Details
-
Bug
-
Status: Closed (View Workflow)
-
Major
-
Resolution: Fixed
-
5.3.12, 5.5.34, 10.0.6
-
None
-
None
-
centos 6.2 running a debug build of mariadb-5.5.34
Description
When running this RQG test:
./runall.pl --grammar=conf/partitioning/partitions-99-hash.yy --duration=300 --queries=100000000 --threads=64 --basedir=/home/prohaska/mariadb-5.5.34-install/ --vardir=/home/prohaska/vardir.part99.3 --mask-level=0 --mask=0 --seed=1 --mysqld=--general_log --sqltrace
|
The code hits an assert in safe_mutex_unlock. With this code in mi_keycache.c that i added:
lock = &share->key_cache->op_lock;
|
pthread_mutex_lock(&share->key_cache->op_lock);
|
if (flush_key_blocks(share->key_cache, share->kfile, &share->dirty_part_map,
|
FLUSH_RELEASE))
|
{
|
error= my_errno;
|
mi_print_error(info->s, HA_ERR_CRASHED);
|
mi_mark_crashed(info); /* Mark that table must be checked */
|
}
|
=>assert(lock == &share->key_cache->op_lock);
|
pthread_mutex_unlock(&share->key_cache->op_lock);
|
Hits the assert since some other thread probably changed the share->key_cache pointer:
#8 0x0000000000c3988a in mi_assign_to_key_cache (info=0x7fb600b1ba60, key_map=1, key_cache=0x7fb60c83ebe0) at /home/prohaska/mariadb-5.5.34/storage/myisam/mi_keycache.c:89
|
#9 0x0000000000c147ff in ha_myisam::assign_to_keycache (this=0x7fb6005d3e78, thd=0x7fb5f194c060, check_opt=0x7fb5efb22ce0) at /home/prohaska/mariadb-5.5.34/storage/myisam/ha_myisam.cc:1237
|
#10 0x0000000000c752a5 in ha_partition::handle_opt_part (this=0x7fb5f66ffa78, thd=0x7fb5f194c060, check_opt=0x7fb5efb22ce0, part_id=5, flag=5) at /home/prohaska/mariadb-5.5.34/sql/ha_partition.cc:1084
|
#11 0x0000000000c75978 in ha_partition::handle_opt_partitions (this=0x7fb5f66ffa78, thd=0x7fb5f194c060, check_opt=0x7fb5efb22ce0, flag=5) at /home/prohaska/mariadb-5.5.34/sql/ha_partition.cc:1239
|
#12 0x0000000000c75083 in ha_partition::assign_to_keycache (this=0x7fb5f66ffa78, thd=0x7fb5f194c060, check_opt=0x7fb5efb22ce0) at /home/prohaska/mariadb-5.5.34/sql/ha_partition.cc:1011
|
#13 0x00000000007159c0 in mysql_admin_table(THD *, TABLE_LIST *, HA_CHECK_OPT *, const char *, thr_lock_type, bool, bool, uint, int (*)(THD *, TABLE_LIST *, HA_CHECK_OPT *), struct {...}, int (*)(THD *, \
|
TABLE_LIST *)) (thd=0x7fb5f194c060, tables=0x7fb5f11da168, check_opt=0x7fb5efb22ce0, operator_name=0xd7efcd "assign_to_keycache", lock_type=TL_READ_NO_INSERT, open_for_modify=false, repair_table_use_frm=\
|
false, extra_open_options=0, prepare_func=0x0, operator_func=&virtual table offset 504, view_operator_func=0x0) at /home/prohaska/mariadb-5.5.34/sql/sql_admin.cc:649
|
#14 0x0000000000716c27 in mysql_assign_to_keycache (thd=0x7fb5f194c060, tables=0x7fb5f11da168, key_cache_name=0x7fb5f194fe48) at /home/prohaska/mariadb-5.5.34/sql/sql_admin.cc:992
|
#15 0x000000000060a1dc in mysql_execute_command (thd=0x7fb5f194c060) at /home/prohaska/mariadb-5.5.34/sql/sql_parse.cc:2349
|
