variable, corresponding to
innodb_old_blocks_time. The value 0 is the default behaviour: no
timeout before making blocks 'new'.

bpage->accessed, bpage->LRU_position, buf_pool->ulint_clock: Remove.

bpage->access_time: New field, replacing bpage->accessed. Protected by
buf_pool_mutex instead of bpage->mutex. Updated when a page is created
or accessed the first time in the buffer pool.

buf_LRU_old_ratio, innobase_old_blocks_pct: New variables,
corresponding to innodb_old_blocks_pct

buf_LRU_old_ratio_update(), innobase_old_blocks_pct_update(): Update
functions for buf_LRU_old_ratio, innobase_old_blocks_pct.

buf_page_peek_if_too_old(): Compare ut_time_ms() to bpage->access_time
if buf_LRU_old_threshold_ms && bpage->old.  Else observe
buf_LRU_old_ratio and bpage->freed_page_clock.

buf_pool_t: Add n_pages_made_young, n_pages_not_made_young,
n_pages_made_young_old, n_pages_not_made_young, for statistics.

buf_print(): Display buf_pool->n_pages_made_young,
buf_pool->n_pages_not_made_young.  This function is only for crash
diagnostics.

buf_print_io(): Display buf_pool->LRU_old_len and quantities derived
from buf_pool->n_pages_made_young, buf_pool->n_pages_not_made_young.
This function is invoked by SHOW ENGINE INNODB STATUS.

rb://129 approved by Heikki Tuuri.  This addresses Bug #45015.
r5704 | marko | 2009-08-27 03:31:17 -0500 (Thu, 27 Aug 2009) | 32 lines
branches/zip: Fix a critical bug in fast index creation that could
corrupt the created indexes.

row_merge(): Make "half" an in/out parameter. Determine the offset of
half the output file. Copy the last blocks record-by-record instead of
block-by-block, so that the records can be counted. Check that the
input and output have matching n_rec.

row_merge_sort(): Do not assume that two blocks of size N are merged
into a block of size 2*N. The output block can be shorter than the
input if the last page of each input block is almost empty. Use an
accurate termination condition, based on the "half" computed by
row_merge().

row_merge_read(), row_merge_write(), row_merge_blocks(): Add debug output.

merge_file_t, row_merge_file_create(): Add n_rec, the number of records
in the merge file.

row_merge_read_clustered_index(): Update n_rec.

row_merge_blocks(): Update and check n_rec.

row_merge_blocks_copy(): New function, for copying the last blocks in
row_merge().  Update and check n_rec.

This bug was discovered with a user-supplied test case that creates an
index where the initial temporary file is 249 one-megabyte blocks and
the merged files become smaller. In the test, possible merge record
sizes are 10, 18, and 26 bytes.

rb://150 approved by Sunny Bains.  This addresses Issue #320.
r5705 | marko | 2009-08-27 06:56:24 -0500 (Thu, 27 Aug 2009) | 11 lines
branches/zip: dict_index_find_cols(): On column name lookup failure,
return DB_CORRUPTION (HA_ERR_CRASHED) instead of abnormally
terminating the server.  Also, disable the previously added diagnostic
output to the error log, because mysql-test-run does not like extra
output in the error log.  (Bug #44571)

dict_index_add_to_cache(): Handle errors from dict_index_find_cols().

mysql-test/innodb_bug44571.test: A test case for triggering the bug.

rb://135 approved by Sunny Bains.
r5706 | inaam | 2009-08-27 11:00:27 -0500 (Thu, 27 Aug 2009) | 20 lines
branches/zip rb://147

Done away with following two status variables:

innodb_buffer_pool_read_ahead_rnd
innodb_buffer_pool_read_ahead_seq

Introduced two new status variables:
innodb_buffer_pool_read_ahead = number of pages read as part of
readahead since server startup
innodb_buffer_pool_read_ahead_evicted = number of pages that are read
in as readahead but were evicted before ever being accessed since
server startup i.e.: a measure of how badly our readahead is
performing

SHOW INNODB STATUS will show two extra numbers in buffer pool section:
pages read ahead/sec and pages evicted without access/sec

Approved by: Marko

r5707 | inaam | 2009-08-27 11:20:35 -0500 (Thu, 27 Aug 2009) | 6 lines
branches/zip

Remove unused macros as we erased the random readahead code in r5703.
Also fixed some comments.


r5708 | inaam | 2009-08-27 17:43:32 -0500 (Thu, 27 Aug 2009) | 4 lines
branches/zip

Remove redundant TRUE : FALSE from the return statement

r5709 | inaam | 2009-08-28 01:22:46 -0500 (Fri, 28 Aug 2009) | 5 lines
branches/zip rb://152

Disable display of deprecated parameter innodb_file_io_threads in
'show variables'.

r5714 | marko | 2009-08-31 01:10:10 -0500 (Mon, 31 Aug 2009) | 5 lines
branches/zip: buf_chunk_not_freed(): Do not acquire block->mutex unless
block->page.state == BUF_BLOCK_FILE_PAGE.  Check that block->page.state
makes sense.

Approved by Sunny Bains over the IM.
r5716 | vasil | 2009-08-31 02:47:49 -0500 (Mon, 31 Aug 2009) | 9 lines
branches/zip:

Fix Bug#46718 InnoDB plugin incompatible with gcc 4.1 (at least: on PPC): "Undefined symbol"

by implementing our own check in plug.in instead of using the result from
the check from MySQL because it is insufficient.

Approved by:	Marko (rb://154)

mysql-test/suite/innodb/r/innodb_bug44571.result

+CREATE TABLE bug44571 (foo INT) ENGINE=InnoDB;
+ALTER TABLE bug44571 CHANGE foo bar INT;
+ALTER TABLE bug44571 ADD INDEX bug44571b (foo);
+ERROR 42000: Key column 'foo' doesn't exist in table
+ALTER TABLE bug44571 ADD INDEX bug44571b (bar);
+ERROR HY000: Incorrect key file for table 'bug44571'; try to repair it
+CREATE INDEX bug44571b ON bug44571 (bar);
+ERROR HY000: Incorrect key file for table 'bug44571'; try to repair it
+DROP TABLE bug44571;

mysql-test/suite/innodb/t/innodb_bug44571.test

+#
+# Bug#44571 InnoDB Plugin crashes on ADD INDEX
+# http://bugs.mysql.com/44571
+#
+-- source include/have_innodb.inc
+-- source suite/innodb/include/have_innodb_plugin.inc
+
+CREATE TABLE bug44571 (foo INT) ENGINE=InnoDB;
+ALTER TABLE bug44571 CHANGE foo bar INT;
+-- error ER_KEY_COLUMN_DOES_NOT_EXITS
+ALTER TABLE bug44571 ADD INDEX bug44571b (foo);
+# The following will fail, because the CHANGE foo bar was
+# not communicated to InnoDB.
+--error ER_NOT_KEYFILE
+ALTER TABLE bug44571 ADD INDEX bug44571b (bar);
+--error ER_NOT_KEYFILE
+CREATE INDEX bug44571b ON bug44571 (bar);
+DROP TABLE bug44571;

storage/innodb_plugin/ChangeLog

+2009-08-27	The InnoDB Team
+
+	* dict/dict0dict.c, include/dict0dict.h,
+	mysql-test/innodb_bug44571.result, mysql-test/innodb_bug44571.test:
+	Fix Bug#44571 InnoDB Plugin crashes on ADD INDEX
+
+2009-08-27	The InnoDB Team
+
+	* row/row0merge.c:
+	Fix a bug in the merge sort that can corrupt indexes in fast index
+	creation.  Add some consistency checks.  Check that the number of
+	records remains constant in every merge sort pass.
+
+2009-08-27	The InnoDB Team
+
+	* buf/buf0buf.c, buf/buf0lru.c, buf/buf0rea.c,
+	handler/ha_innodb.cc, include/buf0buf.h, include/buf0buf.ic,
+	include/buf0lru.h, include/ut0ut.h, ut/ut0ut.c:
+	Make it possible to tune the buffer pool LRU eviction policy to be
+	more resistant against index scans.  Introduce the settable global
+	variables innodb_old_blocks_pct and innodb_old_blocks_time for
+	controlling the buffer pool eviction policy.  The parameter
+	innodb_old_blocks_pct (5..95) controls the desired amount of "old"
+	blocks in the LRU list.  The default is 37, corresponding to the
+	old fixed ratio of 3/8.  Each time a block is accessed, it will be
+	moved to the "new" blocks if its first access was at least
+	innodb_old_blocks_time milliseconds ago (default 0, meaning every
+	block).  The idea is that in index scans, blocks will be accessed
+	a few times within innodb_old_blocks_time, and they will remain in
+	the "old" section of the LRU list.  Thus, when
+	innodb_old_blocks_time is nonzero, blocks retrieved for one-time
+	index scans will be more likely candidates for eviction than
+	blocks that are accessed in random patterns.
+
 2009-08-26	The InnoDB Team
 
 	* handler/ha_innodb.cc, os/os0file.c:

storage/innodb_plugin/btr/btr0sea.c

@@ -957,7 +957,7 @@ btr_search_guess_on_hash(
 	/* Increment the page get statistics though we did not really
 	fix the page: for user info only */
 
-	buf_pool->n_page_gets++;
+	buf_pool->stat.n_page_gets++;
 
 	return(TRUE);
 

storage/innodb_plugin/buf/buf0buf.c

@@ -837,16 +837,35 @@ buf_chunk_not_freed(
 	block = chunk->blocks;
 
 	for (i = chunk->size; i--; block++) {
+		ibool	ready;
+
+		switch (buf_block_get_state(block)) {
+		case BUF_BLOCK_ZIP_FREE:
+		case BUF_BLOCK_ZIP_PAGE:
+		case BUF_BLOCK_ZIP_DIRTY:
+			/* The uncompressed buffer pool should never
+			contain compressed block descriptors. */
+			ut_error;
+			break;
+		case BUF_BLOCK_NOT_USED:
+		case BUF_BLOCK_READY_FOR_USE:
+		case BUF_BLOCK_MEMORY:
+		case BUF_BLOCK_REMOVE_HASH:
+			/* Skip blocks that are not being used for
+			file pages. */
+			break;
+		case BUF_BLOCK_FILE_PAGE:
 			mutex_enter(&block->mutex);
+			ready = buf_flush_ready_for_replace(&block->page);
+			mutex_exit(&block->mutex);
 
-		if (buf_block_get_state(block) == BUF_BLOCK_FILE_PAGE
-		    && !buf_flush_ready_for_replace(&block->page)) {
+			if (!ready) {
 
-			mutex_exit(&block->mutex);
 				return(block);
 			}
 
-		mutex_exit(&block->mutex);
+			break;
+		}
 	}
 
 	return(NULL);
@@ -966,8 +985,6 @@ buf_pool_init(void)
 		buf_pool->no_flush[i] = os_event_create(NULL);
 	}
 
-	buf_pool->ulint_clock = 1;
-
 	/* 3. Initialize LRU fields
 	--------------------------- */
 	/* All fields are initialized by mem_zalloc(). */
@@ -1470,31 +1487,6 @@ buf_pool_resize(void)
 	buf_pool_page_hash_rebuild();
 }
 
-/********************************************************************//**
-Moves the block to the start of the LRU list if there is a danger
-that the block would drift out of the buffer pool. */
-UNIV_INLINE
-void
-buf_block_make_young(
-/*=================*/
-	buf_page_t*	bpage)	/*!< in: block to make younger */
-{
-	ut_ad(!buf_pool_mutex_own());
-
-	/* Note that we read freed_page_clock's without holding any mutex:
-	this is allowed since the result is used only in heuristics */
-
-	if (buf_page_peek_if_too_old(bpage)) {
-
-		buf_pool_mutex_enter();
-		/* There has been freeing activity in the LRU list:
-		best to move to the head of the LRU list */
-
-		buf_LRU_make_block_young(bpage);
-		buf_pool_mutex_exit();
-	}
-}
-
 /********************************************************************//**
 Moves a page to the start of the buffer pool LRU list. This high-level
 function can be used to prevent an important page from from slipping out of
@@ -1649,7 +1641,7 @@ buf_page_get_zip(
 #ifndef UNIV_LOG_DEBUG
 	ut_ad(!ibuf_inside());
 #endif
-	buf_pool->n_page_gets++;
+	buf_pool->stat.n_page_gets++;
 
 	for (;;) {
 		buf_pool_mutex_enter();
@@ -1713,13 +1705,15 @@ buf_page_get_zip(
 got_block:
 	must_read = buf_page_get_io_fix(bpage) == BUF_IO_READ;
 
-	buf_pool_mutex_exit();
+	if (buf_page_peek_if_too_old(bpage)) {
+		buf_LRU_make_block_young(bpage);
+	}
 
-	buf_page_set_accessed(bpage, TRUE);
+	buf_page_set_accessed(bpage);
 
-	mutex_exit(block_mutex);
+	buf_pool_mutex_exit();
 
-	buf_block_make_young(bpage);
+	mutex_exit(block_mutex);
 
 #ifdef UNIV_DEBUG_FILE_ACCESSES
 	ut_a(!bpage->file_page_was_freed);
@@ -2000,7 +1994,7 @@ buf_page_get_gen(
 	mtr_t*		mtr)	/*!< in: mini-transaction */
 {
 	buf_block_t*	block;
-	ibool		accessed;
+	unsigned	access_time;
 	ulint		fix_type;
 	ibool		must_read;
 
@@ -2016,7 +2010,7 @@ buf_page_get_gen(
 #ifndef UNIV_LOG_DEBUG
 	ut_ad(!ibuf_inside() || ibuf_page(space, zip_size, offset, NULL));
 #endif
-	buf_pool->n_page_gets++;
+	buf_pool->stat.n_page_gets++;
 loop:
 	block = guess;
 	buf_pool_mutex_enter();
@@ -2243,17 +2237,20 @@ buf_page_get_gen(
 	UNIV_MEM_ASSERT_RW(&block->page, sizeof block->page);
 
 	buf_block_buf_fix_inc(block, file, line);
-	buf_pool_mutex_exit();
+
+	mutex_exit(&block->mutex);
 
