mariadb/mysql-test/suite/innodb/t/lock_update_split_rc.test
Sergei Golubchik bead24b7f3 mariadb-test: wait on disconnect
Remove one of the major sources of race condiitons in mariadb-test.
Normally, mariadb_close() sends COM_QUIT to the server and immediately
disconnects. In mariadb-test it means the test can switch to another
connection and sends queries to the server before the server even
started parsing the COM_QUIT packet and these queries can see the
connection as fully active, as it didn't reach dispatch_command yet.

This is a major source of instability in tests and many - but not all,
still less than a half - tests employ workarounds. The correct one
is a pair count_sessions.inc/wait_until_count_sessions.inc.
Also very popular was wait_until_disconnected.inc, which was completely
useless, because it verifies that the connection is closed, and after
disconnect it always is, it didn't verify whether the server processed
COM_QUIT. Sadly the placebo was as widely used as the real thing.

Let's fix this by making mariadb-test `disconnect` command _to wait_ for
the server to confirm. This makes almost all workarounds redundant.

In some cases count_sessions.inc/wait_until_count_sessions.inc is still
needed, though, as only `disconnect` command is changed:

 * after external tools, like `exec $MYSQL`
 * after failed `connect` command
 * replication, after `STOP SLAVE`
 * Federated/CONNECT/SPIDER/etc after `DROP TABLE`

and also in some XA tests, because an XA transaction is dissociated from
the THD very late, after the server has closed the client connection.

Collateral cleanups: fix comments, remove some redundant statements:
 * DROP IF EXISTS if nothing is known to exist
 * DROP table/view before DROP DATABASE
 * REVOKE privileges before DROP USER
 etc
2025-07-16 09:14:33 +07:00

73 lines
2.2 KiB
Text

--source include/have_innodb.inc
--source include/have_debug.inc
CREATE TABLE t (
`a` INT NOT NULL,
`b` INT NOT NULL,
PRIMARY KEY (`a`)
) ENGINE=InnoDB;
--disable_query_log
SET @old_innodb_limit_optimistic_insert_debug = @@innodb_limit_optimistic_insert_debug;
--enable_query_log
SET GLOBAL innodb_limit_optimistic_insert_debug = 3;
INSERT INTO t VALUES(10, 0);
INSERT INTO t VALUES(20, 0);
INSERT INTO t VALUES(30, 0);
SET TRANSACTION ISOLATION LEVEL READ COMMITTED;
XA START '1';
REPLACE INTO t VALUES(10, 1);
REPLACE INTO t VALUES(20, 1);
# We need the following sync point because mysql_insert() resets
# trx->duplicates with the following condition:
#
# if (duplic == DUP_REPLACE &&
# (!table->triggers || !table->triggers->has_delete_triggers()))
# table->file->extra(HA_EXTRA_WRITE_CANNOT_REPLACE);
#
# and ha_innobase::extra() resets trx_t::duplicates, but we need
# lock_update_split_right() to be invoked when trx->duplicates is set to
# repeat the bug. So the transaction will hang just after
# row_insert_for_mysql() call until another transaction inserts new row and
# splits the page.
SET DEBUG_SYNC= 'ib_after_row_insert SIGNAL inserted WAIT_FOR cont';
--send REPLACE INTO t VALUES(30, 1)
connect (con1,localhost,root);
SET TRANSACTION ISOLATION LEVEL READ COMMITTED;
XA START '2';
SET DEBUG_SYNC= 'now WAIT_FOR inserted';
# The following statement will cause page split and (20, ...) will be split
# record. As the previous REPLACE set non-gap X-lock on it,
# lock_update_split_right() and lock_rec_inherit_to_gap() will 'inherit' the
# lock from the very first (20, ...) new right page record to the supremum of
# the old left page, what should not be for READ COMMITTED isolation level
INSERT INTO t VALUES(40, 2);
SET DEBUG_SYNC= 'now SIGNAL cont';
--connection default
--reap
XA END '1';
# This will cause the assertion failure, because the supremum of the left page
# has X-lock.
XA PREPARE '1';
--connection default
XA COMMIT '1';
--connection con1
XA END '2';
XA PREPARE '2';
XA COMMIT '2';
--disconnect con1
--connection default
SET DEBUG_SYNC= "RESET";
DROP TABLE t;
--disable_query_log
SET GLOBAL innodb_limit_optimistic_insert_debug = @old_innodb_limit_optimistic_insert_debug;
--enable_query_log