Details
-
Bug
-
Status: Stalled (View Workflow)
-
Major
-
Resolution: Unresolved
-
10.9(EOL), 10.10(EOL), 10.11
-
None
Description
https://buildbot.mariadb.org/#/builders/168/builds/14289
main.lock_sync 'innodb' w6 [ fail ] timeout after 480 seconds
|
Test ended at 2022-09-06 08:22:41
|
Test case timeout after 480 seconds
|
== /buildbot/amd64-ubuntu-1804-clang10-asan/build/mysql-test/var/6/log/lock_sync.log ==
|
disconnect con1;
|
disconnect con2;
|
DROP TABLES t1, t2;
|
#
|
# MDEV-28567 Assertion `0' in open_tables upon function-related operation
|
#
|
CREATE TABLE t1 (a INT);
|
CREATE TABLE t2 (b INT);
|
CREATE TRIGGER tr1 BEFORE INSERT ON t1 FOR EACH ROW UPDATE t2 SET b = 0;
|
CREATE TRIGGER tr2 BEFORE INSERT ON t2 FOR EACH ROW UPDATE t1 SET a = 6;
|
CREATE VIEW v1 AS SELECT * FROM t1;
|
SET AUTOCOMMIT=OFF;
|
SELECT * FROM t1;
|
a
|
DROP TRIGGER tr1;
|
INSERT INTO t2 SELECT * FROM t2;
|
SELECT f() FROM t2;
|
ERROR 42000: FUNCTION test.f does not exist
|
set debug_sync= 'after_open_table_mdl_shared signal s1';
|
ALTER VIEW v1 AS SELECT f() FROM t1;
|
== /buildbot/amd64-ubuntu-1804-clang10-asan/build/mysql-test/var/6/tmp/analyze-timeout-mysqld.1.err ==
|
mysqltest: Could not open connection 'default' after 500 attempts: 2002 Can't connect to local server through socket '/buildbot/amd64-ubuntu-1804-clang10-asan/build/mysql-test/var/tm' (111)
|
Attachments
Issue Links
- is caused by
-
-
- Closed
-
- is duplicated by
-
-
- Closed
-
- is part of
-
MDEV-36647 No red leaves in the forest
-
- Open
-
Activity
The test seem to be broken in quite a few places. Specifically connection switching after send must normally be accompanied by waits for queries to reach appropriate state. Instead it has things like --error 0, ER_LOCK_DEADLOCK to silence its deficiencies. There're at least 3 places that must be inspected, the patch attempts fixing one of them.
I also don't see why we have to issue 8
sync points in MDL_context::acquire_lock(). IIUC mdl_acquire_lock_wait sync point could have been used for the purpose of fixing this bug.
Being more specific:
diff --git a/mysql-test/main/lock_sync.test b/mysql-test/main/lock_sync.test
|
index 998c6529462..7cedf8fbeb2 100644
|
--- a/mysql-test/main/lock_sync.test
|
+++ b/mysql-test/main/lock_sync.test
|
@@ -1100,6 +1100,7 @@ SELECT * FROM t1;
|
# Seized: test/t1 (MDL_SHARED_READ)
|
|
|
--connect (con1,localhost,root,,test)
|
+set debug_sync='mdl_seized_MDL_SHARED_NO_WRITE_t1 SIGNAL con1_seized';
|
--send
|
DROP TRIGGER tr1;
|
# T@7
|
@@ -1109,6 +1110,7 @@ SELECT * FROM t1;
|
# Deadlock: test/t1 (MDL_SHARED_WRITE)
|
|
|
--connection default
|
+set debug_sync='now WAIT_FOR con1_seized';
|
--error 0, ER_LOCK_DEADLOCK
|
INSERT INTO t2 SELECT * FROM t2;
|
# T@6
|
First of all --error 0, ER_LOCK_DEADLOCK should be replaced with --error ER_LOCK_DEADLOCK
Secondly, DROP TRIGGER tr1 can leave required state or even release locks before INSERT INTO t2 SELECT * FROM t2 starts.
Test must ensure the following sequence is preserved:
con1: takes MDL_EXCLUSIVE on tr1
|
default: takes MDL_SHARED_WRITE on t2
|
con1: attempts taking MDL_SHARED_NO_WRITE on t2, waits
|
default: attempts taking MDL_SHARED on tr1, deadlocks
|
I believe existing sync points are more than enough to make it happen.
@@ -1172,6 +1174,7 @@ CREATE FUNCTION f() RETURNS INT RETURN 1;
|
--connection default
|
set debug_sync= 'now wait_for s1';
|
--disable_result_log
|
+--error 0, ER_LOCK_DEADLOCK
|
SELECT * FROM ( SELECT * FROM v1 ) sq;
|
--enable_result_log
|
# T@6
|
Meh, no more recorded unexpected behaviour in this test please. Make proper sequence of calls instead similarly to the above.
Here is a recent failure:
10.11 8283948846740a22f96bbe7bccf250708406d5d9
main.lock_sync 'innodb' w55 [ fail ]Test ended at 2022-11-21 10:44:52CURRENT_TEST: main.lock_syncmysqltest: At line 1222: query 'DROP FUNCTION f' failed: ER_SP_DOES_NOT_EXIST (1305): FUNCTION test.f does not exist