branches/zip: Minor improvements.

buf_LRU_block_remove_hashed_page(): Return the new state of the block.

Only call buf_LRU_block_free_hashed_page()
if buf_LRU_block_remove_hashed_page() did not return BUF_BLOCK_ZIP_FREE,
that is, the control block was not freed.

buf_LRU_insert_zip_clean(): New function, for inserting a compressed-only
block into buf_pool->zip_clean in the LRU order.

buf_LRU_block_remove_hashed_page(), buf_LRU_free_block():
Add the flag "zip" for controlling if the compressed page of an uncompressed
page should be removed.  For now, assume zip==TRUE.

buf_LRU_get_free_block(): Replace the test for UT_LIST_GET_LEN(buf_pool->free)
with a test for the return value of buf_LRU_get_free_only().  Do not
free zip.data, as it must already have been freed.

buf_flush_insert_into_flush_list(), buf_flush_insert_sorted_into_flush_list():
Remove compressed-only blocks from the buf_pool->zip_clean list.

buf_flush_remove(): Restore compressed-only blocks to
the buf_pool->zip_clean list.

buf_page_init_for_read(): Uncompress compressed-only blocks when possible.
Currently, there cannot be any compressed-only blocks in the buffer pool;
they would be inserted by buf_LRU_free_block(bpage, zip=FALSE).

buf/buf0buddy.c

@@ -165,7 +165,7 @@ buf_buddy_alloc_clean(
 
 		for (; j--; bpage = UT_LIST_GET_NEXT(list, bpage)) {
 			if (bpage->zip.ssize != dummy_zip.ssize
-			    || !buf_LRU_free_block(bpage)) {
+			    || !buf_LRU_free_block(bpage, FALSE)) {
 
 				continue;
 			}
@@ -197,7 +197,8 @@ buf_buddy_alloc_clean(
 
 		void* ret;
 
-		if (!buf_LRU_free_block(bpage)) {
+		/* Keep the compressed pages of uncompressed blocks. */
+		if (!buf_LRU_free_block(bpage, FALSE)) {
 
 			continue;
 		}

buf/buf0buf.c

@@ -26,6 +26,7 @@ Created 11/5/1995 Heikki Tuuri
 #include "buf0buf.ic"
 #endif
 
+#include "buf0buddy.h"
 #include "mem0mem.h"
 #include "btr0btr.h"
 #include "fil0fil.h"
@@ -1030,7 +1031,7 @@ buf_pool_shrink(
 
 				buf_LRU_make_block_old(&block->page);
 				dirty++;
-			} else if (!buf_LRU_free_block(&block->page)) {
+			} else if (!buf_LRU_free_block(&block->page, TRUE)) {
 				nonfree++;
 			}
 
@@ -1938,6 +1939,7 @@ buf_page_init_for_read(
 	ulint		offset)	/* in: page number */
 {
 	buf_block_t*	block;
+	buf_page_t*	bpage;
 	mtr_t		mtr;
 
 	ut_ad(buf_pool);
@@ -1972,27 +1974,94 @@ buf_page_init_for_read(
 	if (fil_tablespace_deleted_or_being_deleted_in_mem(
 		    space, tablespace_version)) {
 		*err = DB_TABLESPACE_DELETED;
+
+		goto err_exit;
 	}
 
-	if (*err == DB_TABLESPACE_DELETED
-	    || NULL != buf_page_hash_get(space, offset)) {
+	bpage = buf_page_hash_get(space, offset);
+
+	if (UNIV_LIKELY_NULL(bpage)) {
+
+		switch (buf_page_get_state(bpage)) {
+		case BUF_BLOCK_ZIP_PAGE:
+		case BUF_BLOCK_ZIP_DIRTY:
+			ut_a(page_zip_get_size(&bpage->zip) == zip_size);
+
+			if (bpage->buf_fix_count
+			    || buf_page_get_io_fix(bpage)
+			    != BUF_IO_NONE) {
+
+				goto err_exit;
+			}
+
+			/* Move the compressed page from bpage to block,
+			and uncompress it. */
+
+			buf_buddy_free(block->page.zip.data, zip_size);
+
+			mutex_enter(&buf_pool->zip_mutex);
+			memcpy(&block->page, bpage, sizeof *bpage);
+			block->page.state = BUF_BLOCK_FILE_PAGE;
 
