4 files changed, 106 insertions, 43 deletions

diff --git a/innobase/buf/Makefile.am b/innobase/buf/Makefile.am
index b1463c2220e..3f56c8b02d7 100644
--- a/innobase/buf/Makefile.am
+++ b/innobase/buf/Makefile.am
@@ -17,7 +17,7 @@
 
 include ../include/Makefile.i
 
-libs_LIBRARIES =	libbuf.a
+noinst_LIBRARIES =	libbuf.a
 
 libbuf_a_SOURCES =	buf0buf.c buf0flu.c buf0lru.c buf0rea.c
 
diff --git a/innobase/buf/buf0buf.c b/innobase/buf/buf0buf.c
index 14d538a14bc..1cdddaf6cb4 100644
--- a/innobase/buf/buf0buf.c
+++ b/innobase/buf/buf0buf.c
@@ -597,8 +597,9 @@ buf_pool_init(
 			/* Wipe contents of frame to eliminate a Purify
 			warning */
 
+#ifdef HAVE_purify
 			memset(block->frame, '\0', UNIV_PAGE_SIZE);
-
+#endif
 			if (srv_use_awe) {
 				/* Add to the list of blocks mapped to
 				frames */
@@ -1837,7 +1838,7 @@ buf_pool_invalidate(void)
 	freed = TRUE;
 
 	while (freed) {
-		freed = buf_LRU_search_and_free_block(0);
+		freed = buf_LRU_search_and_free_block(100);
 	}
 	
 	mutex_enter(&(buf_pool->mutex));
@@ -2057,6 +2058,29 @@ buf_get_n_pending_ios(void)
 }
 
 /*************************************************************************
+Returns the ratio in percents of modified pages in the buffer pool /
+database pages in the buffer pool. */
+
+ulint
+buf_get_modified_ratio_pct(void)
+/*============================*/
+{
+	ulint	ratio;
+
+	mutex_enter(&(buf_pool->mutex));
+
+	ratio = (100 * UT_LIST_GET_LEN(buf_pool->flush_list))
+		     / (1 + UT_LIST_GET_LEN(buf_pool->LRU)
+		        + UT_LIST_GET_LEN(buf_pool->free));
+
+		       /* 1 + is there to avoid division by zero */   
+
+	mutex_exit(&(buf_pool->mutex));
+
+	return(ratio);
+}
+
+/*************************************************************************
 Prints info of the buffer i/o. */
 
 void
@@ -2109,8 +2133,10 @@ buf_print_io(
 
 	buf += sprintf(buf,
 		"Pending writes: LRU %lu, flush list %lu, single page %lu\n",
-		buf_pool->n_flush[BUF_FLUSH_LRU],
-		buf_pool->n_flush[BUF_FLUSH_LIST],
+		buf_pool->n_flush[BUF_FLUSH_LRU]
+				+ buf_pool->init_flush[BUF_FLUSH_LRU],
+		buf_pool->n_flush[BUF_FLUSH_LIST]
+				+ buf_pool->init_flush[BUF_FLUSH_LIST],
 		buf_pool->n_flush[BUF_FLUSH_SINGLE_PAGE]);
 
 	current_time = time(NULL);
@@ -2144,7 +2170,7 @@ buf_print_io(
 		/ (buf_pool->n_page_gets - buf_pool->n_page_gets_old)));
 	} else {
 		buf += sprintf(buf,
-			"No buffer pool activity since the last printout\n");
+			"No buffer pool page gets since the last printout\n");
 	}
 
 	buf_pool->n_page_gets_old = buf_pool->n_page_gets;
diff --git a/innobase/buf/buf0flu.c b/innobase/buf/buf0flu.c
index 02587487a92..47ac9c6b041 100644
--- a/innobase/buf/buf0flu.c
+++ b/innobase/buf/buf0flu.c
@@ -107,7 +107,7 @@ buf_flush_ready_for_replace(
 				BUF_BLOCK_FILE_PAGE and in the LRU list */
 {
 	ut_ad(mutex_own(&(buf_pool->mutex)));
-	ut_ad(block->state == BUF_BLOCK_FILE_PAGE);
+	ut_a(block->state == BUF_BLOCK_FILE_PAGE);
 
 	if ((ut_dulint_cmp(block->oldest_modification, ut_dulint_zero) > 0)
 	    || (block->buf_fix_count != 0)
@@ -227,7 +227,9 @@ buf_flush_buffered_writes(void)
 	}
 
