mariadb/mysql-test/t/rpl_sp.test

267 lines
6.7 KiB
Text
Raw Normal View History

Approximative fixes for BUG#2610,2611,9100 i.e. WL#2146 binlogging/replication of routines (stored procs and functions). Approximative, because it's using our binlogging way (what we call "query"-level) and this is not as good as record-level binlog (5.1) would be. It imposes several limitations to routines, and has caveats (which I'll document, and for which the server will try to issue errors but that is not always possible). Reason I don't propagate caller info to the binlog as planned is that on master and slave users may be different; even with that some caveats would remain. mysql-test/mysql-test-run.sh: In the testsuite we know what we do, we are not creating nasty routines, and breaking binlog is ok except in rpl_sp. mysql-test/r/blackhole.result: Updating results now that 4.1 has been merged mysql-test/valgrind.supp: Some suppressions for Valgrind (useful on my machine Suse 9.1); this is just adding to the already existing suppressions of pthread and dl. sql/item_func.cc: Don't binlog the substatements when executing a function. If the function is declared to modify data and does not complete, warning "broken binlog". Note that SELECT myfunc() will not be binlogged even if myfunc() updates data (will be documented); but INSERT INTO t VALUES(myfunc()) will be binlogged (what decides is if the caller gets binlogged; the function changes nothing to binlogging). sql/log_event.cc: Just making functions which can be re-used when we binlog more strings in status_vars in Query_log_event (e.g. one day "user", "host"). sql/log_event.h: comment sql/mysql_priv.h: --log-bin-trust-routine-creators sql/mysqld.cc: --log-bin-trust-routine-creators sql/set_var.cc: --log-bin-trust-routine-creators sql/share/errmsg.txt: error messages to warn about problems with routines and binlog sql/slave.cc: If in a routine, replication table inclusion/exclusion rules always answer "replicate!" (see comment in code). sql/sp.cc: If binlog is on: errors if one wants to create a non-deterministic update routine (repeatability problem - note that the test is not perfect for functions) or does not have SUPER (because routines can easily be made to destroy slave's data with just CREATE ROUTINE and EXECUTE priv on master). --log-bin-trust-routine-creators removes these errors. Binlogging of CREATE PROCEDURE|FUNCTION. sql/sql_acl.cc: No thd==0 in tables_ok(). sql/sql_parse.cc: Binlogging of CALL (and not of the substatements of the SP). If SP returns error, we don't binlog it (see comment); we push warning in this case. Binlogging of ALTER|DROP PROCEDURE|FUNCTION with safety messages.
2005-05-05 14:20:53 +02:00
# Test of replication of stored procedures (WL#2146 for MySQL 5.0)
source include/master-slave.inc;
# First let's test replication of current_user() (that's a related thing)
# we need a db != test, where we don't have automatic grants
create database if not exists mysqltest1;
use mysqltest1;
create table t1 (a varchar(100));
sync_slave_with_master;
use mysqltest1;
# ********************** PART 1 : STORED PROCEDURES ***************
# Does the same proc as on master get inserted into mysql.proc ?
# (same definer, same properties...)
connection master;
# cleanup
--disable_warnings
drop procedure if exists foo;
drop procedure if exists foo2;
drop procedure if exists foo3;
drop procedure if exists foo4;
drop procedure if exists bar;
drop function if exists fn1;
--enable_warnings
delimiter |;
--error 1418; # not deterministic
create procedure foo()
begin
declare b int;
set b = 8;
insert into t1 values (b);
insert into t1 values (unix_timestamp());
end|
--replace_column 2 # 5 #
show binlog events from 98| # check that not there
create procedure foo() deterministic
begin
declare b int;
set b = 8;
insert into t1 values (b);
insert into t1 values (unix_timestamp());
end|
delimiter ;|
# we replace columns having times
# (even with fixed timestamp displayed time may changed based on TZ)
--replace_result localhost.localdomain localhost 127.0.0.1 localhost
--replace_column 13 # 14 #
select * from mysql.proc where name='foo' and db='mysqltest1';
sync_slave_with_master;
--replace_result localhost.localdomain localhost 127.0.0.1 localhost
--replace_column 13 # 14 #
select * from mysql.proc where name='foo' and db='mysqltest1';
