auto-ROLLBACK if binlog was not closed properly

auto-commit on Xid_log_event


client/mysqlbinlog.cc:
  auto-ROLLBACK if binlog was not closed properly.
mysql-test/r/ctype_ucs.result:
  results updated
mysql-test/r/mix_innodb_myisam_binlog.result:
  results updated
mysql-test/r/mysqlbinlog2.result:
  results updated
mysql-test/r/rpl_relayrotate.result:
  results updated
mysql-test/r/user_var.result:
  results updated
mysql-test/t/ctype_ucs.test:
  finalize binlog before calling mysqlbinlog
mysql-test/t/user_var.test:
  finalize binlog before calling mysqlbinlog
sql/log_event.cc:
  commit at Xid_log_event
  comments edited
sql/mysqld.cc:
  free(0) fixed
sql/slave.cc:
  rollback at fake Rotate_log_event
sql/sql_class.h:
  no commit_or_rollback argument for binlog->write(THD *thd, IO_CACHE *cache)
sql/log.cc:
  don't write "COMMIT" query, Xid_log_event is enough
sql/log_event.h:
  more comments for LOG_EVENT_BINLOG_IN_USE_F
  LOG_EVENT_FORCE_ROLLBACK_F added
sql/sql_repl.cc:
  rollback at Rotate_log_event.
  don't consider binlog corrupted if it was open when we read Formar_description but closed when we got to the end
sql/sql_repl.h:
  style fix

client/mysqlbinlog.cc

@@ -1086,7 +1086,7 @@ at offset %lu ; this could be a log format error or read error",
           /* EOF can't be hit here normally, so it's a real error */
           die("Could not read a Rotate_log_event event \
 at offset %lu ; this could be a log format error or read error",
-              tmp_pos); 
+              tmp_pos);
       }
       else
         break;
@@ -1157,9 +1157,16 @@ static int dump_local_log_entries(const char* logname)
     Log_event* ev = Log_event::read_log_event(file, description_event);
     if (!ev)
     {
-      if (file->error)
+      /*
+        if binlog wasn't closed properly ("in use" flag is set) don't complain
+        about a corruption, but issue a "ROLLBACK" to annihilate half-logged
+        transaction. Otherwise, treat it as EOF and move to the next binlog.
+      */
+      if (description_event->flags & LOG_EVENT_BINLOG_IN_USE_F)
+        fprintf(result_file, "ROLLBACK;\n");
+      else if (file->error)
       {
-	fprintf(stderr,
+        fprintf(stderr,
                 "Could not read entry at offset %s:"
                 "Error in log format or read error\n",
                 llstr(old_off,llbuff));

mysql-test/r/ctype_ucs.result

@@ -527,6 +527,7 @@ show binlog events from 96;
 Log_name	Pos	Event_type	Server_id	End_log_pos	Info
 master-bin.000001	96	User var	1	136	@`v`=_ucs2 0x006100620063 COLLATE ucs2_general_ci
 master-bin.000001	136	Query	1	219	use `test`; insert into t2 values (@v)
+flush logs;
 /*!40019 SET @@session.max_insert_delayed_threads=0*/;
 SET @`v`:=_ucs2 0x006100620063 COLLATE ucs2_general_ci;
 use test;

