Debian Codegraph @

mysql-test/suite/rpl/t/rpl_parallel_optimistic_error_stop.test

raw
1 --source include/have_innodb.inc
2 --source include/have_debug.inc
3 --source include/have_debug_sync.inc
4 --let $rpl_topology=1->2
5 --source include/rpl_init.inc
6
7 call mtr.add_suppression("Slave: Commit failed due to failure of an earlier commit");
8 call mtr.add_suppression("Slave: Duplicate entry '99'");
9
10 --connection server_1
11 ALTER TABLE mysql.gtid_slave_pos ENGINE=InnoDB;
12 CREATE TABLE t1 (a int PRIMARY KEY, b INT) ENGINE=InnoDB;
13 INSERT INTO t1 VALUES(1,1); # hit a dup entry on slave
14 INSERT INTO t1 VALUES(2,1); # races to "win" the last exit
15 INSERT INTO t1 VALUES(3,1);
16 INSERT INTO t1 VALUES(4,1); # make W3 race over W1
17 --save_master_pos
18
19 --connection server_2
20 --sync_with_master
21 SET @old_parallel_threads=@@GLOBAL.slave_parallel_threads;
22 --source include/stop_slave.inc
23 SET @old_debug_dbug = @@global.debug_dbug;
24 # In a group of W1,W2,W3 of the same batch W2 simulates slowness.
25 SET @@global.debug_dbug = "d,hold_worker2_favor_worker3";
26 SET GLOBAL slave_parallel_threads=4;
27 CHANGE MASTER TO master_use_gtid=slave_pos;
28 SET @old_parallel_mode=@@GLOBAL.slave_parallel_mode;
29 SET GLOBAL slave_parallel_mode='optimistic';
30
31 # MDEV-30780 optimistic parallel slave hangs after hit an error
32 # Test workers hang scenario to prove it's no more neither
33 # out-of-order access to the active gco list.
34 #
35 # The test provides how to reproduce on the OLD version, false by default.
36 # That branch approximates the original hang with an assert that
37 # confirms the OLD version indeed could access already reclaimed gco.
38 --let $old_version_regression=0
39
40
41 --connection server_1
42
43 # Let W1,W2,W3,W4 parallel workers that are going to execute
44 # the following transaction.
45 # W1 holds on with the 1st statement
46 # then crashes W3 with the 2nd into retry,
47 # finally hits with the 3rd a dup entry, on slave.
48 SET @@gtid_seq_no = 2001;
49 BEGIN;
50 UPDATE t1 SET b = 11 WHERE a = 4;
51 UPDATE t1 SET b = 11 WHERE a = 3;
52 UPDATE t1 SET a = 99 WHERE a = 1;
53 COMMIT;
54 # In the buggy version W2 races to "win" the exit last (of W1..3)
55 # and by that to access last a gco struct, garbage-collected.
56 UPDATE t1 SET b = 2 WHERE a = 2;
57 # W3 garbage-collects the gco struct in the buggy version.
58 UPDATE t1 SET b = 3 WHERE a = 3;
59 # W4 resides in following "singleton" batch to a W2 replacement
60 # in the buggy version to allow W3 reclaim the batch's gco.
61 DROP TABLE IF EXISTS phantom_1;
62
63 --source include/save_master_gtid.inc
64
65 --connect (slave_local_0, 127.0.0.1, root,, test, $SLAVE_MYPORT,)
66 begin;
67 UPDATE t1 set b = 11 where a = 4;
68 --connect (slave_local_1, 127.0.0.1, root,, test, $SLAVE_MYPORT,)
69 begin;
70 INSERT INTO t1 VALUES (99, 11);
71
72 --connect (slave_local_2, 127.0.0.1, root,, test, $SLAVE_MYPORT,)
73 begin;
74 UPDATE t1 SET b = 12 WHERE a = 2;
75
76 --connect (slave_local_3, 127.0.0.1, root,, test, $SLAVE_MYPORT,)
77 begin;
78 UPDATE t1 SET b = 13 WHERE a = 3;
79
80 --connection server_2
81 --source include/start_slave.inc
82
83 --echo # W4 is waiting to start its DROP
84
85 --let $wait_condition= SELECT count(*) = 1 FROM information_schema.processlist WHERE state LIKE "Waiting for prior transaction to start commit%"
86 --source include/wait_condition.inc
87
88 --connection slave_local_3
89 # make W3 to set E.cc <- 1
90 rollback;
91 --let $wait_condition= SELECT count(*) = 1 FROM information_schema.processlist WHERE state LIKE "Waiting for prior transaction to commit%"
92 --source include/wait_condition.inc
93
94 --connection slave_local_0
95 # make W3 into retry and delay it to let W1 hit a dupicate error first,
96 # see 'commit' by slave_local_1.
