| 1 |
--source include/have_debug.inc |
| 2 |
--source include/have_innodb.inc |
| 3 |
--source include/have_binlog_format_statement.inc |
| 4 |
--let $rpl_topology=1->2 |
| 5 |
--source include/rpl_init.inc |
| 6 |
|
| 7 |
--echo *** MDEV-5509: Incorrect value for Seconds_Behind_Master if parallel replication *** |
| 8 |
|
| 9 |
--connection server_2 |
| 10 |
SET @old_parallel_threads=@@GLOBAL.slave_parallel_threads; |
| 11 |
set @old_parallel_mode= @@GLOBAL.slave_parallel_mode; |
| 12 |
--source include/stop_slave.inc |
| 13 |
SET GLOBAL slave_parallel_threads=5; |
| 14 |
set global slave_parallel_mode= optimistic; |
| 15 |
--source include/start_slave.inc |
| 16 |
|
| 17 |
--connection server_1 |
| 18 |
CREATE TABLE t1 (a INT PRIMARY KEY, b INT); |
| 19 |
CALL mtr.add_suppression("Unsafe statement written to the binary log using statement format since BINLOG_FORMAT = STATEMENT. Statement is unsafe because it uses a system function that may return a different value on the slave"); |
| 20 |
--save_master_pos |
| 21 |
|
| 22 |
--connection server_2 |
| 23 |
--sync_with_master |
| 24 |
|
| 25 |
--connection server_1 |
| 26 |
INSERT INTO t1 VALUES (1,sleep(2)); |
| 27 |
--save_master_pos |
| 28 |
|
| 29 |
--connection server_2 |
| 30 |
--sync_with_master |
| 31 |
|
| 32 |
# The slave position (which --sync_with_master waits for) is updated just |
| 33 |
# before the Seconds_Behind_Master. So we have to wait for the zero status |
| 34 |
# to appear, otherwise there is a small window between --sync_with_master |
| 35 |
# and SHOW SLAVE STATUS where we can see a non-zero value. |
| 36 |
--let $slave_param= Seconds_Behind_Master |
| 37 |
--let $slave_param_value= 0 |
| 38 |
--source include/wait_for_slave_param.inc |
| 39 |
--echo Seconds_Behind_Master should be zero here because the slave is fully caught up and idle. |
| 40 |
--let $status_items= Seconds_Behind_Master |
| 41 |
--source include/show_slave_status.inc |
| 42 |
|
| 43 |
|
| 44 |
--echo *** MDEV-8294: Inconsistent behavior of slave parallel threads at runtime *** |
| 45 |
|
| 46 |
--connection server_1 |
| 47 |
INSERT INTO t1 VALUES (10,0); |
| 48 |
# Force a duplicate key error on the slave. |
| 49 |
SET sql_log_bin= 0; |
| 50 |
DELETE FROM t1 WHERE a=10; |
| 51 |
SET sql_log_bin= 1; |
| 52 |
INSERT INTO t1 VALUES (10,0); |
| 53 |
--save_master_pos |
| 54 |
SELECT * FROM t1 WHERE a >= 10 ORDER BY a; |
| 55 |
|
| 56 |
--connection server_2 |
| 57 |
--let $slave_sql_errno= 1062 |
| 58 |
--source include/wait_for_slave_sql_error.inc |
| 59 |
|
| 60 |
# At this point, the worker threads should have stopped also. |
| 61 |
--let $wait_condition= SELECT COUNT(*)=0 FROM information_schema.processlist WHERE User = "system user" AND State = "Waiting for work from SQL thread"; |
| 62 |
--source include/wait_condition.inc |
| 63 |
|
| 64 |
# Check that the pool can still be resized, but remains inactive as no slave |
| 65 |
# SQL thread is running. |
| 66 |
SET GLOBAL slave_parallel_threads=8; |
| 67 |
