branches/zip: Introduce two new dictionary operation modes for transactions.

enum trx_dict_op: dictionary operation modes

trx_get_dict_operation(), trx_set_dict_operation(): Accessors for
trx->dict_operation.

lock_table_enqueue_waiting(), lock_rec_enqueue_waiting(): Do not complain
about lock waits if the dictionary mode is TRX_DICT_OP_INDEX_MAY_WAIT.

row_merge_lock_table(): Remove the work-around for avoiding the warning
in lock_table_enqueue_waiting().

trx_undo_mark_as_dict_operation(): Do not write trx->table_id to the
undo log unless the dict_operation is TRX_DICT_OP_TABLE.

ha_innobase::add_index(): Set the dict_operation mode initially to
TRX_DICT_OP_INDEX_MAY_WAIT, then lock the table exclusively, and set the
mode to TRX_DICT_OP_INDEX, and optionally to TRX_DICT_OP_TABLE when
creating a temporary table.

handler/handler0alter.cc

@@ -629,7 +629,7 @@ ha_innobase::add_index(
 	mem_heap_t*     heap;		/* Heap for index definitions */
 	trx_t*		trx;		/* Transaction */
 	ulint		num_of_idx;
-	ulint		num_created;
+	ulint		num_created	= 0;
 	ibool		dict_locked	= FALSE;
 	ulint		new_primary;
 	ulint		error;
@@ -684,6 +684,8 @@ err_exit:
 	index_defs = innobase_create_key_def(
 		trx, innodb_table, heap, key_info, num_of_idx);
 
+	new_primary = DICT_CLUSTERED & index_defs[0].ind_type;
+
 	/* Allocate memory for dictionary index definitions */
 
 	index = (dict_index_t**) mem_heap_alloc(
@@ -695,23 +697,31 @@ err_exit:
 	row_mysql_lock_data_dictionary(trx);
 	dict_locked = TRUE;
 
-	/* Flag this transaction as a dictionary operation, so that the
-	data dictionary will be locked in crash recovery.  Clear the
-	table_id, so that no table will be dropped in crash recovery,
-	unless a new primary key is defined. */
-	trx->dict_operation = TRUE;
-	trx->table_id = ut_dulint_zero;
+	/* Flag this transaction as a dictionary operation, so that
+	the data dictionary will be locked in crash recovery.  Prevent
+	warnings if row_merge_lock_table() results in a lock wait. */
+	trx_set_dict_operation(trx, TRX_DICT_OP_INDEX_MAY_WAIT);
+
+	/* Acquire an exclusive lock on the table
+	before creating any indexes. */
+	error = row_merge_lock_table(trx, innodb_table);
+
+	if (UNIV_UNLIKELY(error != DB_SUCCESS)) {
+
+		goto error_handling;
+	}
+
+	trx_set_dict_operation(trx, TRX_DICT_OP_INDEX);
 
 	/* If a new primary key is defined for the table we need
 	to drop the original table and rebuild all indexes. */
 
-	new_primary = DICT_CLUSTERED & index_defs[0].ind_type;
-
 	if (UNIV_UNLIKELY(new_primary)) {
 		char*	new_table_name = innobase_create_temporary_tablename(
 			heap, '1', innodb_table->name);
 
 		/* Clone the table. */
+		trx_set_dict_operation(trx, TRX_DICT_OP_TABLE);
 		indexed_table = row_merge_create_temporary_table(
 			new_table_name, index_defs, innodb_table, trx);
 
@@ -737,19 +747,6 @@ err_exit:
 		trx->table_id = indexed_table->id;
 	}
 
-	ut_ad(!error);
-
-	/* Acquire an exclusive lock on the table
-	before creating any indexes. */
-	error = row_merge_lock_table(trx, innodb_table);
-
-	if (UNIV_UNLIKELY(error != DB_SUCCESS)) {
-
-		goto error_handling;
-	}
-
-	num_created = 0;
-
 	/* Create the indexes in SYS_INDEXES and load into dictionary. */
 
 	for (ulint i = 0; i < num_of_idx; i++) {

include/trx0trx.h

@@ -357,6 +357,45 @@ trx_print(
 	ulint	max_query_len);	/* in: max query length to print, or 0 to
 				   use the default max length */
 
