branches/zip: Implement resizeable buffer pool.

mysql.patch: Patch to change or add variables to MySQL

innodb.patch: Patch to make the master thread poll requests to resize
the buffer pool.

Replace srv_pool_size and innobase_buffer_pool_size
with srv_buf_pool_size, srv_buf_pool_old_size,
and srv_buf_pool_curr_size.

Add buf_chunk_t, a collection of buf_block_t.

buf/buf0buf.c

@@ -215,6 +215,15 @@ ibool		buf_debug_prints = FALSE; /* If this is set TRUE,
 					read-ahead or flush occurs */
 #endif /* UNIV_DEBUG */
 
+/* A chunk of buffers.  The buffer pool is allocated in chunks. */
+struct buf_chunk_struct{
+	ulint		mem_size;	/* allocated size of the chunk */
+	ulint		size;		/* size of frames[] and blocks[] */
+	void*		mem;		/* pointer to the memory area which
+					was allocated for the frames */
+	buf_block_t*	blocks;		/* array of buffer control blocks */
+};
+
 /************************************************************************
 Calculates a page checksum which is stored to the page when it is written
 to a file. Note that we must be careful to calculate the same value on
@@ -627,70 +636,220 @@ buf_block_init(
 }
 
 /************************************************************************
-Creates the buffer pool. */
-
-buf_pool_t*
-buf_pool_init(
-/*==========*/
-				/* out, own: buf_pool object, NULL if not
-				enough memory or error */
-	ulint	curr_size)	/* in: current size to use */
+Allocates a chunk of buffer frames. */
+static
+buf_chunk_t*
+buf_chunk_init(
+/*===========*/
+					/* out: chunk, or NULL on failure */
+	buf_chunk_t*	chunk,		/* out: chunk of buffers */
+	ulint		mem_size)	/* in: requested size in bytes */
 {
+	buf_block_t*	block;
 	byte*		frame;
 	ulint		i;
-	buf_block_t*	block;
 
-	buf_pool = mem_alloc(sizeof(buf_pool_t));
+	/* Round down to a multiple of page size,
+	although it already should be. */
+	mem_size = ut_2pow_round(mem_size, UNIV_PAGE_SIZE);
+	/* Reserve space for the block descriptors. */
+	mem_size += ut_2pow_round((mem_size / UNIV_PAGE_SIZE) * (sizeof *block)
+				  + (UNIV_PAGE_SIZE - 1), UNIV_PAGE_SIZE);
 
-	/* 1. Initialize general fields
-	------------------------------- */
-	mutex_create(&buf_pool->mutex, SYNC_BUF_POOL);
-
-	mutex_enter(&(buf_pool->mutex));
-
-	buf_pool->frame_mem_size = (curr_size + 1) * UNIV_PAGE_SIZE;
-
-	buf_pool->frame_mem = os_mem_alloc_large(&buf_pool->frame_mem_size);
+	chunk->mem_size = mem_size;
+	chunk->mem = os_mem_alloc_large(&chunk->mem_size);
 
-	if (UNIV_UNLIKELY(buf_pool->frame_mem == NULL)) {
+	if (UNIV_UNLIKELY(chunk->mem == NULL)) {
 
 		return(NULL);
 	}
 
-	/* Align pointer to the first frame */
-
-	frame = ut_align(buf_pool->frame_mem, UNIV_PAGE_SIZE);
+	/* Allocate the block descriptors from
+	the start of the memory block. */
+	chunk->blocks = chunk->mem;
 
-	curr_size = buf_pool->frame_mem_size / UNIV_PAGE_SIZE
-		- (frame != buf_pool->frame_mem);
+	/* Align pointer to the first frame */
 
-	buf_pool->blocks = ut_malloc(sizeof(buf_block_t) * curr_size);
+	frame = ut_align(chunk->mem, UNIV_PAGE_SIZE);
+	chunk->size = chunk->mem_size / UNIV_PAGE_SIZE
+		- (frame != chunk->mem);
 
-	if (UNIV_UNLIKELY(buf_pool->blocks == NULL)) {
+	/* Subtract the space needed for block descriptors. */
+	{
+		ulint	size = chunk->size;
 
-		os_mem_free_large(buf_pool->frame_mem,
-				  buf_pool->frame_mem_size);
-		buf_pool->frame_mem = NULL;
+		while (frame < (byte*) (chunk->blocks + size)) {
+			frame += UNIV_PAGE_SIZE;
+			size--;
+		}
 
-		return(NULL);
+		chunk->size = size;
 	}
 
-	buf_pool->curr_size = curr_size;
-
 	/* Init block structs and assign frames for them. Then we
 	assign the frames to the first blocks (we already mapped the
 	memory above). */
 
-	for (i = 0; i < curr_size; i++) {
+	block = chunk->blocks;
 
-		block = buf_pool_get_nth_block(buf_pool, i);
+	for (i = chunk->size; i--; ) {
 
 		buf_block_init(block, frame);
 
+#ifdef HAVE_purify
+		/* Wipe contents of frame to eliminate a Purify warning */
+		memset(block->frame, '\0', UNIV_PAGE_SIZE);
+#endif
+		/* Add the block to the free list */
+		UT_LIST_ADD_LAST(free, buf_pool->free, block);
+		block->in_free_list = TRUE;
+
+		block++;
 		frame += UNIV_PAGE_SIZE;
 	}
 