-		    /* The page belongs to a space which has been
-		    deleted or is being deleted, or the page is
-		    already in buf_pool, return */
+			buf_relocate(bpage, &block->page);
+
+			if (buf_page_get_state(bpage) == BUF_BLOCK_ZIP_PAGE) {
+				UT_LIST_REMOVE(list, buf_pool->zip_clean,
+					       bpage);
+			} else {
+				/* Relocate buf_pool->flush_list. */
+				buf_page_t*	b;
+
+				b = UT_LIST_GET_PREV(list, bpage);
+				UT_LIST_REMOVE(list, buf_pool->flush_list,
+					       bpage);
+
+				if (b) {
+					UT_LIST_INSERT_AFTER(
+						list, buf_pool->flush_list, b,
+						&block->page);
+				} else {
+					UT_LIST_ADD_FIRST(
+						list, buf_pool->flush_list,
+						&block->page);
+				}
+			}
+
+			bpage->zip.data = NULL;
+			page_zip_set_size(&bpage->zip, 0);
+
+			buf_buddy_free(bpage, sizeof *bpage);
+
+			mutex_exit(&buf_pool->zip_mutex);
+			break;
+		case BUF_BLOCK_FILE_PAGE:
+			break;
+		case BUF_BLOCK_ZIP_FREE:
+		case BUF_BLOCK_NOT_USED:
+		case BUF_BLOCK_READY_FOR_USE:
+		case BUF_BLOCK_MEMORY:
+		case BUF_BLOCK_REMOVE_HASH:
+			ut_error;
+			break;
+		}
+
+err_exit:
+		/* The page belongs to a space which has been
+		deleted or is being deleted, or the page is
+		already in buf_pool, return */
 
 		mutex_exit(&block->mutex);
-		    mutex_exit(&(buf_pool->mutex));
+		mutex_exit(&(buf_pool->mutex));
 
-		    buf_block_free(block);
+		buf_block_free(block);
 
-		    if (mode == BUF_READ_IBUF_PAGES_ONLY) {
+		if (mode == BUF_READ_IBUF_PAGES_ONLY) {
 
-			    mtr_commit(&mtr);
-		    }
+			mtr_commit(&mtr);
+		}
 
-		    return(NULL);
-	    }
+		return(NULL);
+	}
 
 	ut_ad(block);
 

buf/buf0flu.c

@@ -55,13 +55,29 @@ buf_flush_insert_into_flush_list(
 	ut_ad(mutex_own(&(buf_pool->mutex)));
 #endif /* UNIV_SYNC_DEBUG */
 
-	ut_a(buf_page_in_file(bpage));
-
 	ut_ad((UT_LIST_GET_FIRST(buf_pool->flush_list) == NULL)
 	      || (UT_LIST_GET_FIRST(buf_pool->flush_list)->oldest_modification
 		  <= bpage->oldest_modification));
 
-	UT_LIST_ADD_FIRST(list, buf_pool->flush_list, bpage);
+	switch (buf_page_get_state(bpage)) {
+	case BUF_BLOCK_ZIP_PAGE:
+		mutex_enter(&buf_pool->zip_mutex);
+		buf_page_set_state(bpage, BUF_BLOCK_ZIP_DIRTY);
+		mutex_exit(&buf_pool->zip_mutex);
+		UT_LIST_REMOVE(list, buf_pool->zip_clean, bpage);
+		/* fall through */
+	case BUF_BLOCK_ZIP_DIRTY:
+	case BUF_BLOCK_FILE_PAGE:
+		UT_LIST_ADD_FIRST(list, buf_pool->flush_list, bpage);
+		break;
+	case BUF_BLOCK_ZIP_FREE:
+	case BUF_BLOCK_NOT_USED:
+	case BUF_BLOCK_READY_FOR_USE:
+	case BUF_BLOCK_MEMORY:
+	case BUF_BLOCK_REMOVE_HASH:
+		ut_error;
+		return;
+	}
 
 #if defined UNIV_DEBUG || defined UNIV_BUF_DEBUG
 	ut_a(buf_flush_validate_low());
@@ -85,6 +101,25 @@ buf_flush_insert_sorted_into_flush_list(
 	ut_ad(mutex_own(&(buf_pool->mutex)));
 #endif /* UNIV_SYNC_DEBUG */
 
