~tsarev/percona-server/5.5-processlist_rows_stats-sporadic_fails-fix

# name       : innodb_fake_changes.patch

# introduced : 5.5.12

# maintainer : Yasufumi

#

#!!! notice !!!

# Any small change to this file in the main branch

# should be done or reviewed by the maintainer!

diff -ruN a/storage/innobase/btr/btr0cur.c b/storage/innobase/btr/btr0cur.c

--- a/storage/innobase/btr/btr0cur.c	2011-06-01 21:04:26.346416429 +0900

+++ b/storage/innobase/btr/btr0cur.c	2011-06-01 21:04:53.718415294 +0900

@@ -1167,6 +1167,11 @@

 	rec_t*		rec;

 	roll_ptr_t	roll_ptr;

+	if (thr && thr_get_trx(thr)->fake_changes) {

+		/* skip LOCK, UNDO */

+		return(DB_SUCCESS);

+	}

+

 	/* Check if we have to wait for a lock: enqueue an explicit lock

 	request if yes */

@@ -1298,7 +1303,7 @@

 	}

 #endif /* UNIV_DEBUG */

-	ut_ad(mtr_memo_contains(mtr, block, MTR_MEMO_PAGE_X_FIX));

+	ut_ad((thr && thr_get_trx(thr)->fake_changes) || mtr_memo_contains(mtr, block, MTR_MEMO_PAGE_X_FIX));

 	max_size = page_get_max_insert_size_after_reorganize(page, 1);

 	leaf = page_is_leaf(page);

@@ -1393,6 +1398,12 @@

 		goto fail_err;

 	}

+	if (thr && thr_get_trx(thr)->fake_changes) {

+		/* skip CHANGE, LOG */

+		*big_rec = big_rec_vec;

+		return(err); /* == DB_SUCCESS */

+	}

+

 	page_cursor = btr_cur_get_page_cur(cursor);

 	/* Now, try the insert */

@@ -1535,10 +1546,10 @@

 	*big_rec = NULL;

-	ut_ad(mtr_memo_contains(mtr,

+	ut_ad((thr && thr_get_trx(thr)->fake_changes) || mtr_memo_contains(mtr,

 				dict_index_get_lock(btr_cur_get_index(cursor)),

 				MTR_MEMO_X_LOCK));

-	ut_ad(mtr_memo_contains(mtr, btr_cur_get_block(cursor),

+	ut_ad((thr && thr_get_trx(thr)->fake_changes) || mtr_memo_contains(mtr, btr_cur_get_block(cursor),

 				MTR_MEMO_PAGE_X_FIX));

 	/* Try first an optimistic insert; reset the cursor flag: we do not

@@ -1604,6 +1615,16 @@

 		}

 	}

+	if (thr && thr_get_trx(thr)->fake_changes) {

+		/* skip CHANGE, LOG */

+		if (n_extents > 0) {

+			fil_space_release_free_extents(index->space,

+						       n_reserved);

+		}

+		*big_rec = big_rec_vec;

+		return(DB_SUCCESS);

+	}

+

 	if (dict_index_get_page(index)

 	    == buf_block_get_page_no(btr_cur_get_block(cursor))) {

@@ -1660,6 +1681,11 @@

 	ut_ad(cursor && update && thr && roll_ptr);

+	if (thr && thr_get_trx(thr)->fake_changes) {

+		/* skip LOCK, UNDO */

+		return(DB_SUCCESS);

+	}

+

 	rec = btr_cur_get_rec(cursor);

 	index = cursor->index;

@@ -1958,6 +1984,14 @@

 		return(err);

 	}

+	if (trx->fake_changes) {

+		/* skip CHANGE, LOG */

+		if (UNIV_LIKELY_NULL(heap)) {

+			mem_heap_free(heap);

+		}

+		return(err); /* == DB_SUCCESS */

+	}

+

 	if (block->is_hashed) {

 		/* The function row_upd_changes_ord_field_binary works only

 		if the update vector was built for a clustered index, we must

@@ -2061,7 +2095,7 @@

 	rec = btr_cur_get_rec(cursor);

 	index = cursor->index;

 	ut_ad(!!page_rec_is_comp(rec) == dict_table_is_comp(index->table));

-	ut_ad(mtr_memo_contains(mtr, block, MTR_MEMO_PAGE_X_FIX));

