Merge mysql-5.1-innodb -> mysql-5.1

storage/innobase/include/sync0arr.h

@@ -93,10 +93,13 @@ sync_arr_wake_threads_if_sema_free(void);
 Prints warnings of long semaphore waits to stderr. */
 
 ibool
-sync_array_print_long_waits(void);
-/*=============================*/
+sync_array_print_long_waits(
+/*========================*/
 				/* out: TRUE if fatal semaphore wait threshold
 				was exceeded */
+	os_thread_id_t*	waiter,	/* out: longest waiting thread */
+	const void**	sema)	/* out: longest-waited-for semaphore */
+	__attribute__((nonnull));
 /************************************************************************
 Validates the integrity of the wait array. Checks
 that the number of reserved cells equals the count variable. */

storage/innobase/srv/srv0srv.c

@@ -2183,6 +2183,12 @@ srv_error_monitor_thread(
 	ulint		fatal_cnt	= 0;
 	dulint		old_lsn;
 	dulint		new_lsn;
+	/* longest waiting thread for a semaphore */
+	os_thread_id_t	waiter		= os_thread_get_curr_id();
+	os_thread_id_t	old_waiter	= waiter;
+	/* the semaphore that is being waited for */
+	const void*	sema		= NULL;
+	const void*	old_sema	= NULL;
 
 	old_lsn = srv_start_lsn;
 
@@ -2227,7 +2233,8 @@ loop:
 
 	sync_arr_wake_threads_if_sema_free();
 
-	if (sync_array_print_long_waits()) {
+	if (sync_array_print_long_waits(&waiter, &sema)
+	    && sema == old_sema && os_thread_eq(waiter, old_waiter)) {
 		fatal_cnt++;
 		if (fatal_cnt > 10) {
 
@@ -2242,6 +2249,8 @@ loop:
 		}
 	} else {
 		fatal_cnt = 0;
+		old_waiter = waiter;
+		old_sema = sema;
 	}
 
 	/* Flush stderr so that a database user gets the output

storage/innobase/sync/sync0arr.c

@@ -916,10 +916,12 @@ sync_arr_wake_threads_if_sema_free(void)
 Prints warnings of long semaphore waits to stderr. */
 
 ibool
-sync_array_print_long_waits(void)
-/*=============================*/
+sync_array_print_long_waits(
+/*========================*/
 				/* out: TRUE if fatal semaphore wait threshold
 				was exceeded */
+	os_thread_id_t*	waiter,	/* out: longest waiting thread */
+	const void**	sema)	/* out: longest-waited-for semaphore */
 {
 	sync_cell_t*	cell;
 	ibool		old_val;
@@ -927,24 +929,40 @@ sync_array_print_long_waits(void)
 	ulint		i;
 	ulint		fatal_timeout = srv_fatal_semaphore_wait_threshold;
 	ibool		fatal = FALSE;
+	double		longest_diff = 0;
 
 	for (i = 0; i < sync_primary_wait_array->n_cells; i++) {
 
+		double	diff;
+		void*	wait_object;
+
 		cell = sync_array_get_nth_cell(sync_primary_wait_array, i);
 
-		if (cell->wait_object != NULL && cell->waiting
-		    && difftime(time(NULL), cell->reservation_time) > 240) {
+		wait_object = cell->wait_object;
+
+		if (wait_object == NULL || !cell->waiting) {
+
+			continue;
+		}
+
+		diff = difftime(time(NULL), cell->reservation_time);
+
+		if (diff > 240) {
 			fputs("InnoDB: Warning: a long semaphore wait:\n",
 			      stderr);
 			sync_array_cell_print(stderr, cell);
 			noticed = TRUE;
 		}
 
