Details
-
Bug
-
Status: Closed (View Workflow)
-
Major
-
Resolution: Fixed
-
5.5(EOL), 10.0(EOL), 10.1(EOL)
-
10.0.23, 10.0.25
Description
|
Stack trace from 5.5 commit ce7d8c5ee8c459ac692715994fa73b8f29e9e48a |
5.5/sql/handler.cc:5107: int handler::ha_external_lock(THD*, int): Assertion `next_insert_id == 0' failed.
|
151002 21:20:47 [ERROR] mysqld got signal 6 ;
|
|
|
#6 0x00007f3390ef7311 in *__GI___assert_fail (assertion=0xe0ef02 "next_insert_id == 0", file=<optimized out>, line=5107, function=0xe0fb00 "int handler::ha_external_lock(THD*, int)") at assert.c:81
|
#7 0x00000000007f3a46 in handler::ha_external_lock (this=0x7f337cd1a878, thd=0x7f337ccb4080, lock_type=2) at 5.5/sql/handler.cc:5107
|
#8 0x00000000008c0f44 in unlock_external (thd=0x7f337ccb4080, table=0x7f337449b328, count=1) at 5.5/sql/lock.cc:673
|
#9 0x00000000008c04a2 in mysql_unlock_tables (thd=0x7f337ccb4080, sql_lock=0x7f337449b300, free_lock=true) at 5.5/sql/lock.cc:384
|
#10 0x00000000005d576a in close_thread_tables (thd=0x7f337ccb4080) at 5.5/sql/sql_base.cc:1580
|
#11 0x000000000061ddef in handle_delayed_insert (arg=0x7f337ccb4060) at 5.5/sql/sql_insert.cc:2947
|
#12 0x00007f3392cf1b50 in start_thread (arg=<optimized out>) at pthread_create.c:304
|
#13 0x00007f3390fa795d in clone () at ../sysdeps/unix/sysv/linux/x86_64/clone.S:112
|
|
|
Trying to get some variables.
|
Some pointers may be invalid and cause the dump to abort.
|
Query (0x7f3377d96b20):
|
Connection ID (thread ID): 1123
|
Status: KILL_CONNECTION
|
|
RQG grammar test.yy |
thread1:
|
SELECT Id INTO @kill_id FROM INFORMATION_SCHEMA.PROCESSLIST WHERE Info LIKE 'REPLACE DELAYED%' LIMIT 1; KILL QUERY @kill_id ;
|
|
|
thread2:
|
ALTER TABLE _table FORCE ;
|
|
|
query:
|
REPLACE DELAYED INTO _table ( _field ) VALUES ( NULL ) ;
|
|
RQG command line |
perl .runall-new.pl --threads=3 --duration=600 --queries=100M --grammar=test.yy --engine=MyISAM --basedir1=<your basedir> --vardir1=<your vardir>
|
For RQG, I recommend to use lp:~elenst/randgen/mariadb-patches
Attachments
Activity
I failed to create reasonable test case because it needs signal from delayed thread. The following test works but pollutes debug_sync state because signal is issued unconditionally:
diff --git a/mysql-test/r/delayed_sync.result b/mysql-test/r/delayed_sync.result
|
new file mode 100644
|
index 0000000..c6d6395
|
--- /dev/null
|
+++ b/mysql-test/r/delayed_sync.result
|
@@ -0,0 +1,11 @@
|
+#
|
+# MDEV-8889 - Assertion `next_insert_id == 0' failed in
|
+# handler::ha_external_lock
|
+#
|
+CREATE TABLE t1(a INT PRIMARY KEY AUTO_INCREMENT);
|
+SET debug_sync='after_write_delayed SIGNAL rows_written WAIT_FOR insert_delayed_killed';
|
+INSERT DELAYED INTO t1 VALUES(NULL);
|
+SET debug_sync='now WAIT_FOR rows_written';
|
+FLUSH TABLE t1;
|
+SET debug_sync='RESET';
|
+DROP TABLE t1;
|
diff --git a/mysql-test/t/delayed_sync.test b/mysql-test/t/delayed_sync.test
|
new file mode 100644
|
index 0000000..02d3501
|
--- /dev/null
|
+++ b/mysql-test/t/delayed_sync.test
|
@@ -0,0 +1,21 @@
|
+# delayed works differently in embedded server
|
+--source include/not_embedded.inc
|
+--source include/have_debug_sync.inc
|
+
|
+--echo #
|
+--echo # MDEV-8889 - Assertion `next_insert_id == 0' failed in
|
+--echo # handler::ha_external_lock
|
+--echo #
|
+CREATE TABLE t1(a INT PRIMARY KEY AUTO_INCREMENT);
|
+SET debug_sync='after_write_delayed SIGNAL rows_written WAIT_FOR insert_delayed_killed';
|
+send INSERT DELAYED INTO t1 VALUES(NULL);
|
+
|
+connect(con2, localhost, root);
|
+SET debug_sync='now WAIT_FOR rows_written';
|
+FLUSH TABLE t1;
|
+disconnect con2;
|
+
|
+connection default;
|
+reap;
|
+SET debug_sync='RESET';
|
+DROP TABLE t1;
|
diff --git a/sql/sql_insert.cc b/sql/sql_insert.cc
|
index 42d8839..cd16803 100644
|
--- a/sql/sql_insert.cc
|
+++ b/sql/sql_insert.cc
|
@@ -2862,8 +2862,13 @@ pthread_handler_t handle_delayed_insert(void *arg)
|
di->unlink();
|
lock_count=di->lock_count();
|
mysql_mutex_unlock(&LOCK_delayed_insert);
|
+#ifdef ENABLED_DEBUG_SYNC
|
+ if (debug_sync_service)
|
+ debug_sync_set_action(thd, STRING_WITH_LEN("now SIGNAL insert_delayed_killed"));
|
+#endif
|
mysql_mutex_lock(&di->mutex);
|
if (!lock_count && !di->tables_in_use && !di->stacked_inserts)
|
break; // Time to die
|
}
|
|
The problem is in handle_delayed_insert():
if (thd->killed){uint lock_count;/*Remove this from delay insert list so that no one can request atable from this*/mysql_mutex_unlock(&di->mutex);/* !!! Another (killed) thread may call end_delayed_insert() in this gap and change tables_in_use so that we exit this loop inadvertent. */mysql_mutex_lock(&LOCK_delayed_insert);di->unlink();lock_count=di->lock_count();mysql_mutex_unlock(&LOCK_delayed_insert);mysql_mutex_lock(&di->mutex);if (!lock_count && !di->tables_in_use && !di->stacked_inserts)break; // Time to die}