-	buf_pool->page_hash = hash_create(2 * curr_size);
+	return(chunk);
+}
+
+/*************************************************************************
+Checks that all file pages in the buffer chunk are in a replaceable state. */
+static
+const buf_block_t*
+buf_chunk_not_freed(
+/*================*/
+				/* out: address of a non-free block,
+				or NULL if all freed */
+	buf_chunk_t*	chunk)	/* in: chunk being checked */
+{
+	buf_block_t*	block;
+	ulint		i;
+
+	ut_ad(buf_pool);
+#ifdef UNIV_SYNC_DEBUG
+	ut_ad(mutex_own(&(buf_pool->mutex)));
+#endif /* UNIV_SYNC_DEBUG */
+
+	block = chunk->blocks;
+
+	for (i = chunk->size; i--; block++) {
+		mutex_enter(&block->mutex);
+
+		if (block->state == BUF_BLOCK_FILE_PAGE
+		    && !buf_flush_ready_for_replace(block)) {
+
+			mutex_exit(&block->mutex);
+			return(block);
+		}
+
+		mutex_exit(&block->mutex);
+	}
+
+	return(NULL);
+}
+
+/*************************************************************************
+Checks that all blocks in the buffer chunk are in BUF_BLOCK_NOT_USED state. */
+static
+ibool
+buf_chunk_all_free(
+/*===============*/
+					/* out: TRUE if all freed */
+	const buf_chunk_t*	chunk)	/* in: chunk being checked */
+{
+	const buf_block_t*	block;
+	ulint			i;
+
+	ut_ad(buf_pool);
+#ifdef UNIV_SYNC_DEBUG
+	ut_ad(mutex_own(&(buf_pool->mutex)));
+#endif /* UNIV_SYNC_DEBUG */
+
+	block = chunk->blocks;
+
+	for (i = chunk->size; i--; block++) {
+
+		if (block->state != BUF_BLOCK_NOT_USED) {
+
+			return(FALSE);
+		}
+	}
+
+	return(TRUE);
+}
+
+/************************************************************************
+Frees a chunk of buffer frames. */
+static
+void
+buf_chunk_free(
+/*===========*/
+	buf_chunk_t*	chunk)		/* out: chunk of buffers */
+{
+	buf_block_t*		block;
+	const buf_block_t*	block_end;
+
+#ifdef UNIV_SYNC_DEBUG
+	ut_ad(mutex_own(&(buf_pool->mutex)));
+#endif /* UNIV_SYNC_DEBUG */
+
+	block_end = chunk->blocks + chunk->size;
+
+	for (block = chunk->blocks; block < block_end; block++) {
+		ut_a(block->state == BUF_BLOCK_NOT_USED);
+		ut_a(!block->page_zip.data);
+
+		/* Remove the block from the free list. */
+		ut_a(block->in_free_list);
+		UT_LIST_REMOVE(free, buf_pool->free, block);
+
+		/* Free the latches. */
+		mutex_free(&block->mutex);
+		rw_lock_free(&block->lock);
+#ifdef UNIV_SYNC_DEBUG
+		rw_lock_free(&block->debug_latch);
+#endif /* UNIV_SYNC_DEBUG */
+	}
+
+	os_mem_free_large(chunk->mem, chunk->mem_size);
+}
+
+/************************************************************************
+Creates the buffer pool. */
+
+buf_pool_t*
+buf_pool_init(void)
+/*===============*/
+				/* out, own: buf_pool object, NULL if not
+				enough memory or error */
+{
+	buf_chunk_t*	chunk;
+	ulint		i;
+
+	buf_pool = mem_alloc(sizeof(buf_pool_t));
+
+	/* 1. Initialize general fields
+	------------------------------- */
+	mutex_create(&buf_pool->mutex, SYNC_BUF_POOL);
+
+	mutex_enter(&(buf_pool->mutex));
+
+	buf_pool->n_chunks = 1;
+	buf_pool->chunks = chunk = mem_alloc(sizeof *chunk);
+
+	UT_LIST_INIT(buf_pool->free);
+
+	if (!buf_chunk_init(chunk, srv_buf_pool_size)) {
+		mem_free(chunk);
+		mem_free(buf_pool);
+		buf_pool = NULL;
+		return(NULL);
+	}
+
+	srv_buf_pool_old_size = srv_buf_pool_size;
+	buf_pool->curr_size = chunk->size;
+	srv_buf_pool_curr_size = buf_pool->curr_size * UNIV_PAGE_SIZE;
+
+	buf_pool->page_hash = hash_create(2 * buf_pool->curr_size);
 
 	buf_pool->n_pend_reads = 0;
 
@@ -727,31 +886,128 @@ buf_pool_init(
 
 	buf_pool->LRU_old = NULL;
 
-	/* Add control blocks to the free list */
-	UT_LIST_INIT(buf_pool->free);
+	mutex_exit(&(buf_pool->mutex));
 
-	for (i = 0; i < curr_size; i++) {
+	btr_search_sys_create(buf_pool->curr_size
+			      * UNIV_PAGE_SIZE / sizeof(void*) / 64);
 