 	/* Check if this is the first access to the page */
 
-	accessed = buf_page_is_accessed(&block->page);
+	access_time = buf_page_is_accessed(&block->page);
 
-	buf_page_set_accessed(&block->page, TRUE);
+	if (buf_page_peek_if_too_old(&block->page)) {
+		buf_LRU_make_block_young(&block->page);
+	}
 
-	mutex_exit(&block->mutex);
+	buf_page_set_accessed(&block->page);
 
-	buf_block_make_young(&block->page);
+	buf_pool_mutex_exit();
 
 #ifdef UNIV_DEBUG_FILE_ACCESSES
 	ut_a(!block->page.file_page_was_freed);
@@ -2306,7 +2303,7 @@ buf_page_get_gen(
 
 	mtr_memo_push(mtr, block, fix_type);
 
-	if (!accessed) {
+	if (!access_time) {
 		/* In the case of a first access, try to apply linear
 		read-ahead */
 
@@ -2336,7 +2333,7 @@ buf_page_optimistic_get_func(
 	ulint		line,	/*!< in: line where called */
 	mtr_t*		mtr)	/*!< in: mini-transaction */
 {
-	ibool		accessed;
+	unsigned	access_time;
 	ibool		success;
 	ulint		fix_type;
 
@@ -2353,14 +2350,21 @@ buf_page_optimistic_get_func(
 	}
 
 	buf_block_buf_fix_inc(block, file, line);
-	accessed = buf_page_is_accessed(&block->page);
-	buf_page_set_accessed(&block->page, TRUE);
 
 	mutex_exit(&block->mutex);
 
-	buf_block_make_young(&block->page);
+	buf_pool_mutex_enter();
 
 	/* Check if this is the first access to the page */
+	access_time = buf_page_is_accessed(&block->page);
+
+	if (buf_page_peek_if_too_old(&block->page)) {
+		buf_LRU_make_block_young(&block->page);
+	}
+
+	buf_page_set_accessed(&block->page);
+
+	buf_pool_mutex_exit();
 
 	ut_ad(!ibuf_inside()
 	      || ibuf_page(buf_block_get_space(block),
@@ -2412,7 +2416,7 @@ buf_page_optimistic_get_func(
 #ifdef UNIV_DEBUG_FILE_ACCESSES
 	ut_a(block->page.file_page_was_freed == FALSE);
 #endif
-	if (UNIV_UNLIKELY(!accessed)) {
+	if (UNIV_UNLIKELY(!access_time)) {
 		/* In the case of a first access, try to apply linear
 		read-ahead */
 
@@ -2425,7 +2429,7 @@ buf_page_optimistic_get_func(
 	ut_a(ibuf_count_get(buf_block_get_space(block),
 			    buf_block_get_page_no(block)) == 0);
 #endif
-	buf_pool->n_page_gets++;
+	buf_pool->stat.n_page_gets++;
 
 	return(TRUE);
 }
@@ -2473,10 +2477,16 @@ buf_page_get_known_nowait(
 
 	mutex_exit(&block->mutex);
 
-	if (mode == BUF_MAKE_YOUNG) {
-		buf_block_make_young(&block->page);
+	buf_pool_mutex_enter();
+
+	if (mode == BUF_MAKE_YOUNG && buf_page_peek_if_too_old(&block->page)) {
+		buf_LRU_make_block_young(&block->page);
 	}
 
+	buf_page_set_accessed(&block->page);
+
+	buf_pool_mutex_exit();
+
 	ut_ad(!ibuf_inside() || (mode == BUF_KEEP_OLD));
 
 	if (rw_latch == RW_S_LATCH) {
@@ -2513,7 +2523,7 @@ buf_page_get_known_nowait(
 	     || (ibuf_count_get(buf_block_get_space(block),
 				buf_block_get_page_no(block)) == 0));
 #endif
-	buf_pool->n_page_gets++;
+	buf_pool->stat.n_page_gets++;
 
 	return(TRUE);
 }
@@ -2589,7 +2599,7 @@ buf_page_try_get_func(
 #endif /* UNIV_DEBUG_FILE_ACCESSES */
 	buf_block_dbg_add_level(block, SYNC_NO_ORDER_CHECK);
 
-	buf_pool->n_page_gets++;
+	buf_pool->stat.n_page_gets++;
 
 #ifdef UNIV_IBUF_COUNT_DEBUG
 	ut_a(ibuf_count_get(buf_block_get_space(block),
@@ -2608,10 +2618,10 @@ buf_page_init_low(
 	buf_page_t*	bpage)	/*!< in: block to init */
 {
 	bpage->flush_type = BUF_FLUSH_LRU;
-	bpage->accessed = FALSE;
 	bpage->io_fix = BUF_IO_NONE;
 	bpage->buf_fix_count = 0;
 	bpage->freed_page_clock = 0;
+	bpage->access_time = 0;
 	bpage->newest_modification = 0;
 	bpage->oldest_modification = 0;
 	HASH_INVALIDATE(bpage, hash);
@@ -2953,7 +2963,7 @@ buf_page_create(
 	buf_LRU_add_block(&block->page, FALSE);
 
 	buf_block_buf_fix_inc(block, __FILE__, __LINE__);
-	buf_pool->n_pages_created++;
+	buf_pool->stat.n_pages_created++;
 
 	if (zip_size) {
 		void*	data;
@@ -2990,12 +3000,12 @@ buf_page_create(
 		rw_lock_x_unlock(&block->lock);
 	}
 
+	buf_page_set_accessed(&block->page);
+
 	buf_pool_mutex_exit();
 
 	mtr_memo_push(mtr, block, MTR_MEMO_BUF_FIX);
 
-	buf_page_set_accessed(&block->page, TRUE);
-
 	mutex_exit(&block->mutex);
 
 	/* Delete possible entries for the page from the insert buffer:
@@ -3201,7 +3211,7 @@ buf_page_io_complete(
 
 		ut_ad(buf_pool->n_pend_reads > 0);
 		buf_pool->n_pend_reads--;
-		buf_pool->n_pages_read++;
+		buf_pool->stat.n_pages_read++;
 
 		if (uncompressed) {
 			rw_lock_x_unlock_gen(&((buf_block_t*) bpage)->lock,
@@ -3221,7 +3231,7 @@ buf_page_io_complete(
 					     BUF_IO_WRITE);
 		}
 
-		buf_pool->n_pages_written++;
+		buf_pool->stat.n_pages_written++;
 
 		break;
 
@@ -3528,6 +3538,7 @@ buf_print(void)
 		"n pending decompressions %lu\n"
 		"n pending reads %lu\n"
 		"n pending flush LRU %lu list %lu single page %lu\n"
+		"pages made young %lu, not young %lu\n"
 		"pages read %lu, created %lu, written %lu\n",
 		(ulong) size,
 		(ulong) UT_LIST_GET_LEN(buf_pool->LRU),
@@ -3538,8 +3549,11 @@ buf_print(void)
 		(ulong) buf_pool->n_flush[BUF_FLUSH_LRU],
 		(ulong) buf_pool->n_flush[BUF_FLUSH_LIST],
 		(ulong) buf_pool->n_flush[BUF_FLUSH_SINGLE_PAGE],
-		(ulong) buf_pool->n_pages_read, buf_pool->n_pages_created,
-		(ulong) buf_pool->n_pages_written);
+		(ulong) buf_pool->stat.n_pages_made_young,
+		(ulong) buf_pool->stat.n_pages_not_made_young,
+		(ulong) buf_pool->stat.n_pages_read,
+		(ulong) buf_pool->stat.n_pages_created,
+		(ulong) buf_pool->stat.n_pages_written);
 
 	/* Count the number of blocks belonging to each index in the buffer */
 
@@ -3744,10 +3758,9 @@ buf_print_io(
 {
 	time_t	current_time;
 	double	time_elapsed;
-	ulint	size;
+	ulint	n_gets_diff;
 
 	ut_ad(buf_pool);
-	size = buf_pool->curr_size;
 
 	buf_pool_mutex_enter();
 
@@ -3755,12 +3768,14 @@ buf_print_io(
 		"Buffer pool size   %lu\n"
 		"Free buffers       %lu\n"
 		"Database pages     %lu\n"
+		"Old database pages %lu\n"
 		"Modified db pages  %lu\n"
 		"Pending reads %lu\n"
 		"Pending writes: LRU %lu, flush list %lu, single page %lu\n",
-		(ulong) size,
+		(ulong) buf_pool->curr_size,
 		(ulong) UT_LIST_GET_LEN(buf_pool->free),
 		(ulong) UT_LIST_GET_LEN(buf_pool->LRU),
+		(ulong) buf_pool->LRU_old_len,
 		(ulong) UT_LIST_GET_LEN(buf_pool->flush_list),
 		(ulong) buf_pool->n_pend_reads,
 		(ulong) buf_pool->n_flush[BUF_FLUSH_LRU]
@@ -3772,37 +3787,66 @@ buf_print_io(
 	current_time = time(NULL);
 	time_elapsed = 0.001 + difftime(current_time,
 					buf_pool->last_printout_time);
-	buf_pool->last_printout_time = current_time;
 
 	fprintf(file,
+		"Pages made young %lu, not young %lu\n"
+		"%.2f youngs/s, %.2f non-youngs/s\n"
 		"Pages read %lu, created %lu, written %lu\n"
 		"%.2f reads/s, %.2f creates/s, %.2f writes/s\n",
-		(ulong) buf_pool->n_pages_read,
-		(ulong) buf_pool->n_pages_created,
-		(ulong) buf_pool->n_pages_written,
-		(buf_pool->n_pages_read - buf_pool->n_pages_read_old)
+		(ulong) buf_pool->stat.n_pages_made_young,
+		(ulong) buf_pool->stat.n_pages_not_made_young,
+		(buf_pool->stat.n_pages_made_young
+		 - buf_pool->old_stat.n_pages_made_young)
+		/ time_elapsed,
+		(buf_pool->stat.n_pages_not_made_young
+		 - buf_pool->old_stat.n_pages_not_made_young)
+		/ time_elapsed,
+		(ulong) buf_pool->stat.n_pages_read,
+		(ulong) buf_pool->stat.n_pages_created,
+		(ulong) buf_pool->stat.n_pages_written,
+		(buf_pool->stat.n_pages_read
+		 - buf_pool->old_stat.n_pages_read)
 		/ time_elapsed,
-		(buf_pool->n_pages_created - buf_pool->n_pages_created_old)
+		(buf_pool->stat.n_pages_created
+		 - buf_pool->old_stat.n_pages_created)
 		/ time_elapsed,
-		(buf_pool->n_pages_written - buf_pool->n_pages_written_old)
+		(buf_pool->stat.n_pages_written
+		 - buf_pool->old_stat.n_pages_written)
 		/ time_elapsed);
 
-	if (buf_pool->n_page_gets > buf_pool->n_page_gets_old) {
-		fprintf(file, "Buffer pool hit rate %lu / 1000\n",
+	n_gets_diff = buf_pool->stat.n_page_gets - buf_pool->old_stat.n_page_gets;
+
+	if (n_gets_diff) {
+		fprintf(file,
+			"Buffer pool hit rate %lu / 1000,"
+			" young-making rate %lu / 1000 not %lu / 1000\n",
 			(ulong)
-			(1000 - ((1000 * (buf_pool->n_pages_read
-					  - buf_pool->n_pages_read_old))
-				 / (buf_pool->n_page_gets
-				    - buf_pool->n_page_gets_old))));
+			(1000 - ((1000 * (buf_pool->stat.n_pages_read
+					  - buf_pool->old_stat.n_pages_read))
+				 / (buf_pool->stat.n_page_gets
+				    - buf_pool->old_stat.n_page_gets))),
+			(ulong)
+			(1000 * (buf_pool->stat.n_pages_made_young
+				 - buf_pool->old_stat.n_pages_made_young)
+			 / n_gets_diff),
+			(ulong)
+			(1000 * (buf_pool->stat.n_pages_not_made_young
+				 - buf_pool->old_stat.n_pages_not_made_young)
+			 / n_gets_diff));
 	} else {
 		fputs("No buffer pool page gets since the last printout\n",
 		      file);
 	}
 
-	buf_pool->n_page_gets_old = buf_pool->n_page_gets;
-	buf_pool->n_pages_read_old = buf_pool->n_pages_read;
-	buf_pool->n_pages_created_old = buf_pool->n_pages_created;
-	buf_pool->n_pages_written_old = buf_pool->n_pages_written;
+	/* Statistics about read ahead algorithm */
+	fprintf(file, "Pages read ahead %.2f/s,"
+		" evicted without access %.2f/s\n",
+		(buf_pool->stat.n_ra_pages_read
+		- buf_pool->old_stat.n_ra_pages_read)
+		/ time_elapsed,
+		(buf_pool->stat.n_ra_pages_evicted
+		- buf_pool->old_stat.n_ra_pages_evicted)
+		/ time_elapsed);
 
 	/* Print some values to help us with visualizing what is
 	happening with LRU eviction. */
@@ -3814,6 +3858,7 @@ buf_print_io(
 		buf_LRU_stat_sum.io, buf_LRU_stat_cur.io,
 		buf_LRU_stat_sum.unzip, buf_LRU_stat_cur.unzip);
 
+	buf_refresh_io_stats();
 	buf_pool_mutex_exit();
 }
 
@@ -3825,10 +3870,7 @@ buf_refresh_io_stats(void)
 /*======================*/
 {
 	buf_pool->last_printout_time = time(NULL);
-	buf_pool->n_page_gets_old = buf_pool->n_page_gets;
-	buf_pool->n_pages_read_old = buf_pool->n_pages_read;
-	buf_pool->n_pages_created_old = buf_pool->n_pages_created;
-	buf_pool->n_pages_written_old = buf_pool->n_pages_written;
+	buf_pool->old_stat = buf_pool->stat;
 }
 
 /*********************************************************************//**

storage/innodb_plugin/buf/buf0lru.c

@@ -49,14 +49,23 @@ Created 11/5/1995 Heikki Tuuri
 #include "log0recv.h"
 #include "srv0srv.h"
 