+	switch (buf_page_get_state(bpage)) {
+	case BUF_BLOCK_ZIP_PAGE:
+		mutex_enter(&buf_pool->zip_mutex);
+		buf_page_set_state(bpage, BUF_BLOCK_ZIP_DIRTY);
+		mutex_exit(&buf_pool->zip_mutex);
+		UT_LIST_REMOVE(list, buf_pool->zip_clean, bpage);
+		/* fall through */
+	case BUF_BLOCK_ZIP_DIRTY:
+	case BUF_BLOCK_FILE_PAGE:
+		break;
+	case BUF_BLOCK_ZIP_FREE:
+	case BUF_BLOCK_NOT_USED:
+	case BUF_BLOCK_READY_FOR_USE:
+	case BUF_BLOCK_MEMORY:
+	case BUF_BLOCK_REMOVE_HASH:
+		ut_error;
+		return;
+	}
+
 	prev_b = NULL;
 	b = UT_LIST_GET_FIRST(buf_pool->flush_list);
 
@@ -202,7 +237,9 @@ buf_flush_remove(
 		mutex_enter(&buf_pool->zip_mutex);
 		buf_page_set_state(bpage, BUF_BLOCK_ZIP_PAGE);
 		mutex_exit(&buf_pool->zip_mutex);
-		/* fall through */
+		UT_LIST_REMOVE(list, buf_pool->flush_list, bpage);
+		buf_LRU_insert_zip_clean(bpage);
+		break;
 	case BUF_BLOCK_FILE_PAGE:
 		UT_LIST_REMOVE(list, buf_pool->flush_list, bpage);
 		break;

buf/buf0lru.c

@@ -49,15 +49,20 @@ frames in the buffer pool, we set this to TRUE */
 ibool	buf_lru_switched_on_innodb_mon	= FALSE;
 
 /**********************************************************************
-Takes a block out of the LRU list and page hash table and sets the block
-state to BUF_BLOCK_REMOVE_HASH. */
+Takes a block out of the LRU list and page hash table. */
 static
-void
+enum buf_page_state
 buf_LRU_block_remove_hashed_page(
 /*=============================*/
-	buf_page_t*	bpage);	/* in: block, must contain a file page and
+				/* out: the new state of the block
+				(BUF_BLOCK_ZIP_FREE if the state was
+				BUF_BLOCK_ZIP_PAGE, or BUF_BLOCK_REMOVE_HASH
+				otherwise) */
+	buf_page_t*	bpage,	/* in: block, must contain a file page and
 				be in a state where it can be freed; there
 				may or may not be a hash index to the page */
+	ibool		zip);	/* in: TRUE if should remove also the
+				compressed page of an uncompressed page */
 /**********************************************************************
 Puts a file page whose has no hash index to the free list. */
 static
@@ -139,8 +144,10 @@ buf_LRU_invalidate_tablespace(
 			}
 
 			/* Remove from the LRU list */
-			buf_LRU_block_remove_hashed_page(bpage);
-			buf_LRU_block_free_hashed_page(bpage);
+			if (buf_LRU_block_remove_hashed_page(bpage, TRUE)
+			    != BUF_BLOCK_ZIP_FREE) {
+				buf_LRU_block_free_hashed_page(bpage);
+			}
 		}
 next_page:
 		mutex_exit(block_mutex);
@@ -191,6 +198,40 @@ buf_LRU_get_recent_limit(void)
 	return(limit);
 }
 
+/************************************************************************
+Insert a compressed block into buf_pool->zip_clean in the LRU order. */
+
+void
+buf_LRU_insert_zip_clean(
+/*=====================*/
+	buf_page_t*	bpage)	/* in: pointer to the block in question */
+{
+	buf_page_t*	b;
+
+#ifdef UNIV_SYNC_DEBUG
+	ut_a(mutex_own(&buf_pool->mutex));
+#endif /* UNIV_SYNC_DEBUG */
+	ut_ad(buf_page_get_state(bpage) == BUF_BLOCK_ZIP_PAGE);
+
+	/* Find the first successor of bpage in the LRU list
+	that is in the zip_clean list. */
+	b = bpage;
+	do {
+		b = UT_LIST_GET_NEXT(LRU, b);
+	} while (b && buf_page_get_state(b) != BUF_BLOCK_ZIP_PAGE);
+
+	/* Insert bpage before b, i.e., after the predecessor of b. */
+	if (b) {
+		b = UT_LIST_GET_PREV(list, b);
+	}
+
+	if (b) {
+		UT_LIST_INSERT_AFTER(list, buf_pool->zip_clean, b, bpage);
+	} else {
+		UT_LIST_ADD_FIRST(list, buf_pool->zip_clean, bpage);
+	}
+}
+
 /**********************************************************************
 Try to free a block. */
 