97 rollback;
98 --let $wait_condition= SELECT count(*) = 1 FROM information_schema.processlist WHERE state LIKE "debug sync point: now"
99 --source include/wait_condition.inc
100 SELECT count(*) = 0 as "W3 undid its commit state" FROM information_schema.processlist WHERE state LIKE "Waiting for prior transaction to commit%";
101
102
103 --connection slave_local_2
104 rollback;
105 # wait for W2 to start committing E.cc <- 2
106 --let $wait_condition= SELECT count(*) = 1 FROM information_schema.processlist WHERE state like "Waiting for prior transaction to commit"
107 --source include/wait_condition.inc
108
109 --connection slave_local_1
110
111 # W1 errors out
112 # A. to alert W3
113 # B. W3 will *not* wake up W4 in the fixed version, having to wait for W2 demise.
114 # C. W2 will notify W3 that releases W4 as it would do in normal cases.
115 commit;
116
117 if (!$old_version_regression)
118 {
119 # A. In the fixed version show-processlist W4 is still in the ordered waiting
120 SELECT COUNT(*) = 1 as "W4 remains with the same status" FROM information_schema.processlist WHERE state LIKE "Waiting for prior transaction to start commit%";
121 --let $status= query_get_value("show slave status", Slave_SQL_Running, 1)
122 --echo # Slave_SQL_Running YES = $status
123
124 # B. In the fixed version W3 is waiting for W2,...
125 --let $wait_condition= SELECT count(*) = 1 as "W4 is waiting" FROM information_schema.processlist WHERE state LIKE "Waiting for prior transaction to commit%"
126 --source include/wait_condition.inc
127 --echo # while W2 is held back ...
128 --let $wait_condition= SELECT count(*) = 1 as "W2 simulates slowness" FROM information_schema.processlist WHERE state LIKE "debug sync point: now"
129 --source include/wait_condition.inc
130
131 # C. # ...until NOW.
132 SET DEBUG_SYNC = 'now SIGNAL cont_worker2';
133
134 }
135
136 # To reproduce the hang on the OLD version ...
137 if ($old_version_regression)
138 {
139 # replace the actual fixes block with checking W3,W4 have actually committed,
140 # followed by signaling to W2 like on behalf of W4 which would end up in the hang.
141 --let $wait_condition= SELECT COUNT(*) = 0 as "W4 has moved on" FROM information_schema.processlist WHERE state like "Waiting for prior transaction to start commit"
142 --source include/wait_condition.inc
143 --let $wait_condition= SELECT count(*) = 0 as "W3 does not wait on W2" FROM information_schema.processlist WHERE state LIKE "Waiting for prior transaction to commit%"
144 --source include/wait_condition.inc
145
146 --let $wait_condition= SELECT count(*) = 1 as "W2 simulates slowness" FROM information_schema.processlist WHERE state LIKE "debug sync point: now"
147 --source include/wait_condition.inc
148
149 # Like above, but signaling is done after W4 is done to violate the commit order
150 # that must fire a debug assert.
151 SET DEBUG_SYNC = 'now SIGNAL cont_worker2';
152 }
153
154 --let $slave_sql_errno= 1062
155 --source include/wait_for_slave_sql_error.inc
156
157 # Restore the slave data and resume with replication
158 DELETE FROM t1 WHERE a=99;
159 --source include/start_slave.inc
160 --source include/sync_with_master_gtid.inc
161
162 #
163 # Clean up.
164 #
165 --connection server_2
166 --source include/stop_slave.inc
167 SET GLOBAL slave_parallel_mode=@old_parallel_mode;
168 SET GLOBAL slave_parallel_threads=@old_parallel_threads;
169 SET @@global.debug_dbug = @old_debug_dbug;
170 SET debug_sync = RESET;
171 --source include/start_slave.inc
172
173 --connection server_1
174 DROP TABLE t1;
175 --source include/save_master_gtid.inc
176
177 --connection server_2
178 --source include/sync_with_master_gtid.inc
179
180 --source include/rpl_end.inc