Matthias Leich added a comment - 2019-05-22 15:25 - edited 19555.log – Protocol of my RQG run including backtrace MDEV-19555 .yy - Simplified RQG grammar I am trying to develop some MTR based replay test.

Elena Stepanova added a comment - 2019-08-30 16:40 Note: The test case is non-deterministic, run with --repeat=N. It always fails for me within 5 attempts, but it can vary on different machines and builds. --source include/have_innodb.inc   CREATE TABLE t1 (a INT ) ENGINE=InnoDB;   --connect (con1,localhost,root,,test) FLUSH TABLES WITH READ LOCK;   --connect (con2,localhost,root,,test) --send UPDATE t1 SET a = 1;   --connection default --send CREATE OR REPLACE TABLE t1 (b INT ) ENGINE=InnoDB;   --connection con1 CHECK TABLE t1;   # Cleanup UNLOCK TABLES; --disconnect con1 --connection con2 --error 0,ER_BAD_FIELD_ERROR --reap --disconnect con2 --connection default --reap DROP TABLE t1; 10.4 77e44282 mysqld: /data/src/10.4/sql/sql_error.h:1036: uint Diagnostics_area::sql_errno() const: Assertion `m_status == DA_ERROR' failed. 190830 19:38:21 [ERROR] mysqld got signal 6 ;   #7 0x00007f4b0fc4df12 in __GI___assert_fail (assertion=0x5621b78c3b94 "m_status == DA_ERROR", file=0x5621b78c3b00 "/data/src/10.4/sql/sql_error.h", line=1036, function=0x5621b78c46c0 <Diagnostics_area::sql_errno() const::__PRETTY_FUNCTION__> "uint Diagnostics_area::sql_errno() const") at assert.c:101 #8 0x00005621b6bd0c02 in Diagnostics_area::sql_errno (this=0x7f4ab40064c8) at /data/src/10.4/sql/sql_error.h:1036 #9 0x00005621b702af7e in ha_rollback_trans (thd=0x7f4ab4000b00, all=true) at /data/src/10.4/sql/handler.cc:1919 #10 0x00005621b6e759f5 in trans_rollback_implicit (thd=0x7f4ab4000b00) at /data/src/10.4/sql/transaction.cc:387 #11 0x00005621b6e738d8 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 *, HA_CHECK_OPT *)) (thd=0x7f4ab4000b00, tables=0x7f4ab4011e20, check_opt=0x7f4ab4005cd0, operator_name=0x5621b7934851 "check", lock_type=TL_READ_NO_INSERT, org_open_for_modify=false, repair_table_use_frm=false, extra_open_options=32, prepare_func=0x0, operator_func=(int (handler::*)(handler * const, THD *, HA_CHECK_OPT *)) 0x5621b7032488 <handler::ha_check(THD*, st_ha_check_opt*)>, view_operator_func=0x5621b6dfed79 <view_check(THD*, TABLE_LIST*, st_ha_check_opt*)>) at /data/src/10.4/sql/sql_admin.cc:1177 #12 0x00005621b6e741e6 in Sql_cmd_check_table::execute (this=0x7f4ab40124d8, thd=0x7f4ab4000b00) at /data/src/10.4/sql/sql_admin.cc:1352 #13 0x00005621b6cf3ae9 in mysql_execute_command (thd=0x7f4ab4000b00) at /data/src/10.4/sql/sql_parse.cc:6098 #14 0x00005621b6cf8d59 in mysql_parse (thd=0x7f4ab4000b00, rawbuf=0x7f4ab4011d68 "CHECK TABLE t1", length=14, parser_state=0x7f4b08fd3170, is_com_multi=false, is_next_command=false) at /data/src/10.4/sql/sql_parse.cc:7908 #15 0x00005621b6ce5002 in dispatch_command (command=COM_QUERY, thd=0x7f4ab4000b00, packet=0x7f4ab4008331 "", packet_length=14, is_com_multi=false, is_next_command=false) at /data/src/10.4/sql/sql_parse.cc:1843 #16 0x00005621b6ce3748 in do_command (thd=0x7f4ab4000b00) at /data/src/10.4/sql/sql_parse.cc:1360 #17 0x00005621b6e5d74e in do_handle_one_connection (connect=0x5621b9a24640) at /data/src/10.4/sql/sql_connect.cc:1404 #18 0x00005621b6e5d49d in handle_one_connection (arg=0x5621b9a24640) at /data/src/10.4/sql/sql_connect.cc:1306 #19 0x00005621b778b0ab in pfs_spawn_thread (arg=0x5621b99a3350) at /data/src/10.4/storage/perfschema/pfs.cc:1862 #20 0x00007f4b117c24a4 in start_thread (arg=0x7f4b08fd4700) at pthread_create.c:456 #21 0x00007f4b0fd0ad0f in clone () at ../sysdeps/unix/sysv/linux/x86_64/clone.S:97 Not reproducible on 10.3. No obvious effect on a non-debug build. May or may not be related to MDEV-19840 .