-		block = buf_pool_get_nth_block(buf_pool, i);
+	return(buf_pool);
+}
 
-		if (block->frame) {
-			/* Wipe contents of frame to eliminate a Purify
-			warning */
+/************************************************************************
+Resizes the buffer pool. */
 
-#ifdef HAVE_purify
-			memset(block->frame, '\0', UNIV_PAGE_SIZE);
-#endif
+void
+buf_pool_resize(void)
+/*=================*/
+{
+	buf_chunk_t*	chunks;
+	buf_chunk_t*	chunk;
+
+	mutex_enter(&buf_pool->mutex);
+
+	if (srv_buf_pool_old_size == srv_buf_pool_size) {
+
+		goto func_exit;
+	}
+
+	if (srv_buf_pool_curr_size + 1048576 > srv_buf_pool_size) {
+
+		/* Shrink the buffer pool by at least one megabyte */
+
+		ulint		chunk_size
+			= (srv_buf_pool_curr_size - srv_buf_pool_size) 
+			/ UNIV_PAGE_SIZE;
+		ulint		max_size;
+		buf_chunk_t*	max_chunk;
+
+shrink_again:
+		if (buf_pool->n_chunks <= 1) {
+
+			/* Cannot shrink if there is only one chunk */
+			goto func_done;
 		}
 
-		UT_LIST_ADD_LAST(free, buf_pool->free, block);
-		block->in_free_list = TRUE;
+		/* Search for the largest free chunk
+		not larger than the size difference */
+		chunks = buf_pool->chunks;
+		chunk = chunks + buf_pool->n_chunks;
+		max_size = 0;
+		max_chunk = NULL;
+
+		while (--chunk >= chunks) {
+			if (chunk->size <= chunk_size
+			    && chunk->size > max_size
+			    && buf_chunk_all_free(chunk)) {
+				max_size = chunk->size;
+				max_chunk = chunk;
+			}
+		}
+
+		if (!max_size) {
+
+			/* Cannot shrink: try again later
+			(do not assign srv_buf_pool_old_size) */
+			goto func_exit;
+		}
+
+		srv_buf_pool_old_size = srv_buf_pool_size;
+
+		/* Rewrite buf_pool->chunks.  Copy everything but max_chunk. */
+		chunks = mem_alloc((buf_pool->n_chunks - 1) * sizeof *chunks);
+		memcpy(chunks, buf_pool->chunks,
+		       (max_chunk - buf_pool->chunks) * sizeof *chunks);
+		memcpy(chunks + (max_chunk - buf_pool->chunks),
+		       max_chunk + 1,
+		       buf_pool->chunks + buf_pool->n_chunks
+		       - (max_chunk + 1));
+		ut_a(buf_pool->curr_size > max_chunk->size);
+		buf_pool->curr_size -= max_chunk->size;
+		srv_buf_pool_curr_size = buf_pool->curr_size * UNIV_PAGE_SIZE;
+		chunk_size -= max_chunk->size;
+		buf_chunk_free(max_chunk);
+		mem_free(buf_pool->chunks);
+		buf_pool->chunks = chunks;
+		buf_pool->n_chunks--;
+
+		/* Allow a slack of one megabyte. */
+		if (chunk_size > 1048576 / UNIV_PAGE_SIZE) {
+
+			goto shrink_again;
+		}
+	} else if (srv_buf_pool_curr_size + 1048576 < srv_buf_pool_size) {
+
+		/* Enlarge the buffer pool by at least one megabyte */
+
+		ulint		mem_size
+			= srv_buf_pool_size - srv_buf_pool_curr_size;
+
+		chunks = mem_alloc((buf_pool->n_chunks + 1) * sizeof *chunks);
+
+		memcpy(chunks, buf_pool->chunks, buf_pool->n_chunks
+		       * sizeof *chunks);
+
+		chunk = &chunks[buf_pool->n_chunks];
+
+		if (!buf_chunk_init(chunk, mem_size)) {
+			mem_free(chunks);
+		} else {
+			buf_pool->curr_size += chunk->size;
+			srv_buf_pool_curr_size = buf_pool->curr_size
+				* UNIV_PAGE_SIZE;
+			mem_free(buf_pool->chunks);
+			buf_pool->chunks = chunks;
+			buf_pool->n_chunks++;
+		}
 	}
 
-	mutex_exit(&(buf_pool->mutex));
+	/* TODO: reinitialize buf_pool->page_hash */
 
-	btr_search_sys_create(curr_size * UNIV_PAGE_SIZE / sizeof(void*) / 64);
+func_done:
+	srv_buf_pool_old_size = srv_buf_pool_size;
 
-	return(buf_pool);
+func_exit:
+	mutex_exit(&buf_pool->mutex);
 }
 
 /************************************************************************
@@ -1979,7 +2235,7 @@ ibool
 buf_validate(void)
 /*==============*/
 {
-	buf_block_t*	block;
+	buf_chunk_t*	chunk;
 	ulint		i;
 	ulint		n_single_flush	= 0;
 	ulint		n_lru_flush	= 0;
@@ -1993,58 +2249,67 @@ buf_validate(void)
 
 	mutex_enter(&(buf_pool->mutex));
 
-	for (i = 0; i < buf_pool->curr_size; i++) {
+	chunk = buf_pool->chunks;
 
-		block = buf_pool_get_nth_block(buf_pool, i);
+	for (i = buf_pool->n_chunks; i--; chunk++) {
 
