mirror of
https://github.com/MariaDB/server.git
synced 2025-08-01 03:47:19 +03:00
bead24b7f3
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
236 lines
7.3 KiB
Plaintext
236 lines
7.3 KiB
Plaintext
#
|
|
# Test for Bug#30113362 : BTR_CUR_WILL_MODIFY_TREE() IS INSUFFICIENT FOR HIGHER TREE LEVEL
|
|
#
|
|
|
|
--source include/have_innodb.inc
|
|
--source include/have_debug.inc
|
|
--source include/have_debug_sync.inc
|
|
--source include/have_innodb_16k.inc
|
|
# debug sync points times out when using valgrind
|
|
--source include/not_valgrind.inc
|
|
|
|
SET @old_innodb_limit_optimistic_insert_debug = @@innodb_limit_optimistic_insert_debug;
|
|
SET @old_innodb_adaptive_hash_index = @@innodb_adaptive_hash_index;
|
|
SET @old_innodb_stats_persistent = @@innodb_stats_persistent;
|
|
|
|
SET GLOBAL innodb_adaptive_hash_index = false;
|
|
SET GLOBAL innodb_stats_persistent = false;
|
|
|
|
connect (purge_control,localhost,root,,);
|
|
START TRANSACTION WITH CONSISTENT SNAPSHOT;
|
|
|
|
--connect (con2,localhost,root,,)
|
|
|
|
CREATE TABLE t1 (
|
|
a00 CHAR(255) NOT NULL DEFAULT 'a',
|
|
a01 CHAR(255) NOT NULL DEFAULT 'a',
|
|
a02 CHAR(255) NOT NULL DEFAULT 'a',
|
|
b INT NOT NULL DEFAULT 0,
|
|
PRIMARY KEY(a00, a01, a02)
|
|
) charset latin1 ENGINE = InnoDB COMMENT='MERGE_THRESHOLD=45';
|
|
|
|
#
|
|
# Prepare primary key index tree to be used for this test.
|
|
#
|
|
|
|
SET GLOBAL innodb_limit_optimistic_insert_debug = 3;
|
|
|
|
delimiter |;
|
|
CREATE PROCEDURE data_load_t1()
|
|
BEGIN
|
|
DECLARE c1 INT DEFAULT 97;
|
|
DECLARE c2 INT DEFAULT 97;
|
|
DECLARE c3 INT DEFAULT 97;
|
|
|
|
WHILE c1 < 102 DO
|
|
WHILE c2 < 123 DO
|
|
WHILE c3 < 123 DO
|
|
INSERT INTO t1 (a00) VALUES (CHAR(c1,c2,c3));
|
|
SET c3 = c3 + 1;
|
|
END WHILE;
|
|
SET c3 = 97;
|
|
SET c2 = c2 + 1;
|
|
END WHILE;
|
|
SET c2 = 97;
|
|
SET c1 = c1 + 1;
|
|
END WHILE;
|
|
END |
|
|
delimiter ;|
|
|
call data_load_t1();
|
|
DROP PROCEDURE data_load_t1;
|
|
|
|
# all node pages are sparse (max 3 node_ptrs)
|
|
ANALYZE TABLE t1;
|
|
SELECT CLUST_INDEX_SIZE FROM information_schema.INNODB_SYS_TABLESTATS WHERE NAME = 'test/t1';
|
|
|
|
connection con2;
|
|
DELETE FROM t1 WHERE a00 = 'cnm';
|
|
COMMIT;
|
|
BEGIN;
|
|
INSERT INTO t1 SET a00 = 'cnm';
|
|
# causes "domino falling" merges to upper level
|
|
connection purge_control;
|
|
COMMIT;
|
|
connection con2;
|
|
SET GLOBAL innodb_limit_optimistic_insert_debug = 0;
|
|
ROLLBACK;
