Merge lp:~sergei.glushchenko/percona-xtrabackup/xb20-bug932623 into lp:percona-xtrabackup/2.0

[Sergei Glushchenko]

Bug #932623: RENAME TABLE causes incremental prepare to fail
Fixed by applying .delta changes to tablespace with matching ID.
Matching tablespace also renamed to match .delta name.
When tablespace with matching name but different ID exists in full
backup, it renamed to tmp#ID.ibd. When no matching tablespace found,
new one created.

http://jenkins.percona.com/view/XtraBackup/job/percona-xtrabackup-2.0-param/219/

#24669

[Alexey Kopytov (akopytov)]

Sergei,

Tables may be renamed into another database directory, which this fix doesn't take into account. I think we should do space ID verification in a different way:

let's initialize fil_system, in the same --backup or --prepare does it and then use fil_space_get_by_id() and possibly fil_space_get_by_name().

Now the problem is that fil_system will also be initialized later in --prepare when innodb_init() is called, and initializing fil_system multiple times in the same process is tricky. I had to write some extra code for that in the compact backups branch, see xb_data_files_init() and xb_data_files_close() in lp:~akopytov/percona-xtrabackup/compact-backups.

Other minor things:
- s/matchong/matching
- "cat $topdir/inc/ibdata1.meta" in the test case was probably just for debugging and doesn't make any sense as far as regression testing is concerned.

review: Needs Fixing

[Alexey Kopytov (akopytov)]

Forgot to mention that the following code can be simplified:

14 if (fscanf(fp, "page_size = %lu\n", &info->page_size) != 1)
15 r= FALSE;
16 + if (!r || fscanf(fp, "space_id = %lu\n", &info->space_id) != 1)
17 + r= FALSE;

and

31 snprintf(buf, sizeof(buf), "page_size = %lu\n", info->page_size);
32 -
33 len = strlen(buf);
34 +
35 + snprintf(buf + len, sizeof(buf) - len, "space_id = %lu\n", info->space_id);
36 + len += strlen(buf + len);
37 +

[Sergei Glushchenko (sergei.glushchenko)]

It was really tricky to initialize fil_system, especially for built-in InnoDB and for debug builds.

Jenkins build:
http://jenkins.percona.com/view/XtraBackup/job/percona-xtrabackup-2.0-param/225/

[Alexey Kopytov (akopytov)]

Sergei,

Can you explain the problems with initializing fil_system (including the ones in built-in InnoDB) and how you solved them?

[Sergei Glushchenko (sergei.glushchenko)]

Alexey,

1. fil_space_get_by_id and fil_space_get_by_name are for internal usage. I have to modify patches to make them non-static and not inline. Built-in InnoDB does not have fil_space_get_by_id and fil_space_get_by_name at all. Besides we must hold fil_system->mutex when calling these functions.

2. fil_space use hash tables, which are not working without calling mem_init on debug builds.

3. When we are applying deltas we initialize some parts of InnoDB, then releasing and closing them, and then initializing them and some other parts again. The problem with built-it InnoDB is that it doesn't have proper functions to uninitialize fil, mem, and sync. Plugin have some tricks to do it properly (for example we should close sync, which frees all muteness, then close fil, which checks that sync is closed and only after that been done, we close men. But when closing men it use men_hash_mutex, which wasn't closed in sync_close by using special hack). In case of xtrabackup I cheated and not freed mem in many *_close functions, because it will be freed when ut_free_all_mem called.

4. io_handler_thread had an infinite loop:

 for (i = 0;; i++) {
  fil_aio_wait(segment);
 }
and there was no way to finish this thread. I inserted check of srv_shutdown_state.
 for (i = 0; srv_shutdown_state != SRV_SHUTDOWN_EXIT_THREADS; i++) {
    fil_aio_wait(segment);
 }

[Sergei Glushchenko (sergei.glushchenko)]

Spellchecker did bad thing by replacing s/mem/men/ :)

[Alexey Kopytov (akopytov)]

Sergei,

Thanks for clarifications. My comments on the new version of the fix:

   - I wonder if it would be less invasive to define our own simple
     wrappers around "HASH_SEARCH(hash, fil_system->spaces, id, space,
     space->id == id);" so we don't have to patch InnoDB or provide our
     own implementations for built-in InnoDB, and mess with
     mutexes. I.e. can we have xb_space_get_by_*()?

  - the code in xtrabackup's version of io_handler_thread() doesn't use
    the 'i' variable, why not just copy the code from InnoDB, i.e.:

 while (srv_shutdown_state != SRV_SHUTDOWN_EXIT_THREADS) {
  fil_aio_wait(segment);
 }

  - fil_free_all_spaces() seems to be defined for built-in InnoDB, but
    is not used anywhere?

  - in fact, ut_free_all_mem() will not free anything when the system
    malloc() is used instead of the InnoDB own malloc manager, i.e. when
    srv_use_sys_malloc == TRUE, which is the default. I have some
    thoughts on how to fix that, but I suggest we'll do it later as a
    part of bigger refactoring.

