| Bug #70854 | Tc_log_page_size should be unflushable or server crashes if 2 XA SEs installed | ||
|---|---|---|---|
| Submitted: | 7 Nov 2013 10:40 | Modified: | 3 Apr 2014 15:15 |
| Reporter: | Laurynas Biveinis (OCA) | Email Updates: | |
| Status: | Closed | Impact on me: | |
| Category: | MySQL Server: Logging | Severity: | S2 (Serious) |
| Version: | 5.7 | OS: | Any |
| Assigned to: | CPU Architecture: | Any | |
| Tags: | innodb, SE, tc_log_page_size, TokuDB, xa | ||
[7 Nov 2013 10:40]
Laurynas Biveinis
[7 Nov 2013 18:09]
MySQL Verification Team
We believe you that a crash is possible under circumstances that you describe. However, we can not set this bug to "Verified" unless we have a fully repeatable test case for the MySQL server. A test case can be accepted even if it requires up to 10 runs in order for a crash to occur. If you can not come with a fully repeatable test case for MySQL server, we can set this bug to "To be fixed later", with an internal note that this bug should be tested when we have two XA capable storage engines. The other option is that you revisit this bug when we have two XA-enabled stored engines.
[8 Nov 2013 1:53]
Davi Arnaut
Somehow Bug#47134 can be verified, fixed and tested without two XA capable storage engines, but this one can't.
[8 Nov 2013 9:10]
Laurynas Biveinis
OK.
1) Make EXAMPLE a XA engine and fix -DWITH_EXAMPLE_STORAGE_ENGINE=1:
=== modified file 'storage/example/CMakeLists.txt'
--- storage/example/CMakeLists.txt 2011-09-07 10:08:09 +0000
+++ storage/example/CMakeLists.txt 2013-11-08 09:01:25 +0000
@@ -15,4 +15,4 @@
SET(EXAMPLE_PLUGIN_DYNAMIC "ha_example")
SET(EXAMPLE_SOURCES ha_example.cc)
-MYSQL_ADD_PLUGIN(example ${EXAMPLE_SOURCES} STORAGE_ENGINE MODULE_ONLY)
+MYSQL_ADD_PLUGIN(example ${EXAMPLE_SOURCES} STORAGE_ENGINE)
=== modified file 'storage/example/ha_example.cc'
--- storage/example/ha_example.cc 2013-05-19 18:46:26 +0000
+++ storage/example/ha_example.cc 2013-11-08 08:59:43 +0000
@@ -97,6 +97,10 @@
TABLE_SHARE *table,
MEM_ROOT *mem_root);
+static int example_prepare_handler(handlerton *hton,
+ THD *thd,
+ bool all);
+
handlerton *example_hton;
/* Interface to mysqld, to check system tables supported by SE */
@@ -141,6 +145,7 @@
example_hton= (handlerton *)p;
example_hton->state= SHOW_OPTION_YES;
example_hton->create= example_create_handler;
+ example_hton->prepare= example_prepare_handler;
example_hton->flags= HTON_CAN_RECREATE;
example_hton->system_database= example_system_database;
example_hton->is_supported_system_table= example_is_supported_system_table;
@@ -185,6 +190,13 @@
return new (mem_root) ha_example(hton, table);
}
+static int example_prepare_handler(handlerton *hton,
+ THD *thd,
+ bool all)
+{
+ return 0;
+}
+
ha_example::ha_example(handlerton *hton, TABLE_SHARE *table_arg)
:handler(hton, table_arg)
{}
2) Run this:
CREATE TABLE t1 (a INT) ENGINE=EXAMPLE;
CREATE TABLE t2 (a INT) ENGINE=InnoDB;
SHOW STATUS LIKE 'Tc_log_page_size';
BEGIN;
INSERT INTO t1 VALUES (1);
INSERT INTO t2 VALUES (2);
COMMIT;
FLUSH STATUS;
SHOW STATUS LIKE 'Tc_log_page_size';
DROP TABLE t1;
DROP TABLE t2;
3) Observe this:
CREATE TABLE t1 (a INT) ENGINE=EXAMPLE;
CREATE TABLE t2 (a INT) ENGINE=InnoDB;
SHOW STATUS LIKE 'Tc_log_page_size';
Variable_name Value
Tc_log_page_size 4096
BEGIN;
INSERT INTO t1 VALUES (1);
INSERT INTO t2 VALUES (2);
COMMIT;
main.foo [ fail ]
Test ended at 2013-11-08 11:09:09
CURRENT_TEST: main.foo
mysqltest: At line 11: query 'FLUSH STATUS' failed: 2013: Lost connection to MySQL server during query
...