@@ -198,7 +239,9 @@ ibool
 buf_LRU_free_block(
 /*===============*/
 				/* out: TRUE if freed */
-	buf_page_t*	bpage)	/* in: block to be freed */
+	buf_page_t*	bpage,	/* in: block to be freed */
+	ibool		zip)	/* in: TRUE if should remove also the
+				compressed page of an uncompressed page */
 {
 	mutex_t*	block_mutex = buf_page_get_mutex(bpage);
 
@@ -223,12 +266,8 @@ buf_LRU_free_block(
 	}
 #endif /* UNIV_DEBUG */
 
-	buf_LRU_block_remove_hashed_page(bpage);
-
-	switch (buf_page_get_state(bpage)) {
-	case BUF_BLOCK_REMOVE_HASH:
-		/* The state was changed from BUF_BLOCK_FILE_PAGE
-		in buf_LRU_block_remove_hashed_page(bpage). */
+	if (buf_LRU_block_remove_hashed_page(bpage, zip)
+	    != BUF_BLOCK_ZIP_FREE) {
 		mutex_exit(&(buf_pool->mutex));
 		mutex_exit(block_mutex);
 
@@ -241,22 +280,6 @@ buf_LRU_free_block(
 		mutex_enter(block_mutex);
 
 		buf_LRU_block_free_hashed_page(bpage);
-		break;
-
-	case BUF_BLOCK_ZIP_PAGE:
-		ut_ad(!bpage->in_free_list);
-		ut_ad(!bpage->in_LRU_list);
-
-		UT_LIST_REMOVE(list, buf_pool->zip_clean, bpage);
-
-		buf_buddy_free(bpage->zip.data,
-			       page_zip_get_size(&bpage->zip));
-		buf_buddy_free(bpage, sizeof(*bpage));
-		break;
-
-	default:
-		ut_error;
-		break;
 	}
 
 	return(TRUE);
@@ -290,7 +313,7 @@ buf_LRU_search_and_free_block(
 		mutex_t*	block_mutex = buf_page_get_mutex(bpage);
 
 		mutex_enter(block_mutex);
-		freed = buf_LRU_free_block(bpage);
+		freed = buf_LRU_free_block(bpage, n_iterations > 10);
 		mutex_exit(block_mutex);
 
 		if (freed) {
@@ -496,29 +519,23 @@ buf_LRU_get_free_block(
 	}
 
 	/* If there is a block in the free list, take it */
-	if (UT_LIST_GET_LEN(buf_pool->free) > 0) {
-
-		block = buf_LRU_get_free_only();
-		ut_a(block); /* We tested that buf_pool->free is nonempty. */
+	block = buf_LRU_get_free_only();
+	if (block) {
 
-		if (buf_block_get_zip_size(block) != zip_size) {
-			page_zip_set_size(&block->page.zip, zip_size);
 #ifdef UNIV_DEBUG
-			block->page.zip.m_start =
+		block->page.zip.m_start =
 #endif /* UNIV_DEBUG */
-				block->page.zip.m_end =
-				block->page.zip.m_nonempty =
-				block->page.zip.n_blobs = 0;
-			if (block->page.zip.data) {
-				ut_free(block->page.zip.data);
-			}
+			block->page.zip.m_end =
+			block->page.zip.m_nonempty =
+			block->page.zip.n_blobs = 0;
 
-			if (zip_size) {
-				/* TODO: allocate zip from an aligned pool */
-				block->page.zip.data = ut_malloc(zip_size);
-			} else {
-				block->page.zip.data = NULL;
-			}
+		if (zip_size) {
+			page_zip_set_size(&block->page.zip, zip_size);
+			/* TODO: allocate zip from an aligned pool */
+			block->page.zip.data = ut_malloc(zip_size);
+		} else {
+			page_zip_set_size(&block->page.zip, 0);
+			block->page.zip.data = NULL;
 		}
 
 		mutex_exit(&(buf_pool->mutex));
@@ -951,15 +968,20 @@ buf_LRU_block_free_non_file_page(
 }
 