+	ut_ad((thr && thr_get_trx(thr)->fake_changes) || mtr_memo_contains(mtr, block, MTR_MEMO_PAGE_X_FIX));

 	/* The insert buffer tree should never be updated in place. */

 	ut_ad(!dict_index_is_ibuf(index));

@@ -2171,6 +2205,11 @@

 		goto err_exit;

 	}

+	if (thr && thr_get_trx(thr)->fake_changes) {

+		/* skip CHANGE, LOG */

+		goto err_exit; /* == DB_SUCCESS */

+	}

+

 	/* Ok, we may do the replacement. Store on the page infimum the

 	explicit locks on rec, before deleting rec (see the comment in

 	btr_cur_pessimistic_update). */

@@ -2321,9 +2360,9 @@

 	rec = btr_cur_get_rec(cursor);

 	index = cursor->index;

-	ut_ad(mtr_memo_contains(mtr, dict_index_get_lock(index),

+	ut_ad((thr && thr_get_trx(thr)->fake_changes) || mtr_memo_contains(mtr, dict_index_get_lock(index),

 				MTR_MEMO_X_LOCK));

-	ut_ad(mtr_memo_contains(mtr, block, MTR_MEMO_PAGE_X_FIX));

+	ut_ad((thr && thr_get_trx(thr)->fake_changes) || mtr_memo_contains(mtr, block, MTR_MEMO_PAGE_X_FIX));

 #ifdef UNIV_ZIP_DEBUG

 	ut_a(!page_zip || page_zip_validate(page_zip, page));

 #endif /* UNIV_ZIP_DEBUG */

@@ -2411,6 +2450,9 @@

 		ut_ad(big_rec_vec == NULL);

+		/* fake_changes should not cause undo. so never reaches here */

+		ut_ad(!(trx->fake_changes));

+

 		btr_rec_free_updated_extern_fields(

 			index, rec, page_zip, offsets, update,

 			trx_is_recv(trx) ? RB_RECOVERY : RB_NORMAL, mtr);

@@ -2445,6 +2487,12 @@

 		}

 	}

+	if (trx->fake_changes) {

+		/* skip CHANGE, LOG */

+		err = DB_SUCCESS;

+		goto return_after_reservations;

+	}

+

 	/* Store state of explicit locks on rec on the page infimum record,

 	before deleting rec. The page infimum acts as a dummy carrier of the

 	locks, taking care also of lock releases, before we can move the locks

@@ -2746,6 +2794,11 @@

 	ut_ad(dict_index_is_clust(index));

 	ut_ad(!rec_get_deleted_flag(rec, rec_offs_comp(offsets)));

+	if (thr && thr_get_trx(thr)->fake_changes) {

+		/* skip LOCK, UNDO, CHANGE, LOG */

+		return(DB_SUCCESS);

+	}

+

 	err = lock_clust_rec_modify_check_and_lock(flags, block,

 						   rec, index, offsets, thr);

@@ -2883,6 +2936,11 @@

 	rec_t*		rec;

 	ulint		err;

+	if (thr && thr_get_trx(thr)->fake_changes) {

+		/* skip LOCK, CHANGE, LOG */

+		return(DB_SUCCESS);

+	}

+

 	block = btr_cur_get_block(cursor);

 	rec = btr_cur_get_rec(cursor);

diff -ruN a/storage/innobase/handler/ha_innodb.cc b/storage/innobase/handler/ha_innodb.cc

--- a/storage/innobase/handler/ha_innodb.cc	2011-06-01 21:04:26.511385596 +0900

+++ b/storage/innobase/handler/ha_innodb.cc	2011-06-01 21:04:53.723390828 +0900

@@ -484,6 +484,12 @@

   " or 2 (write at commit, flush once per second).",

   NULL, NULL, 1, 0, 2, 0);

+static MYSQL_THDVAR_BOOL(fake_changes, PLUGIN_VAR_OPCMDARG,

+  "In the transaction after enabled, UPDATE, INSERT and DELETE only move the cursor to the records "

+  "and do nothing other operations (no changes, no ibuf, no undo, no transaction log) in the transaction. "

+  "ATTENTION: the transaction started after enabled is affected.",

+  NULL, NULL, FALSE);