-		mutex_enter(&block->mutex);
+		ulint		j;
+		buf_block_t*	block = chunk->blocks;
+
+		for (j = chunk->size; j--; block++) {
+
+			mutex_enter(&block->mutex);
 
-		if (block->state == BUF_BLOCK_FILE_PAGE) {
+			if (block->state == BUF_BLOCK_FILE_PAGE) {
 
-			ut_a(buf_page_hash_get(block->space,
-					       block->offset) == block);
-			n_page++;
+				ut_a(buf_page_hash_get(block->space,
+						       block->offset)
+				     == block);
+				n_page++;
 
 #ifdef UNIV_IBUF_DEBUG
-			ut_a((block->io_fix == BUF_IO_READ)
-			     || ibuf_count_get(block->space, block->offset)
-			     == 0);
+				ut_a((block->io_fix == BUF_IO_READ)
+				     || !ibuf_count_get(block->space,
+							block->offset));
 #endif
-			if (block->io_fix == BUF_IO_WRITE) {
-
-				if (block->flush_type == BUF_FLUSH_LRU) {
-					n_lru_flush++;
-					ut_a(rw_lock_is_locked(
-						     &block->lock,
-						     RW_LOCK_SHARED));
-				} else if (block->flush_type
-					   == BUF_FLUSH_LIST) {
-					n_list_flush++;
-				} else if (block->flush_type
-					   == BUF_FLUSH_SINGLE_PAGE) {
-					n_single_flush++;
-				} else {
-					ut_error;
-				}
+				if (block->io_fix == BUF_IO_WRITE) {
+
+					switch (block->flush_type) {
+					case BUF_FLUSH_LRU:
+						n_lru_flush++;
+						ut_a(rw_lock_is_locked(
+							     &block->lock,
+							     RW_LOCK_SHARED));
+						break;
+					case BUF_FLUSH_LIST:
+						n_list_flush++;
+						break;
+					case BUF_FLUSH_SINGLE_PAGE:
+						n_single_flush++;
+						break;
+					default:
+						ut_error;
+					}
 
-			} else if (block->io_fix == BUF_IO_READ) {
+				} else if (block->io_fix == BUF_IO_READ) {
 
-				ut_a(rw_lock_is_locked(&(block->lock),
-						       RW_LOCK_EX));
-			}
+					ut_a(rw_lock_is_locked(&block->lock,
+							       RW_LOCK_EX));
+				}
 
-			n_lru++;
+				n_lru++;
 
-			if (ut_dulint_cmp(block->oldest_modification,
-					  ut_dulint_zero) > 0) {
-				n_flush++;
+				if (ut_dulint_cmp(block->oldest_modification,
+						  ut_dulint_zero) > 0) {
+					n_flush++;
+				}
+
+			} else if (block->state == BUF_BLOCK_NOT_USED) {
+				n_free++;
 			}
 
-		} else if (block->state == BUF_BLOCK_NOT_USED) {
-			n_free++;
+			mutex_exit(&block->mutex);
 		}
-
-		mutex_exit(&block->mutex);
 	}
 
 	if (n_lru + n_free > buf_pool->curr_size) {
@@ -2090,7 +2355,7 @@ buf_print(void)
 	ulint		j;
 	dulint		id;
 	ulint		n_found;
-	buf_frame_t*	frame;
+	buf_chunk_t*	chunk;
 	dict_index_t*	index;
 
 	ut_ad(buf_pool);
@@ -2125,30 +2390,38 @@ buf_print(void)
 
 	n_found = 0;
 
-	for (i = 0; i < size; i++) {
-		frame = buf_pool_get_nth_block(buf_pool, i)->frame;
+	chunk = buf_pool->chunks;
 
-		if (fil_page_get_type(frame) == FIL_PAGE_INDEX) {
+	for (i = buf_pool->n_chunks; i--; chunk++) {
+		buf_block_t*	block		= chunk->blocks;
+		ulint		n_blocks	= chunk->size;
 
-			id = btr_page_get_index_id(frame);
+		for (; n_blocks--; block++) {
+			const buf_frame_t* frame = block->frame;
 
-			/* Look for the id in the index_ids array */
-			j = 0;
+			if (fil_page_get_type(frame) == FIL_PAGE_INDEX) {
 
-			while (j < n_found) {
+				id = btr_page_get_index_id(frame);
 
-				if (ut_dulint_cmp(index_ids[j], id) == 0) {
-					counts[j]++;
+				/* Look for the id in the index_ids array */
+				j = 0;
 
-					break;
+				while (j < n_found) {
+
+					if (ut_dulint_cmp(index_ids[j],
+							  id) == 0) {
+						counts[j]++;
+
+						break;
+					}
+					j++;
 				}
-				j++;
-			}
 
-			if (j == n_found) {
-				n_found++;
-				index_ids[j] = id;
-				counts[j] = 1;
+				if (j == n_found) {
+					n_found++;
+					index_ids[j] = id;
+					counts[j] = 1;
+				}
 			}
 		}
 	}
@@ -2184,17 +2457,26 @@ Returns the number of latched pages in the buffer pool. */
 ulint
 buf_get_latched_pages_number(void)
 {
-	buf_block_t*	block;
+	buf_chunk_t*	chunk;
 	ulint		i;
 	ulint		fixed_pages_number = 0;
 
 	mutex_enter(&(buf_pool->mutex));
 