mysql-test/r/mix_innodb_myisam_binlog.result

@@ -11,8 +11,7 @@ Log_name	Pos	Event_type	Server_id	End_log_pos	Info
 master-bin.000001	96	Query	1	#	use `test`; BEGIN
 master-bin.000001	158	Query	1	#	use `test`; insert into t1 values(1)
 master-bin.000001	239	Query	1	#	use `test`; insert into t2 select * from t1
-master-bin.000001	327	Xid	1	#	xid=7
-master-bin.000001	354	Query	1	#	use `test`; COMMIT
+master-bin.000001	327	Xid	1	#	COMMIT /* xid=7 */
 delete from t1;
 delete from t2;
 reset master;
@@ -48,8 +47,7 @@ master-bin.000001	239	Query	1	#	use `test`; savepoint my_savepoint
 master-bin.000001	318	Query	1	#	use `test`; insert into t1 values(4)
 master-bin.000001	399	Query	1	#	use `test`; insert into t2 select * from t1
 master-bin.000001	487	Query	1	#	use `test`; rollback to savepoint my_savepoint
-master-bin.000001	578	Xid	1	#	xid=24
-master-bin.000001	605	Query	1	#	use `test`; COMMIT
+master-bin.000001	578	Xid	1	#	COMMIT /* xid=24 */
 delete from t1;
 delete from t2;
 reset master;
@@ -76,8 +74,7 @@ master-bin.000001	318	Query	1	#	use `test`; insert into t1 values(6)
 master-bin.000001	399	Query	1	#	use `test`; insert into t2 select * from t1
 master-bin.000001	487	Query	1	#	use `test`; rollback to savepoint my_savepoint
 master-bin.000001	578	Query	1	#	use `test`; insert into t1 values(7)
-master-bin.000001	659	Xid	1	#	xid=36
-master-bin.000001	686	Query	1	#	use `test`; COMMIT
+master-bin.000001	659	Xid	1	#	COMMIT /* xid=36 */
 delete from t1;
 delete from t2;
 reset master;
@@ -106,9 +103,8 @@ show binlog events from 96;
 Log_name	Pos	Event_type	Server_id	End_log_pos	Info
 master-bin.000001	96	Query	1	#	use `test`; BEGIN
 master-bin.000001	158	Query	1	#	use `test`; insert into t1 values(9)
-master-bin.000001	239	Xid	1	#	xid=59
-master-bin.000001	266	Query	1	#	use `test`; COMMIT
-master-bin.000001	329	Query	1	#	use `test`; insert into t2 select * from t1
+master-bin.000001	239	Xid	1	#	COMMIT /* xid=59 */
+master-bin.000001	266	Query	1	#	use `test`; insert into t2 select * from t1
 delete from t1;
 delete from t2;
 reset master;
@@ -119,22 +115,19 @@ show binlog events from 96;
 Log_name	Pos	Event_type	Server_id	End_log_pos	Info
 master-bin.000001	96	Query	1	#	use `test`; BEGIN
 master-bin.000001	158	Query	1	#	use `test`; insert into t1 values(10)
-master-bin.000001	240	Xid	1	#	xid=65
-master-bin.000001	267	Query	1	#	use `test`; COMMIT
-master-bin.000001	330	Query	1	#	use `test`; insert into t2 select * from t1
+master-bin.000001	240	Xid	1	#	COMMIT /* xid=65 */
+master-bin.000001	267	Query	1	#	use `test`; insert into t2 select * from t1
 insert into t1 values(11);
 commit;
 show binlog events from 96;
 Log_name	Pos	Event_type	Server_id	End_log_pos	Info
 master-bin.000001	96	Query	1	#	use `test`; BEGIN
 master-bin.000001	158	Query	1	#	use `test`; insert into t1 values(10)
-master-bin.000001	240	Xid	1	#	xid=65
-master-bin.000001	267	Query	1	#	use `test`; COMMIT
-master-bin.000001	330	Query	1	#	use `test`; insert into t2 select * from t1
-master-bin.000001	418	Query	1	#	use `test`; BEGIN
-master-bin.000001	480	Query	1	#	use `test`; insert into t1 values(11)
-master-bin.000001	562	Xid	1	#	xid=67
-master-bin.000001	589	Query	1	#	use `test`; COMMIT
+master-bin.000001	240	Xid	1	#	COMMIT /* xid=65 */
+master-bin.000001	267	Query	1	#	use `test`; insert into t2 select * from t1
+master-bin.000001	355	Query	1	#	use `test`; BEGIN
+master-bin.000001	417	Query	1	#	use `test`; insert into t1 values(11)
+master-bin.000001	499	Xid	1	#	COMMIT /* xid=67 */
 alter table t2 engine=INNODB;
 delete from t1;
 delete from t2;
@@ -148,8 +141,7 @@ Log_name	Pos	Event_type	Server_id	End_log_pos	Info
 master-bin.000001	96	Query	1	#	use `test`; BEGIN
 master-bin.000001	158	Query	1	#	use `test`; insert into t1 values(12)
 master-bin.000001	240	Query	1	#	use `test`; insert into t2 select * from t1
-master-bin.000001	328	Xid	1	#	xid=77
-master-bin.000001	355	Query	1	#	use `test`; COMMIT
+master-bin.000001	328	Xid	1	#	COMMIT /* xid=77 */
 delete from t1;
 delete from t2;
 reset master;
@@ -173,8 +165,7 @@ show binlog events from 96;
 Log_name	Pos	Event_type	Server_id	End_log_pos	Info
 master-bin.000001	96	Query	1	#	use `test`; BEGIN
 master-bin.000001	158	Query	1	#	use `test`; insert into t1 values(14)
-master-bin.000001	240	Xid	1	#	xid=93
-master-bin.000001	267	Query	1	#	use `test`; COMMIT
+master-bin.000001	240	Xid	1	#	COMMIT /* xid=93 */
 delete from t1;
 delete from t2;
 reset master;
@@ -195,8 +186,7 @@ Log_name	Pos	Event_type	Server_id	End_log_pos	Info
 master-bin.000001	96	Query	1	#	use `test`; BEGIN
 master-bin.000001	158	Query	1	#	use `test`; insert into t1 values(16)
 master-bin.000001	240	Query	1	#	use `test`; insert into t1 values(18)
-master-bin.000001	322	Xid	1	#	xid=104
-master-bin.000001	349	Query	1	#	use `test`; COMMIT
+master-bin.000001	322	Xid	1	#	COMMIT /* xid=104 */
 delete from t1;
 delete from t2;
 alter table t2 type=MyISAM;