+

 static handler *innobase_create_handler(handlerton *hton,

                                         TABLE_SHARE *table,

@@ -1675,6 +1681,8 @@

 	trx->check_unique_secondary = !thd_test_options(

 		thd, OPTION_RELAXED_UNIQUE_CHECKS);

+	trx->fake_changes = THDVAR(thd, fake_changes);

+

 #ifdef EXTENDED_SLOWLOG

 	if (thd_log_slow_verbosity(thd) & SLOG_V_INNODB) {

 		trx->take_stats = TRUE;

@@ -7486,12 +7494,20 @@

 	parent_trx = check_trx_exists(thd);

+

+	if (parent_trx->fake_changes

+	    && thd_sql_command(thd) != SQLCOM_CREATE_TABLE) {

+		sql_print_error("innodb_fake_changes cannot treat new table during other than CREATE TABLE. (%s)\n", name);

+		DBUG_RETURN(HA_ERR_WRONG_COMMAND);

+	}

+

 	/* In case MySQL calls this in the middle of a SELECT query, release

 	possible adaptive hash latch to avoid deadlocks of threads */

 	trx_search_latch_release_if_reserved(parent_trx);

 	trx = innobase_trx_allocate(thd);

+	trx->fake_changes = FALSE;

 	/* Latch the InnoDB data dictionary exclusively so that no deadlocks

 	or lock waits can happen in it during a table create operation.

@@ -7701,6 +7717,7 @@

 /*==============================*/

 {

 	int		error;

+	ibool		fake_changes_reserved = FALSE;

 	DBUG_ENTER("ha_innobase::truncate");

@@ -7713,10 +7730,21 @@

 		DBUG_RETURN(HA_ERR_CRASHED);

 	}

+	if (prebuilt->trx->fake_changes) {

+		/* disable temporally */

+		prebuilt->trx->fake_changes = FALSE;

+		fake_changes_reserved = TRUE;

+	}

+

 	/* Truncate the table in InnoDB */

 	error = row_truncate_table_for_mysql(prebuilt->table, prebuilt->trx);

+	if (fake_changes_reserved) {

+		prebuilt->trx->fake_changes = TRUE;

+		fake_changes_reserved = FALSE;

+	}

+

 	if (share->ib_table->is_corrupt) {

 		DBUG_RETURN(HA_ERR_CRASHED);

 	}

@@ -7768,6 +7796,7 @@

 	trx_search_latch_release_if_reserved(parent_trx);

 	trx = innobase_trx_allocate(thd);

+	trx->fake_changes = FALSE;

 	name_len = strlen(name);

@@ -7855,6 +7884,7 @@

 	trx->mysql_thd = NULL;

 #else

 	trx = innobase_trx_allocate(thd);

+	trx->fake_changes = FALSE;

 #endif

 	row_drop_database_for_mysql(namebuf, trx);

 	my_free(namebuf);

@@ -7960,6 +7990,7 @@

 	trx_search_latch_release_if_reserved(parent_trx);

 	trx = innobase_trx_allocate(thd);

+	trx->fake_changes = FALSE;

 	error = innobase_rename_table(trx, from, to, TRUE);

@@ -12228,6 +12259,7 @@

   MYSQL_SYSVAR(rollback_segments),

   MYSQL_SYSVAR(corrupt_table_action),

   MYSQL_SYSVAR(lazy_drop_table),

+  MYSQL_SYSVAR(fake_changes),

   NULL

 };

diff -ruN a/storage/innobase/handler/handler0alter.cc b/storage/innobase/handler/handler0alter.cc

--- a/storage/innobase/handler/handler0alter.cc	2011-04-11 19:44:03.000000000 +0900

+++ b/storage/innobase/handler/handler0alter.cc	2011-06-01 21:04:53.728495636 +0900

@@ -641,6 +641,7 @@

 	ibool		dict_locked	= FALSE;

 	ulint		new_primary;

 	int		error;

+	ulint		should_fake_changes	= FALSE;

 	DBUG_ENTER("ha_innobase::add_index");

 	ut_a(table);

@@ -677,11 +678,17 @@

 	}

 	heap = mem_heap_create(1024);

+

+	if (prebuilt->trx->fake_changes) {

+		prebuilt->trx->fake_changes = FALSE;