-		if (cell->wait_object != NULL && cell->waiting
-		    && difftime(time(NULL), cell->reservation_time)
-		    > fatal_timeout) {
+		if (diff > fatal_timeout) {
 			fatal = TRUE;
 		}
+
+		if (diff > longest_diff) {
+			longest_diff = diff;
+			*sema = wait_object;
+			*waiter = cell->thread;
+		}
 	}
 
 	if (noticed) {

storage/innodb_plugin/ChangeLog

+2011-03-30	The InnoDB Team
+
+	* srv/srv0srv.c, sync/sync0arr.h, sync/sync0arr.c:
+	Fix Bug#11877216 InnoDB too eager to commit suicide on a busy server
+
+2011-03-15	The InnoDB Team
+
+	* btr/btr0cur.c, page/page0zip.c:
+	Fix Bug#11849231 inflateInit() invoked without initializing all memory
+
+2011-02-28	The InnoDB Team
+
+	* btr/btr0sea.c, buf/buf0buf.c, buf/buf0lru.c:
+	Fix Bug#58549 Race condition in buf_LRU_drop_page_hash_for_tablespace()
+	and compressed tables
+
 2011-02-15	The InnoDB Team
 
 	* sync/sync0rw.c, innodb_bug59307.test:

storage/innodb_plugin/btr/btr0cur.c

@@ -4627,27 +4627,45 @@ btr_copy_blob_prefix(
 
 /*******************************************************************//**
 Copies the prefix of a compressed BLOB.  The clustered index record
-that points to this BLOB must be protected by a lock or a page latch. */
+that points to this BLOB must be protected by a lock or a page latch.
+@return	number of bytes written to buf */
 static
-void
+ulint
 btr_copy_zblob_prefix(
 /*==================*/
-	z_stream*	d_stream,/*!< in/out: the decompressing stream */
+	byte*		buf,	/*!< out: the externally stored part of
+				the field, or a prefix of it */
+	ulint		len,	/*!< in: length of buf, in bytes */
 	ulint		zip_size,/*!< in: compressed BLOB page size */
 	ulint		space_id,/*!< in: space id of the BLOB pages */
 	ulint		page_no,/*!< in: page number of the first BLOB page */
 	ulint		offset)	/*!< in: offset on the first BLOB page */
 {
 	ulint		page_type = FIL_PAGE_TYPE_ZBLOB;
+	mem_heap_t*	heap;
+	int		err;
+	z_stream	d_stream;
+
+	d_stream.next_out = buf;
+	d_stream.avail_out = len;
+	d_stream.next_in = Z_NULL;
+	d_stream.avail_in = 0;
+
+	/* Zlib inflate needs 32 kilobytes for the default
+	window size, plus a few kilobytes for small objects. */
+	heap = mem_heap_create(40000);
+	page_zip_set_alloc(&d_stream, heap);
 
 	ut_ad(ut_is_2pow(zip_size));
 	ut_ad(zip_size >= PAGE_ZIP_MIN_SIZE);
 	ut_ad(zip_size <= UNIV_PAGE_SIZE);
 	ut_ad(space_id);
 
+	err = inflateInit(&d_stream);
+	ut_a(err == Z_OK);
+
 	for (;;) {
 		buf_page_t*	bpage;
-		int		err;
 		ulint		next_page_no;
 
 		/* There is no latch on bpage directly.  Instead,
@@ -4663,7 +4681,7 @@ btr_copy_zblob_prefix(
 				" compressed BLOB"
 				" page %lu space %lu\n",
 				(ulong) page_no, (ulong) space_id);
-			return;
+			goto func_exit;
 		}
 
 		if (UNIV_UNLIKELY
@@ -4689,13 +4707,13 @@ btr_copy_zblob_prefix(
 			offset += 4;
 		}
 
-		d_stream->next_in = bpage->zip.data + offset;
-		d_stream->avail_in = zip_size - offset;
+		d_stream.next_in = bpage->zip.data + offset;
+		d_stream.avail_in = zip_size - offset;
 
-		err = inflate(d_stream, Z_NO_FLUSH);
+		err = inflate(&d_stream, Z_NO_FLUSH);
 		switch (err) {
 		case Z_OK:
-			if (!d_stream->avail_out) {
+			if (!d_stream.avail_out) {
 				goto end_of_blob;
 			}
 			break;
@@ -4712,13 +4730,13 @@ inflate_error:
 				" compressed BLOB"
 				" page %lu space %lu returned %d (%s)\n",
 				(ulong) page_no, (ulong) space_id,
-				err, d_stream->msg);
+				err, d_stream.msg);
 		case Z_BUF_ERROR:
 			goto end_of_blob;
 		}
 