# Now when we call it, does the CALL() get into binlog,
# or the substatements?
connection master;
# see if timestamp used in SP on slave is same as on master
set timestamp=1000000000;
call foo();
--replace_column 2 # 5 #
show binlog events from 308;
select * from t1;
sync_slave_with_master;
select * from t1;
# Now a SP which is supposed to not update tables (CALL should not be
# binlogged) as it's "read sql data", so should not give error even if
# non-deterministic.
connection master;
delete from t1;
create procedure foo2()
not deterministic
reads sql data
select * from mysqltest1.t1;
call foo2();
# verify CALL is not in binlog
--replace_column 2 # 5 #
show binlog events from 605;
--error 1418;
alter procedure foo2 contains sql;
# SP with definer's right
drop table t1;
create table t1 (a int);
create table t2 like t1;
create procedure foo3()
deterministic
insert into t1 values (15);
# let's create a non-privileged user
grant CREATE ROUTINE, EXECUTE on mysqltest1.* to "zedjzlcsjhd"@127.0.0.1;
grant SELECT on mysqltest1.t1 to "zedjzlcsjhd"@127.0.0.1;
grant SELECT, INSERT on mysqltest1.t2 to "zedjzlcsjhd"@127.0.0.1;
connect (con1,127.0.0.1,zedjzlcsjhd,,mysqltest1,$MASTER_MYPORT,);
connection con1;
--error 1419; # only full-global-privs user can create a routine
create procedure foo4()
deterministic
insert into t1 values (10);
connection master;
set global log_bin_trust_routine_creators=1;
connection con1;
delimiter |;
create procedure foo4()
deterministic
begin
insert into t2 values(3);
insert into t1 values (5);
end|
delimiter ;|
# I add ,0 so that it does not print the error in the test output,
# because this error is hostname-dependent
--error 1142,0;
Approximative fixes for BUG#2610,2611,9100 i.e. WL#2146 binlogging/replication of routines (stored procs and functions). Approximative, because it's using our binlogging way (what we call "query"-level) and this is not as good as record-level binlog (5.1) would be. It imposes several limitations to routines, and has caveats (which I'll document, and for which the server will try to issue errors but that is not always possible). Reason I don't propagate caller info to the binlog as planned is that on master and slave users may be different; even with that some caveats would remain. mysql-test/mysql-test-run.sh: In the testsuite we know what we do, we are not creating nasty routines, and breaking binlog is ok except in rpl_sp. mysql-test/r/blackhole.result: Updating results now that 4.1 has been merged mysql-test/valgrind.supp: Some suppressions for Valgrind (useful on my machine Suse 9.1); this is just adding to the already existing suppressions of pthread and dl. sql/item_func.cc: Don't binlog the substatements when executing a function. If the function is declared to modify data and does not complete, warning "broken binlog". Note that SELECT myfunc() will not be binlogged even if myfunc() updates data (will be documented); but INSERT INTO t VALUES(myfunc()) will be binlogged (what decides is if the caller gets binlogged; the function changes nothing to binlogging). sql/log_event.cc: Just making functions which can be re-used when we binlog more strings in status_vars in Query_log_event (e.g. one day "user", "host"). sql/log_event.h: comment sql/mysql_priv.h: --log-bin-trust-routine-creators sql/mysqld.cc: --log-bin-trust-routine-creators sql/set_var.cc: --log-bin-trust-routine-creators sql/share/errmsg.txt: error messages to warn about problems with routines and binlog sql/slave.cc: If in a routine, replication table inclusion/exclusion rules always answer "replicate!" (see comment in code). sql/sp.cc: If binlog is on: errors if one wants to create a non-deterministic update routine (repeatability problem - note that the test is not perfect for functions) or does not have SUPER (because routines can easily be made to destroy slave's data with just CREATE ROUTINE and EXECUTE priv on master). --log-bin-trust-routine-creators removes these errors. Binlogging of CREATE PROCEDURE|FUNCTION. sql/sql_acl.cc: No thd==0 in tables_ok(). sql/sql_parse.cc: Binlogging of CALL (and not of the substatements of the SP). If SP returns error, we don't binlog it (see comment); we push warning in this case. Binlogging of ALTER|DROP PROCEDURE|FUNCTION with safety messages.