--let $wait_condition= SELECT COUNT(*)=0 FROM information_schema.processlist WHERE User = "system user" AND State = "Waiting for work from SQL thread"; |
| 68 |
--source include/wait_condition.inc |
| 69 |
|
| 70 |
STOP SLAVE; |
| 71 |
# At this point, the worker threads should have stopped. |
| 72 |
--let $wait_condition= SELECT COUNT(*)=0 FROM information_schema.processlist WHERE User = "system user" AND State = "Waiting for work from SQL thread"; |
| 73 |
--source include/wait_condition.inc |
| 74 |
|
| 75 |
|
| 76 |
SET GLOBAL sql_slave_skip_counter= 1; |
| 77 |
--source include/start_slave.inc |
| 78 |
# At this point, the worker threads should have been spawned. |
| 79 |
--let $wait_condition= SELECT COUNT(*)=8 FROM information_schema.processlist WHERE User = "system user" AND State = "Waiting for work from SQL thread"; |
| 80 |
--source include/wait_condition.inc |
| 81 |
--sync_with_master |
| 82 |
SELECT * FROM t1 WHERE a >= 10 ORDER BY a; |
| 83 |
|
| 84 |
|
| 85 |
--echo *** MDEV-7818: Deadlock occurring with parallel replication and FTWRL *** |
| 86 |
|
| 87 |
--connection server_1 |
| 88 |
CREATE TABLE t2 (a INT PRIMARY KEY, b INT) ENGINE=InnoDB; |
| 89 |
INSERT INTO t2 VALUES (1,0), (2,0), (3,0); |
| 90 |
--save_master_pos |
| 91 |
|
| 92 |
--connection server_2 |
| 93 |
--sync_with_master |
| 94 |
--source include/stop_slave.inc |
| 95 |
|
| 96 |
--connection server_1 |
| 97 |
# Create a group commit with two transactions, will be used to provoke the |
| 98 |
# problematic thread interaction with FTWRL on the slave. |
| 99 |
SET @old_dbug= @@SESSION.debug_dbug; |
| 100 |
SET @commit_id= 4242; |
| 101 |
SET SESSION debug_dbug="+d,binlog_force_commit_id"; |
| 102 |
|
| 103 |
BEGIN; |
| 104 |
UPDATE t2 SET b=b+1 WHERE a=2; |
| 105 |
COMMIT; |
| 106 |
|
| 107 |
BEGIN; |
| 108 |
INSERT INTO t2 VALUES (4,10); |
| 109 |
COMMIT; |
| 110 |
|
| 111 |
SET SESSION debug_dbug= @old_dbug; |
| 112 |
|
| 113 |
INSERT INTO t2 VALUES (5,0); |
| 114 |
INSERT INTO t2 VALUES (6,0); |
| 115 |
INSERT INTO t2 VALUES (7,0); |
| 116 |
INSERT INTO t2 VALUES (8,0); |
| 117 |
INSERT INTO t2 VALUES (9,0); |
| 118 |
INSERT INTO t2 VALUES (10,0); |
| 119 |
INSERT INTO t2 VALUES (11,0); |
| 120 |
INSERT INTO t2 VALUES (12,0); |
| 121 |
INSERT INTO t2 VALUES (13,0); |
| 122 |
INSERT INTO t2 VALUES (14,0); |
| 123 |
INSERT INTO t2 VALUES (15,0); |
| 124 |
INSERT INTO t2 VALUES (16,0); |
| 125 |
INSERT INTO t2 VALUES (17,0); |
| 126 |
INSERT INTO t2 VALUES (18,0); |
| 127 |
INSERT INTO t2 VALUES (19,0); |
| 128 |
--save_master_pos |
| 129 |
|
| 130 |
--connection server_2 |
| 131 |
|
| 132 |
--connect (s1, 127.0.0.1, root,, test, $SLAVE_MYPORT,) |
| 133 |
# Block one transaction on a row lock. |
| 134 |
BEGIN; |
| 135 |
SELECT * FROM t2 WHERE a=2 FOR UPDATE; |
| 136 |
|
| 137 |
--connection server_2 |
| 138 |
|
| 139 |
# Wait for slave thread of the other transaction to have the commit lock. |
| 140 |
--source include/start_slave.inc |
| 141 |