#16 0x00000000006132be in mysql_parse (thd=0x7fb5f194c060, rawbuf=0x7fb5f11da078 "CACHE INDEX i IN c4", length=19, parser_state=0x7fb5efb23650) at /home/prohaska/mariadb-5.5.34/sql/sql_parse.cc:5798
|
|
|
(gdb) p lock
|
$1 = (void *) 0x7fb622c1e6b0
|
(gdb) p &share->key_cache->op_lock
|
$2 = (safe_mutex_t *) 0x7fb5ff8308b0
|
I presume the initial problem (on unmodified binaries) was this:
safe_mutex: Trying to unlock mutex keycache->op_lock that wasn't locked at mi_keycache.c, line 86Last used at mf_keycache.c, line: 6048131225 21:21:25 [ERROR] mysqld got signal 6 ;Simplified grammar:
thread1_init:CREATE TABLE IF NOT EXISTS `t1` ( `f` INT, KEY(`f`) ) ENGINE=MyISAM;query:CACHE INDEX t1 IN cache_name |SET GLOBAL cache_name.key_buffer_size = _smallint_unsigned ;cache_name:c1 | c2 ;# End of grammarCommand line:
perl ./runall.pl --grammar=mdev5405.yy --duration=300 --queries=100M --threads=4 --skip-gendata --vardir=<your vardir> --basedir=<your basedir>or with already running server
perl ./gentest.pl --grammar=mdev5405.yy --duration=300 --queries=100M --threads=4 --dsn=dbi:mysql:user=root:port=3306:database=test:host=127.0.0.1Reproducible on 5.3 and 5.5. On 10.0, it tends to hang instead.
I've set it up on perro as usual, for 5.3 and 5.5
cd mdev5405. ./run53#5 0x00007f5572bf7b8b in __GI_abort () at abort.c:91#6 0x0000000000b97cad in safe_mutex_unlock (mp=0x7f550c0d7388, file=0xd88d38 "mi_keycache.c", line=86) at thr_mutex.c:425#7 0x000000000093825e in mi_assign_to_key_cache (info=0x7f550400b7c8, key_map=1, key_cache=0x7f550c0d7338) at mi_keycache.c:86#8 0x000000000090f859 in ha_myisam::assign_to_keycache (this=0x7f550400aca0, thd=0x4d8c0b8, check_opt=0x7f55587eac30) at ha_myisam.cc:1356#9 0x0000000000839cda in mysql_admin_table(THD *, TABLE_LIST *, HA_CHECK_OPT *, const char *, thr_lock_type, bool, bool, uint, int (*)(THD *, TABLE_LIST *, HA_CHECK_OPT *), struct {...}, int (*)(THD *, TABLE_LIST *)) (thd=0x4d8c0b8, tables=0x7f550c02c640, check_opt=0x7f55587eac30, operator_name=0xd5d8d8 "assign_to_keycache", lock_type=TL_READ_NO_INSERT, open_for_modify=false, no_warnings_for_error=false, extra_open_options=0, prepare_func=0, operator_func=&virtual table offset 480, view_operator_func=0) at sql_table.cc:4972#10 0x000000000083b289 in mysql_assign_to_keycache (thd=0x4d8c0b8, tables=0x7f550c02c640, key_cache_name=0x4d8ef38) at sql_table.cc:5305#11 0x000000000068aa58 in mysql_execute_command (thd=0x4d8c0b8) at sql_parse.cc:2459#12 0x000000000069600c in mysql_parse (thd=0x4d8c0b8, rawbuf=0x7f550c02c550 "CACHE INDEX t1 IN c2", length=20, found_semicolon=0x7f55587eb7e0) at sql_parse.cc:6173#13 0x0000000000687a67 in dispatch_command (command=COM_QUERY, thd=0x4d8c0b8, packet=0x4d942a9 "", packet_length=20) at sql_parse.cc:1243#14 0x0000000000686cd8 in do_command (thd=0x4d8c0b8) at sql_parse.cc:923#15 0x0000000000683b72 in handle_one_connection (arg=0x4d8c0b8) at sql_connect.cc:1231#16 0x00007f5573788e9a in start_thread (arg=0x7f55587ec700) at pthread_create.c:308#17 0x00007f5572cb1cbd in clone () at ../sysdeps/unix/sysv/linux/x86_64/clone.S:112