1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
|
###############################################################################
# Bug#25656123 MEMORY LEAK IN MASTER-MASTER GTID REPLICATION WITH
# SYNC_RELAY_LOG_INFO
#
# Problem: Applier thread is leaking memory when it is updating
# repository tables for ignored/skipped events.
#
# Steps to reproduce:
#
# 1) Create M1<->M2 replication topology
#
# 2) Create a sample table to create ignored/skipped insert events
# on the table. And run some sample inserts on both nodes, so that
# initially bytes used will not be counted.
#
# 3) If gtid_mode is OFF, set sql_replica_skip_counter to 50 on M2 node
# and restart sql thread to make the change effective.
#
# If gtid_mode is ON, execute 50 empty transactions on M2 node,
# so that 50 transactions from M1 will be skipped.
#
# 4) Record thd::main_mem_root usage values on both servers before
# starting the experiment.
#
# 5) Insert 100 sample queries on M1 node. First 50 inserts will be ignored
# on M2 node (because sql_replica_skip_counter set to 50 in step-3).
# Second 50 inserts will be executed on M2 and will be replicated back to
# M1 node. But they will be ignored on M1 because they were originated on
# M1 node.
#
# 6) Record thd::main_mem_root usage value on M1 after
# the experiment. Check that thd::main_mem_root value
# did not increased.
#
# 7) Record thd::main_mem_root usage value on M2 after
# the experiment. Check that thd::main_mem_root value
# did not increased.
#
# 8) Cleanup.
###############################################################################
# This test is independent of binlog format. Choosing 'row' format.
--source include/have_binlog_format_row.inc
# sql_replica_skip_counter is not supported with compression
--source include/not_binlog_transaction_compression_on.inc
# Performance schema Query used to test memory growth
--let $perf_query=SELECT high_number_of_bytes_used/1024/1024 FROM performance_schema.memory_summary_global_by_event_name WHERE event_name = 'memory/sql/thd::main_mem_root'
--let $master_uuid=`SELECT @@SERVER_UUID`
--echo #
--echo # Step-1) Create M1<->M2 replication topology
--echo #
--let $rpl_topology= 1->2->1
--source include/rpl_init.inc
--echo #
--echo # Step-2) Create a sample table to create ignored/skipped
--echo # insert events on the table. And run some sample
--echo # inserts on both nodes, so that initial hiccups
--echo # in bytes used will not be counted.
--echo #
--let $rpl_connection_name=server_1
--source include/rpl_connection.inc
CREATE TABLE t1(i INT);
--source include/rpl_sync.inc
--let $rpl_connection_name=server_1
--source include/rpl_connection.inc
INSERT INTO t1 VALUES (1);
INSERT INTO t1 VALUES (2);
--let $rpl_connection_name=server_2
--source include/rpl_connection.inc
INSERT INTO t1 VALUES (3);
INSERT INTO t1 VALUES (4);
--source include/rpl_sync.inc
--echo #
--echo # Step-3) If gtid_mode is OFF, set sql_replica_skip_counter to 50 on
--echo # M2 node and restart sql thread to make the change effective.
--echo #
--echo # If gtid_mode is ON, execute 50 empty transactions on
--echo # M2 node, so that 50 transactions from M1 will be skipped.
--echo #
--let $rpl_connection_name=server_2
--source include/rpl_connection.inc
--source include/stop_slave_sql.inc
--disable_query_log
if ( `SELECT @@GLOBAL.GTID_MODE = "OFF"` )
{
SET GLOBAL sql_replica_skip_counter=50;
}
if ( `SELECT @@GLOBAL.GTID_MODE = "ON"` )
{
--let $trx_no=2
while ($trx_no <= 50)
{
--replace_result $master_uuid MASTER_UUID
--eval SET GTID_NEXT='$master_uuid:$trx_no'
BEGIN;
COMMIT;
--inc $trx_no
}
SET GTID_NEXT='AUTOMATIC';
}
--enable_query_log
--source include/start_slave_sql.inc
--echo #
--echo # Step-4) Record thd::main_mem_root usage values on both servers before
--echo # starting the experiment.
--echo #
--let $rpl_connection_name=server_1
--source include/rpl_connection.inc
--let $thd_mem_root_usage_on_M1_before_experiment=`$perf_query`
--let $rpl_connection_name=server_2
--source include/rpl_connection.inc
--let $thd_mem_root_usage_on_M2_before_experiment=`$perf_query`
--echo #
--echo # Step-5) Insert 100 sample queries on M1 node. First 50 inserts will
--echo # be ignored on M2 node (because sql_replica_skip_counter set to
--echo # 50). Second 50 inserts will be executed on M2 and will be
--echo # replicated back to M1 node. But they will be ignored on M1
--echo # because they were originated on M1 node.
--echo #
--let $rpl_connection_name=server_1
--source include/rpl_connection.inc
--disable_query_log
--let $i=0
while ($i < 100)
{
--eval INSERT INTO t1 VALUES ($i)
--inc $i
}
--enable_query_log
--source include/rpl_sync.inc
# Wait for all the events from M2 reach M1 back (they will be ignored but we
# should wait till all the events reach M1 to verify our experiment).
# Simple sync will not work in case of GTIDs, as the GTID_EXECUTED already contains
# everything, sync will think that everything is applied.
# So to achieve what we wanted, execute a sample query on M2 that will *not* be ignored
# by M1, and wait till M1 executes that particular query (i.e., a query and then sync).
--let $rpl_connection_name=server_2
--source include/rpl_connection.inc
INSERT INTO t1 VALUES (100);
--source include/rpl_sync.inc
--echo #
--echo # Step-6) Record thd::main_mem_root usage value on M1 after
--echo # the experiment. Check that thd::main_mem_root value
--echo # did not increased.
--echo #
--let $rpl_connection_name=server_1
--source include/rpl_connection.inc
--let $thd_mem_root_usage_on_M1_after_experiment=`$perf_query`
--let $assert_text= thd::main_mem_root on M1 should be the same value as before experiment value.
# Tolerate up to 1M change. The usage value is converted to MB earlier by
# (high_number_of_bytes_used/1024/1024) but it is evaluated as double making
# the precision of this check in bytes. As any background thread can add
# to mem root allocation, with zero tolerance it results in intermittent
# failure. Failure observed in ASAN build with 64 bytes more allocation
# after clone GTID background is added.
--let $assert_cond= $thd_mem_root_usage_on_M1_after_experiment < $thd_mem_root_usage_on_M1_before_experiment + 1
--source include/assert.inc
--echo #
--echo # Step-7) Record thd::main_mem_root usage value on M2 after
--echo # the experiment. Check that thd::main_mem_root value
--echo # did not increased.
--echo #
--let $rpl_connection_name=server_2
--source include/rpl_connection.inc
--let $thd_mem_root_usage_on_M2_after_experiment=`$perf_query`
--let $assert_text= thd::main_mem_root on M2 should be the same value as before experiment value.
# Tolerate up to 1M change.
--let $assert_cond= $thd_mem_root_usage_on_M2_after_experiment < $thd_mem_root_usage_on_M2_before_experiment + 1
--source include/assert.inc
--echo #
--echo # Step-8) Cleanup.
--echo #
DROP TABLE t1;
--source include/rpl_end.inc
|
