set_var.cc, mysqld.cc, ha_innodb.cc, sql_class.h:

  Add a settable session variable innodb_support_xa; setting it to 0 can save up to 10 % of CPU time and 150 bytes of space in each undo log
trx0trx.h, trx0undo.c, trx0trx.c, trx0roll.c:
  Enable XA if innodb_support_xa is not set to 0; make prepare to do log fsync's according to innodb_flush_log_at_trx_commit

innobase/include/trx0trx.h

@@ -369,6 +369,11 @@ struct trx_struct{
 	XID		xid;		/* X/Open XA transaction 
 					identification to identify a 
 					transaction branch */
+	ibool		support_xa;	/* normally we do the XA two-phase
+					commit steps, but by setting this to
+					FALSE, one can save CPU time and about
+					150 bytes in the undo log size as then
+					we skip XA steps */
 	dulint		no;		/* transaction serialization number ==
 					max trx id when the transaction is 
 					moved to COMMITTED_IN_MEMORY state */

innobase/trx/trx0roll.c

@@ -441,16 +441,8 @@ trx_rollback_or_clean_all_without_sess(
 			trx = UT_LIST_GET_NEXT(trx_list, trx);
 		} else if (trx->conc_state == TRX_PREPARED) {
 
-			/* Roll back all prepared transactions if
-			innobase_force_recovery > 0 in my.cnf */
-
-			if (srv_force_recovery > 0) {
-				trx->conc_state = TRX_ACTIVE;
-				break;
-			} else {
 			trx->sess = trx_dummy_sess;
 			trx = UT_LIST_GET_NEXT(trx_list, trx);
-			}
 		} else {
 			break;
 		}
@@ -461,7 +453,7 @@ trx_rollback_or_clean_all_without_sess(
 	if (trx == NULL) {
 		ut_print_timestamp(stderr);
 		fprintf(stderr,
-		"  InnoDB: Rollback of uncommitted transactions completed\n");
+		"  InnoDB: Rollback of non-prepared transactions completed\n");
 
  		mem_heap_free(heap);
 

innobase/trx/trx0trx.c

@@ -93,6 +93,8 @@ trx_create(
 	trx->id = ut_dulint_zero;
 	trx->no = ut_dulint_max;
 
+	trx->support_xa = TRUE;
+
 	trx->check_foreigns = TRUE;
 	trx->check_unique_secondary = TRUE;
 
@@ -453,9 +455,15 @@ trx_lists_init_at_db_start(void)
 					ut_dulint_get_high(trx->id),
 					ut_dulint_get_low(trx->id));
 
-					trx->conc_state = TRX_ACTIVE;
+					if (srv_force_recovery == 0) {
 
-					/* trx->conc_state = TRX_PREPARED;*/
+						trx->conc_state = TRX_PREPARED;
+					} else {
+ 						fprintf(stderr,
+"InnoDB: Since innodb_force_recovery > 0, we will rollback it anyway.\n");
+
+						trx->conc_state = TRX_ACTIVE;
+					}
 				} else {
 					trx->conc_state =
 						TRX_COMMITTED_IN_MEMORY;
@@ -516,10 +524,15 @@ trx_lists_init_at_db_start(void)
 					    ut_dulint_get_high(trx->id),
 					    ut_dulint_get_low(trx->id));
 
-						trx->conc_state = TRX_ACTIVE;
+					    if (srv_force_recovery == 0) {
 
-						/* trx->conc_state = 
-							TRX_PREPARED; */
+						trx->conc_state = TRX_PREPARED;
+					    } else {
+ 						fprintf(stderr,
+"InnoDB: Since innodb_force_recovery > 0, we will rollback it anyway.\n");
+
+						trx->conc_state = TRX_ACTIVE;
+					    }
 					} else {
 						trx->conc_state =
 						  TRX_COMMITTED_IN_MEMORY;
@@ -823,9 +836,6 @@ trx_commit_off_kernel(
 		trx->read_view = NULL;
 	}
 
-/*	fprintf(stderr, "Trx %lu commit finished\n",
-		ut_dulint_get_low(trx->id)); */
-
 	if (must_flush_log) {
 
 		mutex_exit(&kernel_mutex);
@@ -871,7 +881,8 @@ trx_commit_off_kernel(
                         if (srv_unix_file_flush_method == SRV_UNIX_NOSYNC) {
                              	/* Write the log but do not flush it to disk */
 