  - POSIX guarantees the result of snprintf() to be always
    zero-terminated, so there's no need for things like:

+ snprintf(dest_dir, FN_REFLEN, "%s/%s",
+ xtrabackup_target_dir, dbname);
+ dest_dir[FN_REFLEN - 1] = 0;

    if you care about Windows, you should be using either my_snprintf()
    or ut_snprintf(), which do the right thing on Windows. But
    currently, xtrabackup.c just uses snprintf() in other places anyway,
    so what's the point to do some windows-specific stuff in this
    patch. We can do s/snprintf/ut_snprintf/ later, if and when we
    have to support Windows.

  - this code just uses sprintf() instead snprintf() for some reasons:

+ sprintf(tmpname, "./%s/tmp#%lu", dbname, fil_space->id);

  - I think the output in xb_delta_open_matching_space() is too
    verbose. Let's only print messages when something unexpected
    happens, i.e. remove these messages:

+ msg("xtrabackup: Looking for tablespace %s with id %lu by name\n",
+ dest_space_name, space_id);

and

+ msg("xtrabackup: Looking for tablespace %s with id %lu by id\n",
+ dest_space_name, space_id);

and

+ msg("xtrabackup: Found matching tablespace %s\n",
+ fil_space->name);

and

+ msg("xtrabackup: Created tablespace %s with id %lu.\n",
+ dest_space_name, space_id);

  - what's the point to rename a tablespace with the same name but a
    different ID to tmp#id? the only case when it happens is when a
    tablespace was recreated between the full backup and an incremental
    one. In which case, if we rename the dropped table, we will end up
    with ghost 'tmp...' table in the full backup.

  - this if() block can be merged:

+ if (xtrabackup_incremental) {
+ err = xb_data_files_init();
+ if (err != DB_SUCCESS) {
+ msg("xtrabackup: error: xb_data_files_init() failed with"
+ "error code %lu\n", err);
+ goto error;
+ }
+
+ if(!xtrabackup_apply_deltas(TRUE)) {
+ xb_data_files_close();
+ goto error;
+ }
+ }
+
+ if (xtrabackup_incremental) {
+ xb_data_files_close();
+ }

  - can we also a add a cross-db rename test to bug932623.sh?

review: Needs Fixing

[Sergei Glushchenko (sergei.glushchenko)]

Alexey,

> - I wonder if it would be less invasive to define our own simple
> wrappers around "HASH_SEARCH(hash, fil_system->spaces, id, space,
> space->id == id);" so we don't have to patch InnoDB or provide our
> own implementations for built-in InnoDB, and mess with
> mutexes. I.e. can we have xb_space_get_by_*()?

I've wrote xb_space_by_* wrappers. The only modification of innodb51_builtin.patch left is mem_close function.

>
> - the code in xtrabackup's version of io_handler_thread() doesn't use
> the 'i' variable, why not just copy the code from InnoDB, i.e.:
>
> while (srv_shutdown_state != SRV_SHUTDOWN_EXIT_THREADS) {
> fil_aio_wait(segment);
> }
>

fixed

> - fil_free_all_spaces() seems to be defined for built-in InnoDB, but
> is not used anywhere?
>

Yes. I forgot to put call in proper place, fixed.

> - in fact, ut_free_all_mem() will not free anything when the system
> malloc() is used instead of the InnoDB own malloc manager, i.e. when
> srv_use_sys_malloc == TRUE, which is the default. I have some
> thoughts on how to fix that, but I suggest we'll do it later as a
> part of bigger refactoring.
>

OK. Hopefully memory leaks will not be critical.

> - POSIX guarantees the result of snprintf() to be always
> zero-terminated, so there's no need for things like:
>
> + snprintf(dest_dir, FN_REFLEN, "%s/%s",
> + xtrabackup_target_dir, dbname);
> + dest_dir[FN_REFLEN - 1] = 0;
>
> if you care about Windows, you should be using either my_snprintf()
> or ut_snprintf(), which do the right thing on Windows. But
> currently, xtrabackup.c just uses snprintf() in other places anyway,
> so what's the point to do some windows-specific stuff in this
> patch. We can do s/snprintf/ut_snprintf/ later, if and when we
> have to support Windows.
>

Great! I removed implicit zero-terminating statement.

> - this code just uses sprintf() instead snprintf() for some reasons:
>
> + sprintf(tmpname, "./%s/tmp#%lu", dbname,
> fil_space->id);
>

Hmm... I expected that generated tmpname cannot be longer than FN_REFLEN. I've changed to snprintf for more safety.

> - I think the output in xb_delta_open_matching_space() is too
> verbose. Let's only print messages when something unexpected
> happens, i.e. remove these messages:
>
> + msg("xtrabackup: Looking for tablespace %s with id %lu by name\n",
> + dest_space_name, space_id);
>
> and
>
> + msg("xtrabackup: Looking for tablespace %s with id %lu by id\n",
> + dest_space_name, space_id);
>
> and
>
> + msg("xtrabackup: Found matching tablespace %s\n",
> + fil_space->name);
>
> and
>
> + msg("xtrabackup: Created tablespace %s with id %lu.\n",
> + dest_space_name, space_id);
>

Removed.

>
> - what's the point to rename a tablespace with the same name but a
> different ID to tmp#id? the only case when it happens is when a
> tablespace was recreated between the full backup and an incremental
> one....

