Current File : //proc/self/root/usr/share/mysql-test/suite/rpl/t/rpl_low_slave_net_time_out.test
###############################################################################
# Bug #19975697 SLAVE IO_THREAD MAY GET STUCK WHEN USING GTID AND LOW
# SLAVE_NET_TIMEOUTS
#
# Problem: Dump thread is not checking the necessity of heartbeat event
# while it is scanning through the binary log files and skipping some groups
# which were already present at Slave.
# Fix: Dump thread will check whether it is time to send an heartbeat event or
# not before skipping an event. If so, it will send one to Slave.
###############################################################################
--source include/have_binlog_format_statement.inc
--source include/have_gtid.inc
--source include/have_debug_sync.inc
--source include/force_restart.inc
# Problem is reproduced only with gtid auto position protocol
--let $use_gtids=1
--let $rpl_skip_start_slave=1;
--source include/master-slave.inc
# set slave_net_timeout to 10 seconds and then start slave threads
--connection slave
CHANGE MASTER TO MASTER_HEARTBEAT_PERIOD=5;
SET @save_old_slave_net_timeout=@@global.slave_net_timeout;
SET @@global.slave_net_timeout=10;
--source include/start_slave.inc
# In order to reproduce the bug with a smaller test data,
# inject 2 seconds delay after an skipped event.
--connection master
SET @save_debug=@@global.debug;
SET GLOBAL DEBUG='+d,inject_2sec_sleep_when_skipping_an_event';
# Have some gtid transactions in binary log
# Below create and two insert statements would have created '6' sub events
# # GTID 1
# # CREATE
# # GTID 2
# # BEGIN
# # INSERT
# # COMMIT
#
CREATE TABLE t1(i INT) engine=innodb;
INSERT INTO t1 VALUES (1);
# Sync it with Slave
--source include/sync_slave_sql_with_master.inc
# Stop the Replication
--source include/stop_slave.inc
# Create another gtid transaction at Master
--connection master
# Also, we introduce 'hold_dump_thread_inside_inner_loop' debug simulation now
# to see what happens if dump thread is killed when the logic is
# in inner while loop. The server code resets this debug point
# after it is trigged once. So the sync of the following 'drop table'
# statement will happen with the second dump thread that will be created
# after slave triggers for another dump request after slave_net_timeout(10)
# seconds.
--source include/stop_dump_threads.inc
SET GLOBAL DEBUG='+d,hold_dump_thread_inside_inner_loop';
DROP TABLE t1;
# Start the Replication
# This will create a dump thread on Master and because of
# debug point, it will wait in inner loop.
--connection slave
--source include/start_slave.inc
--connection master
SET DEBUG_SYNC='now WAIT_FOR signal_inside_inner_loop';
# When a dump thread is waiting inside inner loop,
# restart slave threads which will kill the existing
# dump thread and will recreate it.
--connection slave
--source include/stop_slave.inc
# Reset debug point for the next dump thread to ignore it
--connection master
SET GLOBAL DEBUG='-d,hold_dump_thread_inside_inner_loop';
# Start slave thread
--connection slave
--source include/start_slave.inc
# Signal the zombie dump thread to go ahead.
--connection master
SET DEBUG_SYNC='now signal signal_continue';
# Now Dump thread-I/O thread communitcation protocol will detect that
# it does not have recent gtid transaction(drop table) and Master
# should send that transaction to Slave. This detection will take
# 12 seconds because of the injected 2 seconds delay after every skipped
# event (Please see above) which is more than the slave_net_timeout.
# But now, after the fix, the server will check for necessity of sending an
# heartbeat event even while it is skipping the events, the below
# sync should not be timed out after the fix.
--connection master
--source include/sync_slave_sql_with_master.inc
# Cleanup
--connection master
SET GLOBAL DEBUG=@save_debug;
--connection slave
SET @@global.slave_net_timeout=@save_old_slave_net_timeout;
--source include/rpl_end.inc
Mini Shell