mysql-test/r/mysqlbinlog2.result

@@ -144,6 +144,7 @@ SET TIMESTAMP=1579609943;
 SET @@session.foreign_key_checks=1, @@session.sql_auto_is_null=1, @@session.unique_checks=1;
 SET @@session.sql_mode=0;
 insert into t1 values(null, "f");
+ROLLBACK;
 
 --- offset --
 /*!40019 SET @@session.max_insert_delayed_threads=0*/;
@@ -171,6 +172,7 @@ SET TIMESTAMP=1579609943;
 SET @@session.foreign_key_checks=1, @@session.sql_auto_is_null=1, @@session.unique_checks=1;
 SET @@session.sql_mode=0;
 insert into t1 values(null, "f");
+ROLLBACK;
 
 --- start-position --
 /*!40019 SET @@session.max_insert_delayed_threads=0*/;
@@ -188,6 +190,7 @@ SET TIMESTAMP=1579609943;
 SET @@session.foreign_key_checks=1, @@session.sql_auto_is_null=1, @@session.unique_checks=1;
 SET @@session.sql_mode=0;
 insert into t1 values(null, "f");
+ROLLBACK;
 
 --- stop-position --
 /*!40019 SET @@session.max_insert_delayed_threads=0*/;
@@ -233,6 +236,7 @@ SET TIMESTAMP=1579609943;
 SET @@session.foreign_key_checks=1, @@session.sql_auto_is_null=1, @@session.unique_checks=1;
 SET @@session.sql_mode=0;
 insert into t1 values(null, "f");
+ROLLBACK;
 
 --- stop-datetime --
 /*!40019 SET @@session.max_insert_delayed_threads=0*/;

mysql-test/r/rpl_relayrotate.result

@@ -18,5 +18,5 @@ max(a)
 8000
 show slave status;
 Slave_IO_State	Master_Host	Master_User	Master_Port	Connect_Retry	Master_Log_File	Read_Master_Log_Pos	Relay_Log_File	Relay_Log_Pos	Relay_Master_Log_File	Slave_IO_Running	Slave_SQL_Running	Replicate_Do_DB	Replicate_Ignore_DB	Replicate_Do_Table	Replicate_Ignore_Table	Replicate_Wild_Do_Table	Replicate_Wild_Ignore_Table	Last_Errno	Last_Error	Skip_Counter	Exec_Master_Log_Pos	Relay_Log_Space	Until_Condition	Until_Log_File	Until_Log_Pos	Master_SSL_Allowed	Master_SSL_CA_File	Master_SSL_CA_Path	Master_SSL_Cert	Master_SSL_Cipher	Master_SSL_Key	Seconds_Behind_Master
-#	127.0.0.1	root	MASTER_MYPORT	1	master-bin.000001	687235	#	#	master-bin.000001	Yes	Yes							0		0	687235	#	None		0	No						#
+#	127.0.0.1	root	MASTER_MYPORT	1	master-bin.000001	687172	#	#	master-bin.000001	Yes	Yes							0		0	687172	#	None		0	No						#
 drop table t1;