 		if (next_page_no == FIL_NULL) {
-			if (!d_stream->avail_in) {
+			if (!d_stream.avail_in) {
 				ut_print_timestamp(stderr);
 				fprintf(stderr,
 					"  InnoDB: unexpected end of"
@@ -4727,7 +4745,7 @@ inflate_error:
 					(ulong) page_no,
 					(ulong) space_id);
 			} else {
-				err = inflate(d_stream, Z_FINISH);
+				err = inflate(&d_stream, Z_FINISH);
 				switch (err) {
 				case Z_STREAM_END:
 				case Z_BUF_ERROR:
@@ -4739,7 +4757,7 @@ inflate_error:
 
 end_of_blob:
 			buf_page_release_zip(bpage);
-			return;
+			goto func_exit;
 		}
 
 		buf_page_release_zip(bpage);
@@ -4751,6 +4769,12 @@ end_of_blob:
 		offset = FIL_PAGE_NEXT;
 		page_type = FIL_PAGE_TYPE_ZBLOB2;
 	}
+
+func_exit:
+	inflateEnd(&d_stream);
+	mem_heap_free(heap);
+	UNIV_MEM_ASSERT_RW(buf, d_stream.total_out);
+	return(d_stream.total_out);
 }
 
 /*******************************************************************//**
@@ -4776,28 +4800,8 @@ btr_copy_externally_stored_field_prefix_low(
 	}
 
 	if (UNIV_UNLIKELY(zip_size)) {
-		int		err;
-		z_stream	d_stream;
-		mem_heap_t*	heap;
-
-		/* Zlib inflate needs 32 kilobytes for the default
-		window size, plus a few kilobytes for small objects. */
-		heap = mem_heap_create(40000);
-		page_zip_set_alloc(&d_stream, heap);
-
-		err = inflateInit(&d_stream);
-		ut_a(err == Z_OK);
-
-		d_stream.next_out = buf;
-		d_stream.avail_out = len;
-		d_stream.avail_in = 0;
-
-		btr_copy_zblob_prefix(&d_stream, zip_size,
-				      space_id, page_no, offset);
-		inflateEnd(&d_stream);
-		mem_heap_free(heap);
-		UNIV_MEM_ASSERT_RW(buf, d_stream.total_out);
-		return(d_stream.total_out);
+		return(btr_copy_zblob_prefix(buf, len, zip_size,
+					     space_id, page_no, offset));
 	} else {
 		return(btr_copy_blob_prefix(buf, len, space_id,
 					    page_no, offset));

storage/innodb_plugin/btr/btr0sea.c

@@ -1201,7 +1201,7 @@ btr_search_drop_page_hash_when_freed(
 	having to fear a deadlock. */
 
 	block = buf_page_get_gen(space, zip_size, page_no, RW_S_LATCH, NULL,
-				BUF_GET_IF_IN_POOL, __FILE__, __LINE__,
+				 BUF_PEEK_IF_IN_POOL, __FILE__, __LINE__,
 				 &mtr);
 	/* Because the buffer pool mutex was released by
 	buf_page_peek_if_search_hashed(), it is possible that the

storage/innodb_plugin/buf/buf0buf.c

@@ -1893,16 +1893,19 @@ buf_block_align(
 	/* TODO: protect buf_pool->chunks with a mutex (it will
 	currently remain constant after buf_pool_init()) */
 	for (chunk = buf_pool->chunks, i = buf_pool->n_chunks; i--; chunk++) {
-		lint	offs = ptr - chunk->blocks->frame;
+		ulint	offs;
 
-		if (UNIV_UNLIKELY(offs < 0)) {
+		if (UNIV_UNLIKELY(ptr < chunk->blocks->frame)) {
 
 			continue;
 		}
+		/* else */
+
+		offs = ptr - chunk->blocks->frame;
 
 		offs >>= UNIV_PAGE_SIZE_SHIFT;
 
-		if (UNIV_LIKELY((ulint) offs < chunk->size)) {
+		if (UNIV_LIKELY(offs < chunk->size)) {
 			buf_block_t*	block = &chunk->blocks[offs];
 