-/** The number of blocks from the LRU_old pointer onward, including the block
-pointed to, must be 3/8 of the whole LRU list length, except that the
-tolerance defined below is allowed. Note that the tolerance must be small
-enough such that for even the BUF_LRU_OLD_MIN_LEN long LRU list, the
-LRU_old pointer is not allowed to point to either end of the LRU list. */
+/** The number of blocks from the LRU_old pointer onward, including
+the block pointed to, must be buf_LRU_old_ratio/BUF_LRU_OLD_RATIO_DIV
+of the whole LRU list length, except that the tolerance defined below
+is allowed. Note that the tolerance must be small enough such that for
+even the BUF_LRU_OLD_MIN_LEN long LRU list, the LRU_old pointer is not
+allowed to point to either end of the LRU list. */
 
 #define BUF_LRU_OLD_TOLERANCE	20
 
+/** The minimum amount of non-old blocks when the LRU_old list exists
+(that is, when there are more than BUF_LRU_OLD_MIN_LEN blocks).
+@see buf_LRU_old_adjust_len */
+#define BUF_LRU_NON_OLD_MIN_LEN	5
+#if BUF_LRU_NON_OLD_MIN_LEN >= BUF_LRU_OLD_MIN_LEN
+# error "BUF_LRU_NON_OLD_MIN_LEN >= BUF_LRU_OLD_MIN_LEN"
+#endif
+
 /** The whole LRU list length is divided by this number to determine an
 initial segment in buf_LRU_get_recent_limit */
 
@@ -107,6 +116,15 @@ UNIV_INTERN buf_LRU_stat_t	buf_LRU_stat_sum;
 
 /* @} */
 
+/** @name Heuristics for detecting index scan @{ */
+/** Reserve this much/BUF_LRU_OLD_RATIO_DIV of the buffer pool for
+"old" blocks.  Protected by buf_pool_mutex. */
+UNIV_INTERN uint	buf_LRU_old_ratio;
+/** Move blocks to "new" LRU list only if the first access was at
+least this many milliseconds ago.  Not protected by any mutex or latch. */
+UNIV_INTERN uint	buf_LRU_old_threshold_ms;
+/* @} */
+
 /******************************************************************//**
 Takes a block out of the LRU list and page hash table.
 If the block is compressed-only (BUF_BLOCK_ZIP_PAGE),
@@ -428,42 +446,6 @@ buf_LRU_invalidate_tablespace(
 	}
 }
 
-/******************************************************************//**
-Gets the minimum LRU_position field for the blocks in an initial segment
-(determined by BUF_LRU_INITIAL_RATIO) of the LRU list. The limit is not
-guaranteed to be precise, because the ulint_clock may wrap around.
-@return	the limit; zero if could not determine it */
-UNIV_INTERN
-ulint
-buf_LRU_get_recent_limit(void)
-/*==========================*/
-{
-	const buf_page_t*	bpage;
-	ulint			len;
-	ulint			limit;
-
-	buf_pool_mutex_enter();
-
-	len = UT_LIST_GET_LEN(buf_pool->LRU);
-
-	if (len < BUF_LRU_OLD_MIN_LEN) {
-		/* The LRU list is too short to do read-ahead */
-
-		buf_pool_mutex_exit();
-
-		return(0);
-	}
-
-	bpage = UT_LIST_GET_FIRST(buf_pool->LRU);
-
-	limit = buf_page_get_LRU_position(bpage);
-	len /= BUF_LRU_INITIAL_RATIO;
-
-	buf_pool_mutex_exit();
-
-	return(limit > len ? (limit - len) : 0);
-}
-
 /********************************************************************//**
 Insert a compressed block into buf_pool->zip_clean in the LRU order. */
 UNIV_INTERN
@@ -594,6 +576,7 @@ buf_LRU_free_from_common_LRU_list(
 	     bpage = UT_LIST_GET_PREV(LRU, bpage), distance--) {
 
 		enum buf_lru_free_block_status	freed;
+		unsigned			accessed;
 		mutex_t*			block_mutex
 			= buf_page_get_mutex(bpage);
 
@@ -601,11 +584,18 @@ buf_LRU_free_from_common_LRU_list(
 		ut_ad(bpage->in_LRU_list);
 
 		mutex_enter(block_mutex);
+		accessed = buf_page_is_accessed(bpage);
 		freed = buf_LRU_free_block(bpage, TRUE, NULL);
 		mutex_exit(block_mutex);
 
 		switch (freed) {
 		case BUF_LRU_FREED:
+			/* Keep track of pages that are evicted without
+			ever being accessed. This gives us a measure of
+			the effectiveness of readahead */
+			if (!accessed) {
+				++buf_pool->stat.n_ra_pages_evicted;
+			}
 			return(TRUE);
 
 		case BUF_LRU_NOT_FREED:
@@ -953,8 +943,10 @@ buf_LRU_old_adjust_len(void)
 
 	ut_a(buf_pool->LRU_old);
 	ut_ad(buf_pool_mutex_own());
-#if 3 * (BUF_LRU_OLD_MIN_LEN / 8) <= BUF_LRU_OLD_TOLERANCE + 5
-# error "3 * (BUF_LRU_OLD_MIN_LEN / 8) <= BUF_LRU_OLD_TOLERANCE + 5"
+	ut_ad(buf_LRU_old_ratio >= BUF_LRU_OLD_RATIO_MIN);
+	ut_ad(buf_LRU_old_ratio <= BUF_LRU_OLD_RATIO_MAX);
+#if BUF_LRU_OLD_RATIO_MIN * BUF_LRU_OLD_MIN_LEN <= BUF_LRU_OLD_RATIO_DIV * (BUF_LRU_OLD_TOLERANCE + 5)
+# error "BUF_LRU_OLD_RATIO_MIN * BUF_LRU_OLD_MIN_LEN <= BUF_LRU_OLD_RATIO_DIV * (BUF_LRU_OLD_TOLERANCE + 5)"
 #endif
 #ifdef UNIV_LRU_DEBUG
 	/* buf_pool->LRU_old must be the first item in the LRU list
@@ -966,34 +958,39 @@ buf_LRU_old_adjust_len(void)
 	     || UT_LIST_GET_NEXT(LRU, buf_pool->LRU_old)->old);
 #endif /* UNIV_LRU_DEBUG */
 
-	for (;;) {
 	old_len = buf_pool->LRU_old_len;
-		new_len = 3 * (UT_LIST_GET_LEN(buf_pool->LRU) / 8);
+	new_len = ut_min(UT_LIST_GET_LEN(buf_pool->LRU)
+			 * buf_LRU_old_ratio / BUF_LRU_OLD_RATIO_DIV,
+			 UT_LIST_GET_LEN(buf_pool->LRU)
+			 - (BUF_LRU_OLD_TOLERANCE
+			    + BUF_LRU_NON_OLD_MIN_LEN));
 
-		ut_ad(buf_pool->LRU_old->in_LRU_list);
-		ut_a(buf_pool->LRU_old);
+	for (;;) {
+		buf_page_t*	LRU_old = buf_pool->LRU_old;
+
+		ut_a(LRU_old);
+		ut_ad(LRU_old->in_LRU_list);
 #ifdef UNIV_LRU_DEBUG
-		ut_a(buf_pool->LRU_old->old);
+		ut_a(LRU_old->old);
 #endif /* UNIV_LRU_DEBUG */
 
 		/* Update the LRU_old pointer if necessary */
 
-		if (old_len < new_len - BUF_LRU_OLD_TOLERANCE) {
+		if (old_len + BUF_LRU_OLD_TOLERANCE < new_len) {
 
-			buf_pool->LRU_old = UT_LIST_GET_PREV(
-				LRU, buf_pool->LRU_old);
+			buf_pool->LRU_old = LRU_old = UT_LIST_GET_PREV(
+				LRU, LRU_old);
 #ifdef UNIV_LRU_DEBUG
-			ut_a(!buf_pool->LRU_old->old);
+			ut_a(!LRU_old->old);
 #endif /* UNIV_LRU_DEBUG */
-			buf_page_set_old(buf_pool->LRU_old, TRUE);
-			buf_pool->LRU_old_len++;
+			buf_page_set_old(LRU_old, TRUE);
+			old_len = ++buf_pool->LRU_old_len;
 
 		} else if (old_len > new_len + BUF_LRU_OLD_TOLERANCE) {
 
-			buf_page_set_old(buf_pool->LRU_old, FALSE);
-			buf_pool->LRU_old = UT_LIST_GET_NEXT(
-				LRU, buf_pool->LRU_old);
-			buf_pool->LRU_old_len--;
+			buf_page_set_old(LRU_old, FALSE);
+			buf_pool->LRU_old = UT_LIST_GET_NEXT(LRU, LRU_old);
+			old_len = --buf_pool->LRU_old_len;
 		} else {
 			return;
 		}
@@ -1021,6 +1018,7 @@ buf_LRU_old_init(void)
 
 	while (bpage != NULL) {
 		ut_ad(bpage->in_LRU_list);
+		ut_ad(buf_page_in_file(bpage));
 		buf_page_set_old(bpage, TRUE);
 		bpage = UT_LIST_GET_NEXT(LRU, bpage);
 	}
@@ -1075,16 +1073,19 @@ buf_LRU_remove_block(
 
 	if (UNIV_UNLIKELY(bpage == buf_pool->LRU_old)) {
 
-		/* Below: the previous block is guaranteed to exist, because
-		the LRU_old pointer is only allowed to differ by the
-		tolerance value from strict 3/8 of the LRU list length. */
+		/* Below: the previous block is guaranteed to exist,
+		because the LRU_old pointer is only allowed to differ
+		by BUF_LRU_OLD_TOLERANCE from strict
+		buf_LRU_old_ratio/BUF_LRU_OLD_RATIO_DIV of the LRU
+		list length. */
+		buf_page_t*	prev_bpage = UT_LIST_GET_PREV(LRU, bpage);
 
-		buf_pool->LRU_old = UT_LIST_GET_PREV(LRU, bpage);
-		ut_a(buf_pool->LRU_old);
+		ut_a(prev_bpage);
 #ifdef UNIV_LRU_DEBUG
-		ut_a(!buf_pool->LRU_old->old);
+		ut_a(!prev_bpage->old);
 #endif /* UNIV_LRU_DEBUG */
-		buf_page_set_old(buf_pool->LRU_old, TRUE);
+		buf_pool->LRU_old = prev_bpage;
+		buf_page_set_old(prev_bpage, TRUE);
 
 		buf_pool->LRU_old_len++;
 	}
@@ -1149,39 +1150,25 @@ buf_LRU_add_block_to_end_low(
 /*=========================*/
 	buf_page_t*	bpage)	/*!< in: control block */
 {
-	buf_page_t*	last_bpage;
-
 	ut_ad(buf_pool);
 	ut_ad(bpage);
 	ut_ad(buf_pool_mutex_own());
 
 	ut_a(buf_page_in_file(bpage));
 
-	last_bpage = UT_LIST_GET_LAST(buf_pool->LRU);
-
-	if (last_bpage) {
-		bpage->LRU_position = last_bpage->LRU_position;
-	} else {
-		bpage->LRU_position = buf_pool_clock_tic();
-	}
-
 	ut_ad(!bpage->in_LRU_list);
 	UT_LIST_ADD_LAST(LRU, buf_pool->LRU, bpage);
 	ut_d(bpage->in_LRU_list = TRUE);
 
 	buf_page_set_old(bpage, TRUE);
 
-	if (UT_LIST_GET_LEN(buf_pool->LRU) >= BUF_LRU_OLD_MIN_LEN) {
-
-		buf_pool->LRU_old_len++;
-	}
-
 	if (UT_LIST_GET_LEN(buf_pool->LRU) > BUF_LRU_OLD_MIN_LEN) {
 
 		ut_ad(buf_pool->LRU_old);
 
 		/* Adjust the length of the old block list if necessary */
 
+		buf_pool->LRU_old_len++;
 		buf_LRU_old_adjust_len();
 
 	} else if (UT_LIST_GET_LEN(buf_pool->LRU) == BUF_LRU_OLD_MIN_LEN) {
@@ -1189,6 +1176,7 @@ buf_LRU_add_block_to_end_low(
 		/* The LRU list is now long enough for LRU_old to become
 		defined: init it */
 
+		buf_pool->LRU_old_len++;
 		buf_LRU_old_init();
 	}
 
@@ -1222,7 +1210,6 @@ buf_LRU_add_block_low(
 
 		UT_LIST_ADD_FIRST(LRU, buf_pool->LRU, bpage);
 
-		bpage->LRU_position = buf_pool_clock_tic();
 		bpage->freed_page_clock = buf_pool->freed_page_clock;
 	} else {
 #ifdef UNIV_LRU_DEBUG
@@ -1237,11 +1224,6 @@ buf_LRU_add_block_low(
 		UT_LIST_INSERT_AFTER(LRU, buf_pool->LRU, buf_pool->LRU_old,
 				     bpage);
 		buf_pool->LRU_old_len++;
-
-		/* We copy the LRU position field of the previous block
-		to the new block */
-
-		bpage->LRU_position = (buf_pool->LRU_old)->LRU_position;
 	}
 
 	ut_d(bpage->in_LRU_list = TRUE);
@@ -1295,6 +1277,9 @@ buf_LRU_make_block_young(
 /*=====================*/
 	buf_page_t*	bpage)	/*!< in: control block */
 {
+	ut_ad(buf_pool_mutex_own());
+	buf_pool->stat.n_pages_made_young++;
+
 	buf_LRU_remove_block(bpage);
 	buf_LRU_add_block_low(bpage, FALSE);
 }
@@ -1847,6 +1832,48 @@ buf_LRU_block_free_hashed_page(
 	buf_LRU_block_free_non_file_page(block);
 }
 