|
|
|
|
# at this moment, in the tree,
|
|
# ...
|
|
# level 4: ...(ast,avw,ayz)(bcc,bff,bii,bll,boo,brr,buu,bxx,cba,ced,cqp,cts)(cwv,czy,ddb)...
|
|
# ...
|
|
|
|
--echo # Test start
|
|
|
|
# (1) Similar case to the first reported corefile at bug#30113362
|
|
# - Deleting 'bii' causes "domino falling" merges and the node_ptr becomes left_most of level 4.
|
|
# So, the operation needs upper level pages' X-latch, though doesn't cause merge more.
|
|
|
|
connection purge_control;
|
|
START TRANSACTION WITH CONSISTENT SNAPSHOT;
|
|
connection con2;
|
|
DELETE FROM t1 WHERE a00 = 'bii';
|
|
COMMIT;
|
|
BEGIN;
|
|
INSERT INTO t1 SET a00 = 'bii';
|
|
SET DEBUG_SYNC = 'rollback_undo_pk SIGNAL roll1_wait WAIT_FOR roll2';
|
|
SET DEBUG_SYNC = 'rollback_purge_clust SIGNAL rollback_waiting WAIT_FOR resume';
|
|
send ROLLBACK;
|
|
|
|
connection purge_control;
|
|
SET DEBUG_SYNC = 'now WAIT_FOR roll1_wait';
|
|
COMMIT;
|
|
SET DEBUG_SYNC = 'now SIGNAL roll2';
|
|
|
|
connect (con1,localhost,root,,);
|
|
# FIXME: This occasionally times out!
|
|
--disable_warnings
|
|
SET DEBUG_SYNC = 'now WAIT_FOR rollback_waiting TIMEOUT 1';
|
|
--enable_warnings
|
|
SET DEBUG_SYNC = 'rw_s_lock_waiting SIGNAL lockwait1';
|
|
send SELECT a00 FROM t1 WHERE a00 = 'bii';
|
|
|
|
connection default;
|
|
# FIXME: This occasionally times out!
|
|
--disable_warnings
|
|
SET DEBUG_SYNC = 'now WAIT_FOR lockwait1 TIMEOUT 1';
|
|
--enable_warnings
|
|
# bug#30113362 caused deadlock
|
|
SET DEBUG_SYNC = 'now SIGNAL resume';
|
|
|
|
connection con1;
|
|
reap;
|
|
connection con2;
|
|
reap;
|
|
connection default;
|
|
|
|
ANALYZE TABLE t1;
|
|
SELECT CLUST_INDEX_SIZE FROM information_schema.INNODB_SYS_TABLESTATS WHERE NAME = 'test/t1';
|
|
|
|
# (2) Confirm blocking domain caused by DELETE modify_tree for tall index tree
|
|
|
|
# at this moment, in the tree,
|
|
# ...
|
|
# level 4: ...(ajk,amn,apq)(ast,avw,ayz,bll,boo,brr,buu,bxx,cba,ced,cqp,cts)(cwv,czy,ddb)(dge,djh,dmk)(dpn,dsq,dvt)(dyw,ebz,efc)...
|
|
# ...
|
|
|
|
# makes >17 records in level4 [(2^(4-1))*2 + 1]. (causes never left_most records)
|
|
DELETE FROM t1 WHERE a00 = 'dpn';
|
|
COMMIT;
|
|
INSERT INTO t1 SET a00 = 'dpn';
|
|
ROLLBACK;
