Bug#38205 Row-based Replication (RBR) causes inconsistencies: HA_ERR_FOUND_DUP

Bug#319  if while a non-transactional slave is replicating a transaction possible problem 

It is impossible to roll back a mixed engines transaction when one of the engine is
non-transaction. In replication that fact is crucial because the slave can not safely
re-apply a transction that was interrupted with STOP SLAVE.

Fixed with making STOP SLAVE not be effective immediately in the case the current
group of replication events has modified a non-transaction table. In order for slave to leave
either the group needs finishing or the user issues KILL QUERY|CONNECTION slave_thread_id.

mysql-test/suite/bugs/t/rpl_bug38205.test

@@ -0,0 +1,165 @@
+#
+# Bug #38205 Row-based Replication (RBR) causes inconsistencies: HA_ERR_FOUND_DUPP_KEY
+#
+# Verifying the fact that STOP SLAVE in the middle of a group execution waits
+# for the end of the group before the slave sql thread will stop.
+# The patch refines STOP SLAVE to not interrupt a transaction or other type of 
+# the replication events group (the part I).
+# Killing the sql thread continues to provide a "hard" stop (the part II).
+#
+# Non-deterministic tests
+#
+
+source include/master-slave.inc;
+source include/have_innodb.inc;
+
+
+#
+# Part II, killed sql slave leaves instantly
+#
+
+# A. multi-statement transaction as the replication group
+
+connection master;
+
+create table t1i(n int primary key) engine=innodb;
+create table t2m(n int primary key) engine=myisam;
+
+sync_slave_with_master;
+
+connection master;
+
+begin;
+insert into t1i values (1);
+insert into t1i values (2);
+insert into t1i values (3);
+commit;
+
+sync_slave_with_master;
+
+#
+# todo: first challenge is to find out the SQL thread id
+# the following is not fully reliable
+#
+
+let $id=`SELECT id from information_schema.processlist where user like 'system user' and state like '%Has read all relay log%' or user like 'system user' and state like '%Reading event from the relay log%'`;
+connection slave;
+begin;
+insert into t1i values (5);
+
+connection master;
+let $pos0_master= query_get_value(SHOW MASTER STATUS, Position, 1);
+begin;
+insert into t1i values (4);
+insert into t2m values (1); # non-ta update
+update t1i set n = 5 where n = 4; # to block at. can't be played with killed
+commit;
+let $pos1_master= query_get_value(SHOW MASTER STATUS, Position, 1);
+
+connection slave;
+# slave sql thread must be locked out by the conn `slave' explicit lock
+let $pos0_slave= query_get_value(SHOW SLAVE STATUS, Exec_Master_Log_Pos, 1);
+--disable_query_log
+eval select $pos0_master - $pos0_slave as zero;
+--enable_query_log
+
+connection slave1;
+
+let $count= 1;
+let $table= t2m;
+source include/wait_until_rows_count.inc;
+#
+# todo: may fail as said above
+#
+--echo *** kill sql thread ***
+--disable_query_log
+eval kill connection $id;
+--enable_query_log
+
+connection slave;
+rollback; # release the sql thread
+
+connection slave1;
+
+source include/wait_for_slave_sql_to_stop.inc;
+let $sql_status= query_get_value(SHOW SLAVE STATUS, Slave_SQL_Running, 1);
+--echo *** sql thread is *not* running: $sql_status ***
+let $pos1_slave= query_get_value(SHOW SLAVE STATUS, Exec_Master_Log_Pos, 1);
+
+connection slave;
+--echo *** the prove: the killed slave has not finished the current transaction ***
+
+--disable_query_log
+select count(*) as three from t1i;
+eval select $pos1_master > $pos1_slave as one;
+eval select $pos1_slave - $pos0_slave as zero;
+--enable_query_log
+
+delete from t2m; # remove the row to be able to replay
+start slave sql_thread;
+
+#
+# Part I: B The homogenous transaction remains interuptable in between
+#
+
+connection master;
+delete from t1i;
+delete from t2m;
+
+sync_slave_with_master;
+begin;
+insert into t1i values (5);
+
+connection master;
+let $pos0_master= query_get_value(SHOW MASTER STATUS, Position, 1);
+begin;
+insert into t1i values (4);
+update t1i set n = 5 where n = 4; # to block at. not to be played
+commit;
+let $pos1_master= query_get_value(SHOW MASTER STATUS, Position, 1);
+
+
+connection slave1;
+# slave sql can't advance as must be locked by the conn `slave' trans
+let $pos0_slave= query_get_value(SHOW SLAVE STATUS, Exec_Master_Log_Pos, 1);
+--disable_query_log
+eval select $pos0_master - $pos0_slave as zero;
+--enable_query_log
+
+#
+# the replicated trans is blocked by the slave's local.
+# However, it's not easy to catch the exact moment when it happens.
+# The test issues sleep which makes the test either non-deterministic or
+# wasting too much time.
+#
+--sleep 3
+
+send stop slave sql_thread;
+
+connection slave;
+rollback; # release the sql thread
+
+connection slave1;
+reap;
+source include/wait_for_slave_sql_to_stop.inc;
+let $sql_status= query_get_value(SHOW SLAVE STATUS, Slave_SQL_Running, 1);
+--echo *** sql thread is running: $sql_status ***
+
+let $pos1_slave= query_get_value(SHOW SLAVE STATUS, Exec_Master_Log_Pos, 1);
+
+--echo *** the prove: the stopped slave has rolled back the current transaction ***
+
+--disable_query_log
+select count(*) as zero from t1i;
+eval select $pos0_master - $pos0_slave as zero;
+eval select $pos1_master > $pos0_slave as one;
+--enable_query_log
+
+start slave sql_thread;
+
+# clean-up
+
+connection master;
+drop table t1i, t2m;
+
+sync_slave_with_master;