+/**********************************************************************//**
+Updates buf_LRU_old_ratio.
+@return	updated old_pct */
+UNIV_INTERN
+uint
+buf_LRU_old_ratio_update(
+/*=====================*/
+	uint	old_pct,/*!< in: Reserve this percentage of
+			the buffer pool for "old" blocks. */
+	ibool	adjust)	/*!< in: TRUE=adjust the LRU list;
+			FALSE=just assign buf_LRU_old_ratio
+			during the initialization of InnoDB */
+{
+	uint	ratio;
+
+	ratio = old_pct * BUF_LRU_OLD_RATIO_DIV / 100;
+	if (ratio < BUF_LRU_OLD_RATIO_MIN) {
+		ratio = BUF_LRU_OLD_RATIO_MIN;
+	} else if (ratio > BUF_LRU_OLD_RATIO_MAX) {
+		ratio = BUF_LRU_OLD_RATIO_MAX;
+	}
+
+	if (adjust) {
+		buf_pool_mutex_enter();
+
+		if (ratio != buf_LRU_old_ratio) {
+			buf_LRU_old_ratio = ratio;
+
+			if (UT_LIST_GET_LEN(buf_pool->LRU)
+			    >= BUF_LRU_OLD_MIN_LEN) {
+				buf_LRU_old_adjust_len();
+			}
+		}
+
+		buf_pool_mutex_exit();
+	} else {
+		buf_LRU_old_ratio = ratio;
+	}
+
+	return(ratio * 100 / BUF_LRU_OLD_RATIO_DIV);
+}
+
 /********************************************************************//**
 Update the historical stats that we are collecting for LRU eviction
 policy at the end of each interval. */
@@ -1896,7 +1923,6 @@ buf_LRU_validate(void)
 	buf_block_t*	block;
 	ulint		old_len;
 	ulint		new_len;
-	ulint		LRU_pos;
 
 	ut_ad(buf_pool);
 	buf_pool_mutex_enter();
@@ -1905,7 +1931,11 @@ buf_LRU_validate(void)
 
 		ut_a(buf_pool->LRU_old);
 		old_len = buf_pool->LRU_old_len;
-		new_len = 3 * (UT_LIST_GET_LEN(buf_pool->LRU) / 8);
+		new_len = ut_min(UT_LIST_GET_LEN(buf_pool->LRU)
+				 * buf_LRU_old_ratio / BUF_LRU_OLD_RATIO_DIV,
+				 UT_LIST_GET_LEN(buf_pool->LRU)
+				 - (BUF_LRU_OLD_TOLERANCE
+				    + BUF_LRU_NON_OLD_MIN_LEN));
 		ut_a(old_len >= new_len - BUF_LRU_OLD_TOLERANCE);
 		ut_a(old_len <= new_len + BUF_LRU_OLD_TOLERANCE);
 	}
@@ -1943,16 +1973,7 @@ buf_LRU_validate(void)
 			ut_a(buf_pool->LRU_old == bpage);
 		}
 
-		LRU_pos	= buf_page_get_LRU_position(bpage);
-
 		bpage = UT_LIST_GET_NEXT(LRU, bpage);
-
-		if (bpage) {
-			/* If the following assert fails, it may
-			not be an error: just the buf_pool clock
-			has wrapped around */
-			ut_a(LRU_pos >= buf_page_get_LRU_position(bpage));
-		}
 	}
 
 	if (buf_pool->LRU_old) {
@@ -2000,9 +2021,6 @@ buf_LRU_print(void)
 	ut_ad(buf_pool);
 	buf_pool_mutex_enter();
 
-	fprintf(stderr, "Pool ulint clock %lu\n",
-		(ulong) buf_pool->ulint_clock);
-
 	bpage = UT_LIST_GET_FIRST(buf_pool->LRU);
 
 	while (bpage != NULL) {
@@ -2033,18 +2051,16 @@ buf_LRU_print(void)
 			const byte*	frame;
 		case BUF_BLOCK_FILE_PAGE:
 			frame = buf_block_get_frame((buf_block_t*) bpage);
-			fprintf(stderr, "\nLRU pos %lu type %lu"
+			fprintf(stderr, "\ntype %lu"
 				" index id %lu\n",
-				(ulong) buf_page_get_LRU_position(bpage),
 				(ulong) fil_page_get_type(frame),
 				(ulong) ut_dulint_get_low(
 					btr_page_get_index_id(frame)));
 			break;
 		case BUF_BLOCK_ZIP_PAGE:
 			frame = bpage->zip.data;
-			fprintf(stderr, "\nLRU pos %lu type %lu size %lu"
+			fprintf(stderr, "\ntype %lu size %lu"
 				" index id %lu\n",
-				(ulong) buf_page_get_LRU_position(bpage),
 				(ulong) fil_page_get_type(frame),
 				(ulong) buf_page_get_zip_size(bpage),
 				(ulong) ut_dulint_get_low(
@@ -2052,8 +2068,7 @@ buf_LRU_print(void)
 			break;
 
 		default:
-			fprintf(stderr, "\nLRU pos %lu !state %lu!\n",
-				(ulong) buf_page_get_LRU_position(bpage),
+			fprintf(stderr, "\n!state %lu!\n",
 				(ulong) buf_page_get_state(bpage));
 			break;
 		}

storage/innodb_plugin/buf/buf0rea.c

@@ -38,14 +38,6 @@ Created 11/5/1995 Heikki Tuuri
 #include "srv0start.h"
 #include "srv0srv.h"
 
-/** The size in blocks of the area where the random read-ahead algorithm counts
-the accessed pages when deciding whether to read-ahead */
-#define	BUF_READ_AHEAD_RANDOM_AREA	BUF_READ_AHEAD_AREA
-
-/** There must be at least this many pages in buf_pool in the area to start
-a random read-ahead */
-#define BUF_READ_AHEAD_RANDOM_THRESHOLD	(1 + BUF_READ_AHEAD_RANDOM_AREA / 2)
-
 /** The linear read-ahead area size */
 #define	BUF_READ_AHEAD_LINEAR_AREA	BUF_READ_AHEAD_AREA
 
@@ -62,7 +54,8 @@ flag is cleared and the x-lock released by an i/o-handler thread.
 @return 1 if a read request was queued, 0 if the page already resided
 in buf_pool, or if the page is in the doublewrite buffer blocks in
 which case it is never read into the pool, or if the tablespace does
-not exist or is being dropped */
+not exist or is being dropped 
+@return 1 if read request is issued. 0 if it is not */
 static
 ulint
 buf_read_page_low(
@@ -164,175 +157,14 @@ buf_read_page_low(
 	return(1);
 }
 
-/********************************************************************//**
-Applies a random read-ahead in buf_pool if there are at least a threshold
-value of accessed pages from the random read-ahead area. Does not read any
-page, not even the one at the position (space, offset), if the read-ahead
-mechanism is not activated. NOTE 1: the calling thread may own latches on
-pages: to avoid deadlocks this function must be written such that it cannot
-end up waiting for these latches! NOTE 2: the calling thread must want
-access to the page given: this rule is set to prevent unintended read-aheads
-performed by ibuf routines, a situation which could result in a deadlock if
-the OS does not support asynchronous i/o.
-@return number of page read requests issued; NOTE that if we read ibuf
-pages, it may happen that the page at the given page number does not
-get read even if we return a positive value! */
-static
-ulint
-buf_read_ahead_random(
-/*==================*/
-	ulint	space,	/*!< in: space id */
-	ulint	zip_size,/*!< in: compressed page size in bytes, or 0 */
-	ulint	offset)	/*!< in: page number of a page which the current thread
-			wants to access */
-{
-	ib_int64_t	tablespace_version;
-	ulint		recent_blocks	= 0;
-	ulint		count;
-	ulint		LRU_recent_limit;
-	ulint		ibuf_mode;
-	ulint		low, high;
-	ulint		err;
-	ulint		i;
-	ulint		buf_read_ahead_random_area;
-
-	/* We have currently disabled random readahead */
-	return(0);
-
-	if (srv_startup_is_before_trx_rollback_phase) {
-		/* No read-ahead to avoid thread deadlocks */
-		return(0);
-	}
-
-	if (ibuf_bitmap_page(zip_size, offset)
-	    || trx_sys_hdr_page(space, offset)) {
-
-		/* If it is an ibuf bitmap page or trx sys hdr, we do
-		no read-ahead, as that could break the ibuf page access
-		order */
-
-		return(0);
-	}
-
-	/* Remember the tablespace version before we ask te tablespace size
-	below: if DISCARD + IMPORT changes the actual .ibd file meanwhile, we
-	do not try to read outside the bounds of the tablespace! */
-
-	tablespace_version = fil_space_get_version(space);
-
-	buf_read_ahead_random_area = BUF_READ_AHEAD_RANDOM_AREA;
-
-	low  = (offset / buf_read_ahead_random_area)
-		* buf_read_ahead_random_area;
-	high = (offset / buf_read_ahead_random_area + 1)
-		* buf_read_ahead_random_area;
-	if (high > fil_space_get_size(space)) {
-
-		high = fil_space_get_size(space);
-	}
-
-	/* Get the minimum LRU_position field value for an initial segment
-	of the LRU list, to determine which blocks have recently been added
-	to the start of the list. */
-
-	LRU_recent_limit = buf_LRU_get_recent_limit();
-
-	buf_pool_mutex_enter();
-
-	if (buf_pool->n_pend_reads
-	    > buf_pool->curr_size / BUF_READ_AHEAD_PEND_LIMIT) {
-		buf_pool_mutex_exit();
-
-		return(0);
-	}
-
-	/* Count how many blocks in the area have been recently accessed,
-	that is, reside near the start of the LRU list. */
-
-	for (i = low; i < high; i++) {
-		const buf_page_t*	bpage = buf_page_hash_get(space, i);
-
-		if (bpage
-		    && buf_page_is_accessed(bpage)
-		    && (buf_page_get_LRU_position(bpage) > LRU_recent_limit)) {
-
-			recent_blocks++;
-
-			if (recent_blocks >= BUF_READ_AHEAD_RANDOM_THRESHOLD) {
-
-				buf_pool_mutex_exit();
-				goto read_ahead;
-			}
-		}
-	}
-
-	buf_pool_mutex_exit();
-	/* Do nothing */
-	return(0);
-
-read_ahead:
-	/* Read all the suitable blocks within the area */
-
-	if (ibuf_inside()) {
-		ibuf_mode = BUF_READ_IBUF_PAGES_ONLY;
-	} else {
-		ibuf_mode = BUF_READ_ANY_PAGE;
-	}
-
-	count = 0;
-
-	for (i = low; i < high; i++) {
-		/* It is only sensible to do read-ahead in the non-sync aio
-		mode: hence FALSE as the first parameter */
-
-		if (!ibuf_bitmap_page(zip_size, i)) {
-			count += buf_read_page_low(
-				&err, FALSE,
-				ibuf_mode | OS_AIO_SIMULATED_WAKE_LATER,
-				space, zip_size, FALSE,
-				tablespace_version, i);
-			if (err == DB_TABLESPACE_DELETED) {
-				ut_print_timestamp(stderr);
-				fprintf(stderr,
-					"  InnoDB: Warning: in random"
-					" readahead trying to access\n"
-					"InnoDB: tablespace %lu page %lu,\n"
-					"InnoDB: but the tablespace does not"
-					" exist or is just being dropped.\n",
-					(ulong) space, (ulong) i);
-			}
-		}
-	}
-
-	/* In simulated aio we wake the aio handler threads only after
-	queuing all aio requests, in native aio the following call does
-	nothing: */
-
-	os_aio_simulated_wake_handler_threads();
-
-#ifdef UNIV_DEBUG
-	if (buf_debug_prints && (count > 0)) {
-		fprintf(stderr,
-			"Random read-ahead space %lu offset %lu pages %lu\n",
-			(ulong) space, (ulong) offset,
-			(ulong) count);
-	}
-#endif /* UNIV_DEBUG */
-
-	++srv_read_ahead_rnd;
-	return(count);
-}
-
 /********************************************************************//**
 High-level function which reads a page asynchronously from a file to the
 buffer buf_pool if it is not already there. Sets the io_fix flag and sets
 an exclusive lock on the buffer frame. The flag is cleared and the x-lock
-released by the i/o-handler thread. Does a random read-ahead if it seems
-sensible.
-@return number of page read requests issued: this can be greater than
-1 if read-ahead occurred */
+released by the i/o-handler thread.
+@return TRUE if page has been read in, FALSE in case of failure */
 UNIV_INTERN
-ulint
+ibool
 buf_read_page(
 /*==========*/
 	ulint	space,	/*!< in: space id */
@@ -341,20 +173,17 @@ buf_read_page(
 {
 	ib_int64_t	tablespace_version;
 	ulint		count;
-	ulint		count2;
 	ulint		err;
 
 	tablespace_version = fil_space_get_version(space);
 
-	count = buf_read_ahead_random(space, zip_size, offset);
-
 	/* We do the i/o in the synchronous aio mode to save thread
 	switches: hence TRUE */
 
-	count2 = buf_read_page_low(&err, TRUE, BUF_READ_ANY_PAGE, space,
+	count = buf_read_page_low(&err, TRUE, BUF_READ_ANY_PAGE, space,
 				  zip_size, FALSE,
 				  tablespace_version, offset);
-	srv_buf_pool_reads+= count2;
+	srv_buf_pool_reads += count;
 	if (err == DB_TABLESPACE_DELETED) {
 		ut_print_timestamp(stderr);
 		fprintf(stderr,
@@ -371,7 +200,7 @@ buf_read_page(
 	/* Increment number of I/O operations used for LRU policy. */
 	buf_LRU_stat_inc_io();
 
-	return(count + count2);
+	return(count > 0);
 }
 
 /********************************************************************//**
@@ -498,9 +327,17 @@ buf_read_ahead_linear(
 			fail_count++;
 
 		} else if (pred_bpage) {
-			int res = (ut_ulint_cmp(
-				       buf_page_get_LRU_position(bpage),
-				       buf_page_get_LRU_position(pred_bpage)));
+			/* Note that buf_page_is_accessed() returns
+			the time of the first access.  If some blocks
+			of the extent existed in the buffer pool at
+			the time of a linear access pattern, the first
+			access times may be nonmonotonic, even though
+			the latest access times were linear.  The
+			threshold (srv_read_ahead_factor) should help
+			a little against this. */
+			int res = ut_ulint_cmp(
+				buf_page_is_accessed(bpage),
+				buf_page_is_accessed(pred_bpage));
 			/* Accesses not in the right order */
 			if (res != 0 && res != asc_or_desc) {
 				fail_count++;
@@ -643,7 +480,7 @@ buf_read_ahead_linear(
 	LRU policy decision. */
 	buf_LRU_stat_inc_io();
 