-	for (i = 0; i < buf_pool->curr_size; i++) {
+	chunk = buf_pool->chunks;
+
+	for (i = buf_pool->n_chunks; i--; chunk++) {
+		buf_block_t*	block;
+		ulint		j;
+
+		block = chunk->blocks;
 
-		block = buf_pool_get_nth_block(buf_pool, i);
+		for (j = chunk->size; j--; block++) {
+			if (block->magic_n != BUF_BLOCK_MAGIC_N) {
+
+				continue;
+			}
 
-		if (block->magic_n == BUF_BLOCK_MAGIC_N) {
 			mutex_enter(&block->mutex);
 
 			if (block->buf_fix_count != 0 || block->io_fix != 0) {
@@ -2340,32 +2622,26 @@ ibool
 buf_all_freed(void)
 /*===============*/
 {
-	buf_block_t*	block;
+	buf_chunk_t*	chunk;
 	ulint		i;
 
 	ut_ad(buf_pool);
 
 	mutex_enter(&(buf_pool->mutex));
 
-	for (i = 0; i < buf_pool->curr_size; i++) {
-
-		block = buf_pool_get_nth_block(buf_pool, i);
-
-		mutex_enter(&block->mutex);
+	chunk = buf_pool->chunks;
 
-		if (block->state == BUF_BLOCK_FILE_PAGE) {
+	for (i = buf_pool->n_chunks; i--; chunk++) {
 
-			if (!buf_flush_ready_for_replace(block)) {
+		const buf_block_t* block = buf_chunk_not_freed(chunk);
 
-				fprintf(stderr,
-					"Page %lu %lu still fixed or dirty\n",
-					(ulong) block->space,
-					(ulong) block->offset);
-				ut_error;
-			}
+		if (UNIV_LIKELY_NULL(block)) {
+			fprintf(stderr,
+				"Page %lu %lu still fixed or dirty\n",
+				(ulong) block->space,
+				(ulong) block->offset);
+			ut_error;
 		}
-
-		mutex_exit(&block->mutex);
 	}
 
 	mutex_exit(&(buf_pool->mutex));

handler/ha_innodb.cc

@@ -157,7 +157,7 @@ long innobase_mirrored_log_groups, innobase_log_files_in_group,
 	innobase_lock_wait_timeout, innobase_force_recovery,
 	innobase_open_files;
 
-longlong innobase_buffer_pool_size, innobase_log_file_size;
+longlong innobase_log_file_size;
 
 /* The default values for the following char* start-up parameters
 are determined in innobase_init below: */
@@ -1404,7 +1404,7 @@ innobase_init(void *p)
 
 	/* Check that values don't overflow on 32-bit systems. */
 	if (sizeof(ulint) == 4) {
-		if (innobase_buffer_pool_size > UINT_MAX32) {
+		if (srv_buf_pool_size > UINT_MAX32) {
 			sql_print_error(
 				"innobase_buffer_pool_size can't be over 4GB"
 				" on 32-bit systems");
@@ -1533,14 +1533,6 @@ innobase_init(void *p)
 #endif /* UNIV_LOG_ARCHIVE */
 	srv_log_buffer_size = (ulint) innobase_log_buffer_size;
 
-	/* We set srv_pool_size here in units of 1 kB. InnoDB internally
-	changes the value so that it becomes the number of database pages. */
-
-	/* Careful here: we first convert the signed long int to ulint
-	   and only after that divide */
-
-	srv_pool_size = ((ulint) innobase_buffer_pool_size) / 1024;
-
 	srv_mem_pool_size = (ulint) innobase_additional_mem_pool_size;
 
 	srv_n_file_io_threads = (ulint) innobase_file_io_threads;

handler/ha_innodb.h

@@ -207,7 +207,7 @@ extern SHOW_VAR innodb_status_variables[];
 extern ulong innobase_fast_shutdown;
 extern ulong innobase_large_page_size;
 extern long innobase_mirrored_log_groups, innobase_log_files_in_group;
-extern longlong innobase_buffer_pool_size, innobase_log_file_size;
+extern longlong innobase_log_file_size;
 extern long innobase_log_buffer_size;
 extern long innobase_additional_mem_pool_size;
 extern long innobase_file_io_threads, innobase_lock_wait_timeout;
@@ -225,6 +225,8 @@ extern my_bool innobase_log_archive,
 	innobase_file_per_table, innobase_locks_unsafe_for_binlog,
 	innobase_create_status_file;
 extern "C" {
+extern ulong srv_buf_pool_curr_size;
+extern ulong srv_buf_pool_size;
 extern ulong srv_max_buf_pool_modified_pct;
 extern ulong srv_max_purge_lag;
 extern ulong srv_auto_extend_increment;

include/buf0buf.h

@@ -69,11 +69,16 @@ extern ulint srv_buf_pool_write_requests; /* variable to count write request
 Creates the buffer pool. */
 