-                               log_write_up_to(lsn, LOG_WAIT_ONE_GROUP, FALSE);
+                               	log_write_up_to(lsn, LOG_WAIT_ONE_GROUP,
+									FALSE);
                         } else {
                                	/* Write the log to the log files AND flush
                                	them to disk */
@@ -1747,12 +1758,11 @@ Prepares a transaction. */
 
 void
 trx_prepare_off_kernel(
-/*==================*/
+/*===================*/
 	trx_t*	trx)	/* in: transaction */
 {
 	page_t*		update_hdr_page;
 	trx_rseg_t*	rseg;
-	trx_undo_t*	undo;
 	ibool		must_flush_log	= FALSE;
 	dulint		lsn;
 	mtr_t		mtr;
@@ -1779,19 +1789,18 @@ trx_prepare_off_kernel(
 		mutex_enter(&(rseg->mutex));
 			
 		if (trx->insert_undo != NULL) {
-			trx_undo_set_state_at_prepare(trx, trx->insert_undo,
-							  		&mtr);
-		}
-
-		undo = trx->update_undo;
 
-		if (undo) {
 			/* It is not necessary to obtain trx->undo_mutex here
 			because only a single OS thread is allowed to do the
 			transaction prepare for this transaction. */
 
+			trx_undo_set_state_at_prepare(trx, trx->insert_undo,
+							  		&mtr);
+		}
+
+		if (trx->update_undo) {
 			update_hdr_page = trx_undo_set_state_at_prepare(trx,
-								undo, &mtr);
+						trx->update_undo, &mtr);
 		}
 
 		mutex_exit(&(rseg->mutex));
@@ -1815,16 +1824,47 @@ trx_prepare_off_kernel(
 	/*--------------------------------------*/
 
 	if (must_flush_log) {
+                /* Depending on the my.cnf options, we may now write the log
+                buffer to the log files, making the prepared state of the
+		transaction durable if the OS does not crash. We may also
+		flush the log files to disk, making the prepared state of the
+		transaction durable also at an OS crash or a power outage.
+
+                The idea in InnoDB's group prepare is that a group of
+                transactions gather behind a trx doing a physical disk write
+                to log files, and when that physical write has been completed,
+                one of those transactions does a write which prepares the whole
+                group. Note that this group prepare will only bring benefit if
+                there are > 2 users in the database. Then at least 2 users can
+                gather behind one doing the physical log write to disk.
+
+		TODO: find out if MySQL holds some mutex when calling this.
+		That would spoil our group prepare algorithm. */
 
 		mutex_exit(&kernel_mutex);
 
-		/* Write the log to the log files AND flush them to disk */
+                if (srv_flush_log_at_trx_commit == 0) {
+                        /* Do nothing */
+                } else if (srv_flush_log_at_trx_commit == 1) {
+                   	if (srv_unix_file_flush_method == SRV_UNIX_NOSYNC) {
+                        	/* Write the log but do not flush it to disk */
 
-		/*-------------------------------------*/
+                        	log_write_up_to(lsn, LOG_WAIT_ONE_GROUP,
+								FALSE);
+                        } else {
+                               	/* Write the log to the log files AND flush
+                               	them to disk */
 
                                	log_write_up_to(lsn, LOG_WAIT_ONE_GROUP, TRUE);
+                        }
+                } else if (srv_flush_log_at_trx_commit == 2) {
 
-		/*-------------------------------------*/
+                        /* Write the log but do not flush it to disk */
+
+                        log_write_up_to(lsn, LOG_WAIT_ONE_GROUP, FALSE);
+                } else {
+                        ut_error;
+                }
 
 		mutex_enter(&kernel_mutex);
 	}

innobase/trx/trx0undo.c

@@ -596,7 +596,7 @@ trx_undo_read_xid(
 }
 
 /*******************************************************************
-Adds the XA XID after an undo log old-style header. */
+Adds space for the XA XID after an undo log old-style header. */
 static
 void
 trx_undo_header_add_space_for_xid(
@@ -1488,6 +1488,7 @@ trx_undo_create(
 /*============*/
 				/* out: undo log object, NULL if did not
 				succeed: out of space */
+	trx_t*		trx,	/* in: transaction */
 	trx_rseg_t*	rseg,	/* in: rollback segment memory copy */
 	ulint		type,	/* in: type of the log: TRX_UNDO_INSERT or
 				TRX_UNDO_UPDATE */
@@ -1530,7 +1531,10 @@ trx_undo_create(
 	
 	offset = trx_undo_header_create(undo_page, trx_id, mtr);
 