 	for (i = 0; i < trx_doublewrite->first_free; i++) {
+
 		block = trx_doublewrite->buf_block_arr[i];
+	        ut_a(block->state == BUF_BLOCK_FILE_PAGE);
 
 		if (block->check_index_page_at_flush
 				&& !page_simple_validate(block->frame)) {
@@ -236,10 +238,12 @@ buf_flush_buffered_writes(void)
 
 			ut_print_timestamp(stderr);
 			fprintf(stderr,
-	"  InnoDB: Apparent corruption of an index page\n"
+	"  InnoDB: Apparent corruption of an index page n:o %lu in space %lu\n"
 	"InnoDB: to be written to data file. We intentionally crash server\n"
 	"InnoDB: to prevent corrupt data from ending up in data\n"
-	"InnoDB: files.\n");
+	"InnoDB: files.\n",
+			block->offset, block->space);
+
 			ut_a(0);
 		}
 	}
@@ -394,7 +398,7 @@ buf_flush_write_block_low(
 	"Warning: cannot force log to disk in the log debug version!\n");
 #else
 	/* Force the log to the disk before writing the modified block */
-	log_flush_up_to(block->newest_modification, LOG_WAIT_ALL_GROUPS);
+	log_write_up_to(block->newest_modification, LOG_WAIT_ALL_GROUPS, TRUE);
 #endif	
 	buf_flush_init_for_writing(block->frame, block->newest_modification,
 						block->space, block->offset);
@@ -432,6 +436,8 @@ buf_flush_try_page(
 
 	block = buf_page_hash_get(space, offset);
 
+	ut_a(block->state == BUF_BLOCK_FILE_PAGE);
+
 	if (flush_type == BUF_FLUSH_LIST
 	    && block && buf_flush_ready_for_flush(block, flush_type)) {
 	
@@ -567,7 +573,8 @@ buf_flush_try_page(
 		rw_lock_s_lock_gen(&(block->lock), BUF_IO_WRITE);
 
 		if (buf_debug_prints) {
-			printf("Flushing single page space %lu, page no %lu \n",
+			printf(
+			"Flushing single page space %lu, page no %lu \n",
 						block->space, block->offset);
 		}
 
@@ -608,15 +615,7 @@ buf_flush_try_neighbors(
 	
 		low = offset;
 		high = offset + 1;
-	} else if (flush_type == BUF_FLUSH_LIST) {
-		/* Since semaphore waits require us to flush the
-		doublewrite buffer to disk, it is best that the
-		search area is just the page itself, to minimize
-		chances for semaphore waits */
-
-		low = offset;
-		high = offset + 1;
-	}		
+	}
 
 	/* printf("Flush area: low %lu high %lu\n", low, high); */
 	
@@ -633,13 +632,20 @@ buf_flush_try_neighbors(
 		if (block && flush_type == BUF_FLUSH_LRU && i != offset
 		    && !block->old) {
 
-		  /* We avoid flushing 'non-old' blocks in an LRU flush,
-		     because the flushed blocks are soon freed */
+		        /* We avoid flushing 'non-old' blocks in an LRU flush,
+		        because the flushed blocks are soon freed */
 
-		  continue;
+		        continue;
 		}
 
-		if (block && buf_flush_ready_for_flush(block, flush_type)) {
+		if (block && buf_flush_ready_for_flush(block, flush_type)
+		   && (i == offset || block->buf_fix_count == 0)) {
+			/* We only try to flush those neighbors != offset
+			where the buf fix count is zero, as we then know that
+			we probably can latch the page without a semaphore
+			wait. Semaphore waits are expensive because we must
+			flush the doublewrite buffer before we start
+			waiting. */
 
 			mutex_exit(&(buf_pool->mutex));
 
@@ -758,7 +764,6 @@ buf_flush_batch(
 				page_count +=
 					buf_flush_try_neighbors(space, offset,
 								flush_type);
-
 				/* printf(
 				"Flush type %lu, page no %lu, neighb %lu\n",
 				flush_type, offset,
@@ -884,11 +889,19 @@ buf_flush_free_margin(void)
 /*=======================*/
 {
 	ulint	n_to_flush;
+	ulint	n_flushed;
 
 	n_to_flush = buf_flush_LRU_recommendation();
 	
 	if (n_to_flush > 0) {
-		buf_flush_batch(BUF_FLUSH_LRU, n_to_flush, ut_dulint_zero);
+		n_flushed = buf_flush_batch(BUF_FLUSH_LRU, n_to_flush,
+							ut_dulint_zero);
+		if (n_flushed == ULINT_UNDEFINED) {
+			/* There was an LRU type flush batch already running;
+			let us wait for it to end */
+		   
+		        buf_flush_wait_batch_end(BUF_FLUSH_LRU);
+		}
 	}
 }
 