 buf_pool_t*
-buf_pool_init(
-/*==========*/
+buf_pool_init(void);
+/*===============*/
 				/* out, own: buf_pool object, NULL if not
 				enough memory or error */
-	ulint	curr_size);	/* in: current size to use */
+/************************************************************************
+Resizes the buffer pool. */
+
+void
+buf_pool_resize(void);
+/*=================*/
 /*************************************************************************
 Gets the current size of buffer buf_pool in bytes. */
 UNIV_INLINE
@@ -833,11 +838,8 @@ struct buf_pool_struct{
 	mutex_t		mutex;		/* mutex protecting the buffer pool
 					struct and control blocks, except the
 					read-write lock in them */
-	byte*		frame_mem;	/* pointer to the memory area which
-					was allocated for the frames */
-	ulint		frame_mem_size;	/* allocated length of frame_mem
-					in bytes */
-	buf_block_t*	blocks;		/* array of buffer control blocks */
+	ulint		n_chunks;	/* number of buffer pool chunks */
+	buf_chunk_t*	chunks;		/* buffer pool chunks */
 	ulint		curr_size;	/* current pool size in pages */
 	hash_table_t*	page_hash;	/* hash table of the file pages */
 

include/buf0buf.ic

@@ -42,22 +42,6 @@ buf_pool_get_curr_size(void)
 	return(buf_pool->curr_size * UNIV_PAGE_SIZE);
 }
 
-/***********************************************************************
-Accessor function for block array. */
-UNIV_INLINE
-buf_block_t*
-buf_pool_get_nth_block(
-/*===================*/
-				/* out: pointer to block */
-	buf_pool_t*	buf_pool,/* in: buf_pool */
-	ulint		i)	/* in: index of the block */
-{
-	ut_ad(buf_pool);
-	ut_ad(i < buf_pool->curr_size);
-
-	return(i + buf_pool->blocks);
-}
-
 /************************************************************************
 Gets the smallest oldest_modification lsn for any page in the pool. Returns
 ut_dulint_zero if all modified pages have been flushed to disk. */
@@ -114,8 +98,6 @@ buf_block_get_frame(
 	buf_block_t*	block)	/* in: pointer to the control block */
 {
 	ut_ad(block);
-	ut_ad(block >= buf_pool->blocks);
-	ut_ad(block < buf_pool->blocks + buf_pool->curr_size);
 	ut_ad(block->state != BUF_BLOCK_NOT_USED);
 	ut_ad((block->state != BUF_BLOCK_FILE_PAGE)
 	      || (block->buf_fix_count > 0));
@@ -133,8 +115,6 @@ buf_block_get_space(
 	const buf_block_t*	block)	/* in: pointer to the control block */
 {
 	ut_ad(block);
-	ut_ad(block >= buf_pool->blocks);
-	ut_ad(block < buf_pool->blocks + buf_pool->curr_size);
 	ut_a(block->state == BUF_BLOCK_FILE_PAGE);
 	ut_ad(block->buf_fix_count > 0);
 
@@ -151,8 +131,6 @@ buf_block_get_page_no(
 	const buf_block_t*	block)	/* in: pointer to the control block */
 {
 	ut_ad(block);
-	ut_ad(block >= buf_pool->blocks);
-	ut_ad(block < buf_pool->blocks + buf_pool->curr_size);
 	ut_a(block->state == BUF_BLOCK_FILE_PAGE);
 	ut_ad(block->buf_fix_count > 0);
 

include/buf0flu.h

@@ -94,8 +94,9 @@ ibool
 buf_flush_ready_for_replace(
 /*========================*/
 				/* out: TRUE if can replace immediately */
-	buf_block_t*	block);	/* in: buffer control block, must be in state
-				BUF_BLOCK_FILE_PAGE and in the LRU list */
+	buf_block_t*	block);	/* in: buffer control block, must
+				be in state BUF_BLOCK_FILE_PAGE
+				and in the LRU list */
 /**********************************************************************
 Validates the flush list. */
 

include/buf0types.h

@@ -10,6 +10,7 @@ Created 11/17/1995 Heikki Tuuri
 #define buf0types_h
 
 typedef	struct buf_block_struct		buf_block_t;
+typedef struct buf_chunk_struct		buf_chunk_t;
 typedef	struct buf_pool_struct		buf_pool_t;
 
 /* The 'type' used of a buffer frame */

include/srv0srv.h

@@ -84,7 +84,9 @@ extern ulong	srv_flush_log_at_trx_commit;
 
 extern byte	srv_latin1_ordering[256];/* The sort order table of the latin1
 					character set */
-extern ulint	srv_pool_size;
+extern ulong	srv_buf_pool_size;	/* requested size in kilobytes */
+extern ulong	srv_buf_pool_old_size;	/* previously requested size */
+extern ulong	srv_buf_pool_curr_size;	/* current size in kilobytes */
 extern ulint	srv_mem_pool_size;
 extern ulint	srv_lock_table_size;
 
@@ -153,7 +155,6 @@ extern ulong	srv_thread_sleep_delay;
 extern ulint	srv_spin_wait_delay;
 extern ibool	srv_priority_boost;
 