+		should_fake_changes = TRUE;

+	}

 	trx_start_if_not_started(prebuilt->trx);

 	/* Create a background transaction for the operations on

 	the data dictionary tables. */

 	trx = innobase_trx_allocate(user_thd);

+	trx->fake_changes = FALSE;

 	trx_start_if_not_started(trx);

 	/* Create table containing all indexes to be built in this

@@ -760,6 +767,9 @@

 			row_mysql_unlock_data_dictionary(trx);

 			trx_free_for_mysql(trx);

 			trx_commit_for_mysql(prebuilt->trx);

+			if (should_fake_changes) {

+				prebuilt->trx->fake_changes = TRUE;

+			}

 			DBUG_RETURN(error);

 		}

@@ -925,6 +935,9 @@

 	trx_commit_for_mysql(trx);

 	if (prebuilt->trx) {

 		trx_commit_for_mysql(prebuilt->trx);

+		if (should_fake_changes) {

+			prebuilt->trx->fake_changes = TRUE;

+		}

 	}

 	if (dict_locked) {

@@ -1172,6 +1185,7 @@

 	/* Create a background transaction for the operations on

 	the data dictionary tables. */

 	trx = innobase_trx_allocate(user_thd);

+	trx->fake_changes = FALSE;

 	trx_start_if_not_started(trx);

 	/* Flag this transaction as a dictionary operation, so that

diff -ruN a/storage/innobase/ibuf/ibuf0ibuf.c b/storage/innobase/ibuf/ibuf0ibuf.c

--- a/storage/innobase/ibuf/ibuf0ibuf.c	2011-06-01 21:04:26.529383815 +0900

+++ b/storage/innobase/ibuf/ibuf0ibuf.c	2011-06-01 21:04:53.732416483 +0900

@@ -3496,6 +3496,8 @@

 	ut_a(trx_sys_multiple_tablespace_format);

+	ut_ad(!(thr_get_trx(thr)->fake_changes));

+

 	do_merge = FALSE;

 	/* Perform dirty reads of ibuf->size and ibuf->max_size, to

diff -ruN a/storage/innobase/include/trx0trx.h b/storage/innobase/include/trx0trx.h

--- a/storage/innobase/include/trx0trx.h	2011-06-01 21:04:25.762416618 +0900

+++ b/storage/innobase/include/trx0trx.h	2011-06-01 21:04:53.736387944 +0900

@@ -512,6 +512,7 @@

 					FALSE, one can save CPU time and about

 					150 bytes in the undo log size as then

 					we skip XA steps */

+	ulint		fake_changes;

 	ulint		flush_log_later;/* In 2PC, we hold the

 					prepare_commit mutex across

 					both phases. In that case, we

diff -ruN a/storage/innobase/lock/lock0lock.c b/storage/innobase/lock/lock0lock.c

--- a/storage/innobase/lock/lock0lock.c	2011-06-01 21:04:26.543396960 +0900

+++ b/storage/innobase/lock/lock0lock.c	2011-06-01 21:04:53.739416526 +0900

@@ -3909,6 +3909,10 @@

 	trx = thr_get_trx(thr);

+	if (trx->fake_changes && mode == LOCK_IX) {

+		mode = LOCK_IS;

+	}

+

 	lock_mutex_enter_kernel();

 	/* Look for stronger locks the same trx already has on the table */

@@ -5107,6 +5111,11 @@

 	}

 	trx = thr_get_trx(thr);

+

+	if (trx->fake_changes) {

+		return(DB_SUCCESS);

+	}

+

 	next_rec = page_rec_get_next_const(rec);

 	next_rec_heap_no = page_rec_get_heap_no(next_rec);

@@ -5275,6 +5284,10 @@

 		return(DB_SUCCESS);

 	}

+	if (thr && thr_get_trx(thr)->fake_changes) {

+		return(DB_SUCCESS);

+	}

+

 	heap_no = rec_offs_comp(offsets)

 		? rec_get_heap_no_new(rec)

 		: rec_get_heap_no_old(rec);

@@ -5333,6 +5346,10 @@

 		return(DB_SUCCESS);

 	}

+	if (thr && thr_get_trx(thr)->fake_changes) {

+		return(DB_SUCCESS);

+	}

+

 	heap_no = page_rec_get_heap_no(rec);