+/** Type of data dictionary operation */
+enum trx_dict_op {
+	/** The transaction is not modifying the data dictionary. */
+	TRX_DICT_OP_NONE = 0,
+	/** The transaction is creating a table or an index, or
+	dropping a table.  The table must be dropped in crash
+	recovery.  This and TRX_DICT_OP_NONE are the only possible
+	operation modes in crash recovery. */
+	TRX_DICT_OP_TABLE = 1,
+	/** The transaction is creating an index in an existing table,
+	In crash recovery, the the data dictionary must be locked, but
+	the table must not be dropped. */
+	TRX_DICT_OP_INDEX = 2,
+	/** The transaction is creating an index in an existing table,
+	In crash recovery, the the data dictionary must be locked, but
+	the table must not be dropped.  A lock wait timeout is allowed
+	to occur. */
+	TRX_DICT_OP_INDEX_MAY_WAIT = 3
+};
+
+/**************************************************************************
+Determine if a transaction is a dictionary operation. */
+UNIV_INLINE
+enum trx_dict_op
+trx_get_dict_operation(
+/*===================*/
+				/* out: dictionary operation mode */
+	const trx_t*	trx)	/* in: transaction */
+	__attribute__((pure));
+/**************************************************************************
+Flag a transaction a dictionary operation. */
+UNIV_INLINE
+void
+trx_set_dict_operation(
+/*===================*/
+	trx_t*			trx,	/* in/out: transaction */
+	enum trx_dict_op	op);	/* in: operation, not
+					TRX_DICT_OP_NONE */
+
 #ifndef UNIV_HOTBACKUP
 /**************************************************************************
 Determines if the currently running transaction has been interrupted. */
@@ -471,12 +510,7 @@ struct trx_struct{
 					were modifications by the transaction;
 					in that case we must flush the log
 					in trx_commit_complete_for_mysql() */
-	unsigned	dict_operation:1;/* nonzero if the trx is used
-					to create a table, create an
-					index, or drop a table. This
-					is a hint that the table may
-					need to be dropped in crash
-					recovery. */
+	unsigned	dict_operation:2;/**< @see enum trx_dict_op */
 	unsigned	duplicates:2;	/* TRX_DUP_IGNORE | TRX_DUP_REPLACE */
 	unsigned	active_trans:2;	/* 1 - if a transaction in MySQL
 					is active. 2 - if prepare_commit_mutex

include/trx0trx.ic

@@ -145,3 +145,61 @@ trx_get_que_state_str(
 		return("UNKNOWN");
 	}
 }
+
+/**************************************************************************
+Determine if a transaction is a dictionary operation. */
+UNIV_INLINE
+enum trx_dict_op
+trx_get_dict_operation(
+/*===================*/
+				/* out: dictionary operation mode */
+	const trx_t*	trx)	/* in: transaction */
+{
+	enum trx_dict_op op = (enum trx_dict_op) trx->dict_operation;
+
+#ifdef UNIV_DEBUG
+	switch (op) {
+	case TRX_DICT_OP_NONE:
+	case TRX_DICT_OP_TABLE:
+	case TRX_DICT_OP_INDEX:
+	case TRX_DICT_OP_INDEX_MAY_WAIT:
+		return(op);
+	}
+	ut_error;
+#endif /* UNIV_DEBUG */
+	return((enum trx_dict_op) UNIV_EXPECT(op, TRX_DICT_OP_NONE));
+}
+/**************************************************************************
+Flag a transaction a dictionary operation. */
+UNIV_INLINE
+void
+trx_set_dict_operation(
+/*===================*/
+	trx_t*			trx,	/* in/out: transaction */
+	enum trx_dict_op	op)	/* in: operation, not
+					TRX_DICT_OP_NONE */
+{
+#ifdef UNIV_DEBUG
+	enum trx_dict_op	old_op = trx_get_dict_operation(trx);
+
+	switch (op) {
+	case TRX_DICT_OP_NONE:
+		ut_error;
+		break;
+	case TRX_DICT_OP_TABLE:
+		ut_ad(old_op == TRX_DICT_OP_NONE
+		      || old_op == TRX_DICT_OP_INDEX);
+		break;
+	case TRX_DICT_OP_INDEX:
+		ut_ad(old_op == TRX_DICT_OP_NONE
+		      || old_op == TRX_DICT_OP_INDEX_MAY_WAIT);
+		break;
+	case TRX_DICT_OP_INDEX_MAY_WAIT:
+		ut_ad(old_op == TRX_DICT_OP_NONE
+		      || old_op == TRX_DICT_OP_INDEX);
+		break;
+	}
+#endif /* UNIV_DEBUG */
+
+	trx->dict_operation = op;
+}