-extern	ulint	srv_pool_size;
 extern	ulint	srv_mem_pool_size;
 extern	ulint	srv_lock_table_size;
 

innodb.patch

+Index: srv/srv0srv.c
+===================================================================
+--- srv/srv0srv.c	(revision 1010)
++++ srv/srv0srv.c	(working copy)
+@@ -2183,6 +2182,12 @@ loop:
+ 	/* ---- When there is database activity by users, we cycle in this
+ 	loop */
+ 
++	if (UNIV_UNLIKELY(srv_buf_pool_size != srv_buf_pool_old_size)) {
++		srv_main_thread_op_info = "resizing buffer pool";
++
++		buf_pool_resize();
++	}
++
+ 	srv_main_thread_op_info = "reserving kernel mutex";
+ 
+ 	n_ios_very_old = log_sys->n_log_ios + buf_pool->n_pages_read
+@@ -2543,6 +2548,12 @@ flush_loop:
+ 	master thread to wait for more server activity */
+ 
+ suspend_thread:
++	if (UNIV_UNLIKELY(srv_buf_pool_size != srv_buf_pool_old_size)) {
++		srv_main_thread_op_info = "resizing buffer pool";
++
++		buf_pool_resize();
++	}
++
+ 	srv_main_thread_op_info = "suspending";
+ 
+ 	mutex_enter(&kernel_mutex);
+@@ -2553,7 +2564,9 @@ suspend_thread:
+ 		goto loop;
+ 	}
+ 
++#if 0
+ 	event = srv_suspend_thread();
++#endif
+ 
+ 	mutex_exit(&kernel_mutex);
+ 
+@@ -2563,7 +2576,11 @@ suspend_thread:
+ 	manual also mentions this string in several places. */
+ 	srv_main_thread_op_info = "waiting for server activity";
+ 
++#if 0
+ 	os_event_wait(event);
++#else
++	os_thread_sleep(1000000);
++#endif
+ 
+ 	if (srv_shutdown_state == SRV_SHUTDOWN_EXIT_THREADS) {
+ 		/* This is only extra safety, the thread should exit

mysql.patch

+diff -pu mysql-5.1-bk/sql/mysqld.cc mysql-5.1-zip/sql/mysqld.cc
+--- mysql-5.1-bk/sql/mysqld.cc	2006-11-09 16:01:19.000000000 +0200
++++ mysql-5.1-zip/sql/mysqld.cc	2006-11-13 12:54:04.000000000 +0200
+@@ -370,7 +370,6 @@ extern long innobase_lock_scan_time;
+ extern long innobase_mirrored_log_groups, innobase_log_files_in_group;
+ extern longlong innobase_log_file_size;
+ extern long innobase_log_buffer_size;
+-extern longlong innobase_buffer_pool_size;
+ extern long innobase_additional_mem_pool_size;
+ extern long innobase_file_io_threads, innobase_lock_wait_timeout;
+ extern long innobase_force_recovery;
+@@ -390,6 +389,8 @@ extern "C" {
+ extern ulong srv_max_buf_pool_modified_pct;
+ extern ulong srv_max_purge_lag;
+ extern ulong srv_auto_extend_increment;
++extern ulong srv_buf_pool_curr_size;
++extern ulong srv_buf_pool_size;
+ extern ulong srv_n_spin_wait_rounds;
+ extern ulong srv_n_free_tickets_to_enter;
+ extern ulong srv_thread_sleep_delay;
+@@ -5801,9 +5802,9 @@ log and this option does nothing anymore
+    (gptr*) &srv_auto_extend_increment,
+    0, GET_LONG, REQUIRED_ARG, 8L, 1L, 1000L, 0, 1L, 0},
+   {"innodb_buffer_pool_size", OPT_INNODB_BUFFER_POOL_SIZE,
+-   "The size of the memory buffer InnoDB uses to cache data and indexes of its tables.",
+-   (gptr*) &innobase_buffer_pool_size, (gptr*) &innobase_buffer_pool_size, 0,
+-   GET_LL, REQUIRED_ARG, 8*1024*1024L, 1024*1024L, LONGLONG_MAX, 0,
++   "The requested size of the memory buffer InnoDB uses to cache data and indexes of its tables.",
++   (gptr*) &srv_buf_pool_size, (gptr*) &srv_buf_pool_size, 0,
++   GET_LL, REQUIRED_ARG, 8*1024*1024L, 5*1024*1024L, LONGLONG_MAX, 0,
+    1024*1024L, 0},
+   {"innodb_commit_concurrency", OPT_INNODB_COMMIT_CONCURRENCY,
+    "Helps in performance tuning in heavily concurrent environments.",
+diff -pu mysql-5.1-bk/sql/set_var.cc mysql-5.1-zip/sql/set_var.cc
+--- mysql-5.1-bk/sql/set_var.cc	2006-11-09 16:01:19.000000000 +0200
++++ mysql-5.1-zip/sql/set_var.cc	2006-11-13 12:54:05.000000000 +0200
+@@ -65,7 +65,6 @@ extern ulong innobase_fast_shutdown;
+ extern long innobase_mirrored_log_groups, innobase_log_files_in_group;
+ extern longlong innobase_log_file_size;
+ extern long innobase_log_buffer_size;
+-extern longlong innobase_buffer_pool_size;
+ extern long innobase_additional_mem_pool_size;
+ extern long innobase_file_io_threads, innobase_lock_wait_timeout;
+ extern long innobase_force_recovery;
+@@ -81,6 +80,8 @@ extern my_bool innobase_log_archive,
+                innobase_locks_unsafe_for_binlog;
+ 
+ extern "C" {
++extern ulong srv_buf_pool_curr_size;
++extern ulong srv_buf_pool_size;
+ extern ulong srv_max_buf_pool_modified_pct;
+ extern ulong srv_max_purge_lag;
+ extern ulong srv_auto_extend_increment;
+@@ -499,6 +500,8 @@ sys_var_thd_bool	sys_innodb_support_xa("
+                                                &SV::innodb_support_xa);
+ sys_var_long_ptr	sys_innodb_autoextend_increment("innodb_autoextend_increment",
+ 							&srv_auto_extend_increment);
++sys_var_long_ptr	sys_innodb_buffer_pool_size("innodb_buffer_pool_size",
++						    &srv_buf_pool_size);
+ sys_var_long_ptr	sys_innodb_sync_spin_loops("innodb_sync_spin_loops",
+                                              &srv_n_spin_wait_rounds);
+ sys_var_long_ptr  sys_innodb_concurrency_tickets("innodb_concurrency_tickets",
+@@ -818,7 +821,8 @@ SHOW_VAR init_vars[]= {
+ #ifdef WITH_INNOBASE_STORAGE_ENGINE
+   {"innodb_additional_mem_pool_size", (char*) &innobase_additional_mem_pool_size, SHOW_LONG },
+   {sys_innodb_autoextend_increment.name, (char*) &sys_innodb_autoextend_increment, SHOW_SYS},
+-  {"innodb_buffer_pool_size", (char*) &innobase_buffer_pool_size, SHOW_LONGLONG },
++  {sys_innodb_buffer_pool_size.name, (char*) &sys_innodb_buffer_pool_size, SHOW_SYS },
++  {"innodb_buffer_pool_curr_size", (char*) &srv_buf_pool_curr_size, SHOW_LONGLONG },
+   {"innodb_checksums", (char*) &innobase_use_checksums, SHOW_MY_BOOL},
+   {sys_innodb_commit_concurrency.name, (char*) &sys_innodb_commit_concurrency, SHOW_SYS},
+   {sys_innodb_concurrency_tickets.name, (char*) &sys_innodb_concurrency_tickets, SHOW_SYS},