-	++srv_read_ahead_seq;
+	buf_pool->stat.n_ra_pages_read += count;
 	return(count);
 }
 

storage/innodb_plugin/dict/dict0dict.c

@@ -82,9 +82,10 @@ static char	dict_ibfk[] = "_ibfk_";
 
 /*******************************************************************//**
 Tries to find column names for the index and sets the col field of the
-index. */
+index.
+@return TRUE if the column names were found */
 static
-void
+ibool
 dict_index_find_cols(
 /*=================*/
 	dict_table_t*	table,	/*!< in: table */
@@ -1431,7 +1432,7 @@ dict_index_too_big_for_tree(
 
 /**********************************************************************//**
 Adds an index to the dictionary cache.
-@return	DB_SUCCESS or DB_TOO_BIG_RECORD */
+@return	DB_SUCCESS, DB_TOO_BIG_RECORD, or DB_CORRUPTION */
 UNIV_INTERN
 ulint
 dict_index_add_to_cache(
@@ -1457,7 +1458,10 @@ dict_index_add_to_cache(
 	ut_a(!dict_index_is_clust(index)
 	     || UT_LIST_GET_LEN(table->indexes) == 0);
 
-	dict_index_find_cols(table, index);
+	if (!dict_index_find_cols(table, index)) {
+
+		return(DB_CORRUPTION);
+	}
 
 	/* Build the cache internal representation of the index,
 	containing also the added system fields */
@@ -1665,9 +1669,10 @@ dict_index_remove_from_cache(
 
 /*******************************************************************//**
 Tries to find column names for the index and sets the col field of the
-index. */
+index.
+@return TRUE if the column names were found */
 static
-void
+ibool
 dict_index_find_cols(
 /*=================*/
 	dict_table_t*	table,	/*!< in: table */
@@ -1692,17 +1697,21 @@ dict_index_find_cols(
 			}
 		}
 
+#ifdef UNIV_DEBUG
 		/* It is an error not to find a matching column. */
 		fputs("InnoDB: Error: no matching column for ", stderr);
 		ut_print_name(stderr, NULL, FALSE, field->name);
 		fputs(" in ", stderr);
 		dict_index_name_print(stderr, NULL, index);
 		fputs("!\n", stderr);
-		ut_error;
+#endif /* UNIV_DEBUG */
+		return(FALSE);
 
 found:
 		;
 	}
+
+	return(TRUE);
 }
 #endif /* !UNIV_HOTBACKUP */
 

storage/innodb_plugin/handler/ha_innodb.cc

@@ -72,6 +72,7 @@ with this program; if not, write to the Free Software Foundation, Inc.,
 /* Include necessary InnoDB headers */
 extern "C" {
 #include "univ.i"
+#include "buf0lru.h"
 #include "btr0sea.h"
 #include "os0file.h"
 #include "os0thread.h"
@@ -152,6 +153,10 @@ static ulong innobase_write_io_threads;
 
 static long long innobase_buffer_pool_size, innobase_log_file_size;
 
+/** Percentage of the buffer pool to reserve for 'old' blocks.
+Connected to buf_LRU_old_ratio. */
+static uint innobase_old_blocks_pct;
+
 /* The default values for the following char* start-up parameters
 are determined in innobase_init below: */
 
@@ -490,10 +495,10 @@ static SHOW_VAR innodb_status_variables[]= {
   (char*) &export_vars.innodb_buffer_pool_pages_misc,	  SHOW_LONG},
   {"buffer_pool_pages_total",
   (char*) &export_vars.innodb_buffer_pool_pages_total,	  SHOW_LONG},
-  {"buffer_pool_read_ahead_rnd",
-  (char*) &export_vars.innodb_buffer_pool_read_ahead_rnd, SHOW_LONG},
-  {"buffer_pool_read_ahead_seq",
-  (char*) &export_vars.innodb_buffer_pool_read_ahead_seq, SHOW_LONG},
+  {"buffer_pool_read_ahead",
+  (char*) &export_vars.innodb_buffer_pool_read_ahead,	  SHOW_LONG},
+  {"buffer_pool_read_ahead_evicted",
+  (char*) &export_vars.innodb_buffer_pool_read_ahead_evicted, SHOW_LONG},
   {"buffer_pool_read_requests",
   (char*) &export_vars.innodb_buffer_pool_read_requests,  SHOW_LONG},
   {"buffer_pool_reads",
@@ -2204,6 +2209,9 @@ innobase_init(
 	ut_a(0 == strcmp(my_charset_latin1.name, "latin1_swedish_ci"));
 	srv_latin1_ordering = my_charset_latin1.sort_order;
 
+	innobase_old_blocks_pct = buf_LRU_old_ratio_update(
+		innobase_old_blocks_pct, FALSE);
+
 	innobase_commit_concurrency_init_default();
 
 	/* Since we in this module access directly the fields of a trx
@@ -9610,6 +9618,25 @@ innodb_adaptive_hash_index_update(
 	}
 }
 
+/****************************************************************//**
+Update the system variable innodb_old_blocks_pct using the "saved"
+value. This function is registered as a callback with MySQL. */
+static
+void
+innodb_old_blocks_pct_update(
+/*=========================*/
+	THD*				thd,	/*!< in: thread handle */
+	struct st_mysql_sys_var*	var,	/*!< in: pointer to
+						system variable */
+	void*				var_ptr,/*!< out: where the
+						formal string goes */
+	const void*			save)	/*!< in: immediate result
+						from check function */
+{
+	innobase_old_blocks_pct = buf_LRU_old_ratio_update(
+		*static_cast<const uint*>(save), TRUE);
+}
+
 /*************************************************************//**
 Check if it is a valid value of innodb_change_buffering.  This function is
 registered as a callback with MySQL.
@@ -9847,7 +9874,7 @@ static MYSQL_SYSVAR_ULONG(concurrency_tickets, srv_n_free_tickets_to_enter,
   NULL, NULL, 500L, 1L, ~0L, 0);
 
 static MYSQL_SYSVAR_LONG(file_io_threads, innobase_file_io_threads,
-  PLUGIN_VAR_RQCMDARG | PLUGIN_VAR_READONLY,
+  PLUGIN_VAR_RQCMDARG | PLUGIN_VAR_READONLY | PLUGIN_VAR_NOSYSVAR,
   "Number of file I/O threads in InnoDB.",
   NULL, NULL, 4, 4, 64, 0);
 
@@ -9886,6 +9913,18 @@ static MYSQL_SYSVAR_LONG(mirrored_log_groups, innobase_mirrored_log_groups,
   "Number of identical copies of log groups we keep for the database. Currently this should be set to 1.",
   NULL, NULL, 1, 1, 10, 0);
 
+static MYSQL_SYSVAR_UINT(old_blocks_pct, innobase_old_blocks_pct,
+  PLUGIN_VAR_RQCMDARG,
+  "Percentage of the buffer pool to reserve for 'old' blocks.",
+  NULL, innodb_old_blocks_pct_update, 100 * 3 / 8, 5, 95, 0);
+
+static MYSQL_SYSVAR_UINT(old_blocks_time, buf_LRU_old_threshold_ms,
+  PLUGIN_VAR_RQCMDARG,
+  "Move blocks to the 'new' end of the buffer pool if the first access"
+  " was at least this many milliseconds ago."
+  " The timeout is disabled if 0 (the default).",
+  NULL, NULL, 0, 0, UINT_MAX32, 0);
+
 static MYSQL_SYSVAR_LONG(open_files, innobase_open_files,
   PLUGIN_VAR_RQCMDARG | PLUGIN_VAR_READONLY,
   "How many files at the maximum InnoDB keeps open at the same time.",
@@ -9984,6 +10023,8 @@ static struct st_mysql_sys_var* innobase_system_variables[]= {
   MYSQL_SYSVAR(adaptive_flushing),
   MYSQL_SYSVAR(max_purge_lag),
   MYSQL_SYSVAR(mirrored_log_groups),
+  MYSQL_SYSVAR(old_blocks_pct),
+  MYSQL_SYSVAR(old_blocks_time),
   MYSQL_SYSVAR(open_files),
   MYSQL_SYSVAR(rollback_on_timeout),
   MYSQL_SYSVAR(stats_on_metadata),

storage/innodb_plugin/include/buf0buf.h

@@ -707,15 +707,6 @@ buf_page_belongs_to_unzip_LRU(
 /*==========================*/
 	const buf_page_t*	bpage)	/*!< in: pointer to control block */
 	__attribute__((pure));
-/*********************************************************************//**
-Determine the approximate LRU list position of a block.
-@return	LRU list position */
-UNIV_INLINE
-ulint
-buf_page_get_LRU_position(
-/*======================*/
-	const buf_page_t*	bpage)	/*!< in: control block */
-	__attribute__((pure));
 
 /*********************************************************************//**
 Gets the mutex of a block.
@@ -816,22 +807,22 @@ buf_page_set_old(
 	buf_page_t*	bpage,	/*!< in/out: control block */
 	ibool		old);	/*!< in: old */
 /*********************************************************************//**
-Determine if a block has been accessed in the buffer pool.
-@return	TRUE if accessed */
+Determine the time of last access a block in the buffer pool.
+@return	ut_time_ms() at the time of last access, 0 if not accessed */
 UNIV_INLINE
-ibool
+unsigned
 buf_page_is_accessed(
 /*=================*/
 	const buf_page_t*	bpage)	/*!< in: control block */
-	__attribute__((pure));
+	__attribute__((nonnull, pure));
 /*********************************************************************//**
 Flag a block accessed. */
 UNIV_INLINE
 void
 buf_page_set_accessed(
 /*==================*/
-	buf_page_t*	bpage,		/*!< in/out: control block */
-	ibool		accessed);	/*!< in: accessed */
+	buf_page_t*	bpage)		/*!< in/out: control block */
+	__attribute__((nonnull));
 /*********************************************************************//**
 Gets the buf_block_t handle of a buffered file block if an uncompressed
 page frame exists, or NULL.
@@ -1017,14 +1008,6 @@ buf_block_hash_get(
 /*===============*/
 	ulint	space,	/*!< in: space id */
 	ulint	offset);/*!< in: offset of the page within space */
-/*******************************************************************//**
-Increments the pool clock by one and returns its new value. Remember that
-in the 32 bit version the clock wraps around at 4 billion!
-@return	new clock value */
-UNIV_INLINE
-ulint
-buf_pool_clock_tic(void);
-/*====================*/
 /*********************************************************************//**
 Gets the current length of the free list of buffer blocks.
 @return	length of the free list */
@@ -1064,16 +1047,10 @@ struct buf_page_struct{
 					flushed to disk, this tells the
 					flush_type.
 					@see enum buf_flush */
-	unsigned	accessed:1;	/*!< TRUE if the page has been accessed
-					while in the buffer pool: read-ahead
-					may read in pages which have not been
-					accessed yet; a thread is allowed to
-					read this for heuristic purposes
-					without holding any mutex or latch */
 	unsigned	io_fix:2;	/*!< type of pending I/O operation;
 					also protected by buf_pool_mutex
 					@see enum buf_io_fix */
-	unsigned	buf_fix_count:24;/*!< count of how manyfold this block
+	unsigned	buf_fix_count:25;/*!< count of how manyfold this block
 					is currently bufferfixed */
 	/* @} */
 #endif /* !UNIV_HOTBACKUP */
@@ -1152,17 +1129,7 @@ struct buf_page_struct{
 #endif /* UNIV_DEBUG */
 	unsigned	old:1;		/*!< TRUE if the block is in the old
 					blocks in the LRU list */
-	unsigned	LRU_position:31;/*!< value which monotonically
-					decreases (or may stay
-					constant if old==TRUE) toward
-					the end of the LRU list, if
-					buf_pool->ulint_clock has not
-					wrapped around: NOTE that this
-					value can only be used in
-					heuristic algorithms, because
-					of the possibility of a
-					wrap-around! */
-	unsigned	freed_page_clock:32;/*!< the value of
+	unsigned	freed_page_clock:31;/*!< the value of
 					buf_pool->freed_page_clock
 					when this block was the last
 					time put to the head of the
@@ -1170,6 +1137,9 @@ struct buf_page_struct{
 					to read this for heuristic
 					purposes without holding any
 					mutex or latch */
+	unsigned	access_time:32;	/*!< time of first access, or
+					0 if the block was never accessed
+					in the buffer pool */
 	/* @} */
 # ifdef UNIV_DEBUG_FILE_ACCESSES
 	ibool		file_page_was_freed;
@@ -1314,6 +1284,31 @@ 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))
 /* @} */
 
+/** @brief The buffer pool statistics structure. */
+struct buf_pool_stat_struct{
+	ulint	n_page_gets;	/*!< number of page gets performed;
+				also successful searches through
+				the adaptive hash index are
+				counted as page gets; this field
+				is NOT protected by the buffer
+				pool mutex */
+	ulint	n_pages_read;	/*!< number read operations */
+	ulint	n_pages_written;/*!< number write operations */
+	ulint	n_pages_created;/*!< number of pages created
+				in the pool with no read */
+	ulint	n_ra_pages_read;/*!< number of pages read in
+				as part of read ahead */
+	ulint	n_ra_pages_evicted;/*!< number of read ahead
+				pages that are evicted without
+				being accessed */
+	ulint	n_pages_made_young; /*!< number of pages made young, in
+				calls to buf_LRU_make_block_young() */
+	ulint	n_pages_not_made_young; /*!< number of pages not made
+				young because the first access
+				was not long enough ago, in
+				buf_page_peek_if_too_old() */
+};
+
 /** @brief The buffer pool structure.
 