 	/* Another transaction cannot have an implicit lock on the record,

@@ -5420,6 +5437,10 @@

 		return(DB_SUCCESS);

 	}

+	if (thr && thr_get_trx(thr)->fake_changes && mode == LOCK_X) {

+		mode = LOCK_S;

+	}

+

 	heap_no = page_rec_get_heap_no(rec);

 	lock_mutex_enter_kernel();

@@ -5496,6 +5517,10 @@

 		return(DB_SUCCESS);

 	}

+	if (thr && thr_get_trx(thr)->fake_changes && mode == LOCK_X) {

+		mode = LOCK_S;

+	}

+

 	heap_no = page_rec_get_heap_no(rec);

 	lock_mutex_enter_kernel();

diff -ruN a/storage/innobase/que/que0que.c b/storage/innobase/que/que0que.c

--- a/storage/innobase/que/que0que.c	2011-06-01 21:04:25.553416889 +0900

+++ b/storage/innobase/que/que0que.c	2011-06-01 21:04:53.743419001 +0900

@@ -1417,6 +1417,12 @@

 	ut_a(trx->error_state == DB_SUCCESS);

+	if (trx->fake_changes) {

+		/* fake_changes should not access to system tables */

+		fprintf(stderr, "InnoDB: ERROR: innodb_fake_changes tried to access to system tables.\n");

+		return(DB_ERROR);

+	}

+

 	if (reserve_dict_mutex) {

 		mutex_enter(&dict_sys->mutex);

 	}

diff -ruN a/storage/innobase/row/row0ins.c b/storage/innobase/row/row0ins.c

--- a/storage/innobase/row/row0ins.c	2011-06-01 21:04:26.099449453 +0900

+++ b/storage/innobase/row/row0ins.c	2011-06-01 21:04:53.746383371 +0900

@@ -1499,6 +1499,11 @@

 	if (UNIV_LIKELY_NULL(heap)) {

 		mem_heap_free(heap);

 	}

+

+	if (trx->fake_changes) {

+		err = DB_SUCCESS;

+	}

+

 	return(err);

 }

@@ -2004,7 +2009,7 @@

 	}

 	btr_cur_search_to_nth_level(index, 0, entry, PAGE_CUR_LE,

-				    search_mode,

+				    thr_get_trx(thr)->fake_changes ? BTR_SEARCH_LEAF : search_mode,

 				    &cursor, 0, __FILE__, __LINE__, &mtr);

 	if (cursor.flag == BTR_CUR_INSERT_TO_IBUF) {

@@ -2065,7 +2070,7 @@

 			btr_cur_search_to_nth_level(index, 0, entry,

 						    PAGE_CUR_LE,

-						    mode | BTR_INSERT,

+						    thr_get_trx(thr)->fake_changes ? BTR_SEARCH_LEAF : (mode | BTR_INSERT),

 						    &cursor, 0,

 						    __FILE__, __LINE__, &mtr);

 		}

@@ -2119,6 +2124,22 @@

 	if (UNIV_LIKELY_NULL(big_rec)) {

 		rec_t*	rec;

 		ulint*	offsets;

+

+		if (thr_get_trx(thr)->fake_changes) {

+			/* skip store extern */

+			if (modify) {

+				dtuple_big_rec_free(big_rec);

+			} else {

+				dtuple_convert_back_big_rec(index, entry, big_rec);

+			}

+

+			if (UNIV_LIKELY_NULL(heap)) {

+				mem_heap_free(heap);

+			}

+

+			return(err);

+		}

+

 		mtr_start(&mtr);

 		btr_cur_search_to_nth_level(index, 0, entry, PAGE_CUR_LE,

diff -ruN a/storage/innobase/row/row0mysql.c b/storage/innobase/row/row0mysql.c

--- a/storage/innobase/row/row0mysql.c	2011-06-01 21:04:26.287385626 +0900

+++ b/storage/innobase/row/row0mysql.c	2011-06-01 21:04:53.749372325 +0900

@@ -1245,6 +1245,7 @@

 		prebuilt->table->stat_n_rows--;

 	}

+	if (!(trx->fake_changes))

 	row_update_statistics_if_needed(prebuilt->table);

 	trx->op_info = "";