-	trx_undo_header_add_space_for_xid(undo_page, undo_page + offset, mtr);
+	if (trx->support_xa) {
+		trx_undo_header_add_space_for_xid(undo_page,
+					undo_page + offset, mtr);
+	}
 
 	undo = trx_undo_mem_create(rseg, id, type, trx_id, xid,
 							page_no, offset);
@@ -1547,6 +1551,7 @@ trx_undo_reuse_cached(
 /*==================*/
 				/* out: the undo log memory object, NULL if
 				none cached */
+	trx_t*		trx,	/* in: transaction */
 	trx_rseg_t*	rseg,	/* in: rollback segment memory object */
 	ulint		type,	/* in: type of the log: TRX_UNDO_INSERT or
 				TRX_UNDO_UPDATE */
@@ -1597,16 +1602,22 @@ trx_undo_reuse_cached(
 
 	if (type == TRX_UNDO_INSERT) {
 		offset = trx_undo_insert_header_reuse(undo_page, trx_id, mtr);
-		trx_undo_header_add_space_for_xid(undo_page, undo_page + offset,
-									mtr);
+		
+		if (trx->support_xa) {
+			trx_undo_header_add_space_for_xid(undo_page,
+						undo_page + offset, mtr);
+		}
 	} else {
 		ut_a(mach_read_from_2(undo_page + TRX_UNDO_PAGE_HDR
 					+ TRX_UNDO_PAGE_TYPE)
 						== TRX_UNDO_UPDATE);
 
 		offset = trx_undo_header_create(undo_page, trx_id, mtr);
-		trx_undo_header_add_space_for_xid(undo_page, undo_page + offset,
-									mtr);
+
+		if (trx->support_xa) {
+			trx_undo_header_add_space_for_xid(undo_page,
+						undo_page + offset, mtr);
+		}
 	}
 	
 	trx_undo_mem_init_for_reuse(undo, trx_id, xid, offset);
@@ -1674,11 +1685,11 @@ trx_undo_assign_undo(
 #endif /* UNIV_SYNC_DEBUG */
 	mutex_enter(&(rseg->mutex));
 
-	undo = trx_undo_reuse_cached(rseg, type, trx->id, &trx->xid, &mtr);
-
+	undo = trx_undo_reuse_cached(trx, rseg, type, trx->id, &trx->xid,
+									&mtr);
 	if (undo == NULL) {
-		undo = trx_undo_create(rseg, type, trx->id, &trx->xid, &mtr);
-
+		undo = trx_undo_create(trx, rseg, type, trx->id, &trx->xid,
+									&mtr);
 		if (undo == NULL) {
 			/* Did not succeed */
 

sql/ha_innodb.cc

@@ -690,6 +690,10 @@ check_trx_exists(
 		trx->mysql_query_str = &(thd->query);
                 trx->active_trans = 0;
 
+		/* Update the info whether we should skip XA steps that eat
+		CPU time */
+		trx->support_xa = (ibool)(thd->variables.innodb_support_xa);
+
                 thd->ha_data[innobase_hton.slot] = trx;
 	} else {
 		if (trx->magic_n != TRX_MAGIC_N) {
@@ -1434,6 +1438,9 @@ innobase_commit(
 
 	trx = check_trx_exists(thd);
 
+	/* Update the info whether we should skip XA steps that eat CPU time */
+	trx->support_xa = (ibool)(thd->variables.innodb_support_xa);
+
 	/* Release a possible FIFO ticket and search latch. Since we will
 	reserve the kernel mutex, we have to release the search system latch
 	first to obey the latching order. */
@@ -1620,6 +1627,9 @@ innobase_rollback(
 
 	trx = check_trx_exists(thd);
 