 NOTE! The definition appears here only for other modules of this
@@ -1338,28 +1333,16 @@ struct buf_pool_struct{
 	ulint		n_pend_reads;	/*!< number of pending read operations */
 	ulint		n_pend_unzip;	/*!< number of pending decompressions */
 
-	time_t		last_printout_time; /*!< when buf_print was last time
+	time_t		last_printout_time;
+					/*!< when buf_print_io was last time
 					called */
-	ulint		n_pages_read;	/*!< number read operations */
-	ulint		n_pages_written;/*!< number write operations */
-	ulint		n_pages_created;/*!< number of pages created
-					in the pool with no read */
-	ulint		n_page_gets;	/*!< number of page gets performed;
-					also successful searches through
-					the adaptive hash index are
-					counted as page gets; this field
-					is NOT protected by the buffer
-					pool mutex */
-	ulint		n_page_gets_old;/*!< n_page_gets when buf_print was
-					last time called: used to calculate
-					hit rate */
-	ulint		n_pages_read_old;/*!< n_pages_read when buf_print was
-					last time called */
-	ulint		n_pages_written_old;/*!< number write operations */
-	ulint		n_pages_created_old;/*!< number of pages created in
-					the pool with no read */
+	buf_pool_stat_t	stat;		/*!< current statistics */
+	buf_pool_stat_t	old_stat;	/*!< old statistics */
+
 	/* @} */
+
 	/** @name Page flushing algorithm fields */
+
 	/* @{ */
 
 	UT_LIST_BASE_NODE_T(buf_page_t) flush_list;
@@ -1375,10 +1358,6 @@ struct buf_pool_struct{
 					/*!< this is in the set state
 					when there is no flush batch
 					of the given type running */
-	ulint		ulint_clock;	/*!< a sequence number used to count
-					time. NOTE! This counter wraps
-					around at 4 billion (if ulint ==
-					32 bits)! */
 	ulint		freed_page_clock;/*!< a sequence number used
 					to count the number of buffer
 					blocks removed from the end of
@@ -1402,9 +1381,11 @@ struct buf_pool_struct{
 					block list */
 	UT_LIST_BASE_NODE_T(buf_page_t) LRU;
 					/*!< base node of the LRU list */
-	buf_page_t*	LRU_old;	/*!< pointer to the about 3/8 oldest
-					blocks in the LRU list; NULL if LRU
-					length less than BUF_LRU_OLD_MIN_LEN;
+	buf_page_t*	LRU_old;	/*!< pointer to the about
+					buf_LRU_old_ratio/BUF_LRU_OLD_RATIO_DIV
+					oldest blocks in the LRU list;
+					NULL if LRU length less than
+					BUF_LRU_OLD_MIN_LEN;
 					NOTE: when LRU_old != NULL, its length
 					should always equal LRU_old_len */
 	ulint		LRU_old_len;	/*!< length of the LRU list from

storage/innodb_plugin/include/buf0buf.ic

@@ -72,9 +72,24 @@ buf_page_peek_if_too_old(
 /*=====================*/
 	const buf_page_t*	bpage)	/*!< in: block to make younger */
 {
-	return(buf_pool->freed_page_clock
-	       >= buf_page_get_freed_page_clock(bpage)
-	       + 1 + (buf_pool->curr_size / 4));
+	if (buf_LRU_old_threshold_ms && bpage->old) {
+		unsigned	access_time = buf_page_is_accessed(bpage);
+
+		if (access_time && ut_time_ms() - access_time
+		    >= buf_LRU_old_threshold_ms) {
+			return(TRUE);
+		}
+
+		buf_pool->stat.n_pages_not_made_young++;
+		return(FALSE);
+	} else {
+		/* FIXME: bpage->freed_page_clock is 31 bits */
+		return((buf_pool->freed_page_clock & ~(1 << 31))
+		       > bpage->freed_page_clock
+		       + (buf_pool->curr_size
+			  * (BUF_LRU_OLD_RATIO_DIV - buf_LRU_old_ratio)
+			  / (BUF_LRU_OLD_RATIO_DIV * 4)));
+	}
 }
 
 /*********************************************************************//**
@@ -118,22 +133,6 @@ buf_pool_get_oldest_modification(void)
 
 	return(lsn);
 }
-
-/*******************************************************************//**
-Increments the buf_pool clock by one and returns its new value. Remember
-that in the 32 bit version the clock wraps around at 4 billion!
-@return	new clock value */
-UNIV_INLINE
-ulint
-buf_pool_clock_tic(void)
-/*====================*/
-{
-	ut_ad(buf_pool_mutex_own());
-
-	buf_pool->ulint_clock++;
-
-	return(buf_pool->ulint_clock);
-}
 #endif /* !UNIV_HOTBACKUP */
 
 /*********************************************************************//**
@@ -279,21 +278,6 @@ buf_page_belongs_to_unzip_LRU(
 	       && buf_page_get_state(bpage) == BUF_BLOCK_FILE_PAGE);
 }
 
-/*********************************************************************//**
-Determine the approximate LRU list position of a block.
-@return	LRU list position */
-UNIV_INLINE
-ulint
-buf_page_get_LRU_position(
-/*======================*/
-	const buf_page_t*	bpage)	/*!< in: control block */
-{
-	ut_ad(buf_page_in_file(bpage));
-	ut_ad(buf_pool_mutex_own());
-
-	return(bpage->LRU_position);
-}
-
 /*********************************************************************//**
 Gets the mutex of a block.
 @return	pointer to mutex protecting bpage */
@@ -487,17 +471,17 @@ buf_page_set_old(
 }
 
 /*********************************************************************//**
-Determine if a block has been accessed in the buffer pool.
-@return	TRUE if accessed */
+Determine the time of last access a block in the buffer pool.
+@return	ut_time_ms() at the time of last access, 0 if not accessed */
 UNIV_INLINE
-ibool
+unsigned
 buf_page_is_accessed(
 /*=================*/
 	const buf_page_t*	bpage)	/*!< in: control block */
 {
 	ut_ad(buf_page_in_file(bpage));
 
-	return(bpage->accessed);
+	return(bpage->access_time);
 }
 
 /*********************************************************************//**
@@ -506,13 +490,15 @@ UNIV_INLINE
 void
 buf_page_set_accessed(
 /*==================*/
-	buf_page_t*	bpage,		/*!< in/out: control block */
-	ibool		accessed)	/*!< in: accessed */
+	buf_page_t*	bpage)		/*!< in/out: control block */
 {
 	ut_a(buf_page_in_file(bpage));
-	ut_ad(mutex_own(buf_page_get_mutex(bpage)));
+	ut_ad(buf_pool_mutex_own());
 
-	bpage->accessed = accessed;
+	if (!bpage->access_time) {
+		/* Make this the time of the first access. */
+		bpage->access_time = ut_time_ms();
+	}
 }
 
 /*********************************************************************//**

storage/innodb_plugin/include/buf0lru.h

@@ -69,7 +69,7 @@ These are low-level functions
 #########################################################################*/
 
 /** Minimum LRU list length for which the LRU_old pointer is defined */
-#define BUF_LRU_OLD_MIN_LEN	80
+#define BUF_LRU_OLD_MIN_LEN	512	/* 8 megabytes of 16k pages */
 
 /** Maximum LRU list search length in buf_flush_LRU_recommendation() */
 #define BUF_LRU_FREE_SEARCH_LEN		(5 + 2 * BUF_READ_AHEAD_AREA)
@@ -84,15 +84,6 @@ void
 buf_LRU_invalidate_tablespace(
 /*==========================*/
 	ulint	id);	/*!< in: space id */
-/******************************************************************//**
-Gets the minimum LRU_position field for the blocks in an initial segment
-(determined by BUF_LRU_INITIAL_RATIO) of the LRU list. The limit is not
-guaranteed to be precise, because the ulint_clock may wrap around.
-@return	the limit; zero if could not determine it */
-UNIV_INTERN
-ulint
-buf_LRU_get_recent_limit(void);
-/*==========================*/
 /********************************************************************//**
 Insert a compressed block into buf_pool->zip_clean in the LRU order. */
 UNIV_INTERN
@@ -201,6 +192,18 @@ void
 buf_LRU_make_block_old(
 /*===================*/
 	buf_page_t*	bpage);	/*!< in: control block */
+/**********************************************************************//**
+Updates buf_LRU_old_ratio.
+@return	updated old_pct */
+UNIV_INTERN
+uint
+buf_LRU_old_ratio_update(
+/*=====================*/
+	uint	old_pct,/*!< in: Reserve this percentage of
+			the buffer pool for "old" blocks. */
+	ibool	adjust);/*!< in: TRUE=adjust the LRU list;
+			FALSE=just assign buf_LRU_old_ratio
+			during the initialization of InnoDB */
 /********************************************************************//**
 Update the historical stats that we are collecting for LRU eviction
 policy at the end of each interval. */
@@ -227,6 +230,35 @@ buf_LRU_print(void);
 /*===============*/
 #endif /* UNIV_DEBUG_PRINT || UNIV_DEBUG || UNIV_BUF_DEBUG */
 
+/** @name Heuristics for detecting index scan @{ */
+/** Reserve this much/BUF_LRU_OLD_RATIO_DIV of the buffer pool for
+"old" blocks.  Protected by buf_pool_mutex. */
+extern uint	buf_LRU_old_ratio;
+/** The denominator of buf_LRU_old_ratio. */
+#define BUF_LRU_OLD_RATIO_DIV	1024
+/** Maximum value of buf_LRU_old_ratio.
+@see buf_LRU_old_adjust_len
+@see buf_LRU_old_ratio_update */
+#define BUF_LRU_OLD_RATIO_MAX	BUF_LRU_OLD_RATIO_DIV
+/** Minimum value of buf_LRU_old_ratio.
+@see buf_LRU_old_adjust_len
+@see buf_LRU_old_ratio_update
+The minimum must exceed
+(BUF_LRU_OLD_TOLERANCE + 5) * BUF_LRU_OLD_RATIO_DIV / BUF_LRU_OLD_MIN_LEN. */
+#define BUF_LRU_OLD_RATIO_MIN	51
+
+#if BUF_LRU_OLD_RATIO_MIN >= BUF_LRU_OLD_RATIO_MAX
+# error "BUF_LRU_OLD_RATIO_MIN >= BUF_LRU_OLD_RATIO_MAX"
+#endif
+#if BUF_LRU_OLD_RATIO_MAX > BUF_LRU_OLD_RATIO_DIV
+# error "BUF_LRU_OLD_RATIO_MAX > BUF_LRU_OLD_RATIO_DIV"
+#endif
+
+/** Move blocks to "new" LRU list only if the first access was at
+least this many milliseconds ago.  Not protected by any mutex or latch. */
+extern uint	buf_LRU_old_threshold_ms;
+/* @} */
+
 /** @brief Statistics for selecting the LRU list for eviction.
 
 These statistics are not 'of' LRU but 'for' LRU.  We keep count of I/O

storage/innodb_plugin/include/buf0rea.h

@@ -33,12 +33,10 @@ Created 11/5/1995 Heikki Tuuri
 High-level function which reads a page asynchronously from a file to the
 buffer buf_pool if it is not already there. Sets the io_fix flag and sets
 an exclusive lock on the buffer frame. The flag is cleared and the x-lock
-released by the i/o-handler thread. Does a random read-ahead if it seems
-sensible.
-@return number of page read requests issued: this can be greater than
-1 if read-ahead occurred */
+released by the i/o-handler thread.
+@return TRUE if page has been read in, FALSE in case of failure */
 UNIV_INTERN
-ulint
+ibool
 buf_read_page(
 /*==========*/
 	ulint	space,	/*!< in: space id */

storage/innodb_plugin/include/buf0types.h

@@ -34,6 +34,8 @@ typedef	struct buf_block_struct		buf_block_t;
 typedef struct buf_chunk_struct		buf_chunk_t;
 /** Buffer pool comprising buf_chunk_t */
 typedef	struct buf_pool_struct		buf_pool_t;
+/** Buffer pool statistics struct */
+typedef	struct buf_pool_stat_struct	buf_pool_stat_t;
 
 /** A buffer frame. @see page_t */
 typedef	byte	buf_frame_t;

storage/innodb_plugin/include/dict0dict.h

@@ -712,7 +712,7 @@ dict_index_find_on_id_low(
 	dulint	id);	/*!< in: index id */
 /**********************************************************************//**
 Adds an index to the dictionary cache.
-@return	DB_SUCCESS or error code */
+@return	DB_SUCCESS, DB_TOO_BIG_RECORD, or DB_CORRUPTION */
 UNIV_INTERN
 ulint
 dict_index_add_to_cache(

storage/innodb_plugin/include/os0sync.h

@@ -285,7 +285,7 @@ os_fast_mutex_free(
 /**********************************************************//**
 Atomic compare-and-swap and increment for InnoDB. */
 
-#ifdef HAVE_GCC_ATOMIC_BUILTINS
+#ifdef HAVE_IB_GCC_ATOMIC_BUILTINS
 /**********************************************************//**
 Returns true if swapped, ptr is pointer to target, old_val is value to
 compare to, new_val is the value to swap in. */
@@ -377,7 +377,7 @@ InterlockedExchange() operates on LONG, and the LONG will be
 clobbered */
 # define os_atomic_test_and_set_byte(ptr, new_val) \
 	((byte) InterlockedExchange(ptr, new_val))
-#endif /* HAVE_GCC_ATOMIC_BUILTINS */
+#endif /* HAVE_IB_GCC_ATOMIC_BUILTINS */
 
 #ifndef UNIV_NONINL
 #include "os0sync.ic"

storage/innodb_plugin/include/srv0srv.h

@@ -315,10 +315,6 @@ extern ulint srv_buf_pool_flushed;
 /** Number of buffer pool reads that led to the
 reading of a disk page */
 extern ulint srv_buf_pool_reads;
-/** Number of sequential read-aheads */
-extern ulint srv_read_ahead_seq;
-/** Number of random read-aheads */
-extern ulint srv_read_ahead_rnd;
 