mysql-test/r/user_var.result

@@ -179,6 +179,7 @@ master-bin.000001	96	User var	1	137	@`a b`=_latin1 0x68656C6C6F COLLATE latin1_s
 master-bin.000001	137	Query	1	223	use `test`; INSERT INTO t1 VALUES(@`a b`)
 master-bin.000001	223	User var	1	265	@`var1`=_latin1 0x273B616161 COLLATE latin1_swedish_ci
 master-bin.000001	265	Query	1	351	use `test`; insert into t1 values (@var1)
+flush logs;
 /*!40019 SET @@session.max_insert_delayed_threads=0*/;
 SET @`a b`:=_latin1 0x68656C6C6F COLLATE latin1_swedish_ci;
 use test;

mysql-test/t/ctype_ucs.test

@@ -339,6 +339,7 @@ set @v=convert('abc' using ucs2);
 reset master;
 insert into t2 values (@v);
 show binlog events from 96;
+flush logs;
 # more important than SHOW BINLOG EVENTS, mysqlbinlog (where we
 # absolutely need variables names to be quoted and strings to be
 # escaped).

mysql-test/t/user_var.test

@@ -110,6 +110,7 @@ INSERT INTO t1 VALUES(@`a b`);
 set @var1= "';aaa";
 insert into t1 values (@var1);
 show binlog events from 96;
+flush logs;
 # more important than SHOW BINLOG EVENTS, mysqlbinlog (where we
 # absolutely need variables names to be quoted and strings to be
 # escaped).

sql/log.cc

@@ -119,7 +119,7 @@ static int binlog_commit(THD *thd, bool all)
   }
 
   /* Update the binary log as we have cached some queries */
-  error= mysql_bin_log.write(thd, trans_log, 1);
+  error= mysql_bin_log.write(thd, trans_log);
   binlog_cleanup_trans(trans_log);
   DBUG_RETURN(error);
 }
@@ -142,7 +142,11 @@ static int binlog_rollback(THD *thd, bool all)
      non-transactional table inside a transaction...)
   */
   if (unlikely(thd->options & OPTION_STATUS_NO_TRANS_UPDATE))
-    error= mysql_bin_log.write(thd, trans_log, 0);
+  {
+    Query_log_event qev(thd, "ROLLBACK", 8, TRUE, FALSE);
+    qev.write(trans_log);
+    error= mysql_bin_log.write(thd, trans_log);
+  }
   binlog_cleanup_trans(trans_log);
   DBUG_RETURN(error);
 }
@@ -425,7 +429,6 @@ const char *MYSQL_LOG::generate_name(const char *log_name,
                                      const char *suffix,
                                      bool strip_ext, char *buff)
 {
-  DBUG_ASSERT(!strip_ext || (log_name && log_name[0]));
   if (!log_name || !log_name[0])
   {
     /*
@@ -611,6 +614,7 @@ bool MYSQL_LOG::open(const char *log_name,
         even if this is not the very first binlog.
       */
       Format_description_log_event s(BINLOG_VERSION);
+      s.flags|= LOG_EVENT_BINLOG_IN_USE_F;
       if (!s.is_valid())
         goto err;
       if (null_created_arg)
@@ -1779,8 +1783,6 @@ uint MYSQL_LOG::next_file_id()
     write()
     thd
     cache		The cache to copy to the binlog
-    is_commit           If true, will write "COMMIT" in the end, if false will
-                        write "ROLLBACK".
 
   NOTE
     - We only come here if there is something in the cache.
@@ -1799,7 +1801,7 @@ uint MYSQL_LOG::next_file_id()
       same updates are run on the slave.
 */
 
-bool MYSQL_LOG::write(THD *thd, IO_CACHE *cache, bool is_commit)
+bool MYSQL_LOG::write(THD *thd, IO_CACHE *cache)
 {
   VOID(pthread_mutex_lock(&LOCK_log));
   DBUG_ENTER("MYSQL_LOG::write(cache");
@@ -1809,18 +1811,10 @@ bool MYSQL_LOG::write(THD *thd, IO_CACHE *cache, bool is_commit)
     uint length;
 