 			/* The function buf_chunk_init() invokes
@@ -2028,7 +2031,7 @@ buf_page_get_gen(
 	ulint		rw_latch,/*!< in: RW_S_LATCH, RW_X_LATCH, RW_NO_LATCH */
 	buf_block_t*	guess,	/*!< in: guessed block or NULL */
 	ulint		mode,	/*!< in: BUF_GET, BUF_GET_IF_IN_POOL,
-				BUF_GET_NO_LATCH */
+				BUF_PEEK_IF_IN_POOL, BUF_GET_NO_LATCH */
 	const char*	file,	/*!< in: file name */
 	ulint		line,	/*!< in: line where called */
 	mtr_t*		mtr)	/*!< in: mini-transaction */
@@ -2044,9 +2047,19 @@ buf_page_get_gen(
 	ut_ad((rw_latch == RW_S_LATCH)
 	      || (rw_latch == RW_X_LATCH)
 	      || (rw_latch == RW_NO_LATCH));
-	ut_ad((mode != BUF_GET_NO_LATCH) || (rw_latch == RW_NO_LATCH));
-	ut_ad((mode == BUF_GET) || (mode == BUF_GET_IF_IN_POOL)
-	      || (mode == BUF_GET_NO_LATCH));
+#ifdef UNIV_DEBUG
+	switch (mode) {
+	case BUF_GET_NO_LATCH:
+		ut_ad(rw_latch == RW_NO_LATCH);
+		break;
+	case BUF_GET:
+	case BUF_GET_IF_IN_POOL:
+	case BUF_PEEK_IF_IN_POOL:
+		break;
+	default:
+		ut_error;
+	}
+#endif /* UNIV_DEBUG */
 	ut_ad(zip_size == fil_space_get_zip_size(space));
 	ut_ad(ut_is_2pow(zip_size));
 #ifndef UNIV_LOG_DEBUG
@@ -2088,7 +2101,8 @@ loop2:
 
 		buf_pool_mutex_exit();
 
-		if (mode == BUF_GET_IF_IN_POOL) {
+		if (mode == BUF_GET_IF_IN_POOL
+		    || mode == BUF_PEEK_IF_IN_POOL) {
 
 			return(NULL);
 		}
@@ -2127,7 +2141,8 @@ loop2:
 
 	must_read = buf_block_get_io_fix(block) == BUF_IO_READ;
 
-	if (must_read && mode == BUF_GET_IF_IN_POOL) {
+	if (must_read && (mode == BUF_GET_IF_IN_POOL
+			  || mode == BUF_PEEK_IF_IN_POOL)) {
 		/* The page is only being read to buffer */
 		buf_pool_mutex_exit();
 
@@ -2245,6 +2260,7 @@ wait_until_unfixed:
 		mutex_exit(&buf_pool_zip_mutex);
 		buf_pool->n_pend_unzip++;
 
+		bpage->state = BUF_BLOCK_ZIP_FREE;
 		buf_buddy_free(bpage, sizeof *bpage);
 
 		buf_pool_mutex_exit();
@@ -2321,7 +2337,9 @@ wait_until_unfixed:
 
 	buf_pool_mutex_exit();
 
+	if (UNIV_LIKELY(mode != BUF_PEEK_IF_IN_POOL)) {
 		buf_page_set_accessed_make_young(&block->page, access_time);
+	}
 
 #if defined UNIV_DEBUG_FILE_ACCESSES || defined UNIV_DEBUG
 	ut_a(!block->page.file_page_was_freed);
@@ -2374,7 +2392,7 @@ wait_until_unfixed:
 
 	mtr_memo_push(mtr, block, fix_type);
 
-	if (!access_time) {
+	if (UNIV_LIKELY(mode != BUF_PEEK_IF_IN_POOL) && !access_time) {
 		/* In the case of a first access, try to apply linear
 		read-ahead */
 
@@ -2923,6 +2941,7 @@ err_exit:
 		    && UNIV_LIKELY_NULL(buf_page_hash_get(space, offset))) {
 