--let $wait_condition= SELECT COUNT(*) > 0 FROM information_schema.processlist WHERE state = "Waiting for prior transaction to commit" |
| 142 |
--source include/wait_condition.inc |
| 143 |
|
| 144 |
--connect (s2, 127.0.0.1, root,, test, $SLAVE_MYPORT,) |
| 145 |
send FLUSH TABLES WITH READ LOCK; |
| 146 |
# The bug was that at this point we were deadlocked. |
| 147 |
# The FTWRL command would wait forever for T2 to commit. |
| 148 |
# T2 would wait for T1 to commit first, but T1 is waiting for |
| 149 |
# the global read lock to be released. |
| 150 |
|
| 151 |
--connection s1 |
| 152 |
# Release the lock that blocs T1 from replicating. |
| 153 |
COMMIT; |
| 154 |
|
| 155 |
--connection s1 |
| 156 |
send STOP SLAVE; |
| 157 |
|
| 158 |
--connection s2 |
| 159 |
reap; |
| 160 |
|
| 161 |
--connection server_1 |
| 162 |
SELECT * FROM t2 ORDER BY a; |
| 163 |
|
| 164 |
--connection s2 |
| 165 |
UNLOCK TABLES; |
| 166 |
|
| 167 |
SELECT "after UNLOCK TABLES" as state; |
| 168 |
|
| 169 |
--connection s1 |
| 170 |
reap; |
| 171 |
|
| 172 |
SELECT "after reap of STOP SLAVE" as state; |
| 173 |
|
| 174 |
--connection server_2 |
| 175 |
--source include/wait_for_slave_to_stop.inc |
| 176 |
--source include/start_slave.inc |
| 177 |
--sync_with_master |
| 178 |
|
| 179 |
SELECT * FROM t2 ORDER BY a; |
| 180 |
|
| 181 |
|
| 182 |
|
| 183 |
--echo *** MDEV-8318: Assertion `!pool->busy' failed in pool_mark_busy(rpl_parallel_thread_pool*) on concurrent FTWRL *** |
| 184 |
|
| 185 |
--connection server_1 |
| 186 |
LOCK TABLE t2 WRITE; |
| 187 |
|
| 188 |
|
| 189 |
--connect (m1,localhost,root,,test) |
| 190 |
--connection m1 |
| 191 |
--let $cid=`SELECT CONNECTION_ID()` |
| 192 |
send FLUSH TABLES WITH READ LOCK; |
| 193 |
|
| 194 |
--connect (m2,localhost,root,,test) |
| 195 |
# We cannot force the race with DEBUG_SYNC, because the race does not |
| 196 |
# exist after fixing the bug. At best we could force a debug sync to |
| 197 |
# time out, which is effectively just a sleep. |
| 198 |
# So just put a small sleep here; it is enough to trigger the bug in |
| 199 |
# most run before the bug fix, and the code should work correctly |
| 200 |
# however the thread scheduling happens. |
| 201 |
--sleep 0.1 |
| 202 |
send FLUSH TABLES WITH READ LOCK; |
| 203 |
|
| 204 |
--connection server_1 |
| 205 |
--replace_result $cid CID |
| 206 |
eval KILL QUERY $cid; |
| 207 |
|
| 208 |
--connection m1 |
| 209 |
--error ER_QUERY_INTERRUPTED |
| 210 |
reap; |
| 211 |
|
| 212 |
--connection server_1 |
| 213 |
UNLOCK TABLES; |
| 214 |
|
| 215 |
--connection m2 |
| 216 |
reap; |
| 217 |
UNLOCK TABLES; |
| 218 |
|
| 219 |
|
| 220 |
# Clean up. |
| 221 |
--connection server_2 |
| 222 |
--source include/stop_slave.inc |
| 223 |
SET GLOBAL slave_parallel_threads=@old_parallel_threads; |
| 224 |
set global slave_parallel_mode= @old_parallel_mode; |
| 225 |
--source include/start_slave.inc |
| 226 |
|
| 227 |
--connection server_1 |
| 228 |
DROP TABLE t1, t2; |
| 229 |
|
| 230 |
--source include/rpl_end.inc |