     /*
-      Add the "BEGIN" and "COMMIT" in the binlog around transactions
-      which may contain more than 1 SQL statement. If we run with
-      AUTOCOMMIT=1, then MySQL immediately writes each SQL statement to
-      the binlog when the statement has been completed. No need to add
-      "BEGIN" ... "COMMIT" around such statements. Otherwise, MySQL uses
-      trans_log (that is thd->ha_data[binlog_hton.slot]) to cache
-      the SQL statements until the explicit commit, and at the commit writes
-      the contents in trans_log to the binlog.
-
-      We write the "BEGIN" mark first in the buffer (trans_log) where we
-      store the SQL statements for a transaction. At the transaction commit
-      we will add the "COMMIT mark and write the buffer to the binlog.
+      Log "BEGIN" at the beginning of the transaction.
+      which may contain more than 1 SQL statement.
+      There is no need to append "COMMIT", as  it's already in the 'cache'
+      (in fact, Xid_log_event is there which does the commit on slaves)
     */
     {
       Query_log_event qinfo(thd, "BEGIN", 5, TRUE, FALSE);
@@ -1846,6 +1840,7 @@ bool MYSQL_LOG::write(THD *thd, IO_CACHE *cache, bool is_commit)
     if (reinit_io_cache(cache, READ_CACHE, 0, 0, 0))
       goto err;
     length=my_b_bytes_in_cache(cache);
+    DBUG_EXECUTE_IF("half_binlogged_transaction", length-=100;);
     do
     {
       /* Write data to the binary log file */
@@ -1854,21 +1849,9 @@ bool MYSQL_LOG::write(THD *thd, IO_CACHE *cache, bool is_commit)
       cache->read_pos=cache->read_end;		// Mark buffer used up
     } while ((length=my_b_fill(cache)));
 
-    /*
-      We write the command "COMMIT" as the last SQL command in the
-      binlog segment cached for this transaction
-    */
-
-    {
-      Query_log_event qinfo(thd, 
-                            is_commit ? "COMMIT" : "ROLLBACK",
-                            is_commit ?  6       :  8,
-                            TRUE, FALSE);
-      qinfo.error_code= 0;
-      if (qinfo.write(&log_file) || flush_io_cache(&log_file) ||
-          sync_binlog(&log_file))
+    if (flush_io_cache(&log_file) || sync_binlog(&log_file))
 	goto err;
-    }
+    DBUG_EXECUTE_IF("half_binlogged_transaction", abort(););
     if (cache->error)				// Error on read
     {
       sql_print_error(ER(ER_ERROR_ON_READ), cache->file_name, errno);
@@ -2093,10 +2076,10 @@ void MYSQL_LOG::close(uint exiting)
     end_io_cache(&log_file);
 
     /* don't pwrite in a file opened with O_APPEND - it doesn't work */
-    if (log_file.type == WRITE_CACHE)
+    if (log_file.type == WRITE_CACHE && log_type == LOG_BIN)
     {
       my_off_t offset= BIN_LOG_HEADER_SIZE + FLAGS_OFFSET;
-      char flags=LOG_EVENT_BINLOG_CLOSED_F;
+      char flags=0; // clearing LOG_EVENT_BINLOG_IN_USE_F
       my_pwrite(log_file.file, &flags, 1, offset, MYF(0));
     }
 
@@ -2944,15 +2927,12 @@ int TC_LOG_BINLOG::open(const char *opt_name)
       goto err;
     }
 
-    if (((ev= Log_event::read_log_event(&log, 0, &fdle))) &&
-        (ev->get_type_code() == FORMAT_DESCRIPTION_EVENT))
-    {
-      if (ev->flags & LOG_EVENT_BINLOG_CLOSED_F)
-        error=0;
-      else
+    if ((ev= Log_event::read_log_event(&log, 0, &fdle)) &&
+        ev->get_type_code() == FORMAT_DESCRIPTION_EVENT &&
+        ev->flags & LOG_EVENT_BINLOG_IN_USE_F)
         error= recover(&log, (Format_description_log_event *)ev);
-    }
-    // else nothing to do (probably MySQL 4.x binlog)
+      else
+        error=0;
 
     delete ev;
     end_io_cache(&log);
@@ -3009,6 +2989,8 @@ int TC_LOG_BINLOG::recover(IO_CACHE *log, Format_description_log_event *fdle)
 
   init_alloc_root(&mem_root, tc_log_page_size, tc_log_page_size);
 
