Crash in check_table_is_closed with concurrent workload

[Philip Stoev]

When executing a concurrent workload involving CREATE and DROP on per-thread tables, the server crashed as follows:

# 2011-10-29T06:01:37 #4 0x0000000000971837 in check_table_is_closed (
# 2011-10-29T06:01:37 name=0x7f254c0828e0 "./transforms/where_subselect_31133",
# 2011-10-29T06:01:37 where=0xe36557 "delete") at mi_dbug.c:179
# 2011-10-29T06:01:37 #5 0x000000000098f2e8 in mi_delete_table (
# 2011-10-29T06:01:37 name=0x7f254c0828e0 "./transforms/where_subselect_31133")
# 2011-10-29T06:01:37 at mi_delete_table.c:66
# 2011-10-29T06:01:37 #6 0x000000000093c314 in ha_myisam::delete_table (this=0x7f24c00076f8,
# 2011-10-29T06:01:37 name=0x7f254c0828e0 "./transforms/where_subselect_31133")
# 2011-10-29T06:01:37 at ha_myisam.cc:2004
# 2011-10-29T06:01:37 #7 0x000000000083947e in handler::ha_delete_table (this=0x7f24c00076f8,
# 2011-10-29T06:01:37 name=0x7f254c0828e0 "./transforms/where_subselect_31133")
# 2011-10-29T06:01:37 at handler.cc:3560
# 2011-10-29T06:01:37 #8 0x00000000008363bd in ha_delete_table (thd=0x4927ec8,
# 2011-10-29T06:01:37 table_type=0x358d768,
# 2011-10-29T06:01:37 path=0x7f254c0828e0 "./transforms/where_subselect_31133",
# 2011-10-29T06:01:37 db=0x7f24c00072b8 "transforms",
# 2011-10-29T06:01:37 alias=0x7f24c00072c8 "where_subselect_31133", generate_warning=true)
# 2011-10-29T06:01:37 at handler.cc:2066
# 2011-10-29T06:01:37 #9 0x0000000000859a79 in mysql_rm_table_part2 (thd=0x4927ec8,
# 2011-10-29T06:01:37 tables=0x7f24c0007320, if_exists=false, drop_temporary=false,
# 2011-10-29T06:01:37 drop_view=false, dont_log_query=false) at sql_table.cc:2067
# 2011-10-29T06:01:37 #10 0x000000000085903e in mysql_rm_table (thd=0x4927ec8,
# 2011-10-29T06:01:37 tables=0x7f24c0007320, if_exists=0 '\000', drop_temporary=0 '\000')
# 2011-10-29T06:01:37 at sql_table.cc:1845
# 2011-10-29T06:01:37 #11 0x00000000006c2d6b in mysql_execute_command (thd=0x4927ec8)
# 2011-10-29T06:01:37 at sql_parse.cc:3411
# 2011-10-29T06:01:37 #12 0x00000000006cb25f in mysql_parse (thd=0x4927ec8,
# 2011-10-29T06:01:37 rawbuf=0x7f24c00071f0 "DROP TABLE transforms.where_subselect_31133",
# 2011-10-29T06:01:37 length=43, found_semicolon=0x7f254c083c68) at sql_parse.cc:6112
# 2011-10-29T06:01:37 #13 0x00000000006bcccb in dispatch_command (command=COM_QUERY, thd=0x4927ec8,
# 2011-10-29T06:01:37 packet=0x492bb99 "DROP TABLE transforms.where_subselect_31133",
# 2011-10-29T06:01:37 packet_length=43) at sql_parse.cc:1221
# 2011-10-29T06:01:37 #14 0x00000000006bc04b in do_command (thd=0x4927ec8) at sql_parse.cc:916
# 2011-10-29T06:01:37 #15 0x00000000006b8f27 in handle_one_connection (arg=0x4927ec8)
# 2011-10-29T06:01:37 at sql_connect.cc:1191
# 2011-10-29T06:01:37 #16 0x00000035a7207761 in start_thread () from /lib64/libpthread.so.0
# 2011-10-29T06:01:37 #17 0x00000035a6ee098d in clone () from /lib64/libc.so.6

Each thread operates on a separate set of tables so there are no concurrent DDL statements that involve the same table.

RQG command line:

perl runall.pl --queries=100000000 --debug --mysqld1=--sql_mode=ONLY_FULL_GROUP_BY --queries=100M --duration=600 --threads=4 --seed=time --mysqld1=--log-output=file --no-mask --reporter=QueryTimeout,Backtrace,ErrorLog,Deadlock --mysqld1=--optimizer_switch=derived_merge=ON,derived_with_keys=ON,index_condition_pushdown=ON,mrr=ON,mrr_sort_keys=ON,semijoin=ON,subquery_cache=OFF,join_cache_hashed=ON,materialization=OFF --validator=Transformer --transformers=DisableChosenPlan,ExecuteAsPreparedOnce,ExecuteAsView,ExecuteAsWhereSubquery,ExecuteAsDerived,ConvertTablesToViews,ConvertTablesToDerived --mysqld1=--join_cache_level=1 --views=TEMPTABLE --engine=MyISAM --varchar-length=48 --grammar=conf/optimizer/optimizer_no_subquery_simple.yy --seed=1319856919 --basedir1=/home/philips/bzr/maria-5.3

[Philip Stoev (pstoev-askmonty)]

revision-id: <email address hidden>
date: 2011-10-28 12:38:36 +0400
build-date: 2011-10-31 11:26:39 +0200
revno: 3258
branch-nick: maria-5.3

[Michael Widenius (monty)]

I have now added a mutex to protect the structure that caused a crash.
I tried to repeat the original problem for some time, but was not successful with that.
I also did run the test for couple of hours after adding the patch without any core dumping issues, so I
assume this should now be fixed.