lock/lock0lock.c

@@ -1783,7 +1783,12 @@ lock_rec_enqueue_waiting(
 
 	trx = thr_get_trx(thr);
 
-	if (UNIV_UNLIKELY(trx->dict_operation)) {
+	switch (trx_get_dict_operation(trx)) {
+	case TRX_DICT_OP_NONE:
+	case TRX_DICT_OP_INDEX_MAY_WAIT:
+		break;
+	case TRX_DICT_OP_TABLE:
+	case TRX_DICT_OP_INDEX:
 		ut_print_timestamp(stderr);
 		fputs("  InnoDB: Error: a record lock wait happens"
 		      " in a dictionary operation!\n"
@@ -3641,7 +3646,12 @@ lock_table_enqueue_waiting(
 
 	trx = thr_get_trx(thr);
 
-	if (trx->dict_operation) {
+	switch (trx_get_dict_operation(trx)) {
+	case TRX_DICT_OP_NONE:
+	case TRX_DICT_OP_INDEX_MAY_WAIT:
+		break;
+	case TRX_DICT_OP_TABLE:
+	case TRX_DICT_OP_INDEX:
 		ut_print_timestamp(stderr);
 		fputs("  InnoDB: Error: a table lock wait happens"
 		      " in a dictionary operation!\n"

row/row0merge.c

@@ -1674,16 +1674,8 @@ run_again:
 	thr->run_node = thr;
 	thr->prev_node = thr->common.parent;
 
-	/* Temporarily clear the dict_operation flag in order to
-	avoid a bogus warning in lock_table_enqueue_waiting(). */
-	ut_ad(trx->dict_operation == 1);
-	trx->dict_operation = 0;
-
 	err = lock_table(0, table, LOCK_X, thr);
 
-	/* Restore the dict_operation flag. */
-	trx->dict_operation = 1;
-
 	trx->error_state = err;
 
 	if (UNIV_LIKELY(err == DB_SUCCESS)) {

row/row0mysql.c

@@ -1938,7 +1938,7 @@ row_create_table_for_mysql(
 
 	heap = mem_heap_create(512);
 
-	trx->dict_operation = TRUE;
+	trx_set_dict_operation(trx, TRX_DICT_OP_TABLE);
 
 	node = tab_create_graph_create(table, heap);
 
@@ -2089,7 +2089,7 @@ row_create_index_for_mysql(
 
 	heap = mem_heap_create(512);
 
-	trx->dict_operation = TRUE;
+	trx_set_dict_operation(trx, TRX_DICT_OP_TABLE);
 
 	/* Note that the space id where we store the index is inherited from
 	the table in dict_build_index_def_step() in dict0crea.c. */
@@ -2161,7 +2161,7 @@ row_table_add_foreign_constraints(
 
 	trx_start_if_not_started(trx);
 
-	trx->dict_operation = TRUE;
+	trx_set_dict_operation(trx, TRX_DICT_OP_TABLE);
 
 	err = dict_create_foreign_constraints(trx, sql_string, name,
 					      reject_fks);
@@ -3286,7 +3286,7 @@ check_next_foreign:
 
 	lock_reset_all_on_table(table);
 
