The problem is caused by this sequence of events:
Statement is run:
CREATE TEMPORARY TABLE tmp (a INT);
|
^ This creates a temporary table.
CREATE TABLE non_existing_db.t SELECT SLEEP(2) AS b;
|
^ When running this statement, the following happens:
1. thd->lex->sql_command=SQLCOM_CREATE_TABLE
2. The statement fails (non_existing_db doesn't exist)
3. Statement cleanup is performed, e.g. the SELECT_LEX from
the SELECT part is cleaned up.
--disconnect con1
|
The connection is closed, so we drop the temporary table and try to binlog the
DROP command.
The contents of thd->lex and thd->lex->sql_command remain the same
(SQLCOM_CREATE_TABLE).
We get here:
(gdb) wher
|
#0 Query_log_event::Query_log_event (this=0x7ffff0324760, thd_arg=0x7fffcc000e00, query_arg=0x7ffff0324880 "DROP /*!40005 TEMPORARY */ TABLE IF EXISTS `tmp`,", <incomplete sequence \313>, query_length=48, using_trans=false, direct=true, suppress_use=false, errcode=0) at /home/psergey/dev-git/10.4-cl/sql/log_event.cc:4433
|
#1 0x0000555555fa9a7f in THD::log_events_and_free_tmp_shares (this=0x7fffcc000e00) at /home/psergey/dev-git/10.4-cl/sql/temporary_tables.cc:1382
|
#2 0x0000555555fa7a7a in THD::close_temporary_tables (this=0x7fffcc000e00) at /home/psergey/dev-git/10.4-cl/sql/temporary_tables.cc:523
|
#3 0x0000555555cc2350 in THD::cleanup (this=0x7fffcc000e00) at /home/psergey/dev-git/10.4-cl/sql/sql_class.cc:1525
|
#4 0x0000555555c00700 in unlink_thd (thd=0x7fffcc000e00) at /home/psergey/dev-git/10.4-cl/sql/mysqld.cc:2627
|
#5 0x0000555555c00bde in one_thread_per_connection_end (thd=0x7fffcc000e00, put_in_cache=true) at /home/psergey/dev-git/10.4-cl/sql/mysqld.cc:2776
|
#6 0x0000555555ea65ba in do_handle_one_connection (connect=0x555558171b00) at /home/psergey/dev-git/10.4-cl/sql/sql_connect.cc:1418
|
and execute this code:
switch (lex->sql_command) |
{
|
...
|
|
|
case SQLCOM_CREATE_TABLE: |
case SQLCOM_CREATE_SEQUENCE: |
/* |
If we are using CREATE ... SELECT or if we are a slave
|
executing BEGIN...COMMIT (generated by CREATE...SELECT) we
|
have to use the transactional cache to ensure we don't
|
calculate any checksum for the CREATE part.
|
*/
|
trx_cache= (lex->first_select_lex()->item_list.elements &&
|
thd->is_current_stmt_binlog_format_row()) ||
|
(thd->variables.option_bits & OPTION_GTID_BEGIN);
|
This is apparently wrong. The code assumes that lex->sql_command describes the current command, but this is simply NOT TRUE . (The value of lex->sql_command describes whatever was the last command before the client disconnected, which should not have any effect on how we drop the temporary tables and binlog it)
the value of lex->first_select_lex()->item_list.elements is not valid anymore, either. So if we are trying to read it, it's an error.
The problem doesn't reproduce when I use mysql command line client and disconnect manually.
When I do that, lex->sql_command=SQLCOM_END.
The value of SQLCOM_END is set here in dispatch_command:
/* |
We should always call reset_for_next_command() before a query.
|
mysql_parse() will do this for queries. Ensure it's also done
|
for other commands.
|
*/
|
if (command != COM_QUERY) |
thd->reset_for_next_command();
|
thd->set_command(command);
|
|
|
thd->enable_slow_log= true; |
thd->query_plan_flags= QPLAN_INIT;
|
thd->lex->sql_command= SQLCOM_END; /* to avoid confusing VIEW detectors */ |
thd->reset_kill_query();
|
See the {{thd->lex->sql_command= SQLCOM_END; }} line above.
I assume the fix here would be to set thd-lex->sql_command to SQLCOM_END in THD::cleanup(). But I would like to run this through somebody else knowledgeable in the server runtime module.
