mirror of
https://github.com/MariaDB/server.git
synced 2026-02-02 16:59:08 +01:00
c7a550e9c2
There was a race where a new GTID could be allocated (but not written to the binlog)during the FLUSH, so that the GTID state written at the start of the new binlog file was incorrect. This in turn could lead to duplicate GTID being sent to the slave if it happens to reconnect at that exact point. Signed-off-by: Kristian Nielsen <knielsen@knielsen-hq.org>
120 lines
3.7 KiB
Text
120 lines
3.7 KiB
Text
--source include/have_debug.inc
|
|
--source include/have_debug_sync.inc
|
|
--source include/have_binlog_format_row.inc
|
|
--source include/master-slave.inc
|
|
--source include/have_innodb_binlog.inc
|
|
|
|
--echo *** Test that the GTID state record is written correctly when inside a group commit.
|
|
|
|
# The test needs a group commit to cross the file boundary, so needs the
|
|
# binlog size to be as expected.
|
|
SELECT @@GLOBAL.max_binlog_size;
|
|
|
|
CREATE TABLE t1 (a INT NOT NULL, b INT NOT NULL, c TEXT, PRIMARY KEY(a, b)) ENGINE=InnoDB;
|
|
# Get close to the end of the file.
|
|
--let $i= 0
|
|
while ($i < 10) {
|
|
eval INSERT INTO t1 VALUES (1, $i, REPEAT('$i', 24000));
|
|
inc $i;
|
|
}
|
|
--echo *** Check that we are still in binlog file 0:
|
|
--let $file= query_get_value(SHOW MASTER STATUS, File, 1)
|
|
--echo $file
|
|
|
|
--echo *** Create a group commit that spans into file 1
|
|
--connection master
|
|
--let $i= 1
|
|
while ($i <= 10) {
|
|
--connect (con_$i,127.0.0.1,root,,test,$SERVER_MYPORT_1,)
|
|
eval SET debug_sync='commit_after_release_LOCK_prepare_ordered SIGNAL master_queued$i WAIT_FOR master_cont NO_CLEAR_EVENT';
|
|
send_eval INSERT INTO t1 VALUES (2, $i, REPEAT(CHR(64 + $i), 400));
|
|
--connection master
|
|
eval SET debug_sync='now WAIT_FOR master_queued$i';
|
|
inc $i;
|
|
}
|
|
|
|
SET debug_sync= 'now SIGNAL master_cont';
|
|
|
|
--let $i= 1
|
|
while ($i <= 10) {
|
|
--connection con_$i
|
|
reap;
|
|
--connection master
|
|
--disconnect con_$i
|
|
inc $i;
|
|
}
|
|
SET debug_sync= 'RESET';
|
|
|
|
--echo *** Check that we are now in binlog file 1:
|
|
--let $file= query_get_value(SHOW MASTER STATUS, File, 1)
|
|
--echo $file
|
|
|
|
SELECT a, b, LENGTH(c) FROM t1 ORDER BY a, b;
|
|
|
|
--source include/save_master_gtid.inc
|
|
--connection slave
|
|
--source include/sync_with_master_gtid.inc
|
|
|
|
SELECT a, b, LENGTH(c) FROM t1 ORDER BY a, b;
|
|
|
|
--echo *** Stop and restart the slave.
|
|
# There was a bug that the GTID state record at the start of the file
|
|
# would containt the GTID for the _end_ of the group commit, not of the
|
|
# point in the group commit where it crossed into the next file. And then
|
|
# the slave would replicate from too early and get a duplicate key error.
|
|
--source include/stop_slave.inc
|
|
--source include/start_slave.inc
|
|
|
|
|
|
--echo *** Test FLUSH BINARY LOGS in parallel with GTID allocation.
|
|
|
|
--connection master
|
|
INSERT INTO t1 VALUES (10, 1, 'before race');
|
|
|
|
# Send a FLUSH BINARY LOGS.
|
|
# Make the FLUSH sleep a little after releasing the LOCK_log but before reading
|
|
# the current binlog GTID state, to make the race likely to hit. Using
|
|
# debug_sync here will not be effective after fixing the bug, as the fix
|
|
# will prevent the race and make the debug_sync time out.
|
|
--connect (con1,127.0.0.1,root,,test,$SERVER_MYPORT_1,)
|
|
SET SESSION debug_dbug= '+d,flush_binlog_sleep_after_release_lock_log';
|
|
send FLUSH BINARY LOGS;
|
|
|
|
--connection master
|
|
# The FLUSH will sleep 0.2, sleep 0.1 here to have a high chance to hit the race.
|
|
SET @old_dbug= @@SESSION.debug_dbug;
|
|
SET SESSION debug_dbug= '+d,binlog_sleep_after_alloc_gtid';
|
|
--sleep 0.1
|
|
send INSERT INTO t1 VALUES (10, 2, 'race?');
|
|
|
|
--connection con1
|
|
reap;
|
|
--disconnect con1
|
|
|
|
--connection master
|
|
reap;
|
|
SET SESSION debug_dbug= @old_dbug;
|
|
--source include/save_master_gtid.inc
|
|
|
|
--connection slave
|
|
--source include/sync_with_master_gtid.inc
|
|
|
|
--echo *** Stop and restart the slave, 2.
|
|
# The bug was that the FLUSH could run after the new GTID was allocated, but
|
|
# before it was written into the binlog. This would cause the wrong binlog
|
|
# state at the start of the new binlog file, and a slave that reconnects at
|
|
# that exact GTID will get a duplicate GTID.
|
|
--source include/stop_slave.inc
|
|
--source include/start_slave.inc
|
|
|
|
--connection master
|
|
INSERT INTO t1 VALUES (10, 3, 'check?');
|
|
--source include/save_master_gtid.inc
|
|
--connection slave
|
|
--source include/sync_with_master_gtid.inc
|
|
|
|
|
|
# Cleanup.
|
|
--connection master
|
|
DROP TABLE t1;
|
|
--source include/rpl_end.inc
|