[Sergei Glushchenko (sergei.glushchenko)]

New Jenkins build:
http://jenkins.percona.com/view/XtraBackup/job/percona-xtrabackup-2.0-param/226/

[Alexey Kopytov (akopytov)]

Approved. Please when merging the fix to 2.1 move xb_space_get_by_*() and fil_free_all_spaces() to innodb_int.h and innodb_int.c (because those files were created exactly for things like this).

review: Approve

[Laurynas Biveinis (laurynas-biveinis)]

This patch regresses on compressed tablespace creation: the "flags" arg for the fil_space_create() is *not* zip_size. Please do not resubmit this MP. I have follow-up MPs depending on the code and "Approved" status of this one and I will fix this in a follow-up right now.

=== modified file 'patches/innodb51_builtin.patch'
--- patches/innodb51_builtin.patch	2012-02-10 05:04:04 +0000
+++ patches/innodb51_builtin.patch	2012-07-18 15:36:21 +0000
@@ -613,7 +613,21 @@
  
 --- a/storage/innobase/include/mem0mem.h
 +++ b/storage/innobase/include/mem0mem.h
-@@ -401,6 +401,7 @@
+@@ -63,6 +63,13 @@
+ mem_init(
+ /*=====*/
+ 	ulint	size);	/* in: common pool size in bytes */
++
++/******************************************************************//**
++Closes the memory system. */
++void
++mem_close(void);
++/*===========*/
++
+ /******************************************************************
+ Use this macro instead of the corresponding function! Macro for memory
+ heap creation. */
+@@ -401,6 +409,7 @@
  			allocated buffer frame, which can be appended as a
  			free block to the heap, if we need more space;
  			otherwise, this is NULL */
@@ -654,6 +668,21 @@
  
  	if (heap->free_block) {
  		size += UNIV_PAGE_SIZE;
+--- a/storage/innobase/include/mem0pool.h
++++ b/storage/innobase/include/mem0pool.h
+@@ -42,6 +42,12 @@
+ /*============*/
+ 			/* out: memory pool */
+ 	ulint	size);	/* in: pool size in bytes */
++/********************************************************************//**
++Frees a memory pool. */
++void
++mem_pool_free(
++/*==========*/
++	mem_pool_t*	pool);	/*!< in, own: memory pool */
+ /************************************************************************
+ Allocates memory from a pool. NOTE: This low-level function should only be
+ used in mem0mem.*! */
 --- a/storage/innobase/include/srv0srv.h
 +++ b/storage/innobase/include/srv0srv.h
 @@ -60,6 +60,8 @@
@@ -958,7 +987,7 @@
  /**********************************************************
 --- a/storage/innobase/mem/mem0dbg.c
 +++ b/storage/innobase/mem/mem0dbg.c
-@@ -133,6 +133,14 @@
+@@ -133,9 +133,30 @@
  	mem_hash_initialized = TRUE;
  #endif
  
@@ -973,6 +1002,22 @@
  	mem_comm_pool = mem_pool_create(size);
  }
  
++/******************************************************************//**
++Closes the memory system. */
++void
++mem_close(void)
++/*===========*/
++{
++	mem_pool_free(mem_comm_pool);
++	mem_comm_pool = NULL;
++#ifdef UNIV_MEM_DEBUG
++	mem_hash_initialized = FALSE;
++#endif /* UNIV_MEM_DEBUG */
++}
++
+ #ifdef UNIV_MEM_DEBUG
+ /**********************************************************************
+ Initializes an allocated memory field in the debug version. */
 --- a/storage/innobase/mem/mem0mem.c
 +++ b/storage/innobase/mem/mem0mem.c
 @@ -472,6 +472,7 @@
@@ -1026,7 +1071,25 @@
  	pool = ut_malloc(sizeof(mem_pool_t));
  
  	/* We do not set the memory to zero (FALSE) in the pool,
-@@ -333,6 +332,10 @@
+@@ -244,6 +243,17 @@
+ 	return(pool);
+ }
+ 
++/********************************************************************//**
++Frees a memory pool. */
++void
++mem_pool_free(
++/*==========*/
++	mem_pool_t*	pool)	/*!< in, own: memory pool */
++{
++	ut_free(pool->buf);
++	ut_free(pool);
++}
++
+ /************************************************************************
+ Fills the specified free list. */
+ static
+@@ -333,6 +344,10 @@
  	ulint		n;
  	ibool		ret;
  
@@ -1037,7 +1100,7 @@
  	n = ut_2_log(ut_max(size + MEM_AREA_EXTRA_SIZE, MEM_AREA_MIN_SIZE));
  
  	mutex_enter(&(pool->mutex));
-@@ -465,6 +468,11 @@
+@@ -465,6 +480,11 @@
  	ulint		size;
  	ulint		n;
  
 
=== modified file 'src/xtrabackup.c'
--- src/xtrabackup.c	2012-06-20 07:36:26 +0000
+++ src/xtrabackup.c	2012-07-18 15:36:21 +0000
@@ -655,10 +655,107 @@
 
 typedef struct {
 	ulint	page_size;
+	ulint	space_id;
 } xb_delta_info_t;
 
 extern fil_system_t*   fil_system;
 
