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

167 lines
4.4 KiB
Text
Raw Normal View History

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/r/rpl_bug38205.result: bug#38205 non-deterministic part of tests results. mysql-test/suite/bugs/t/rpl_bug38205.test: bug#38205 non-deterministic part of tests. mysql-test/suite/rpl/r/rpl_start_stop_slave.result: bug#38205 deterministic part of tests results. mysql-test/suite/rpl/t/rpl_start_stop_slave-slave.opt: increasing `innodb_lock_wait_timeout' to make the test pass on slow env w/o timeout expired issue. mysql-test/suite/rpl/t/rpl_start_stop_slave.test: bug#38205 deterministic part of tests. sql/log_event.cc: Augmenting row-based events applying with the notion of thd->transaction.{all,stmt}.modified_non_trans_table. The pair is set and reset according to its specification for the mixed transaction processing. Particualry, once `modified_non_trans_table' is set in the row-events processing loop, it will remain till the commit of the transaction. sql/slave.cc: Consulting `thd->transaction.all.modified_non_trans_table' to decide whether to terminate by the sql thread or to continue even though the sql thread might have been STOP-ed (rli->abort_slave).
2009-03-26 10:25:06 +02:00
#
# Bug #38205 Row-based Replication (RBR) causes inconsistencies: HA_ERR_FOUND_DUPP_KEY
# Bug#319 if while a non-transactional slave is replicating a transaction possible problem
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/r/rpl_bug38205.result: bug#38205 non-deterministic part of tests results. mysql-test/suite/bugs/t/rpl_bug38205.test: bug#38205 non-deterministic part of tests. mysql-test/suite/rpl/r/rpl_start_stop_slave.result: bug#38205 deterministic part of tests results. mysql-test/suite/rpl/t/rpl_start_stop_slave-slave.opt: increasing `innodb_lock_wait_timeout' to make the test pass on slow env w/o timeout expired issue. mysql-test/suite/rpl/t/rpl_start_stop_slave.test: bug#38205 deterministic part of tests. sql/log_event.cc: Augmenting row-based events applying with the notion of thd->transaction.{all,stmt}.modified_non_trans_table. The pair is set and reset according to its specification for the mixed transaction processing. Particualry, once `modified_non_trans_table' is set in the row-events processing loop, it will remain till the commit of the transaction. sql/slave.cc: Consulting `thd->transaction.all.modified_non_trans_table' to decide whether to terminate by the sql thread or to continue even though the sql thread might have been STOP-ed (rli->abort_slave).
2009-03-26 10:25:06 +02:00
#
# 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 *not* running: $sql_status ***
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/r/rpl_bug38205.result: bug#38205 non-deterministic part of tests results. mysql-test/suite/bugs/t/rpl_bug38205.test: bug#38205 non-deterministic part of tests. mysql-test/suite/rpl/r/rpl_start_stop_slave.result: bug#38205 deterministic part of tests results. mysql-test/suite/rpl/t/rpl_start_stop_slave-slave.opt: increasing `innodb_lock_wait_timeout' to make the test pass on slow env w/o timeout expired issue. mysql-test/suite/rpl/t/rpl_start_stop_slave.test: bug#38205 deterministic part of tests. sql/log_event.cc: Augmenting row-based events applying with the notion of thd->transaction.{all,stmt}.modified_non_trans_table. The pair is set and reset according to its specification for the mixed transaction processing. Particualry, once `modified_non_trans_table' is set in the row-events processing loop, it will remain till the commit of the transaction. sql/slave.cc: Consulting `thd->transaction.all.modified_non_trans_table' to decide whether to terminate by the sql thread or to continue even though the sql thread might have been STOP-ed (rli->abort_slave).
2009-03-26 10:25:06 +02:00
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;