2005-05-05 14:20:53 +02:00
call foo4(); # invoker has no INSERT grant on table => failure
show warnings;
connection master;
call foo3(); # success (definer == root)
show warnings;
--replace_result localhost.localdomain localhost 127.0.0.1 localhost
--error 1142,0;
Approximative fixes for BUG#2610,2611,9100 i.e. WL#2146 binlogging/replication of routines (stored procs and functions). Approximative, because it's using our binlogging way (what we call "query"-level) and this is not as good as record-level binlog (5.1) would be. It imposes several limitations to routines, and has caveats (which I'll document, and for which the server will try to issue errors but that is not always possible). Reason I don't propagate caller info to the binlog as planned is that on master and slave users may be different; even with that some caveats would remain. mysql-test/mysql-test-run.sh: In the testsuite we know what we do, we are not creating nasty routines, and breaking binlog is ok except in rpl_sp. mysql-test/r/blackhole.result: Updating results now that 4.1 has been merged mysql-test/valgrind.supp: Some suppressions for Valgrind (useful on my machine Suse 9.1); this is just adding to the already existing suppressions of pthread and dl. sql/item_func.cc: Don't binlog the substatements when executing a function. If the function is declared to modify data and does not complete, warning "broken binlog". Note that SELECT myfunc() will not be binlogged even if myfunc() updates data (will be documented); but INSERT INTO t VALUES(myfunc()) will be binlogged (what decides is if the caller gets binlogged; the function changes nothing to binlogging). sql/log_event.cc: Just making functions which can be re-used when we binlog more strings in status_vars in Query_log_event (e.g. one day "user", "host"). sql/log_event.h: comment sql/mysql_priv.h: --log-bin-trust-routine-creators sql/mysqld.cc: --log-bin-trust-routine-creators sql/set_var.cc: --log-bin-trust-routine-creators sql/share/errmsg.txt: error messages to warn about problems with routines and binlog sql/slave.cc: If in a routine, replication table inclusion/exclusion rules always answer "replicate!" (see comment in code). sql/sp.cc: If binlog is on: errors if one wants to create a non-deterministic update routine (repeatability problem - note that the test is not perfect for functions) or does not have SUPER (because routines can easily be made to destroy slave's data with just CREATE ROUTINE and EXECUTE priv on master). --log-bin-trust-routine-creators removes these errors. Binlogging of CREATE PROCEDURE|FUNCTION. sql/sql_acl.cc: No thd==0 in tables_ok(). sql/sql_parse.cc: Binlogging of CALL (and not of the substatements of the SP). If SP returns error, we don't binlog it (see comment); we push warning in this case. Binlogging of ALTER|DROP PROCEDURE|FUNCTION with safety messages.
2005-05-05 14:20:53 +02:00
call foo4(); # definer's rights => failure
show warnings;
# we test replication of ALTER PROCEDURE
alter procedure foo4 sql security invoker;
call foo4(); # invoker's rights => success
show warnings;
# Check that only successful CALLs are in binlog
--replace_column 2 # 5 #
show binlog events from 841;
# Note that half-failed CALLs are not in binlog, which is a known
# bug. If we compare t2 on master and slave we see they differ:
select * from t1;
select * from t2;
sync_slave_with_master;
select * from t1;
select * from t2;
# Test of DROP PROCEDURE
--replace_result localhost.localdomain localhost 127.0.0.1 localhost
--replace_column 13 # 14 #
select * from mysql.proc where name="foo4" and db='mysqltest1';
connection master;
drop procedure foo4;
select * from mysql.proc where name="foo4" and db='mysqltest1';
sync_slave_with_master;
select * from mysql.proc where name="foo4" and db='mysqltest1';
# ********************** PART 2 : FUNCTIONS ***************
connection master;
drop procedure foo;
drop procedure foo2;
drop procedure foo3;
delimiter |;
create function fn1(x int)
returns int
deterministic
begin
insert into t1 values (x);
return x+2;
end|
delimiter ;|
delete t1,t2 from t1,t2;
select fn1(20);
insert into t2 values(fn1(21));
select * from t1;
select * from t2;
sync_slave_with_master;
select * from t1;
select * from t2;
connection master;
delimiter |;
drop function fn1;
create function fn1()
returns int
deterministic
begin
return unix_timestamp();
end|
delimiter ;|
delete from t1;
set timestamp=1000000000;
insert into t1 values(fn1());
--replace_result localhost.localdomain localhost 127.0.0.1 localhost
--replace_column 13 # 14 #
select * from mysql.proc where db='mysqltest1';
select * from t1;
sync_slave_with_master;
use mysqltest1;
select * from t1;
--replace_result localhost.localdomain localhost 127.0.0.1 localhost
--replace_column 13 # 14 #
select * from mysql.proc where db='mysqltest1';
# And now triggers
connection con1;
--error 1227;
create trigger trg before insert on t1 for each row set new.a= 10;
connection master;