+/** Value of fil_space_struct::magic_n */
+#define	FIL_SPACE_MAGIC_N	89472
+
+/*******************************************************************//**
+Returns the table space by a given id, NULL if not found. */
+fil_space_t*
+xb_space_get_by_id(
+/*================*/
+	ulint	id);	/*!< in: space id */
+
+/*******************************************************************//**
+Returns the table space by a given name, NULL if not found. */
+fil_space_t*
+xb_space_get_by_name(
+/*==================*/
+	const char*	name);	/*!< in: space name */
+
+/*******************************************************************//**
+Returns the table space by a given id, NULL if not found. */
+fil_space_t*
+xb_space_get_by_id(
+/*================*/
+	ulint	id)	/*!< in: space id */
+{
+	fil_space_t*	space;
+
+	ut_ad(mutex_own(&fil_system->mutex));
+
+#ifdef INNODB_VERSION_SHORT
+	HASH_SEARCH(hash, fil_system->spaces, id,
+		    fil_space_t*, space,
+		    ut_ad(space->magic_n == FIL_SPACE_MAGIC_N),
+		    space->id == id);
+#else
+	HASH_SEARCH(hash, fil_system->spaces, id, space, space->id == id);
+#endif
+
+	return(space);
+}
+
+/*******************************************************************//**
+Returns the table space by a given name, NULL if not found. */
+fil_space_t*
+xb_space_get_by_name(
+/*==================*/
+	const char*	name)	/*!< in: space name */
+{
+	fil_space_t*	space;
+	ulint		fold;
+
+	ut_ad(mutex_own(&fil_system->mutex));
+
+#ifdef INNODB_VERSION_SHORT
+	fold = ut_fold_string(name);
+	HASH_SEARCH(name_hash, fil_system->name_hash, fold,
+		    fil_space_t*, space,
+		    ut_ad(space->magic_n == FIL_SPACE_MAGIC_N),
+		    !strcmp(name, space->name));
+#else
+	HASH_SEARCH(name_hash, fil_system->name_hash, ut_fold_string(name),
+		    space, 0 == strcmp(name, space->name));
+#endif
+
+	return(space);
+}
+
+#ifndef INNODB_VERSION_SHORT
+
+/*******************************************************************//**
+Free all spaces in space_list. */
+void
+fil_free_all_spaces(void)
+/*=====================*/
+{
+	fil_space_t*	space;
+
+	mutex_enter(&fil_system->mutex);
+
+	space = UT_LIST_GET_FIRST(fil_system->space_list);
+
+	while (space != NULL) {
+		fil_node_t*	node;
+		fil_space_t*	prev_space = space;
+
+		space = UT_LIST_GET_NEXT(space_list, space);
+
+		fil_space_free(prev_space->id, FALSE);
+	}
+
+	mutex_exit(&fil_system->mutex);
+}
+
+#define SRV_SHUTDOWN_NONE 0
+
+#endif
+
 /* ==end=== definition  at fil0fil.c === */
 
 
@@ -2752,7 +2849,8 @@
 		return(TRUE);
 	}
 
-	if (fscanf(fp, "page_size = %lu\n", &info->page_size) != 1)
+	if (fscanf(fp, "page_size = %lu\nspace_id = %lu\n",
+		&info->page_size, &info->space_id) != 2)
 		r= FALSE;
 
 	fclose(fp);
@@ -2770,14 +2868,16 @@
 {
 	datasink_t	*ds = ds_ctxt->datasink;
 	ds_file_t	*f;
-	char		buf[32];
+	char		buf[64];
 	my_bool		ret;
 	size_t		len;
 	MY_STAT		mystat;
 
-	snprintf(buf, sizeof(buf), "page_size = %lu\n", info->page_size);
-
+	snprintf(buf, sizeof(buf),
+		 "page_size = %lu\nspace_id = %lu\n",
+		 info->page_size, info->space_id);
 	len = strlen(buf);
+
 	mystat.st_size = len;
 	mystat.st_mtime = my_time(0);
 
@@ -2981,6 +3081,7 @@
 	ds_file_t	*dstfile = NULL;
 
 	info.page_size = 0;
+	info.space_id = 0;
 
 #ifdef XTRADB_BASED
 	if (xtrabackup_tables && (!trx_sys_sys_space(node->space->id)))
@@ -3163,6 +3264,7 @@
 		page_in_buffer++;
 
 		info.page_size = page_size;
+		info.space_id = node->space->id;
 	} else
 		info.page_size = 0;
 
@@ -3722,11 +3824,10 @@
 	void*	arg)
 {
 	ulint	segment;
-	ulint	i;
 	
 	segment = *((ulint*)arg);
 
-	for (i = 0;; i++) {
+	while (srv_shutdown_state != SRV_SHUTDOWN_EXIT_THREADS) {
 		fil_aio_wait(segment);
 	}
 