Hello Andrei,
Can you please review the changes for MDEV-19716.
Please find the bb link:
http://buildbot.askmonty.org/buildbot/grid?category=main&branch=bb-10.4-sujatha
https://github.com/MariaDB/server/commit/818504917687b6454b91ac0ac16dd35937a3a3ca
Thank you.
Thanks for a good piece of analysis, Sujatha! The patch is okay, just make sure we
don't push 2 secs extra waiting for the mtr test to run ... times as many as it's going to be launched in the future 
.
"rpl_drop_temp_table_invaid_lex.test" fails ocassionally on build bot with following symptoms.
It needs to be fixed.
http://buildbot.askmonty.org/buildbot/builders/bld-p9-rhel7/builds/4201/steps/mtr/logs/stdio
rpl.rpl_drop_temp_table_invaid_lex 'mix' w30 [ pass ] 3361
|
|
|
MTR's internal check of the test case 'rpl.rpl_drop_temp_table_invaid_lex' failed.
|
This means that the test case does not preserve the state that existed
|
before the test case was executed. Most likely the test case did not
|
do a proper clean-up. It could also be caused by the previous test run
|
by this thread, if the server wasn't restarted.
|
This is the diff of the states of the servers before and after the
|
test case was executed:
|
mysqltest: Logging to '/opt/buildbot-slave/bld-p9-rhel7/build/mysql-test/var/30/tmp/check-mysqld_2.log'.
|
mysqltest: Results saved in '/opt/buildbot-slave/bld-p9-rhel7/build/mysql-test/var/30/tmp/check-mysqld_2.result'.
|
mysqltest: Connecting to server localhost:16561 (socket /opt/buildbot-slave/bld-p9-rhel7/build/mysql-test/var/tmp/30/mysqld.2.sock) as 'root', connection 'default', attempt 0 ...
|
mysqltest: ... Connected.
|
mysqltest: Start processing test commands from './include/check-testcase.test' ...
|
mysqltest: ... Done processing test commands.
|
--- /opt/buildbot-slave/bld-p9-rhel7/build/mysql-test/var/30/tmp/check-mysqld_2.result 2020-06-11 15:23:27.656444761 +0300
|
+++ /opt/buildbot-slave/bld-p9-rhel7/build/mysql-test/var/30/tmp/check-mysqld_2.reject 2020-06-11 15:23:33.476210985 +0300
|
@@ -1032,7 +1032,7 @@
|
DropGeometryColumn def mysql DropGeometryColumn PROCEDURE NULL NULL NULL NULL NULL NULL NULL NULL SQL begin
|
set @qwe= concat('ALTER TABLE ', t_schema, '.', t_name, ' DROP ', geometry_column); PREPARE ls from @qwe; execute ls; deallocate prepare ls; end NULL NULL SQL NO CONTAINS SQL NULL INVOKER 2020-06-11 15:15:22 2020-06-11 15:15:22 mariadb.sys@localhost latin1 latin1_swedish_ci latin1_swedish_ci
|
Variable_name Value
|
-Slave_open_temp_tables 0
|
+Slave_open_temp_tables 1
|
Table Checksum
|
mysql.columns_priv 0
|
mysql.db 204716352
|
Hello Elkin
- "rpl_drop_temp_table_invaid_lex.test" test fails sporadically in build bot, 'check-testcase' fails.
- Test name also needs to be changed from s/rpl_drop_temp_table_invaid_lex.test/rpl_drop_temp_table_invalid_lex.test
I have implemented the changes. Please find the following patch.
Patch: https://github.com/MariaDB/server/commit/9b8dd9690a9a3bcf5f1f6e7929860751e732091b
Thank you.
Hello alice
Can you please review the changes to "rpl_drop_temp_table_invalid_lex.test".
- "rpl_drop_temp_table_invaid_lex.test" test fails sporadically in build bot, 'check-testcase' fails.
- Test name also needs to be changed from s/rpl_drop_temp_table_invaid_lex.test/rpl_drop_temp_table_invalid_lex.test
Patch: https://github.com/MariaDB/server/commit/fa2505dbcccf5b6316a52e36596270d9fcb1f9c1
Thank you.
Patch was reviewed and approved by alice. Test changes are pushed into 10.4.18.