Thread 1 (Thread 0x7f1084356700 (LWP 2015)):
#0 0x00007f10924b3f0c in __pthread_kill (threadid=<optimized out>, signo=8) at ../nptl/sysdeps/unix/sysv/linux/pthread_kill.c:62
#1 0x0000000000b38cbe in my_write_core (sig=8) at /home/laurynas/percona/src/mysql-server/mysys/stacktrace.c:258
#2 0x000000000081827e in handle_fatal_signal (sig=8) at /home/laurynas/percona/src/mysql-server/sql/signal_handler.cc:216
#3 <signal handler called>
#4 0x00000000007db6d5 in TC_LOG_MMAP::unlog (this=0x1927760 <tc_log_mmap>, cookie=8192, xid=25) at /home/laurynas/percona/src/mysql-server/sql/log.cc:2213
#5 0x00000000007db2b4 in TC_LOG_MMAP::commit (this=0x1927760 <tc_log_mmap>, thd=0x7f1064000a50, all=true) at /home/laurynas/percona/src/mysql-server/sql/log.cc:2081
#6 0x00000000009a53b6 in trans_commit_implicit (thd=0x7f1064000a50) at /home/laurynas/percona/src/mysql-server/sql/transaction.cc:284
#7 0x00000000008c3121 in mysql_execute_command (thd=0x7f1064000a50) at /home/laurynas/percona/src/mysql-server/sql/sql_parse.cc:4794
#8 0x00000000008c4459 in mysql_parse (thd=0x7f1064000a50, rawbuf=0x7f106400dd20 "FLUSH STATUS", length=12, parser_state=0x7f1084354ed0) at /home/laurynas/percona/src/mysql-server/sql/sql_parse.cc:5274
#9 0x00000000008b97ad in dispatch_command (command=COM_QUERY, thd=0x7f1064000a50, packet=0x7f1064004a31 "FLUSH STATUS", packet_length=12) at /home/laurynas/percona/src/mysql-server/sql/sql_parse.cc:1328
#10 0x00000000008b88f7 in do_command (thd=0x7f1064000a50) at /home/laurynas/percona/src/mysql-server/sql/sql_parse.cc:1027
#11 0x00000000009c7036 in handle_connection (arg=0x357eea0) at /home/laurynas/percona/src/mysql-server/sql/conn_handler/connection_handler_per_thread.cc:227
#12 0x0000000000be0e8d in pfs_spawn_thread (arg=0x34cd1c0) at /home/laurynas/percona/src/mysql-server/storage/perfschema/pfs.cc:1983
#13 0x00007f10924aef6e in start_thread (arg=0x7f1084356700) at pthread_create.c:311
#14 0x00007f10919bb9cd in clone () at ../sysdeps/unix/sysv/linux/x86_64/clone.S:113
In GDB:
Program received signal SIGFPE, Arithmetic exception.
[Switching to Thread 0x7fffec433700 (LWP 1893)]
0x00000000007db6d5 in TC_LOG_MMAP::unlog (this=0x1927760 <tc_log_mmap>, cookie=8192, xid=7) at /home/laurynas/percona/src/mysql-server/sql/log.cc:2213
2213 PAGE *p= pages + (cookie / tc_log_page_size);
(gdb) print tc_log_page_size
$1 = 0
[8 Nov 2013 14:57]
MySQL Verification Team
One more question only ... I suppose that main.foo is the name of the MTR script that should be run by MTR script ???
[8 Nov 2013 15:12]
Laurynas Biveinis
Sinisa - Yes. I put the SQL from step 2) in mysql-test/t/foo.test and run it through MTR.
[6 Dec 2013 17:41]
MySQL Verification Team
After couple of weeks, I managed to verify this bug. Had problems with interim 5.7 versions, most of all some `cmake` bugs. Small bugs ..... :o) But, now it is fully verified.
[7 Feb 2014 2:01]
Paul DuBois
Noted in 5.7.4 changelog. FLUSH STATUS cleared a variable that could result a subsequent implicit commit of an XA transaction causing a server exit.
[3 Apr 2014 15:15]
Laurynas Biveinis
5.7$ bzr log -r 7195 ------------------------------------------------------------ revno: 7195 committer: Dmitry Shulga <Dmitry.Shulga@oracle.com> branch nick: mysql-trunk-bug17911445 timestamp: Tue 2013-12-31 16:23:39 +0700 message: This patch fixes the bug#17911445 - TC_LOG_PAGE_SIZE SHOULD BE UNFLUSHABLE OR SERVER CRASHES IF 2 XA SES INSTALLED. The status variable tc_log_page_size is directly bound to tc_log_page_size source variable, which is setup once in TC_LOG_MMAP::open(). Running of statement FLUSH STATUS would cause it to be reset to zero. Then during doing implicit commit the TC_LOG_MMAP::unlog() would attempt to divide by it that led to crash. To fix this bug a type of status variable tc_log_page_size was changed to SHOW_LONG_NOFLUSH that denies reset of status variable value to zero. MTR test isn't added since there isn't currently any other engine with XA feature enabled aside from INNODB. I tested my patch using the fake second engine with XA feature enabled that I created from ha_example.cc. I think it won't be correct to modify ha_example.cc just for the sake of patch test coverage.
[3 Apr 2014 15:15]
Laurynas Biveinis
5.7$ bzr log -r 7195 ------------------------------------------------------------ revno: 7195 committer: Dmitry Shulga <Dmitry.Shulga@oracle.com> branch nick: mysql-trunk-bug17911445 timestamp: Tue 2013-12-31 16:23:39 +0700 message: This patch fixes the bug#17911445 - TC_LOG_PAGE_SIZE SHOULD BE UNFLUSHABLE OR SERVER CRASHES IF 2 XA SES INSTALLED. The status variable tc_log_page_size is directly bound to tc_log_page_size source variable, which is setup once in TC_LOG_MMAP::open(). Running of statement FLUSH STATUS would cause it to be reset to zero. Then during doing implicit commit the TC_LOG_MMAP::unlog() would attempt to divide by it that led to crash. To fix this bug a type of status variable tc_log_page_size was changed to SHOW_LONG_NOFLUSH that denies reset of status variable value to zero. MTR test isn't added since there isn't currently any other engine with XA feature enabled aside from INNODB. I tested my patch using the fake second engine with XA feature enabled that I created from ha_example.cc. I think it won't be correct to modify ha_example.cc just for the sake of patch test coverage.