 			/* The block was added by some other thread. */
+			bpage->state = BUF_BLOCK_ZIP_FREE;
 			buf_buddy_free(bpage, sizeof *bpage);
 			buf_buddy_free(data, zip_size);
 

storage/innodb_plugin/buf/buf0lru.c

@@ -246,71 +246,75 @@ buf_LRU_drop_page_hash_for_tablespace(
 	page_arr = ut_malloc(sizeof(ulint)
 			     * BUF_LRU_DROP_SEARCH_HASH_SIZE);
 	buf_pool_mutex_enter();
+	num_entries = 0;
 
 scan_again:
-	num_entries = 0;
 	bpage = UT_LIST_GET_LAST(buf_pool->LRU);
 
 	while (bpage != NULL) {
-		mutex_t*	block_mutex = buf_page_get_mutex(bpage);
 		buf_page_t*	prev_bpage;
+		ibool		is_fixed;
 
-		mutex_enter(block_mutex);
 		prev_bpage = UT_LIST_GET_PREV(LRU, bpage);
 
 		ut_a(buf_page_in_file(bpage));
 
 		if (buf_page_get_state(bpage) != BUF_BLOCK_FILE_PAGE
 		    || bpage->space != id
-		    || bpage->buf_fix_count > 0
 		    || bpage->io_fix != BUF_IO_NONE) {
-			/* We leave the fixed pages as is in this scan.
-			To be dealt with later in the final scan. */
-			mutex_exit(block_mutex);
-			goto next_page;
+			/* Compressed pages are never hashed.
+			Skip blocks of other tablespaces.
+			Skip I/O-fixed blocks (to be dealt with later). */
+next_page:
+			bpage = prev_bpage;
+			continue;
 		}
 
-		if (((buf_block_t*) bpage)->is_hashed) {
+		mutex_enter(&((buf_block_t*) bpage)->mutex);
+		is_fixed = bpage->buf_fix_count > 0
+			|| !((buf_block_t*) bpage)->is_hashed;
+		mutex_exit(&((buf_block_t*) bpage)->mutex);
 
-			/* Store the offset(i.e.: page_no) in the array
-			so that we can drop hash index in a batch
-			later. */
+		if (is_fixed) {
+			goto next_page;
+		}
+
+		/* Store the page number so that we can drop the hash
+		index in a batch later. */
 		page_arr[num_entries] = bpage->offset;
-			mutex_exit(block_mutex);
 		ut_a(num_entries < BUF_LRU_DROP_SEARCH_HASH_SIZE);
 		++num_entries;
 
 		if (num_entries < BUF_LRU_DROP_SEARCH_HASH_SIZE) {
 			goto next_page;
 		}
+
 		/* Array full. We release the buf_pool_mutex to
 		obey the latching order. */
 		buf_pool_mutex_exit();
-
 		buf_LRU_drop_page_hash_batch(id, zip_size, page_arr,
 					     num_entries);
-			num_entries = 0;
 		buf_pool_mutex_enter();
-		} else {
-			mutex_exit(block_mutex);
-		}
+		num_entries = 0;
 
-next_page:
-		/* Note that we may have released the buf_pool mutex
-		above after reading the prev_bpage during processing
-		of a page_hash_batch (i.e.: when the array was full).
-		This means that prev_bpage can change in LRU list.
-		This is OK because this function is a 'best effort'
-		to drop as many search hash entries as possible and
-		it does not guarantee that ALL such entries will be
-		dropped. */
-		bpage = prev_bpage;
+		/* Note that we released the buf_pool mutex above
+		after reading the prev_bpage during processing of a
+		page_hash_batch (i.e.: when the array was full).
+		Because prev_bpage could belong to a compressed-only
+		block, it may have been relocated, and thus the
+		pointer cannot be trusted. Because bpage is of type
+		buf_block_t, it is safe to dereference.
+
+		bpage can change in the LRU list. This is OK because
+		this function is a 'best effort' to drop as many
+		search hash entries as possible and it does not
+		guarantee that ALL such entries will be dropped. */
 