@@ -1504,6 +1505,7 @@

 	that changes indexed columns, UPDATEs that change only non-indexed

 	columns would not affect statistics. */

 	if (node->is_delete || !(node->cmpl_info & UPD_NODE_NO_ORD_CHANGE)) {

+		if (!(trx->fake_changes))

 		row_update_statistics_if_needed(prebuilt->table);

 	}

@@ -1721,6 +1723,7 @@

 		srv_n_rows_updated++;

 	}

+	if (!(trx->fake_changes))

 	row_update_statistics_if_needed(table);

 	return(err);

diff -ruN a/storage/innobase/row/row0upd.c b/storage/innobase/row/row0upd.c

--- a/storage/innobase/row/row0upd.c	2011-06-01 21:04:25.588384975 +0900

+++ b/storage/innobase/row/row0upd.c	2011-06-01 21:04:53.753385366 +0900

@@ -1603,7 +1603,8 @@

 		mode |= BTR_DELETE_MARK;

 	}

-	search_result = row_search_index_entry(index, entry, mode,

+	search_result = row_search_index_entry(index, entry,

+					       trx->fake_changes ? BTR_SEARCH_LEAF : mode,

 					       &pcur, &mtr);

 	btr_cur = btr_pcur_get_btr_cur(&pcur);

@@ -1850,9 +1851,11 @@

 		the previous invocation of this function. Mark the

 		off-page columns in the entry inherited. */

+		if (!(trx->fake_changes)) {

 		change_ownership = row_upd_clust_rec_by_insert_inherit(

 			NULL, NULL, entry, node->update);

 		ut_a(change_ownership);

+		}

 		/* fall through */

 	case UPD_NODE_INSERT_CLUSTERED:

 		/* A lock wait occurred in row_ins_index_entry() in

@@ -1882,7 +1885,7 @@

 		delete-marked old record, mark them disowned by the

 		old record and owned by the new entry. */

-		if (rec_offs_any_extern(offsets)) {

+		if (rec_offs_any_extern(offsets) && !(trx->fake_changes)) {

 			change_ownership = row_upd_clust_rec_by_insert_inherit(

 				rec, offsets, entry, node->update);

@@ -2012,7 +2015,8 @@

 	the same transaction do not modify the record in the meantime.

 	Therefore we can assert that the restoration of the cursor succeeds. */

-	ut_a(btr_pcur_restore_position(BTR_MODIFY_TREE, pcur, mtr));

+	ut_a(btr_pcur_restore_position(thr_get_trx(thr)->fake_changes ? BTR_SEARCH_LEAF : BTR_MODIFY_TREE,

+				       pcur, mtr));

 	ut_ad(!rec_get_deleted_flag(btr_pcur_get_rec(pcur),

 				    dict_table_is_comp(index->table)));

@@ -2022,7 +2026,8 @@

 					 node->cmpl_info, thr, mtr);

 	mtr_commit(mtr);

-	if (err == DB_SUCCESS && big_rec) {

+	/* skip store extern for fake_changes */

+	if (err == DB_SUCCESS && big_rec && !(thr_get_trx(thr)->fake_changes)) {

 		ulint		offsets_[REC_OFFS_NORMAL_SIZE];

 		rec_t*		rec;

 		rec_offs_init(offsets_);

@@ -2146,7 +2151,8 @@

 	ut_a(pcur->rel_pos == BTR_PCUR_ON);

-	success = btr_pcur_restore_position(BTR_MODIFY_LEAF, pcur, mtr);

+	success = btr_pcur_restore_position(thr_get_trx(thr)->fake_changes ? BTR_SEARCH_LEAF : BTR_MODIFY_LEAF,

+					    pcur, mtr);

 	if (!success) {

 		err = DB_RECORD_NOT_FOUND;

diff -ruN a/storage/innobase/trx/trx0trx.c b/storage/innobase/trx/trx0trx.c

--- a/storage/innobase/trx/trx0trx.c	2011-06-01 21:04:26.411537237 +0900

+++ b/storage/innobase/trx/trx0trx.c	2011-06-01 21:04:53.756415956 +0900

@@ -121,6 +121,8 @@

 	trx->support_xa = TRUE;

+	trx->fake_changes = FALSE;

+

 	trx->check_foreigns = TRUE;

 	trx->check_unique_secondary = TRUE;