Details
-
Bug
-
Status: Closed (View Workflow)
-
Minor
-
Resolution: Fixed
-
None
-
None
-
None
Description
If a slave crashes while writing an event group into its binary log, specifically after writing GTID event but before finishing Xid, and if gtid_strict_mode is enabled, the slave cannot resume replication after restart, it aborts with the error "An attempt was made to binlog GTID X-Y-Z which would create an out-of-order sequence number with existing GTID X-Y-Z, and gtid strict mode is enabled".
Note: This is a follow-up on the failure that we discussed earlier on IRC. I can now positively confirm that the failure I observed also happened upon slave crash. And as you suggested, I was able to reproduce it with "crash_before_writing_xid" (see the test case below). In fact, in my case the picture was slightly different, the slave crashed one step earlier, right after writing the GTID event, and before writing anything else. I tried to add a debug crash point there and it causes the same problem, so I suppose crash_before_writing_xid will do just as well.
Test case:
--source include/master-slave.inc
|
--source include/have_innodb.inc
|
|
|
create table t1 (i int) engine=InnoDB; |
insert into t1 values (1),(2); |
|
|
--sync_slave_with_master
|
|
|
--source include/stop_slave.inc
|
set sql_log_bin = 0; |
alter table mysql.gtid_slave_pos engine=InnoDB; |
change master to master_use_gtid=current_pos; |
--source include/start_slave.inc
|
|
|
--let $_server_id= `SELECT @@server_id`
|
--let $_expect_file_name= $MYSQLTEST_VARDIR/tmp/mysqld.$_server_id.expect
|
--write_file $_expect_file_name
|
wait
|
EOF
|
|
|
SET GLOBAL debug_dbug="+d,crash_before_writing_xid"; |
|
|
connection master; |
|
|
insert into t1 values (3),(4); |
|
|
connection slave; |
|
|
--source include/wait_until_disconnected.inc
|
|
|
--append_file $_expect_file_name
|
restart: --gtid_strict_mode=1 |
EOF
|
|
|
--enable_reconnect
|
--source include/wait_until_connected_again.inc
|
|
|
show variables like 'gtid%'; |
|
|
# I intentionally don't use the include file here, |
# because start_slave.inc is indeterministic when a problem occurs on startup. |
# If something goes wrong, the next sync_slave_with_master will indicate that |
|
|
start slave;
|
|
|
connection master; |
drop table t1; |
|
|
--sync_slave_with_master
|
query_vertical show slave status;
|
bzr version-info
revision-id: sergii@pisem.net-20130624185655-3ysky07m0gvet6gl
|
revno: 3669
|
branch-nick: 10.0-base
|
Attachments
Issue Links
- relates to
-
MDEV-26 Global transaction ID
-
- Closed
-