@@ -3744,6 +3845,141 @@
 #endif
 }
 
+#define SRV_N_PENDING_IOS_PER_THREAD 	OS_AIO_N_PENDING_IOS_PER_THREAD
+#define SRV_MAX_N_PENDING_SYNC_IOS	100
+
+/************************************************************************
+Initialize the tablespace memory cache and populate it by scanning for and
+opening data files.
+@returns DB_SUCCESS or error code.*/
+ulint
+xb_data_files_init(void)
+/*====================*/
+{
+	ulint	i;
+	ibool	create_new_db;
+#ifdef XTRADB_BASED
+	ibool	create_new_doublewrite_file;
+#endif
+	ulint	err;
+	LSN64	min_flushed_lsn;
+	LSN64	max_flushed_lsn;
+	ulint   sum_of_new_sizes;
+
+#ifndef INNODB_VERSION_SHORT
+	os_aio_init(8 * SRV_N_PENDING_IOS_PER_THREAD
+		    * srv_n_file_io_threads,
+		    srv_n_file_io_threads,
+		    SRV_MAX_N_PENDING_SYNC_IOS);
+
+	fil_init(srv_max_n_open_files);
+#else
+	srv_n_file_io_threads = 2 + srv_n_read_io_threads +
+		srv_n_write_io_threads;
+
+	os_aio_init(8 * SRV_N_PENDING_IOS_PER_THREAD,
+		    srv_n_read_io_threads,
+		    srv_n_write_io_threads,
+		    SRV_MAX_N_PENDING_SYNC_IOS);
+
+	fil_init(srv_file_per_table ? 50000 : 5000,
+		 srv_max_n_open_files);
+#endif
+
+	fsp_init();
+
+	for (i = 0; i < srv_n_file_io_threads; i++) {
+		thread_nr[i] = i;
+
+		os_thread_create(io_handler_thread, thread_nr + i,
+				 thread_ids + i);
+    	}
+
+	os_thread_sleep(200000); /*0.2 sec*/
+
+	err = open_or_create_data_files(&create_new_db,
+#ifdef XTRADB_BASED
+					&create_new_doublewrite_file,
+#endif
+					&min_flushed_lsn, &max_flushed_lsn,
+					&sum_of_new_sizes);
+	if (err != DB_SUCCESS) {
+		msg("xtrabackup: Could not open or create data files.\n"
+		    "xtrabackup: If you tried to add new data files, and it "
+		    "failed here,\n"
+		    "xtrabackup: you should now edit innodb_data_file_path in "
+		    "my.cnf back\n"
+		    "xtrabackup: to what it was, and remove the new ibdata "
+		    "files InnoDB created\n"
+		    "xtrabackup: in this failed attempt. InnoDB only wrote "
+		    "those files full of\n"
+		    "xtrabackup: zeros, but did not yet use them in any way. "
+		    "But be careful: do not\n"
+		    "xtrabackup: remove old data files which contain your "
+		    "precious data!\n");
+		return(err);
+	}
+
+	/* create_new_db must not be TRUE.. */
+	if (create_new_db) {
+		msg("xtrabackup: could not find data files at the "
+		    "specified datadir\n");
+		return(DB_ERROR);
+	}
+
+	return(fil_load_single_table_tablespaces());
+}
+
+/************************************************************************
+Destroy the tablespace memory cache. */
+void
+xb_data_files_close(void)
+/*====================*/
+{
+	ulint	i;
+
+	/* Shutdown the aio threads. This has been copied from
+	innobase_shutdown_for_mysql(). */
+
+	srv_shutdown_state = SRV_SHUTDOWN_EXIT_THREADS;
+
+	for (i = 0; i < 1000; i++) {
+		os_aio_wake_all_threads_at_shutdown();
+
+		os_mutex_enter(os_sync_mutex);
+
+		if (os_thread_count == 0) {
+
+			os_mutex_exit(os_sync_mutex);
+
+			os_thread_sleep(10000);
+
+			break;
+		}
+
+		os_mutex_exit(os_sync_mutex);
+
+		os_thread_sleep(10000);
+	}
+
+	if (i == 1000) {
+		msg("xtrabackup: Warning: %lu threads created by InnoDB"
+		    " had not exited at shutdown!\n",
+		    (ulong) os_thread_count);
+	}
+
+#ifdef INNODB_VERSION_SHORT
+	os_aio_free();
+#endif
+	fil_close_all_files();
+#ifndef INNODB_VERSION_SHORT
+	fil_free_all_spaces();
+#endif
+	fil_system = NULL;
+
+	srv_shutdown_state = SRV_SHUTDOWN_NONE;
+}
+
 /***************************************************************************
 Creates an output directory for a given tablespace, if it does not exist */
 static
@@ -4050,85 +4286,25 @@
 	srv_general_init();
 
 	{
-	ibool	create_new_db;
-#ifdef XTRADB_BASED
-	ibool	create_new_doublewrite_file;
-#endif
 	ibool	log_file_created;
 	ibool	log_created	= FALSE;
 	ibool	log_opened	= FALSE;
-	LSN64	min_flushed_lsn;
-	LSN64	max_flushed_lsn;
-	ulint   sum_of_new_sizes;
 	ulint	err;
 	ulint	i;
 