Matthias Leich added a comment - 2020-02-07 14:56 10.5.1 commit 42e825dd0a8d25c1d6fa93f5a07115ceed3ee0ff 2020-02-04 compiled with debug – replay 10.5.1 older commit compiled without debug – no replay 10.4.13 commit c1eaa385ffb44bdf6264d2cc4b4cc0e10284c88a 2020-02-03 compiled with debug – replay 10.3.23 commit b0fa30808622fe12d474a70af1838906e60b9897 2020-02-02 compiled with debug – no replay

Roel Van de Paar added a comment - 2020-06-02 09:22 I see this one all over the runs too. A fix would be appreciated. Thank you!

Andrei Elkin added a comment - 2022-01-18 11:41 rucha174 , please consider this patch diff --git a/sql/handler.cc b/sql/handler.cc index 57992d98c04..c08c480e9b9 100644 --- a/sql/handler.cc +++ b/sql/handler.cc @@ - 2203 , 7 + 2203 , 8 @@ int ha_rollback_trans(THD *thd, bool all) Thanks to possibility of MDL deadlock rollback request can come even if transaction hasn't been started in any transactional storage engine. */ - if (thd->transaction_rollback_request) + if (thd->transaction_rollback_request && + thd->transaction->xid_state.is_explicit_XA()) thd->transaction->xid_state.set_error(thd->get_stmt_da()->sql_errno()); thd->has_waiter= false ; According to git-annotate f189f34ed4b8 refactored thd->transaction.xid_state.rm_error= into the current form, but the assignment pre-condition was not preserved. Originally that was to the explicit XA trx, that's what my patch makes.

Matthias Leich added a comment - 2022-01-24 13:23 - edited Hi Andrei, many thanks for your work on MDEV-19555.   I am just running automatic test simplification for some Fulltext issue on the development tree origin/bb-10.6-MDEV-20605-cur-pos-fix 714b3adc82eac16b46a47c4953cbeedaf14c77cb 2022-01-20T18:41:45+03:00. The FTS issue I am looking for occurs on actual main trees too. The corresponding RQG test does neither invoke explicite XA statements nor replication. DB servers get startet with "--sync-binlog=1".   Before applying your patch to the development tree I got 50% of all tests fail because of MDEV-19555 1% fail with the issue I am searching for. After losing the patience and applying your patch it looks far way better. It also does not look like the bad effect "Assertion `m_status == DA_ERROR' failed for CHECK TABLE" is now somehow replaced by some other new bad effect.

Matthias Leich added a comment - 2022-01-25 10:42 The tree origin/10.8, 10.8 5595ed9d9f20aeb463ac26e075c9cba41927e85e 2022-01-24T19:26:09+09:00 + MDEV-19555 patch + some preliminary patch for MDEV-14481 performed well in RQG testing with the test battery for broad range coverage of functionality. No new bad effects.

Andrei Elkin added a comment - 2022-01-26 14:59 Rucha, howdy. The patch merely restores logical flaw in the problematic sources code spot. The setting thd->transaction->xid_state.set_error() is meaningful only for the user (aka explicit) XA. To clear your mind about that I already suggested to time travel with git-annotate. Thank you, Andrei.

Andrei Elkin added a comment - 2022-01-26 17:52 Approved by Sanja. For merging, there's a usual conflict dealing with THD::transaction that is the pointer in higher branches.

Vladislav Lesin added a comment - 2024-10-18 14:52 - edited We still have the stack trace during RQG testing . Here is the analyses(the assertion is caused by MDL deadlock for that specific case, it can be useful for bug reproducing) and bugfix proposal .