 /** Status variables to be passed to MySQL */
 typedef struct export_var_struct export_struc;
@@ -605,13 +601,13 @@ struct export_var_struct{
 #ifdef UNIV_DEBUG
 	ulint innodb_buffer_pool_pages_latched;	/*!< Latched pages */
 #endif /* UNIV_DEBUG */
-	ulint innodb_buffer_pool_read_requests;	/*!< buf_pool->n_page_gets */
+	ulint innodb_buffer_pool_read_requests;	/*!< buf_pool->stat.n_page_gets */
 	ulint innodb_buffer_pool_reads;		/*!< srv_buf_pool_reads */
 	ulint innodb_buffer_pool_wait_free;	/*!< srv_buf_pool_wait_free */
 	ulint innodb_buffer_pool_pages_flushed;	/*!< srv_buf_pool_flushed */
 	ulint innodb_buffer_pool_write_requests;/*!< srv_buf_pool_write_requests */
-	ulint innodb_buffer_pool_read_ahead_seq;/*!< srv_read_ahead_seq */
-	ulint innodb_buffer_pool_read_ahead_rnd;/*!< srv_read_ahead_rnd */
+	ulint innodb_buffer_pool_read_ahead;	/*!< srv_read_ahead */
+	ulint innodb_buffer_pool_read_ahead_evicted;/*!< srv_read_ahead evicted*/
 	ulint innodb_dblwr_pages_written;	/*!< srv_dblwr_pages_written */
 	ulint innodb_dblwr_writes;		/*!< srv_dblwr_writes */
 	ibool innodb_have_atomic_builtins;	/*!< HAVE_ATOMIC_BUILTINS */
@@ -623,9 +619,9 @@ struct export_var_struct{
 	ulint innodb_os_log_pending_writes;	/*!< srv_os_log_pending_writes */
 	ulint innodb_os_log_pending_fsyncs;	/*!< fil_n_pending_log_flushes */
 	ulint innodb_page_size;			/*!< UNIV_PAGE_SIZE */
-	ulint innodb_pages_created;		/*!< buf_pool->n_pages_created */
-	ulint innodb_pages_read;		/*!< buf_pool->n_pages_read */
-	ulint innodb_pages_written;		/*!< buf_pool->n_pages_written */
+	ulint innodb_pages_created;		/*!< buf_pool->stat.n_pages_created */
+	ulint innodb_pages_read;		/*!< buf_pool->stat.n_pages_read */
+	ulint innodb_pages_written;		/*!< buf_pool->stat.n_pages_written */
 	ulint innodb_row_lock_waits;		/*!< srv_n_lock_wait_count */
 	ulint innodb_row_lock_current_waits;	/*!< srv_n_lock_wait_current_count */
 	ib_int64_t innodb_row_lock_time;	/*!< srv_n_lock_wait_time

storage/innodb_plugin/include/univ.i

@@ -125,11 +125,11 @@ if we are compiling on Windows. */
 #  include <sched.h>
 # endif
 
-# if defined(HAVE_GCC_ATOMIC_BUILTINS) || defined(HAVE_SOLARIS_ATOMICS) \
+# if defined(HAVE_IB_GCC_ATOMIC_BUILTINS) || defined(HAVE_SOLARIS_ATOMICS) \
      || defined(HAVE_WINDOWS_ATOMICS)
 /* If atomics are defined we use them in InnoDB mutex implementation */
 #  define HAVE_ATOMIC_BUILTINS
-# endif /* (HAVE_GCC_ATOMIC_BUILTINS) || (HAVE_SOLARIS_ATOMICS)
+# endif /* (HAVE_IB_GCC_ATOMIC_BUILTINS) || (HAVE_SOLARIS_ATOMICS)
 	|| (HAVE_WINDOWS_ATOMICS) */
 
 /* For InnoDB rw_locks to work with atomics we need the thread_id

storage/innodb_plugin/include/ut0ut.h

@@ -239,6 +239,15 @@ ullint
 ut_time_us(
 /*=======*/
 	ullint*	tloc);	/*!< out: us since epoch, if non-NULL */
+/**********************************************************//**
+Returns the number of milliseconds since some epoch.  The
+value may wrap around.  It should only be used for heuristic
+purposes.
+@return	ms since epoch */
+UNIV_INTERN
+uint
+ut_time_ms(void);
+/*============*/
 
 /**********************************************************//**
 Returns the difference of two times in seconds.

storage/innodb_plugin/plug.in

@@ -63,6 +63,57 @@ MYSQL_PLUGIN_ACTIONS(innodb_plugin,  [
 		;;
   esac
   AC_SUBST(INNODB_DYNAMIC_CFLAGS)
+
+  AC_MSG_CHECKING(whether GCC atomic builtins are available)
+  AC_TRY_RUN(
+    [
+      int main()
+      {
+	long	x;
+	long	y;
+	long	res;
+	char	c;
+
+	x = 10;
+	y = 123;
+	res = __sync_bool_compare_and_swap(&x, x, y);
+	if (!res || x != y) {
+          return(1);
+        }
+
+	x = 10;
+	y = 123;
+	res = __sync_bool_compare_and_swap(&x, x + 1, y);
+	if (res || x != 10) {
+          return(1);
+        }
+
+	x = 10;
+	y = 123;
+	res = __sync_add_and_fetch(&x, y);
+	if (res != 123 + 10 || x != 123 + 10) {
+	  return(1);
+	}
+
+	c = 10;
+	res = __sync_lock_test_and_set(&c, 123);
+	if (res != 10 || c != 123) {
+	  return(1);
+	}
+
+	return(0);
+      }
+    ],
+    [
+      AC_DEFINE([HAVE_IB_GCC_ATOMIC_BUILTINS], [1],
+                [GCC atomic builtins are available])
+      AC_MSG_RESULT(yes)
+    ],
+    [
+      AC_MSG_RESULT(no)
+    ]
+    )
+
   AC_MSG_CHECKING(whether pthread_t can be used by GCC atomic builtins)
   AC_TRY_RUN(
     [

storage/innodb_plugin/row/row0merge.c

@@ -60,9 +60,19 @@ Completed by Sunny Bains and Marko Makela
 #ifdef UNIV_DEBUG
 /** Set these in order ot enable debug printout. */
 /* @{ */
+/** Log the outcome of each row_merge_cmp() call, comparing records. */
 static ibool	row_merge_print_cmp;
+/** Log each record read from temporary file. */
 static ibool	row_merge_print_read;
+/** Log each record write to temporary file. */
 static ibool	row_merge_print_write;
+/** Log each row_merge_blocks() call, merging two blocks of records to
+a bigger one. */
+static ibool	row_merge_print_block;
+/** Log each block read from temporary file. */
+static ibool	row_merge_print_block_read;
+/** Log each block read from temporary file. */
+static ibool	row_merge_print_block_write;
 /* @} */
 #endif /* UNIV_DEBUG */
 
@@ -110,7 +120,8 @@ typedef struct row_merge_buf_struct row_merge_buf_t;
 /** Information about temporary files used in merge sort */
 struct merge_file_struct {
 	int		fd;		/*!< file descriptor */
-	ulint	offset;		/*!< file offset */
+	ulint		offset;		/*!< file offset (end of file) */
+	ib_uint64_t	n_rec;		/*!< number of records in the file */
 };
 
 /** Information about temporary files used in merge sort */
@@ -682,6 +693,13 @@ row_merge_read(
 	ib_uint64_t	ofs = ((ib_uint64_t) offset) * sizeof *buf;
 	ibool		success;
 
+#ifdef UNIV_DEBUG
+	if (row_merge_print_block_read) {
+		fprintf(stderr, "row_merge_read fd=%d ofs=%lu\n",
+			fd, (ulong) offset);
+	}
+#endif /* UNIV_DEBUG */
+
 	success = os_file_read_no_error_handling(OS_FILE_FROM_FD(fd), buf,
 						 (ulint) (ofs & 0xFFFFFFFF),
 						 (ulint) (ofs >> 32),
@@ -709,6 +727,13 @@ row_merge_write(
 	ib_uint64_t	ofs = ((ib_uint64_t) offset)
 		* sizeof(row_merge_block_t);
 
+#ifdef UNIV_DEBUG
+	if (row_merge_print_block_write) {
+		fprintf(stderr, "row_merge_write fd=%d ofs=%lu\n",
+			fd, (ulong) offset);
+	}
+#endif /* UNIV_DEBUG */
+
 	return(UNIV_LIKELY(os_file_write("(merge)", OS_FILE_FROM_FD(fd), buf,
 					 (ulint) (ofs & 0xFFFFFFFF),
 					 (ulint) (ofs >> 32),
@@ -718,7 +743,7 @@ row_merge_write(
 /********************************************************************//**
 Read a merge record.
 @return	pointer to next record, or NULL on I/O error or end of list */
-static
+static __attribute__((nonnull))
 const byte*
 row_merge_read_rec(
 /*===============*/
@@ -1070,7 +1095,7 @@ row_merge_cmp(
 Reads clustered index of the table and create temporary files
 containing the index entries for the indexes to be built.
 @return	DB_SUCCESS or error */
-static
+static __attribute__((nonnull))
 ulint
 row_merge_read_clustered_index(
 /*===========================*/
@@ -1233,6 +1258,7 @@ row_merge_read_clustered_index(
 
 			if (UNIV_LIKELY
 			    (row && row_merge_buf_add(buf, row, ext))) {
+				file->n_rec++;
 				continue;
 			}
 
@@ -1274,15 +1300,20 @@ row_merge_read_clustered_index(
 			UNIV_MEM_INVALID(block[0], sizeof block[0]);
 			merge_buf[i] = row_merge_buf_empty(buf);
 
-			/* Try writing the record again, now that
-			the buffer has been written out and emptied. */
+			if (UNIV_LIKELY(row != NULL)) {
+				/* Try writing the record again, now
+				that the buffer has been written out
+				and emptied. */
 
 				if (UNIV_UNLIKELY
-			    (row && !row_merge_buf_add(buf, row, ext))) {
+				    (!row_merge_buf_add(buf, row, ext))) {
 					/* An empty buffer should have enough
 					room for at least one record. */
 					ut_error;
 				}
+
+				file->n_rec++;
+			}
 		}
 
 		mem_heap_empty(row_heap);
@@ -1320,7 +1351,7 @@ row_merge_read_clustered_index(
 		b2 = row_merge_write_rec(&block[2], &buf[2], b2,	\
 					 of->fd, &of->offset,		\
 					 mrec##N, offsets##N);		\
-		if (UNIV_UNLIKELY(!b2)) {				\
+		if (UNIV_UNLIKELY(!b2 || ++of->n_rec > file->n_rec)) {	\
 			goto corrupt;					\
 		}							\
 		b##N = row_merge_read_rec(&block[N], &buf[N],		\
@@ -1336,14 +1367,14 @@ row_merge_read_clustered_index(
 	} while (0)
 
 /*************************************************************//**
-Merge two blocks of linked lists on disk and write a bigger block.
+Merge two blocks of records on disk and write a bigger block.
 @return	DB_SUCCESS or error code */
 static
 ulint
 row_merge_blocks(
 /*=============*/
 	const dict_index_t*	index,	/*!< in: index being created */
-	merge_file_t*		file,	/*!< in/out: file containing
+	const merge_file_t*	file,	/*!< in: file containing
 					index entries */
 	row_merge_block_t*	block,	/*!< in/out: 3 buffers */
 	ulint*			foffs0,	/*!< in/out: offset of first
@@ -1366,6 +1397,17 @@ row_merge_blocks(
 	ulint*		offsets0;/* offsets of mrec0 */
 	ulint*		offsets1;/* offsets of mrec1 */
 
+#ifdef UNIV_DEBUG
+	if (row_merge_print_block) {
+		fprintf(stderr,
+			"row_merge_blocks fd=%d ofs=%lu + fd=%d ofs=%lu"
+			" = fd=%d ofs=%lu\n",
+			file->fd, (ulong) *foffs0,
+			file->fd, (ulong) *foffs1,
+			of->fd, (ulong) of->offset);
+	}
+#endif /* UNIV_DEBUG */
+
 	heap = row_merge_heap_create(index, &offsets0, &offsets1);
 
 	/* Write a record and read the next record.  Split the output
@@ -1437,17 +1479,88 @@ row_merge_blocks(
 	return(b2 ? DB_SUCCESS : DB_CORRUPTION);
 }
 
+/*************************************************************//**
+Copy a block of index entries.
+@return	TRUE on success, FALSE on failure */
+static __attribute__((nonnull))
+ibool
+row_merge_blocks_copy(
+/*==================*/
+	const dict_index_t*	index,	/*!< in: index being created */
+	const merge_file_t*	file,	/*!< in: input file */
+	row_merge_block_t*	block,	/*!< in/out: 3 buffers */
+	ulint*			foffs0,	/*!< in/out: input file offset */
+	merge_file_t*		of)	/*!< in/out: output file */
+{
+	mem_heap_t*	heap;	/*!< memory heap for offsets0, offsets1 */
+
+	mrec_buf_t	buf[3];	/*!< buffer for handling
+				split mrec in block[] */
+	const byte*	b0;	/*!< pointer to block[0] */
+	byte*		b2;	/*!< pointer to block[2] */
+	const mrec_t*	mrec0;	/*!< merge rec, points to block[0] */
+	ulint*		offsets0;/* offsets of mrec0 */
+	ulint*		offsets1;/* dummy offsets */
+
+#ifdef UNIV_DEBUG
+	if (row_merge_print_block) {
+		fprintf(stderr,
+			"row_merge_blocks_copy fd=%d ofs=%lu"
+			" = fd=%d ofs=%lu\n",
+			file->fd, (ulong) foffs0,
+			of->fd, (ulong) of->offset);
+	}
+#endif /* UNIV_DEBUG */
+
+	heap = row_merge_heap_create(index, &offsets0, &offsets1);
+
+	/* Write a record and read the next record.  Split the output
+	file in two halves, which can be merged on the following pass. */
+
+	if (!row_merge_read(file->fd, *foffs0, &block[0])) {
+corrupt:
+		mem_heap_free(heap);
+		return(FALSE);
+	}
+
+	b0 = block[0];
+	b2 = block[2];
+
+	b0 = row_merge_read_rec(&block[0], &buf[0], b0, index, file->fd,
+				foffs0, &mrec0, offsets0);
+	if (UNIV_UNLIKELY(!b0 && mrec0)) {
+
+		goto corrupt;
+	}
+
+	if (mrec0) {
+		/* append all mrec0 to output */
+		for (;;) {
+			ROW_MERGE_WRITE_GET_NEXT(0, goto done0);
+		}
+	}
+done0:
+
+	/* The file offset points to the beginning of the last page
+	that has been read.  Update it to point to the next block. */
+	(*foffs0)++;
+
+	mem_heap_free(heap);
+	return(row_merge_write_eof(&block[2], b2, of->fd, &of->offset)
+	       != NULL);
+}
+
 /*************************************************************//**
 Merge disk files.
 @return	DB_SUCCESS or error code */
-static
+static __attribute__((nonnull))
 ulint
 row_merge(
 /*======*/
 	const dict_index_t*	index,	/*!< in: index being created */
 	merge_file_t*		file,	/*!< in/out: file containing
 					index entries */
-	ulint			half,	/*!< in: half the file */
+	ulint*			half,	/*!< in/out: half the file */
 	row_merge_block_t*	block,	/*!< in/out: 3 buffers */
 	int*			tmpfd,	/*!< in/out: temporary file handle */
 	TABLE*			table)	/*!< in/out: MySQL table, for
@@ -1458,43 +1571,76 @@ row_merge(
 	ulint		foffs1;	/*!< second input offset */
 	ulint		error;	/*!< error code */
 	merge_file_t	of;	/*!< output file */
+	const ulint	ihalf	= *half;
+				/*!< half the input file */
+	ulint		ohalf;	/*!< half the output file */
 