-
-
-
-#define SRV_N_PENDING_IOS_PER_THREAD 	OS_AIO_N_PENDING_IOS_PER_THREAD
-#define SRV_MAX_N_PENDING_SYNC_IOS	100
-
-#ifndef INNODB_VERSION_SHORT
-                os_aio_init(8 * SRV_N_PENDING_IOS_PER_THREAD
-                            * srv_n_file_io_threads,
-                            srv_n_file_io_threads,
-                            SRV_MAX_N_PENDING_SYNC_IOS);
-
-	fil_init(srv_max_n_open_files);
-#else
-	srv_n_file_io_threads = 2 + srv_n_read_io_threads + srv_n_write_io_threads;
-
-	os_aio_init(8 * SRV_N_PENDING_IOS_PER_THREAD,
-		    srv_n_read_io_threads,
-		    srv_n_write_io_threads,
-		    SRV_MAX_N_PENDING_SYNC_IOS);
-
-	fil_init(srv_file_per_table ? 50000 : 5000,
-		 srv_max_n_open_files);
-#endif
-
-	fsp_init();
+	err = xb_data_files_init();
+	if (err != DB_SUCCESS) {
+		msg("xtrabackup: error: xb_data_files_init() failed with"
+		    "error code %lu\n", err);
+		exit(EXIT_FAILURE);
+	}
+
 	log_init();
 
 	lock_sys_create(srv_lock_table_size);
 
-	for (i = 0; i < srv_n_file_io_threads; i++) {
-		thread_nr[i] = i;
-
-		os_thread_create(io_handler_thread, thread_nr + i, thread_ids + i);
-    	}
-
-	os_thread_sleep(200000); /*0.2 sec*/
-
-	err = open_or_create_data_files(&create_new_db,
-#ifdef XTRADB_BASED
-					&create_new_doublewrite_file,
-#endif
-					&min_flushed_lsn, &max_flushed_lsn,
-					&sum_of_new_sizes);
-	if (err != DB_SUCCESS) {
-	        msg(
-"xtrabackup: Could not open or create data files.\n"
-"xtrabackup: If you tried to add new data files, and it failed here,\n"
-"xtrabackup: you should now edit innodb_data_file_path in my.cnf back\n"
-"xtrabackup: to what it was, and remove the new ibdata files InnoDB created\n"
-"xtrabackup: in this failed attempt. InnoDB only wrote those files full of\n"
-"xtrabackup: zeros, but did not yet use them in any way. But be careful: do not\n"
-"xtrabackup: remove old data files which contain your precious data!\n");
-
-		//return((int) err);
-		exit(EXIT_FAILURE);
-	}
-
-	/* create_new_db must not be TRUE.. */
-	if (create_new_db) {
-		msg("xtrabackup: Something wrong with source files...\n");
-		exit(EXIT_FAILURE);
-	}
-
 	for (i = 0; i < srv_n_log_files; i++) {
-		err = open_or_create_log_file(create_new_db, &log_file_created,
+		err = open_or_create_log_file(FALSE, &log_file_created,
 							     log_opened, 0, i);
 		if (err != DB_SUCCESS) {
 
@@ -4141,8 +4317,7 @@
 		} else {
 			log_opened = TRUE;
 		}
-		if ((log_opened && create_new_db)
-			    		|| (log_opened && log_created)) {
+		if ((log_opened && log_created)) {
 			msg(
 	"xtrabackup: Error: all log files must be created at the same time.\n"
 	"xtrabackup: All log files must be created also in database creation.\n"
@@ -4162,8 +4337,6 @@
 		exit(EXIT_FAILURE);
 	}
 
-	fil_load_single_table_tablespaces();
-
 	}
 
 	/* create extra LSN dir if it does not exist. */
@@ -4545,6 +4718,8 @@
 	msg("xtrabackup: Transaction log of lsn (%llu) to (%llu) was copied.\n",
 	    checkpoint_lsn_start, log_copy_scanned_lsn);
 #endif
+
+	xb_data_files_close();
 }
 
 /* ================= stats ================= */
@@ -5434,6 +5609,154 @@
 	return TRUE;
 }
 