diff --git a/innobase/buf/buf0lru.c b/innobase/buf/buf0lru.c
index 051aa0191f6..40f49f1fddc 100644
--- a/innobase/buf/buf0lru.c
+++ b/innobase/buf/buf0lru.c
@@ -104,12 +104,15 @@ ibool
 buf_LRU_search_and_free_block(
 /*==========================*/
 				/* out: TRUE if freed */
-	ulint	n_iterations __attribute__((unused))) /* in: how many times
-                                this has been called repeatedly without
-                                result: a high value means that we should 
-                                search farther */
+	ulint	n_iterations)   /* in: how many times this has been called
+				repeatedly without result: a high value means
+				that we should search farther; if value is
+				k < 10, then we only search k/10 * [number
+				of pages in the buffer pool] from the end
+				of the LRU list */
 {
 	buf_block_t*	block;
+	ulint		distance = 0;
 	ibool		freed;
 
 	mutex_enter(&(buf_pool->mutex));
@@ -152,6 +155,18 @@ buf_LRU_search_and_free_block(
 		}
 
 		block = UT_LIST_GET_PREV(LRU, block);
+		distance++;
+
+		if (!freed && n_iterations <= 10
+		    && distance > 100 + (n_iterations * buf_pool->curr_size)
+					/ 10) {
+
+			buf_pool->LRU_flush_ended = 0;
+
+			mutex_exit(&(buf_pool->mutex));
+			
+			return(FALSE);
+		}
 	}
 
 	if (buf_pool->LRU_flush_ended > 0) {
@@ -186,7 +201,7 @@ buf_LRU_try_free_flushed_blocks(void)
 
 		mutex_exit(&(buf_pool->mutex));
 
-		buf_LRU_search_and_free_block(0);
+		buf_LRU_search_and_free_block(1);
 		
 		mutex_enter(&(buf_pool->mutex));
 	}
@@ -208,7 +223,7 @@ buf_LRU_get_free_block(void)
 {
 	buf_block_t*	block		= NULL;
 	ibool		freed;
-	ulint		n_iterations	= 0;
+	ulint		n_iterations	= 1;
 	ibool		mon_value_was   = 0; /* remove bug */
 	ibool		started_monitor	= FALSE;
 loop:
@@ -236,9 +251,12 @@ loop:
 	   	fprintf(stderr,
 "  InnoDB: WARNING: over 4 / 5 of the buffer pool is occupied by\n"
 "InnoDB: lock heaps or the adaptive hash index! Check that your\n"
-"InnoDB: transactions do not set too many row locks. Starting InnoDB\n"
-"InnoDB: Monitor to print diagnostics, including lock heap and hash index\n"
-"InnoDB: sizes.\n");
+"InnoDB: transactions do not set too many row locks.\n"
+"InnoDB: Your buffer pool size is %lu MB. Maybe you should make\n"
+"InnoDB: the buffer pool bigger?\n"
+"InnoDB: Starting the InnoDB Monitor to print diagnostics, including\n"
+"InnoDB: lock heap and hash index sizes.\n",
+		buf_pool->curr_size / (1024 * 1024 / UNIV_PAGE_SIZE));
 
 		srv_print_innodb_monitor = TRUE;
 
@@ -251,14 +269,6 @@ loop:
 
 		srv_print_innodb_monitor = FALSE;
 	}
-
-	if (buf_pool->LRU_flush_ended > 0) {
-		mutex_exit(&(buf_pool->mutex));
-
-		buf_LRU_try_free_flushed_blocks();
-		
-		mutex_enter(&(buf_pool->mutex));
-	}
 	
 	/* If there is a block in the free list, take it */
 	if (UT_LIST_GET_LEN(buf_pool->free) > 0) {
@@ -340,6 +350,20 @@ loop:
 
 	os_aio_simulated_wake_handler_threads();
 
+	mutex_enter(&(buf_pool->mutex));
+
+	if (buf_pool->LRU_flush_ended > 0) {
+		/* We have written pages in an LRU flush. To make the insert
+		buffer more efficient, we try to move these pages to the free
+		list. */
+
+		mutex_exit(&(buf_pool->mutex));
+
+		buf_LRU_try_free_flushed_blocks();
+	} else {
+		mutex_exit(&(buf_pool->mutex));
+	}
+
 	if (n_iterations > 10) {
 
 		os_thread_sleep(500000);