|
|
|
|
# at this moment, in the tree,
|
|
# (* before "]" and after "[" records are treated as left_most possible records)
|
|
# ...
|
|
# level 4: ...(ajk,amn,apq)(ast,avw,ayz,bll,boo,brr,buu,bxx],cba,ced,[cqp,cts,cwv,czy,ddb,dge,dsq,dvt)(dyw,ebz,efc)...
|
|
# level 3: ...(cba,ccb,cdc)(ced,cfe,cgf,chg],cih,cji,[ckj,clk,con,cpo)(cqp,crq,csr)...
|
|
# level 2: ...(ckj,cks,clb)(clk,clt],cmc,cml,cmu,cnd,[cnv,coe)(con,cow,cpf)...
|
|
# level 1: ...(cmu,cmx,cna)(cnd],cng,cnj,cnp,[cns)(cnv,cny,cob)...
|
|
# level 0: ...(cnd,cne,cnf)(cng,cnh,cni)(cnj,cnk,cnl,cnn,cno)(cnp,cnq,cnr)...
|
|
|
|
# deletes just 'ced' node_ptr only from level 4. doesn't cause merge and never left_most.
|
|
# adjusts MERGE_THRESHOLD to do so.
|
|
ALTER TABLE t1 COMMENT='MERGE_THRESHOLD=35';
|
|
|
|
connection purge_control;
|
|
START TRANSACTION WITH CONSISTENT SNAPSHOT;
|
|
|
|
connection con2;
|
|
DELETE FROM t1 WHERE a00 = 'cnd';
|
|
COMMIT;
|
|
BEGIN;
|
|
INSERT INTO t1 SET a00 = 'cnd';
|
|
SET DEBUG_SYNC = 'rollback_undo_pk SIGNAL roll1_wait WAIT_FOR roll2';
|
|
SET DEBUG_SYNC = 'rollback_purge_clust SIGNAL rollback_waiting WAIT_FOR resume EXECUTE 2';
|
|
send ROLLBACK;
|
|
|
|
connection purge_control;
|
|
SET DEBUG_SYNC = 'now WAIT_FOR roll1_wait';
|
|
START TRANSACTION WITH CONSISTENT SNAPSHOT;
|
|
SET DEBUG_SYNC = 'now SIGNAL roll2';
|
|
|
|
connection con1;
|
|
# FIXME: For some reason, we will not always receive these signals!
|
|
--disable_warnings
|
|
# An optimistic row_undo_mod_remove_clust_low() will fail.
|
|
SET DEBUG_SYNC = 'now WAIT_FOR rollback_waiting TIMEOUT 1';
|
|
SET DEBUG_SYNC = 'now SIGNAL resume';
|
|
# Wait for the pessimistic row_undo_mod_remove_clust_low() attempt.
|
|
SET DEBUG_SYNC = 'now WAIT_FOR rollback_waiting TIMEOUT 1';
|
|
--enable_warnings
|
|
disconnect purge_control;
|
|
|
|
# The expectation should be...
|
|
# level 0: (#cnd#,cne,cnf): causes merge
|
|
# level 1: (#cnd#],cng,cnj,cnp,[cns): left_most
|
|
# level 2: (clk,clt],cmc,cml,cmu,#cnd#,[cnv,coe): causes merge
|
|
# level 3: (ced,cfe,cgf,chg],cih,cji,[ckj,#clk#,con,cpo): left_most possible (not cause merge)
|
|
# level 4: (ast,avw,ayz,bll,boo,brr,buu,bxx],cba,#ced#,[cqp,cts,cwv,czy,ddb,dge,dsq,dvt): no merge, not left_most possible
|
|
# So, the top X-latch page is at level4. (ast~dvt)
|
|
|
|
# blocking domain based on whether its ancestor is latched or not.
|
|
# (*[]: ancestor is X-latched)
|
|
# level 0: ...(asq,asr,ass) [(ast,asu,asv)...(dyt,dyu,dyv)] (dyw,dyx,dyy)...
|
|
|
|
# Not blocked searches
|
|
## In MariaDB, both these will block, because we use different DEBUG_SYNC
|
|
## instrumentation (in rollback, not purge) and the root page (number 3)
|
|
## is being latched in row_undo_mod_remove_clust_low().
|
|
## SELECT a00 FROM t1 WHERE a00 = 'ass';
|
|
## SELECT a00 FROM t1 WHERE a00 = 'dyx';
|
|
|
|
## SET DEBUG_SYNC = 'rw_s_lock_waiting SIGNAL lockwait1';
|
|
## send SELECT a00 FROM t1 WHERE a00 = 'ast';
|
|
|
|
## connection con2;
|
|
## SET DEBUG_SYNC = 'rw_s_lock_waiting SIGNAL lockwait2';
|
|
## send SELECT a00 FROM t1 WHERE a00 = 'dyw';
|
|
|
|
connection default;
|
|
## SET DEBUG_SYNC = 'now WAIT_FOR lockwait1';
|
|
## SET DEBUG_SYNC = 'now WAIT_FOR lockwait2';
|
|
SET DEBUG_SYNC = 'now SIGNAL resume';
|
|
|
|
## connection con1;
|
|
## reap;
|
|
disconnect con1;
|
|
|
|
connection con2;
|
|
reap;
|
|
disconnect con2;
|
|
|
|
connection default;
|
|
ANALYZE TABLE t1;
|
|
SELECT CLUST_INDEX_SIZE FROM information_schema.INNODB_SYS_TABLESTATS WHERE NAME = 'test/t1';
|
|
|
|
|
|
# Cleanup
|
|
SET DEBUG_SYNC = 'RESET';
|
|
|
|
DROP TABLE t1;
|
|
|
|
SET GLOBAL innodb_limit_optimistic_insert_debug = @old_innodb_limit_optimistic_insert_debug;
|
|
SET GLOBAL innodb_adaptive_hash_index = @old_innodb_adaptive_hash_index;
|
|
SET GLOBAL innodb_stats_persistent = @old_innodb_stats_persistent;
|