+/***********************************************************************
+Searches for matching tablespace file for given .delta file and space_id
+in given directory. When matching tablespace found, renames it to match the
+name of .delta file. If there was a tablespace with matching name and
+mismatching ID, renames it to xtrabackup_tmp_#ID.ibd. If there was no
+matching file, creates the new one.
+@return file handle of matched or created file */
+static
+os_file_t
+xb_delta_open_matching_space(
+	const char*	dbname,		/* in: path to destination database dir */
+	const char*	name,		/* in: name of delta file (without .delta) */
+	ulint		space_id,	/* in: space id of delta file */
+	ulint		zip_size,	/* in: zip_size of tablespace */
+	char*		real_name,	/* out: full path of destination file */
+	size_t		real_name_len,	/* out: buffer size for real_name */
+	ibool* 		success)	/* out: indicates error. TRUE = success */
+{
+	char		dest_dir[FN_REFLEN];
+	char		dest_space_name[FN_REFLEN];
+	ibool		ok;
+	fil_space_t*	fil_space;
+	os_file_t	file	= 0;
+
+	ut_a(dbname || space_id == 0);
+
+	*success = FALSE;
+
+	if (dbname) {
+		snprintf(dest_dir, FN_REFLEN, "%s/%s",
+			xtrabackup_target_dir, dbname);
+		srv_normalize_path_for_win(dest_dir);
+
+		snprintf(dest_space_name, FN_REFLEN, "./%s/%s",
+			dbname, name);
+	} else {
+		snprintf(dest_dir, FN_REFLEN, "%s", xtrabackup_target_dir);
+		srv_normalize_path_for_win(dest_dir);
+
+		snprintf(dest_space_name, FN_REFLEN, "./%s", name);
+	}
+
+	snprintf(real_name, real_name_len,
+		 "%s/%s",
+		 xtrabackup_target_dir, dest_space_name);
+	srv_normalize_path_for_win(real_name);
+
+	/* Create the database directory if it doesn't exist yet */
+	if (!os_file_create_directory(dest_dir, FALSE)) {
+		msg("xtrabackup: error: cannot create dir %s\n", dest_dir);
+		return file;
+	}
+
+	mutex_enter(&fil_system->mutex);
+	fil_space = xb_space_get_by_name(dest_space_name);
+	mutex_exit(&fil_system->mutex);
+
+	if (fil_space != NULL) {
+		if (fil_space->id == space_id) {
+			/* we found matching space */
+			goto found;
+		} else {
+
+			char	tmpname[FN_REFLEN];
+
+			snprintf(tmpname, FN_REFLEN, "./%s/xtrabackup_tmp_#%lu",
+				dbname, fil_space->id);
+
+			msg("xtrabackup: Renaming %s to %s.ibd\n",
+				fil_space->name, tmpname);
+
+			if (!fil_rename_tablespace(NULL,
+					fil_space->id, tmpname))
+			{
+				msg("xtrabackup: Cannot rename %s to %s\n",
+					fil_space->name, tmpname);
+				goto exit;
+			}
+		}
+	}
+
+	mutex_enter(&fil_system->mutex);
+	fil_space = xb_space_get_by_id(space_id);
+	mutex_exit(&fil_system->mutex);
+	if (fil_space != NULL) {
+		char	tmpname[FN_REFLEN];
+
+		strncpy(tmpname, dest_space_name, FN_REFLEN);
+		tmpname[strlen(tmpname) - 4] = 0;
+
+		msg("xtrabackup: Renaming %s to %s\n",
+		    fil_space->name, dest_space_name);
+
+		if (!fil_rename_tablespace(NULL,
+				      fil_space->id, tmpname))
+		{
+			msg("xtrabackup: Cannot rename %s to %s\n",
+				fil_space->name, dest_space_name);
+			goto exit;
+		}
+
+		goto found;
+	}
+
+	/* no matching space found. create the new one */
+
+#ifdef INNODB_VERSION_SHORT
+	if (!fil_space_create(dest_space_name, space_id,
+		zip_size, FIL_TABLESPACE)) {
+#else
+	if (!fil_space_create(dest_space_name, space_id,
+		FIL_TABLESPACE)) {
+#endif
+		msg("xtrabackup: Cannot create tablespace %s\n",
+			dest_space_name);
+		goto exit;
+	}
+
+	file = xb_file_create_no_error_handling(real_name, OS_FILE_CREATE,
+					    OS_FILE_READ_WRITE,
+					    &ok);
+
+	if (ok) {
+		*success = TRUE;
+	} else {
+		msg("xtrabackup: Cannot open file %s\n", real_name);
+	}
+
+	goto exit;
+
+found:
+	/* open the file and return it's handle */
+
+	file = xb_file_create_no_error_handling(real_name, OS_FILE_OPEN,
+					    OS_FILE_READ_WRITE,
+					    &ok);
+
+	if (ok) {
+		*success = TRUE;
+	} else {
+		msg("xtrabackup: Cannot open file %s\n", real_name);
+	}
+
+exit:
+
+	return file;
+}
+
 /************************************************************************
 Applies a given .delta file to the corresponding data file.
 @return TRUE on success */
@@ -5451,6 +5774,7 @@
 	char	src_path[FN_REFLEN];
 	char	dst_path[FN_REFLEN];
 	char	meta_path[FN_REFLEN];
+	char	space_name[FN_REFLEN];
 	ibool	success;
 
 	ibool	last_buffer = FALSE;
@@ -5478,6 +5802,9 @@
 	}
 	dst_path[strlen(dst_path) - 6] = '\0';
 
+	strncpy(space_name, filename, FN_REFLEN);
+	space_name[strlen(space_name) -  6] = 0;
+
 	if (!get_meta_path(src_path, meta_path)) {
 		goto error;
 	}
@@ -5517,36 +5844,11 @@
 
 	xb_file_set_nocache(src_file, src_path, "OPEN");
 
-	dst_file = xb_file_create_no_error_handling(dst_path, OS_FILE_OPEN,
-						    OS_FILE_READ_WRITE,
-						    &success);
-again:
+	dst_file = xb_delta_open_matching_space(
+			dbname, space_name, info.space_id,
+			info.page_size == UNIV_PAGE_SIZE ? 0 : info.page_size,
+			dst_path, sizeof(dst_path), &success);
 	if (!success) {
-		ulint errcode = os_file_get_last_error(TRUE);
-
-		if (errcode == OS_FILE_NOT_FOUND) {
-			msg("xtrabackup: target data file %s "
-			    "is not found, creating a new one\n", dst_path);
-			/* Create the database directory if it doesn't exist yet
-			*/
-			if (dbname) {
-				char	dst_dir[FN_REFLEN];
-
-				snprintf(dst_dir, sizeof(dst_dir), "%s/%s",
-					 xtrabackup_real_target_dir, dbname);
-				srv_normalize_path_for_win(dst_dir);
-
-				if (!os_file_create_directory(dst_dir, FALSE))
-					goto error;
-			}
-			dst_file =
-				xb_file_create_no_error_handling(dst_path,
-								 OS_FILE_CREATE,
-								 OS_FILE_READ_WRITE,
-								 &success);
-			goto again;
-		}
-
 		msg("xtrabackup: error: cannot open %s\n", dst_path);
 		goto error;
 	}
@@ -5563,7 +5865,7 @@
 	incremental_buffer = ut_align(incremental_buffer_base,
 				      UNIV_PAGE_SIZE_MAX);
 