srv/srv0srv.c

@@ -159,10 +159,11 @@ byte	srv_latin1_ordering[256]	/* The sort order table of the latin1
 , 0xD8, 0x55, 0x55, 0x55, 0x59, 0x59, 0xDE, 0xFF
 };
 
-ulint	srv_pool_size		= ULINT_MAX;	/* size in pages; MySQL inits
-						this to size in kilobytes but
-						we normalize this to pages in
-						srv_boot() */
+ulong	srv_buf_pool_size	= ULINT_MAX;	/* requested size
+						in kilobytes */
+ulong	srv_buf_pool_old_size;			/* previously requested size */
+ulong	srv_buf_pool_curr_size	= 0;		/* current size
+						in kilobytes */
 ulint	srv_mem_pool_size	= ULINT_MAX;	/* size in bytes */
 ulint	srv_lock_table_size	= ULINT_MAX;
 
@@ -1233,9 +1234,7 @@ srv_normalize_init_values(void)
 
 	srv_log_buffer_size = srv_log_buffer_size / UNIV_PAGE_SIZE;
 
-	srv_pool_size = srv_pool_size / (UNIV_PAGE_SIZE / 1024);
-
-	srv_lock_table_size = 5 * srv_pool_size;
+	srv_lock_table_size = 5 * (srv_buf_pool_size / UNIV_PAGE_SIZE);
 
 	return(DB_SUCCESS);
 }

srv/srv0start.c

@@ -1132,7 +1132,7 @@ innobase_start_or_create_for_mysql(void)
 	}
 
 	/* Note that the call srv_boot() also changes the values of
-	srv_pool_size etc. to the units used by InnoDB internally */
+	some variables to the units used by InnoDB internally */
 
 	/* Set the maximum number of threads which can wait for a semaphore
 	inside InnoDB: this is the 'sync wait array' size, as well as the
@@ -1147,15 +1147,12 @@ innobase_start_or_create_for_mysql(void)
 	NetWare. */
 	srv_max_n_threads = 1000;
 #else
-	if (srv_pool_size >= 1000 * 1024) {
-		/* Here we still have srv_pool_size counted
-		in kilobytes (in 4.0 this was in bytes)
-		srv_boot() converts the value to
-		pages; if buffer pool is less than 1000 MB,
+	if (srv_buf_pool_size >= 1000 * 1024 * 1024) {
+		/* If buffer pool is less than 1000 MB,
 		assume fewer threads. */
 		srv_max_n_threads = 50000;
 
-	} else if (srv_pool_size >= 8 * 1024) {
+	} else if (srv_buf_pool_size >= 8 * 1024 * 1024) {
 
 		srv_max_n_threads = 10000;
 	} else {
@@ -1164,7 +1161,7 @@ innobase_start_or_create_for_mysql(void)
 						computers */
 	}
 #endif
-	err = srv_boot(); /* This changes srv_pool_size to units of a page */
+	err = srv_boot();
 
 	if (err != DB_SUCCESS) {
 
@@ -1230,7 +1227,7 @@ innobase_start_or_create_for_mysql(void)
 
 	fil_init(srv_max_n_open_files);
 
-	ret = buf_pool_init(srv_pool_size);
+	ret = buf_pool_init();
 
 	if (ret == NULL) {
 		fprintf(stderr,