+  fdle->flags&= ~LOG_EVENT_BINLOG_IN_USE_F; // abort on the first error
+
   while ((ev= Log_event::read_log_event(log,0,fdle)) && ev->is_valid())
   {
     if (ev->get_type_code() == XID_EVENT)

sql/log_event.h

@@ -303,15 +303,38 @@ struct sql_ex_info
 #endif
 
 /*
-   This flag only makes sense for Format_description_log_event.
-   It is set not when the event is written, but when a binlog file
-   is closed. It serves as a reliable indicator that binlog was
-   closed correctly. (Stop_log_event is not enough, there's always
-   a small chance that mysqld crashes in the middle of insert
-   and end of the binlog would look like a Stop_log_event)
+   This flag only makes sense for Format_description_log_event. It is set
+   when the event is written, and *reset* when a binlog file is
+   closed (yes, it's the only case when MySQL modifies already written
+   part of binlog).  Thus it is a reliable indicator that binlog was
+   closed correctly.  (Stop_log_event is not enough, there's always a
+   small chance that mysqld crashes in the middle of insert and end of
+   the binlog would look like a Stop_log_event).
+
+   This flag is used to detect a restart after a crash,
+   and to provide "unbreakable" binlog. The problem is that on a crash
+   storage engines rollback automatically, while binlog does not.
+   To solve this we use this flag and automatically append ROLLBACK
+   to every non-closed binlog (append virtually, on reading, file itself
+   is not changed). If this flag is found, mysqlbinlog simply prints "ROLLBACK"
+   Replication master does not abort on binlog corruption, but takes it as EOF,
+   and replication slave forces a rollback in this case (see below).
+
+   Note, that old binlogs does not have this flag set, so we get a
+   a backward-compatible behaviour.
 */
 
-#define LOG_EVENT_BINLOG_CLOSED_F   0x1
+#define LOG_EVENT_BINLOG_IN_USE_F       0x1
+
+/*
+  This flag is only used for fake Rotate_log_event. When a master, doing
+  binlog dump, reaches the end of the binlog and fakes a rotate to make
+  the slave to go to a new file, this flag is used if there was no
+  "natural" Rotate_log_event.
+  If this flag is set, slave will execute ROLLBACK before going further
+*/
+
+#define LOG_EVENT_FORCE_ROLLBACK_F      0x1
 
 /*
    If the query depends on the thread (for example: TEMPORARY TABLE).
@@ -335,21 +358,22 @@ struct sql_ex_info
 #define LOG_EVENT_SUPPRESS_USE_F    0x8
 
 /*
-   OPTIONS_WRITTEN_TO_BIN_LOG are the bits of thd->options which must be written
-   to the binlog. OPTIONS_WRITTEN_TO_BINLOG could be written into the
-   Format_description_log_event, so that if later we don't want to replicate a
-   variable we did replicate, or the contrary, it's doable. But it should not be
-   too hard to decide once for all of what we replicate and what we don't, among
-   the fixed 32 bits of thd->options.
+   OPTIONS_WRITTEN_TO_BIN_LOG are the bits of thd->options which must be
+   written to the binlog. OPTIONS_WRITTEN_TO_BINLOG could be written
+   into the Format_description_log_event, so that if later we don't want
+   to replicate a variable we did replicate, or the contrary, it's
+   doable. But it should not be too hard to decide once for all of what
+   we replicate and what we don't, among the fixed 32 bits of
+   thd->options.
    I (Guilhem) have read through every option's usage, and it looks like
-   OPTION_AUTO_IS_NULL and OPTION_NO_FOREIGN_KEYS are the only ones which alter
-   how the query modifies the table. It's good to replicate
-   OPTION_RELAXED_UNIQUE_CHECKS too because otherwise, the slave may insert data
-   slower than the master, in InnoDB.
+   OPTION_AUTO_IS_NULL and OPTION_NO_FOREIGN_KEYS are the only ones
+   which alter how the query modifies the table. It's good to replicate
+   OPTION_RELAXED_UNIQUE_CHECKS too because otherwise, the slave may
+   insert data slower than the master, in InnoDB.
    OPTION_BIG_SELECTS is not needed (the slave thread runs with
-   max_join_size=HA_POS_ERROR) and OPTION_BIG_TABLES is not needed either, as
-   the manual says (because a too big in-memory temp table is automatically
-   written to disk).
+   max_join_size=HA_POS_ERROR) and OPTION_BIG_TABLES is not needed
+   either, as the manual says (because a too big in-memory temp table is
+   automatically written to disk).
 */
 #define OPTIONS_WRITTEN_TO_BIN_LOG (OPTION_AUTO_IS_NULL | \
 OPTION_NO_FOREIGN_KEY_CHECKS | OPTION_RELAXED_UNIQUE_CHECKS)
@@ -470,14 +494,15 @@ class Log_event
   ulong data_written;
 