-	msg("Applying %s ...\n", src_path);
+	msg("Applying %s to %s...\n", src_path, dst_path);
 
 	while (!last_buffer) {
 		ulint cluster_header;
@@ -5873,6 +6175,8 @@
 static void
 xtrabackup_prepare_func(void)
 {
+	ulint	err;
+
 	/* cd to target-dir */
 
 	if (my_setwd(xtrabackup_real_target_dir,MYF(MY_WME)))
@@ -5943,12 +6247,29 @@
 	if(xtrabackup_init_temp_log())
 		goto error;
 
-	if(xtrabackup_incremental && !xtrabackup_apply_deltas(TRUE))
+	if(innodb_init_param())
 		goto error;
 
+	mem_init(srv_mem_pool_size);
+	if (xtrabackup_incremental) {
+		err = xb_data_files_init();
+		if (err != DB_SUCCESS) {
+			msg("xtrabackup: error: xb_data_files_init() failed with"
+			    "error code %lu\n", err);
+			goto error;
+		}
+
+		if(!xtrabackup_apply_deltas(TRUE)) {
+			xb_data_files_close();
+			goto error;
+		}
+
+		xb_data_files_close();
+	}
 	sync_close();
 	sync_initialized = FALSE;
 	os_sync_free();
+	mem_close();
 	os_sync_mutex = NULL;
 	ut_free_all_mem();
 
 
=== added file 'test/t/bug932623.sh'
--- test/t/bug932623.sh	1970-01-01 00:00:00 +0000
+++ test/t/bug932623.sh	2012-07-18 15:36:21 +0000
@@ -0,0 +1,85 @@
+########################################################################
+# Bug #932623: RENAME TABLE causes incremental prepare to fail
+########################################################################
+
+. inc/common.sh
+
+start_server --innodb_file_per_table
+
+run_cmd $MYSQL $MYSQL_ARGS test <<EOF
+CREATE TABLE t1_old(a INT) ENGINE=InnoDB;
+INSERT INTO t1_old VALUES (1), (2), (3);
+
+CREATE TABLE t1(a INT) ENGINE=InnoDB;
+INSERT INTO t1 VALUES (4), (5), (6);
+EOF
+
+# Full backup
+# backup root directory
+vlog "Starting backup"
+innobackupex  --no-timestamp $topdir/full
+
+vlog "Rotating the table"
+
+run_cmd $MYSQL $MYSQL_ARGS test <<EOF
+CREATE TABLE t1_new(a int) ENGINE=InnoDB;
+INSERT INTO t1_new VALUES (7), (8), (9);
+
+CREATE DATABASE db2;
+RENAME TABLE t1_old TO db2.t1;
+
+RENAME TABLE t1 TO t1_old;
+RENAME TABLE t1_new TO t1;
+
+INSERT INTO t1_old VALUES (10), (11), (12);
+INSERT INTO t1 VALUES (13), (14), (15);
+
+EOF
+
+vlog "Creating incremental backup"
+
+innobackupex --incremental --no-timestamp \
+    --incremental-basedir=$topdir/full $topdir/inc
+
+vlog "Preparing backup"
+
+innobackupex --apply-log --redo-only $topdir/full
+vlog "Log applied to full backup"
+
+innobackupex --apply-log --redo-only --incremental-dir=$topdir/inc \
+    $topdir/full
+vlog "Delta applied to full backup"
+
+innobackupex --apply-log $topdir/full
+vlog "Data prepared for restore"
+
+checksum_t1_a=`checksum_table test t1`
+checksum_t1_old_a=`checksum_table test t1_old`
+checksum_t1_db2_a=`checksum_table db2 t1`
+
+# Destroying mysql data
+stop_server
+rm -rf $mysql_datadir/*
+vlog "Data destroyed"
+
+# Restore backup
+vlog "Copying files"
+
+innobackupex --copy-back $topdir/full
+vlog "Data restored"
+
+start_server --innodb_file_per_table
+
+vlog "Checking checksums"
+checksum_t1_b=`checksum_table test t1`
+checksum_t1_old_b=`checksum_table test t1_old`
+checksum_t1_db2_b=`checksum_table db2 t1`
+
+if [ "$checksum_t1_a" != "$checksum_t1_b" -o "$checksum_t1_old_a" != "$checksum_t1_old_b" \
+	-o "$checksum_t1_db2_a" != "$checksum_t1_db2_b" ]
+then 
+	vlog "Checksums do not match"
+	exit -1
+fi
+
+vlog "Checksums are OK"