Description:
SHOW OPEN TABLES command hangs during highly concurrent deadlock-like load, such as the load described in Bug#39193.
Stacktrace from all threads is available here (internal only):
https://intranet.mysql.com/secure/paste/displaypaste.php?codeid=8491
Stacktrace from the SHOW OPEN TABLES thread:
Thread 2 (Thread 0x7f14a94ae950 (LWP 18924)):
#0 0x00007f14bd47ea94 in __lll_lock_wait () from /lib/libpthread.so.0
#1 0x00007f14bd47a190 in _L_lock_102 () from /lib/libpthread.so.0
#2 0x00007f14bd479a7e in pthread_mutex_lock () from /lib/libpthread.so.0
#3 0x0000000000cbd000 in safe_mutex_lock (mp=0x13974c0, my_flags=0, file=0xe5110d "sql_base.cc", line=740) at thr_mutex.c:225
#4 0x000000000076346f in list_open_tables (thd=0x7f14b40892f8, db=0x0, wild=0x0) at sql_base.cc:740
#5 0x0000000000897918 in fill_open_tables (thd=0x7f14b40892f8, tables=0x3eb79b8, cond=0x0) at sql_show.cc:5757
#6 0x0000000000893273 in get_schema_tables_result (join=0x39b7f10, executed_place=PROCESSED_BY_JOIN_EXEC) at sql_show.cc:6510
#7 0x00000000007ad346 in JOIN::exec (this=0x39b7f10) at sql_select.cc:2398
#8 0x00000000007a9a5b in mysql_select (thd=0x7f14b40892f8, rref_pointer_array=0x7f14b408b2a8, tables=0x3eb79b8, wild_num=0, fields=@0x7f14b408b1c8,
conds=0x0, og_num=0, order=0x0, group=0x0, having=0x0, proc_param=0x0, select_options=2684635648, result=0x3eb8508, unit=0x7f14b408ac58,
select_lex=0x7f14b408b0c0) at sql_select.cc:3067
#9 0x00000000007af39f in handle_select (thd=0x7f14b40892f8, lex=0x7f14b408abb8, result=0x3eb8508, setup_tables_done_option=0) at sql_select.cc:310
#10 0x000000000070949a in execute_sqlcom_select (thd=0x7f14b40892f8, all_tables=0x3eb79b8) at sql_parse.cc:4964
#11 0x000000000070ad8f in mysql_execute_command (thd=0x7f14b40892f8) at sql_parse.cc:2172
#12 0x0000000000713321 in mysql_parse (thd=0x7f14b40892f8, inBuf=0x3eb74b0 "SHOW OPEN TABLES", length=16, found_semicolon=0x7f14a94adb30)
at sql_parse.cc:5979
#13 0x0000000000713f1a in dispatch_command (command=COM_QUERY, thd=0x7f14b40892f8, packet=0x7f14b43dfac9 "SHOW OPEN TABLES", packet_length=16)
at sql_parse.cc:1064
#14 0x00000000007154a1 in do_command (thd=0x7f14b40892f8) at sql_parse.cc:746
#15 0x0000000000702809 in handle_one_connection (arg=0x7f14b40892f8) at sql_connect.cc:1146
#16 0x00007f14bd4783ba in start_thread () from /lib/libpthread.so.0
#17 0x00007f14bc3e4fcd in clone () from /lib/libc.so.6
#18 0x0000000000000000 in ?? ()
Diagnostic commands such as this should continue to work even though the server is "messed up".
How to repeat:
I ran into this while running the Random Query Generator test falcon_deadlock.
Branch & build mysql-6.0-falcon-team tree. Refer to this as FALCON. Version used was: (2759) kevin.lewis@sun.com-20090727153630-g4lz01ad9wpoewn8
Branch the mysql-test-extra-6.0 tree (must be recently up-to-date, since SHOW OPEN TABLES calls were recently added for debugging purposes). Version used was: (987) pstoev@mysql.com-20090728091820-65q25oz42y0bw8m9
cd mysql-test/gentest/
perl ./runall.pl \
--basedir=$FALCON \
--mysqld=--loose-innodb-lock-wait-timeout=5 \
--mysqld=--table-lock-wait-timeout=5 \
--mysqld=--loose-falcon-lock-wait-timeout=5 \
--mysqld=--loose-falcon-debug-mask=2 \
--mysqld=--skip-safemalloc \
--grammar=conf/falcon_recovery.yy \
--gendata=conf/falcon_recovery.zz \
--mysqld=--falcon-checkpoint-schedule=1 1 1 1 1 \
--reporters=Deadlock,ErrorLog,Backtrace,Recovery,WinPackage,Shutdown \
--duration=1200 \
--vardir=$FALCON/vardirs \
--mysqld=--log-output=file \
--queries=100000 \
--engine=falcon
Can probably be repeated in a simpler way with some effort.
Suggested fix:
Make SHOW OPEN TABLES complete (best-effort) within a reasonable amount time.