   /*
-    The master's server id (is preserved in the relay log; used to prevent from
-    infinite loops in circular replication).
+    The master's server id (is preserved in the relay log; used to
+    prevent from infinite loops in circular replication).
   */
   uint32 server_id;
 
   /*
-    Some 16 flags. Look above for LOG_EVENT_TIME_F, LOG_EVENT_FORCED_ROTATE_F,
-    LOG_EVENT_THREAD_SPECIFIC_F, and LOG_EVENT_SUPPRESS_USE_F for notes.
+    Some 16 flags. Look above for LOG_EVENT_TIME_F,
+    LOG_EVENT_FORCED_ROTATE_F, LOG_EVENT_THREAD_SPECIFIC_F, and
+    LOG_EVENT_SUPPRESS_USE_F for notes.
   */
   uint16 flags;
 

sql/mysqld.cc

@@ -2684,7 +2684,7 @@ server.");
     ln= mysql_bin_log.generate_name(opt_bin_logname, "-bin", 1, buf);
     if (ln == buf)
     {
-      my_free(opt_bin_logname, MYF(0));
+      my_free(opt_bin_logname, MYF(MY_ALLOW_ZERO_PTR));
       opt_bin_logname=my_strdup(buf, MYF(0));
     }
     mysql_bin_log.open_index_file(opt_binlog_index_name, ln);

sql/slave.cc

@@ -670,11 +670,11 @@ int terminate_slave_thread(THD* thd, pthread_mutex_t* term_lock,
     }
   }
   DBUG_ASSERT(thd != 0);
+  THD_CHECK_SENTRY(thd);
   /*
-    Is is criticate to test if the slave is running. Otherwise, we might
+    Is is critical to test if the slave is running. Otherwise, we might
     be referening freed memory trying to kick it
   */
-  THD_CHECK_SENTRY(thd);
 
   while (*slave_running)			// Should always be true
   {
@@ -2935,8 +2935,7 @@ static ulong read_event(MYSQL* mysql, MASTER_INFO *mi, bool* suppress_warnings)
       *suppress_warnings= TRUE;
     }
     else
-      sql_print_error("Error reading packet from server: %s (\
-server_errno=%d)",
+      sql_print_error("Error reading packet from server: %s ( server_errno=%d)",
 		      mysql_error(mysql), mysql_errno(mysql));
     return packet_error;
   }
@@ -3167,7 +3166,21 @@ static int exec_relay_log_event(THD* thd, RELAY_LOG_INFO* rli)
     thd->set_time();				// time the query
     thd->lex->current_select= 0;
     if (!ev->when)
+    {
       ev->when = time(NULL);
+      /*
+        fake Rotate: it means that normal execution flow of statements is
+        interrupted.  Let's fake ROLLBACK to undo any half-executed transaction
+      */
+      if (ev->get_type_code() == ROTATE_EVENT &&
+          ev->flags & LOG_EVENT_FORCE_ROLLBACK_F)
+      {
+        ha_rollback_stmt(thd);
+        ha_rollback(thd);
+        thd->options&= ~(ulong) (OPTION_BEGIN | OPTION_STATUS_NO_TRANS_UPDATE);
+        thd->server_status&= ~SERVER_STATUS_IN_TRANS;
+      }
+    }
     ev->thd = thd;
     exec_res = ev->exec_event(rli);
     DBUG_ASSERT(rli->sql_thd==thd);
@@ -3260,7 +3273,6 @@ extern "C" pthread_handler_decl(handle_slave_io,arg)
     goto err;
   }
   