 		/* If, however, bpage has been removed from LRU list
 		to the free list then we should restart the scan.
 		bpage->state is protected by buf_pool mutex. */
-		if (bpage && !buf_page_in_file(bpage)) {
-			ut_a(num_entries == 0);
+		if (bpage
+		    && buf_page_get_state(bpage) != BUF_BLOCK_FILE_PAGE) {
 			goto scan_again;
 		}
 	}
@@ -1799,6 +1803,7 @@ buf_LRU_block_remove_hashed_page(
 		buf_pool_mutex_exit_forbid();
 		buf_buddy_free(bpage->zip.data,
 			       page_zip_get_size(&bpage->zip));
+		bpage->state = BUF_BLOCK_ZIP_FREE;
 		buf_buddy_free(bpage, sizeof(*bpage));
 		buf_pool_mutex_exit_allow();
 		UNIV_MEM_UNDESC(bpage);

storage/innodb_plugin/include/buf0buf.h

@@ -41,6 +41,8 @@ Created 11/5/1995 Heikki Tuuri
 /* @{ */
 #define BUF_GET			10	/*!< get always */
 #define	BUF_GET_IF_IN_POOL	11	/*!< get if in pool */
+#define BUF_PEEK_IF_IN_POOL	12	/*!< get if in pool, do not make
+					the block young in the LRU list */
 #define BUF_GET_NO_LATCH	14	/*!< get and bufferfix, but
 					set no latch; we have
 					separated this case, because
@@ -284,7 +286,7 @@ buf_page_get_gen(
 	ulint		rw_latch,/*!< in: RW_S_LATCH, RW_X_LATCH, RW_NO_LATCH */
 	buf_block_t*	guess,	/*!< in: guessed block or NULL */
 	ulint		mode,	/*!< in: BUF_GET, BUF_GET_IF_IN_POOL,
-				BUF_GET_NO_LATCH */
+				BUF_PEEK_IF_IN_POOL, BUF_GET_NO_LATCH */
 	const char*	file,	/*!< in: file name */
 	ulint		line,	/*!< in: line where called */
 	mtr_t*		mtr);	/*!< in: mini-transaction */

storage/innodb_plugin/include/sync0arr.h

@@ -115,8 +115,11 @@ Prints warnings of long semaphore waits to stderr.
 @return	TRUE if fatal semaphore wait threshold was exceeded */
 UNIV_INTERN
 ibool
-sync_array_print_long_waits(void);
-/*=============================*/
+sync_array_print_long_waits(
+/*========================*/
+	os_thread_id_t*	waiter,	/*!< out: longest waiting thread */
+	const void**	sema)	/*!< out: longest-waited-for semaphore */
+	__attribute__((nonnull));
 /********************************************************************//**
 Validates the integrity of the wait array. Checks
 that the number of reserved cells equals the count variable. */

storage/innodb_plugin/page/page0zip.c

@@ -653,13 +653,13 @@ page_zip_dir_encode(
 Allocate memory for zlib. */
 static
 void*
-page_zip_malloc(
+page_zip_zalloc(
 /*============*/
 	void*	opaque,	/*!< in/out: memory heap */
 	uInt	items,	/*!< in: number of items to allocate */
 	uInt	size)	/*!< in: size of an item in bytes */
 {
-	return(mem_heap_alloc(opaque, items * size));
+	return(mem_heap_zalloc(opaque, items * size));
 }
 
 /**********************************************************************//**
@@ -684,7 +684,7 @@ page_zip_set_alloc(
 {
 	z_stream*	strm = stream;
 
-	strm->zalloc = page_zip_malloc;
+	strm->zalloc = page_zip_zalloc;
 	strm->zfree = page_zip_free;
 	strm->opaque = heap;
 }
@@ -2912,19 +2912,18 @@ zlib_error:
 
 	page_zip_set_alloc(&d_stream, heap);
 
-	if (UNIV_UNLIKELY(inflateInit2(&d_stream, UNIV_PAGE_SIZE_SHIFT)
-			  != Z_OK)) {
-		ut_error;
-	}
-
 	d_stream.next_in = page_zip->data + PAGE_DATA;
 	/* Subtract the space reserved for
 	the page header and the end marker of the modification log. */
 	d_stream.avail_in = page_zip_get_size(page_zip) - (PAGE_DATA + 1);
-
 	d_stream.next_out = page + PAGE_ZIP_START;
 	d_stream.avail_out = UNIV_PAGE_SIZE - PAGE_ZIP_START;
 