 	UNIV_MEM_ASSERT_W(block[0], 3 * sizeof block[0]);
-	ut_ad(half > 0);
+	ut_ad(ihalf > 0);
+	ut_ad(ihalf < file->offset);
 
 	of.fd = *tmpfd;
 	of.offset = 0;
+	of.n_rec = 0;
 
 	/* Merge blocks to the output file. */
+	ohalf = 0;
 	foffs0 = 0;
-	foffs1 = half;
+	foffs1 = ihalf;
+
+	for (; foffs0 < ihalf && foffs1 < file->offset; foffs0++, foffs1++) {
+		ulint	ahalf;	/*!< arithmetic half the input file */
 
-	for (; foffs0 < half && foffs1 < file->offset; foffs0++, foffs1++) {
 		error = row_merge_blocks(index, file, block,
 					 &foffs0, &foffs1, &of, table);
 
 		if (error != DB_SUCCESS) {
 			return(error);
 		}
+
+		/* Record the offset of the output file when
+		approximately half the output has been generated.  In
+		this way, the next invocation of row_merge() will
+		spend most of the time in this loop.  The initial
+		estimate is ohalf==0. */
+		ahalf = file->offset / 2;
+		ut_ad(ohalf <= of.offset);
+
+		/* Improve the estimate until reaching half the input
+		file size, or we can not get any closer to it.  All
+		comparands should be non-negative when !(ohalf < ahalf)
+		because ohalf <= of.offset. */
+		if (ohalf < ahalf || of.offset - ahalf < ohalf - ahalf) {
+			ohalf = of.offset;
+		}
 	}
 
-	/* Copy the last block, if there is one. */
-	while (foffs0 < half) {
-		if (!row_merge_read(file->fd, foffs0++, block)
-		    || !row_merge_write(of.fd, of.offset++, block)) {
+	/* Copy the last blocks, if there are any. */
+
+	while (foffs0 < ihalf) {
+		if (!row_merge_blocks_copy(index, file, block, &foffs0, &of)) {
 			return(DB_CORRUPTION);
 		}
 	}
+
+	ut_ad(foffs0 == ihalf);
+
 	while (foffs1 < file->offset) {
-		if (!row_merge_read(file->fd, foffs1++, block)
-		    || !row_merge_write(of.fd, of.offset++, block)) {
+		if (!row_merge_blocks_copy(index, file, block, &foffs1, &of)) {
 			return(DB_CORRUPTION);
 		}
 	}
 
+	ut_ad(foffs1 == file->offset);
+
+	if (UNIV_UNLIKELY(of.n_rec != file->n_rec)) {
+		return(DB_CORRUPTION);
+	}
+
 	/* Swap file descriptors for the next pass. */
 	*tmpfd = file->fd;
 	*file = of;
+	*half = ohalf;
 
 	UNIV_MEM_INVALID(block[0], 3 * sizeof block[0]);
 
@@ -1517,20 +1663,17 @@ row_merge_sort(
 					reporting erroneous key value
 					if applicable */
 {
-	ulint	blksz;	/*!< block size */
+	ulint	half = file->offset / 2;
 
-	for (blksz = 1; blksz < file->offset; blksz *= 2) {
-		ulint	half;
+	do {
 		ulint	error;
 
-		ut_ad(ut_is_2pow(blksz));
-		half = ut_2pow_round((file->offset + (blksz - 1)) / 2, blksz);
-		error = row_merge(index, file, half, block, tmpfd, table);
+		error = row_merge(index, file, &half, block, tmpfd, table);
 
 		if (error != DB_SUCCESS) {
 			return(error);
 		}
-	}
+	} while (half < file->offset && half > 0);
 
 	return(DB_SUCCESS);
 }
@@ -1909,6 +2052,7 @@ row_merge_file_create(
 {
 	merge_file->fd = innobase_mysql_tmpfile();
 	merge_file->offset = 0;
+	merge_file->n_rec = 0;
 }
 
 /*********************************************************************//**

storage/innodb_plugin/srv/srv0srv.c

@@ -292,12 +292,6 @@ UNIV_INTERN ulint srv_buf_pool_flushed = 0;
 reading of a disk page */
 UNIV_INTERN ulint srv_buf_pool_reads = 0;
 
-/** Number of sequential read-aheads */
-UNIV_INTERN ulint srv_read_ahead_seq = 0;
-
-/** Number of random read-aheads */
-UNIV_INTERN ulint srv_read_ahead_rnd = 0;
-
 /* structure to pass status variables to MySQL */
 UNIV_INTERN export_struc export_vars;
 
@@ -464,8 +458,6 @@ static ulint   srv_main_background_loops	= 0;
 static ulint   srv_main_flush_loops		= 0;
 /* Log writes involving flush. */
 static ulint   srv_log_writes_and_flush		= 0;
-/* Log writes not including flush. */
-static ulint   srv_log_buffer_writes		= 0;
 
 /* This is only ever touched by the master thread. It records the
 time when the last flush of log file has happened. The master
@@ -714,9 +706,8 @@ srv_print_master_thread_info(
 		srv_main_1_second_loops, srv_main_sleeps,
 		srv_main_10_second_loops, srv_main_background_loops,
 		srv_main_flush_loops);
-	fprintf(file, "srv_master_thread log flush and writes: %lu "
-		      " log writes only: %lu\n",
-		      srv_log_writes_and_flush, srv_log_buffer_writes);
+	fprintf(file, "srv_master_thread log flush and writes: %lu\n",
+		      srv_log_writes_and_flush);
 }
 
 /*********************************************************************//**
@@ -1877,14 +1868,16 @@ srv_export_innodb_status(void)
 	export_vars.innodb_data_reads = os_n_file_reads;
 	export_vars.innodb_data_writes = os_n_file_writes;
 	export_vars.innodb_data_written = srv_data_written;
-	export_vars.innodb_buffer_pool_read_requests = buf_pool->n_page_gets;
+	export_vars.innodb_buffer_pool_read_requests = buf_pool->stat.n_page_gets;
 	export_vars.innodb_buffer_pool_write_requests
 		= srv_buf_pool_write_requests;
 	export_vars.innodb_buffer_pool_wait_free = srv_buf_pool_wait_free;
 	export_vars.innodb_buffer_pool_pages_flushed = srv_buf_pool_flushed;
 	export_vars.innodb_buffer_pool_reads = srv_buf_pool_reads;
-	export_vars.innodb_buffer_pool_read_ahead_rnd = srv_read_ahead_rnd;
-	export_vars.innodb_buffer_pool_read_ahead_seq = srv_read_ahead_seq;
+	export_vars.innodb_buffer_pool_read_ahead
+		= buf_pool->stat.n_ra_pages_read;
+	export_vars.innodb_buffer_pool_read_ahead_evicted
+		= buf_pool->stat.n_ra_pages_evicted;
 	export_vars.innodb_buffer_pool_pages_data
 		= UT_LIST_GET_LEN(buf_pool->LRU);
 	export_vars.innodb_buffer_pool_pages_dirty
@@ -1915,9 +1908,9 @@ srv_export_innodb_status(void)
 	export_vars.innodb_log_writes = srv_log_writes;
 	export_vars.innodb_dblwr_pages_written = srv_dblwr_pages_written;
 	export_vars.innodb_dblwr_writes = srv_dblwr_writes;
-	export_vars.innodb_pages_created = buf_pool->n_pages_created;
-	export_vars.innodb_pages_read = buf_pool->n_pages_read;
-	export_vars.innodb_pages_written = buf_pool->n_pages_written;
+	export_vars.innodb_pages_created = buf_pool->stat.n_pages_created;
+	export_vars.innodb_pages_read = buf_pool->stat.n_pages_read;
+	export_vars.innodb_pages_written = buf_pool->stat.n_pages_written;
 	export_vars.innodb_row_lock_waits = srv_n_lock_wait_count;
 	export_vars.innodb_row_lock_current_waits
 		= srv_n_lock_wait_current_count;
@@ -2284,12 +2277,6 @@ srv_sync_log_buffer_in_background(void)
 		log_buffer_sync_in_background(TRUE);
 		srv_last_log_flush_time = current_time;
 		srv_log_writes_and_flush++;
-	} else {
-		/* Actually we don't need to write logs here.
-		We are just being extra safe here by forcing
-		the log buffer to log file. */
-		log_buffer_sync_in_background(FALSE);
-		srv_log_buffer_writes++;
 	}
 }
 
@@ -2340,8 +2327,8 @@ srv_master_thread(
 
 	srv_main_thread_op_info = "reserving kernel mutex";
 
-	n_ios_very_old = log_sys->n_log_ios + buf_pool->n_pages_read
-		+ buf_pool->n_pages_written;
+	n_ios_very_old = log_sys->n_log_ios + buf_pool->stat.n_pages_read
+		+ buf_pool->stat.n_pages_written;
 	mutex_enter(&kernel_mutex);
 
 	/* Store the user activity counter at the start of this loop */
@@ -2361,8 +2348,8 @@ srv_master_thread(
 	skip_sleep = FALSE;
 
 	for (i = 0; i < 10; i++) {
-		n_ios_old = log_sys->n_log_ios + buf_pool->n_pages_read
-			+ buf_pool->n_pages_written;
+		n_ios_old = log_sys->n_log_ios + buf_pool->stat.n_pages_read
+			+ buf_pool->stat.n_pages_written;
 		srv_main_thread_op_info = "sleeping";
 		srv_main_1_second_loops++;
 
@@ -2401,8 +2388,8 @@ srv_master_thread(
 
 		n_pend_ios = buf_get_n_pending_ios()
 			+ log_sys->n_pending_writes;
-		n_ios = log_sys->n_log_ios + buf_pool->n_pages_read
-			+ buf_pool->n_pages_written;
+		n_ios = log_sys->n_log_ios + buf_pool->stat.n_pages_read
+			+ buf_pool->stat.n_pages_written;
 		if (n_pend_ios < SRV_PEND_IO_THRESHOLD
 		    && (n_ios - n_ios_old < SRV_RECENT_IO_ACTIVITY)) {
 			srv_main_thread_op_info = "doing insert buffer merge";
@@ -2418,6 +2405,8 @@ srv_master_thread(
 			/* Try to keep the number of modified pages in the
 			buffer pool under the limit wished by the user */
 
+			srv_main_thread_op_info =
+				"flushing buffer pool pages";
 			n_pages_flushed = buf_flush_batch(BUF_FLUSH_LIST,
 							  PCT_IO(100),
 							  IB_ULONGLONG_MAX);
@@ -2436,6 +2425,8 @@ srv_master_thread(
 			ulint n_flush = buf_flush_get_desired_flush_rate();
 
 			if (n_flush) {
+				srv_main_thread_op_info =
+					"flushing buffer pool pages";
 				n_flush = ut_min(PCT_IO(100), n_flush);
 				n_pages_flushed =
 					buf_flush_batch(
@@ -2473,8 +2464,8 @@ srv_master_thread(
 	are not required, and may be disabled. */
 
 	n_pend_ios = buf_get_n_pending_ios() + log_sys->n_pending_writes;
-	n_ios = log_sys->n_log_ios + buf_pool->n_pages_read
-		+ buf_pool->n_pages_written;
+	n_ios = log_sys->n_log_ios + buf_pool->stat.n_pages_read
+		+ buf_pool->stat.n_pages_written;
 
 	srv_main_10_second_loops++;
 	if (n_pend_ios < SRV_PEND_IO_THRESHOLD

storage/innodb_plugin/srv/srv0start.c

@@ -1106,7 +1106,7 @@ innobase_start_or_create_for_mysql(void)
 			"InnoDB: The InnoDB memory heap is disabled\n");
 	}
 
-#ifdef HAVE_GCC_ATOMIC_BUILTINS
+#ifdef HAVE_IB_GCC_ATOMIC_BUILTINS
 # ifdef INNODB_RW_LOCKS_USE_ATOMICS
 	fprintf(stderr,
 		"InnoDB: Mutexes and rw_locks use GCC atomic builtins.\n");
@@ -1130,10 +1130,10 @@ innobase_start_or_create_for_mysql(void)
 	fprintf(stderr,
 		"InnoDB: Mutexes use Windows interlocked functions.\n");
 # endif /* INNODB_RW_LOCKS_USE_ATOMICS */
-#else /* HAVE_GCC_ATOMIC_BUILTINS */
+#else /* HAVE_IB_GCC_ATOMIC_BUILTINS */
 	fprintf(stderr,
 		"InnoDB: Neither mutexes nor rw_locks use GCC atomic builtins.\n");
-#endif /* HAVE_GCC_ATOMIC_BUILTINS */
+#endif /* HAVE_IB_GCC_ATOMIC_BUILTINS */
 
 	/* Since InnoDB does not currently clean up all its internal data
 	structures in MySQL Embedded Server Library server_end(), we

storage/innodb_plugin/ut/ut0ut.c

@@ -199,6 +199,23 @@ ut_time_us(
 	return(us);
 }
 
+/**********************************************************//**
+Returns the number of milliseconds since some epoch.  The
+value may wrap around.  It should only be used for heuristic
+purposes.
+@return	ms since epoch */
+UNIV_INTERN
+uint
+ut_time_ms(void)
+/*============*/
+{
+	struct timeval	tv;
+
+	ut_gettimeofday(&tv, NULL);
+
+	return((uint) tv.tv_sec * 1000 + tv.tv_usec / 1000);
+}
+
 /**********************************************************//**
 Returns the difference of two times in seconds.
 @return	time2 - time1 expressed in seconds */