-
   thd->proc_info = "Connecting to master";
   // we can get killed during safe_connect
   if (!safe_connect(thd, mysql, mi))
@@ -3354,9 +3366,9 @@ after reconnect");
       bool suppress_warnings= 0;    
       /* 
          We say "waiting" because read_event() will wait if there's nothing to
-         read. But if there's something to read, it will not wait. The important
-         thing is to not confuse users by saying "reading" whereas we're in fact
-         receiving nothing.
+         read. But if there's something to read, it will not wait. The
+         important thing is to not confuse users by saying "reading" whereas
+         we're in fact receiving nothing.
       */
       thd->proc_info = "Waiting for master to send event";
       ulong event_len = read_event(mysql, mi, &suppress_warnings);
@@ -3870,6 +3882,7 @@ static int process_io_rotate(MASTER_INFO *mi, Rotate_log_event *rev)
   if (disconnect_slave_event_count)
     events_till_disconnect++;
 #endif
+
   /*
     If description_event_for_queue is format <4, there is conversion in the
     relay log to the slave's format (4). And Rotate can mean upgrade or
@@ -3893,8 +3906,8 @@ static int process_io_rotate(MASTER_INFO *mi, Rotate_log_event *rev)
 }
 
 /*
-  Reads a 3.23 event and converts it to the slave's format. This code was copied
-  from MySQL 4.0.
+  Reads a 3.23 event and converts it to the slave's format. This code was
+  copied from MySQL 4.0.
 */
 static int queue_binlog_ver_1_event(MASTER_INFO *mi, const char *buf,
 			   ulong event_len)
@@ -4157,9 +4170,9 @@ int queue_event(MASTER_INFO* mi,const char* buf, ulong event_len)
       to write this event again).
     */
     /*
-      We are the only thread which reads/writes description_event_for_queue. The
-      relay_log struct does not move (though some members of it can change), so
-      we needn't any lock (no rli->data_lock, no log lock).
+      We are the only thread which reads/writes description_event_for_queue.
+      The relay_log struct does not move (though some members of it can
+      change), so we needn't any lock (no rli->data_lock, no log lock).
     */
     Format_description_log_event* tmp;
     const char* errmsg;

sql/sql_class.h

@@ -301,7 +301,7 @@ class MYSQL_LOG: public TC_LOG
   bool write(THD *thd, const char *query, uint query_length,
 	     time_t query_start=0);
   bool write(Log_event* event_info); // binary log write
-  bool write(THD *thd, IO_CACHE *cache, bool commit_or_rollback);
+  bool write(THD *thd, IO_CACHE *cache);
 
   /*
     v stands for vector
@@ -314,11 +314,11 @@ class MYSQL_LOG: public TC_LOG
   void make_log_name(char* buf, const char* log_ident);
   bool is_active(const char* log_file_name);
   int update_log_index(LOG_INFO* linfo, bool need_update_threads);
-  int purge_logs(const char *to_log, bool included, 
+  int purge_logs(const char *to_log, bool included,
                  bool need_mutex, bool need_update_threads,
                  ulonglong *decrease_log_space);
   int purge_logs_before_date(time_t purge_time);
-  int purge_first_log(struct st_relay_log_info* rli, bool included); 
+  int purge_first_log(struct st_relay_log_info* rli, bool included);
   bool reset_logs(THD* thd);
   void close(uint exiting);
 

sql/sql_repl.h

@@ -36,7 +36,11 @@ extern I_List<i_string> binlog_do_db, binlog_ignore_db;
 extern int max_binlog_dump_events;
 extern my_bool opt_sporadic_binlog_dump_fail;
 
-#define KICK_SLAVE(thd) { pthread_mutex_lock(&(thd)->LOCK_delete); (thd)->awake(THD::NOT_KILLED); pthread_mutex_unlock(&(thd)->LOCK_delete); }
+#define KICK_SLAVE(thd) do {                                            \
+                          pthread_mutex_lock(&(thd)->LOCK_delete);      \
+                          (thd)->awake(THD::NOT_KILLED);                \
+                          pthread_mutex_unlock(&(thd)->LOCK_delete);    \
+                        } while(0)
 
 int start_slave(THD* thd, MASTER_INFO* mi, bool net_report);
 int stop_slave(THD* thd, MASTER_INFO* mi, bool net_report);