+	/* Update the info whether we should skip XA steps that eat CPU time */
+	trx->support_xa = (ibool)(thd->variables.innodb_support_xa);
+
 	/* Release a possible FIFO ticket and search latch. Since we will
 	reserve the kernel mutex, we have to release the search system latch
 	first to obey the latching order. */
@@ -6308,6 +6318,11 @@ innobase_xa_prepare(
 	int error = 0;
         trx_t* trx;
 
+	if (!thd->variables.innodb_support_xa) {
+
+		return(0);
+	}
+
         trx = check_trx_exists(thd);
 
         trx->xid=thd->transaction.xid;

sql/mysqld.cc

@@ -4211,6 +4211,7 @@ enum options_mysqld
   OPT_INNODB_STATUS_FILE,
   OPT_INNODB_MAX_DIRTY_PAGES_PCT,
   OPT_INNODB_TABLE_LOCKS,
+  OPT_INNODB_SUPPORT_XA,
   OPT_INNODB_OPEN_FILES,
   OPT_INNODB_AUTOEXTEND_INCREMENT,
   OPT_INNODB_SYNC_SPIN_LOOPS,
@@ -4514,6 +4515,11 @@ Disable with --skip-innodb-doublewrite.", (gptr*) &innobase_use_doublewrite,
    (gptr*) &global_system_variables.innodb_table_locks,
    (gptr*) &global_system_variables.innodb_table_locks,
    0, GET_BOOL, OPT_ARG, 1, 0, 0, 0, 0, 0},
+  {"innodb_support_xa", OPT_INNODB_SUPPORT_XA,
+   "Enable InnoDB support for the XA two-phase commit",
+   (gptr*) &global_system_variables.innodb_support_xa,
+   (gptr*) &global_system_variables.innodb_support_xa,
+   0, GET_BOOL, OPT_ARG, 1, 0, 0, 0, 0, 0},
 #endif /* End HAVE_INNOBASE_DB */
   {"isam", OPT_ISAM, "Enable ISAM (if this version of MySQL supports it). \
 Disable with --skip-isam.",

sql/set_var.cc

@@ -392,6 +392,8 @@ sys_var_long_ptr	sys_innodb_max_purge_lag("innodb_max_purge_lag",
 							&srv_max_purge_lag);
 sys_var_thd_bool	sys_innodb_table_locks("innodb_table_locks",
                                                &SV::innodb_table_locks);
+sys_var_thd_bool	sys_innodb_support_xa("innodb_support_xa",
+                                               &SV::innodb_support_xa);
 sys_var_long_ptr	sys_innodb_autoextend_increment("innodb_autoextend_increment",
 							&srv_auto_extend_increment);
 sys_var_long_ptr	sys_innodb_sync_spin_loops("innodb_sync_spin_loops",
@@ -689,6 +691,7 @@ sys_var *sys_variables[]=
   &sys_innodb_max_dirty_pages_pct,
   &sys_innodb_max_purge_lag,
   &sys_innodb_table_locks,
+  &sys_innodb_support_xa,
   &sys_innodb_max_purge_lag,
   &sys_innodb_autoextend_increment,
   &sys_innodb_sync_spin_loops,
@@ -810,6 +813,7 @@ struct show_var_st init_vars[]= {
   {"innodb_open_files", (char*) &innobase_open_files, SHOW_LONG },
   {sys_innodb_sync_spin_loops.name, (char*) &sys_innodb_sync_spin_loops, SHOW_SYS},
   {sys_innodb_table_locks.name, (char*) &sys_innodb_table_locks, SHOW_SYS},
+  {sys_innodb_support_xa.name, (char*) &sys_innodb_support_xa, SHOW_SYS},
   {sys_innodb_thread_concurrency.name, (char*) &sys_innodb_thread_concurrency, SHOW_SYS},
   {sys_innodb_thread_sleep_delay.name, (char*) &sys_innodb_thread_sleep_delay, SHOW_SYS},
 #endif

sql/sql_class.h

@@ -578,6 +578,7 @@ struct system_variables
 #endif /* HAVE_REPLICATION */
 #ifdef HAVE_INNOBASE_DB
   my_bool innodb_table_locks;
+  my_bool innodb_support_xa;
 #endif /* HAVE_INNOBASE_DB */
 #ifdef HAVE_NDBCLUSTER_DB
   ulong ndb_autoincrement_prefetch_sz;