 /**********************************************************************
-Takes a block out of the LRU list and page hash table and sets the block
-state to BUF_BLOCK_REMOVE_HASH. */
+Takes a block out of the LRU list and page hash table. */
 static
-void
+enum buf_page_state
 buf_LRU_block_remove_hashed_page(
 /*=============================*/
-	buf_page_t*	bpage)	/* in: block, must contain a file page and
+				/* out: the new state of the block
+				(BUF_BLOCK_ZIP_FREE if the state was
+				BUF_BLOCK_ZIP_PAGE, or BUF_BLOCK_REMOVE_HASH
+				otherwise) */
+	buf_page_t*	bpage,	/* in: block, must contain a file page and
 				be in a state where it can be freed; there
 				may or may not be a hash index to the page */
+	ibool		zip)	/* in: TRUE if should remove also the
+				compressed page of an uncompressed page */
 {
 	const buf_page_t*	hashed_bpage;
 	ut_ad(bpage);
@@ -982,8 +1004,14 @@ buf_LRU_block_remove_hashed_page(
 		break;
 	case BUF_BLOCK_ZIP_PAGE:
 		break;
-	default:
+	case BUF_BLOCK_ZIP_FREE:
+	case BUF_BLOCK_ZIP_DIRTY:
+	case BUF_BLOCK_NOT_USED:
+	case BUF_BLOCK_READY_FOR_USE:
+	case BUF_BLOCK_MEMORY:
+	case BUF_BLOCK_REMOVE_HASH:
 		ut_error;
+		break;
 	}
 
 	hashed_bpage = buf_page_hash_get(bpage->space, bpage->offset);
@@ -1018,6 +1046,8 @@ buf_LRU_block_remove_hashed_page(
 		    bpage);
 	switch (buf_page_get_state(bpage)) {
 	case BUF_BLOCK_ZIP_PAGE:
+		ut_ad(!bpage->in_free_list);
+		ut_ad(!bpage->in_LRU_list);
 		ut_a(bpage->zip.data);
 		ut_a(buf_page_get_zip_size(bpage));
 		memset(bpage->zip.data + FIL_PAGE_OFFSET, 0xff, 4);
@@ -1030,10 +1060,27 @@ buf_LRU_block_remove_hashed_page(
 		memset(((buf_block_t*) bpage)->frame
 		       + FIL_PAGE_ARCH_LOG_NO_OR_SPACE_ID, 0xff, 4);
 		buf_page_set_state(bpage, BUF_BLOCK_REMOVE_HASH);
+
+		if (bpage->zip.data) {
+			/* Free the compressed page. */
+			ut_free(bpage->zip.data);
+			bpage->zip.data = NULL;
+			page_zip_set_size(&bpage->zip, 0);
+		}
+
+		return(BUF_BLOCK_REMOVE_HASH);
+
+	case BUF_BLOCK_ZIP_FREE:
+	case BUF_BLOCK_ZIP_DIRTY:
+	case BUF_BLOCK_NOT_USED:
+	case BUF_BLOCK_READY_FOR_USE:
+	case BUF_BLOCK_MEMORY:
+	case BUF_BLOCK_REMOVE_HASH:
 		break;
-	default:
-		ut_error;
 	}
+
+	ut_error;
+	return(BUF_BLOCK_ZIP_FREE);
 }
 
 /**********************************************************************

include/buf0lru.h

@@ -65,6 +65,13 @@ ulint
 buf_LRU_get_recent_limit(void);
 /*==========================*/
 			/* out: the limit; zero if could not determine it */
+/************************************************************************
+Insert a compressed block into buf_pool->zip_clean in the LRU order. */
+
+void
+buf_LRU_insert_zip_clean(
+/*=====================*/
+	buf_page_t*	bpage);	/* in: pointer to the block in question */
 /**********************************************************************
 Try to free a block. */
 
@@ -72,7 +79,9 @@ ibool
 buf_LRU_free_block(
 /*===============*/
 				/* out: TRUE if freed */
-	buf_page_t*	block);	/* in: block to be freed */
+	buf_page_t*	block,	/* in: block to be freed */
+	ibool		zip);	/* in: TRUE if should remove also the
+				compressed page of an uncompressed page */
 /**********************************************************************
 Look for a replaceable block from the end of the LRU list and put it to
 the free list if found. */