+	if (UNIV_UNLIKELY(inflateInit2(&d_stream, UNIV_PAGE_SIZE_SHIFT)
+			  != Z_OK)) {
+		ut_error;
+	}
+
 	/* Decode the zlib header and the index information. */
 	if (UNIV_UNLIKELY(inflate(&d_stream, Z_BLOCK) != Z_OK)) {
 

storage/innodb_plugin/srv/srv0srv.c

@@ -2236,6 +2236,12 @@ srv_error_monitor_thread(
 	ulint		fatal_cnt	= 0;
 	ib_uint64_t	old_lsn;
 	ib_uint64_t	new_lsn;
+	/* longest waiting thread for a semaphore */
+	os_thread_id_t	waiter		= os_thread_get_curr_id();
+	os_thread_id_t	old_waiter	= waiter;
+	/* the semaphore that is being waited for */
+	const void*	sema		= NULL;
+	const void*	old_sema	= NULL;
 
 	old_lsn = srv_start_lsn;
 
@@ -2284,7 +2290,8 @@ loop:
 
 	sync_arr_wake_threads_if_sema_free();
 
-	if (sync_array_print_long_waits()) {
+	if (sync_array_print_long_waits(&waiter, &sema)
+	    && sema == old_sema && os_thread_eq(waiter, old_waiter)) {
 		fatal_cnt++;
 		if (fatal_cnt > 10) {
 
@@ -2299,6 +2306,8 @@ loop:
 		}
 	} else {
 		fatal_cnt = 0;
+		old_waiter = waiter;
+		old_sema = sema;
 	}
 
 	/* Flush stderr so that a database user gets the output

storage/innodb_plugin/sync/sync0arr.c

@@ -914,8 +914,10 @@ Prints warnings of long semaphore waits to stderr.
 @return	TRUE if fatal semaphore wait threshold was exceeded */
 UNIV_INTERN
 ibool
-sync_array_print_long_waits(void)
-/*=============================*/
+sync_array_print_long_waits(
+/*========================*/
+	os_thread_id_t*	waiter,	/*!< out: longest waiting thread */
+	const void**	sema)	/*!< out: longest-waited-for semaphore */
 {
 	sync_cell_t*	cell;
 	ibool		old_val;
@@ -923,24 +925,40 @@ sync_array_print_long_waits(void)
 	ulint		i;
 	ulint		fatal_timeout = srv_fatal_semaphore_wait_threshold;
 	ibool		fatal = FALSE;
+	double		longest_diff = 0;
 
 	for (i = 0; i < sync_primary_wait_array->n_cells; i++) {
 
+		double	diff;
+		void*	wait_object;
+
 		cell = sync_array_get_nth_cell(sync_primary_wait_array, i);
 
-		if (cell->wait_object != NULL && cell->waiting
-		    && difftime(time(NULL), cell->reservation_time) > 240) {
+		wait_object = cell->wait_object;
+
+		if (wait_object == NULL || !cell->waiting) {
+
+			continue;
+		}
+
+		diff = difftime(time(NULL), cell->reservation_time);
+
+		if (diff > 240) {
 			fputs("InnoDB: Warning: a long semaphore wait:\n",
 			      stderr);
 			sync_array_cell_print(stderr, cell);
 			noticed = TRUE;
 		}
 
-		if (cell->wait_object != NULL && cell->waiting
-		    && difftime(time(NULL), cell->reservation_time)
-		    > fatal_timeout) {
+		if (diff > fatal_timeout) {
 			fatal = TRUE;
 		}
+
+		if (diff > longest_diff) {
+			longest_diff = diff;
+			*sema = wait_object;
+			*waiter = cell->thread;
+		}
 	}
 
 	if (noticed) {

storage/innodb_plugin/trx/trx0i_s.c

@@ -508,7 +508,7 @@ fill_trx_row(
 		query[stmt_len] = '\0';
 
 		row->trx_query = ha_storage_put_memlim(
-			cache->storage, stmt, stmt_len + 1,
+			cache->storage, query, stmt_len + 1,
 			MAX_ALLOWED_FOR_STORAGE(cache));
 
 		row->trx_query_cs = innobase_get_charset(trx->mysql_thd);