-	trx->dict_operation = TRUE;
+	trx_set_dict_operation(trx, TRX_DICT_OP_TABLE);
 	trx->table_id = table->id;
 
 	/* We use the private SQL parser of Innobase to generate the

trx/trx0roll.c

@@ -447,7 +447,7 @@ trx_rollback_active(
 
 	trx->mysql_process_no = os_proc_get_number();
 
-	if (trx->dict_operation) {
+	if (trx_get_dict_operation(trx) != TRX_DICT_OP_NONE) {
 		row_mysql_lock_data_dictionary(trx);
 		dictionary_locked = TRUE;
 	}
@@ -470,7 +470,8 @@ trx_rollback_active(
 
 	mutex_exit(&kernel_mutex);
 
-	if (trx->dict_operation && !ut_dulint_is_zero(trx->table_id)) {
+	if (trx_get_dict_operation(trx) != TRX_DICT_OP_NONE
+	    && !ut_dulint_is_zero(trx->table_id)) {
 
 		/* If the transaction was for a dictionary operation, we
 		drop the relevant table, if it still exists */

trx/trx0trx.c

@@ -97,7 +97,7 @@ trx_create(
 	trx->flush_log_later = FALSE;
 	trx->must_flush_log_later = FALSE;
 
-	trx->dict_operation = FALSE;
+	trx_set_dict_operation(trx, TRX_DICT_OP_NONE);
 	trx->table_id = ut_dulint_zero;
 
 	trx->mysql_thd = NULL;
@@ -469,7 +469,8 @@ trx_lists_init_at_db_start(void)
 			}
 
 			if (undo->dict_operation) {
-				trx->dict_operation = undo->dict_operation;
+				trx_set_dict_operation(
+					trx, TRX_DICT_OP_TABLE);
 				trx->table_id = undo->table_id;
 			}
 
@@ -548,8 +549,8 @@ trx_lists_init_at_db_start(void)
 				trx_list_insert_ordered(trx);
 
 				if (undo->dict_operation) {
-					trx->dict_operation
-						= undo->dict_operation;
+					trx_set_dict_operation(
+						trx, TRX_DICT_OP_TABLE);
 					trx->table_id = undo->table_id;
 				}
 			}

trx/trx0undo.c

@@ -1674,20 +1674,30 @@ trx_undo_mark_as_dict_operation(
 {
 	page_t*	hdr_page;
 
-	ut_a(trx->dict_operation);
-
 	hdr_page = trx_undo_page_get(undo->space, undo->zip_size,
 				     undo->hdr_page_no, mtr);
 
+	switch (trx_get_dict_operation(trx)) {
+	case TRX_DICT_OP_NONE:
+	case TRX_DICT_OP_INDEX_MAY_WAIT:
+		ut_error;
+	case TRX_DICT_OP_INDEX:
+		/* Do not discard the table on recovery. */
+		undo->table_id = ut_dulint_zero;
+		break;
+	case TRX_DICT_OP_TABLE:
+		undo->table_id = trx->table_id;
+		break;
+	}
+
 	mlog_write_ulint(hdr_page + undo->hdr_offset
 			 + TRX_UNDO_DICT_TRANS,
-			 trx->dict_operation, MLOG_1BYTE, mtr);
+			 TRUE, MLOG_1BYTE, mtr);
 
 	mlog_write_dulint(hdr_page + undo->hdr_offset + TRX_UNDO_TABLE_ID,
-			  trx->table_id, mtr);
+			  undo->table_id, mtr);
 
-	undo->dict_operation = trx->dict_operation;
-	undo->table_id = trx->table_id;
+	undo->dict_operation = TRUE;
 }
 
 /**************************************************************************
@@ -1743,7 +1753,7 @@ trx_undo_assign_undo(
 		trx->update_undo = undo;
 	}
 
-	if (trx->dict_operation) {
+	if (trx_get_dict_operation(trx) != TRX_DICT_OP_NONE) {
 		trx_undo_mark_as_dict_operation(trx, undo, &mtr);
 	}