# fn1() above uses timestamps, so in !ps-protocol, the timezone will be
# binlogged, but in --ps-protocol it will not be (BUG#9359) so
# the binlog offsets get shifted which spoils SHOW BINLOG EVENTS.
# To be immune, we take a new binlog.
flush logs;
delete from t1;
# TODO: when triggers can contain an update, test that this update
# does not go into binlog.
# I'm not setting user vars in the trigger, because replication of user vars
# would take care of propagating the user var's value to slave, so even if
# the trigger was not executed on slave it would not be discovered.
create trigger trg before insert on t1 for each row set new.a= 10;
insert into t1 values (1);
select * from t1;
sync_slave_with_master;
select * from t1;
connection master;
delete from t1;
drop trigger t1.trg;
insert into t1 values (1);
select * from t1;
--replace_column 2 # 5 #
show binlog events in 'master-bin.000002' from 98;
sync_slave_with_master;
select * from t1;
Approximative fixes for BUG#2610,2611,9100 i.e. WL#2146 binlogging/replication of routines (stored procs and functions). Approximative, because it's using our binlogging way (what we call "query"-level) and this is not as good as record-level binlog (5.1) would be. It imposes several limitations to routines, and has caveats (which I'll document, and for which the server will try to issue errors but that is not always possible). Reason I don't propagate caller info to the binlog as planned is that on master and slave users may be different; even with that some caveats would remain. mysql-test/mysql-test-run.sh: In the testsuite we know what we do, we are not creating nasty routines, and breaking binlog is ok except in rpl_sp. mysql-test/r/blackhole.result: Updating results now that 4.1 has been merged mysql-test/valgrind.supp: Some suppressions for Valgrind (useful on my machine Suse 9.1); this is just adding to the already existing suppressions of pthread and dl. sql/item_func.cc: Don't binlog the substatements when executing a function. If the function is declared to modify data and does not complete, warning "broken binlog". Note that SELECT myfunc() will not be binlogged even if myfunc() updates data (will be documented); but INSERT INTO t VALUES(myfunc()) will be binlogged (what decides is if the caller gets binlogged; the function changes nothing to binlogging). sql/log_event.cc: Just making functions which can be re-used when we binlog more strings in status_vars in Query_log_event (e.g. one day "user", "host"). sql/log_event.h: comment sql/mysql_priv.h: --log-bin-trust-routine-creators sql/mysqld.cc: --log-bin-trust-routine-creators sql/set_var.cc: --log-bin-trust-routine-creators sql/share/errmsg.txt: error messages to warn about problems with routines and binlog sql/slave.cc: If in a routine, replication table inclusion/exclusion rules always answer "replicate!" (see comment in code). sql/sp.cc: If binlog is on: errors if one wants to create a non-deterministic update routine (repeatability problem - note that the test is not perfect for functions) or does not have SUPER (because routines can easily be made to destroy slave's data with just CREATE ROUTINE and EXECUTE priv on master). --log-bin-trust-routine-creators removes these errors. Binlogging of CREATE PROCEDURE|FUNCTION. sql/sql_acl.cc: No thd==0 in tables_ok(). sql/sql_parse.cc: Binlogging of CALL (and not of the substatements of the SP). If SP returns error, we don't binlog it (see comment); we push warning in this case. Binlogging of ALTER|DROP PROCEDURE|FUNCTION with safety messages.
2005-05-05 14:20:53 +02:00
# Clean up
connection master;
drop function fn1;
drop database mysqltest1;
drop user "zedjzlcsjhd"@127.0.0.1;
sync_slave_with_master;