handle_fatal_signal (sig=6) in row_ins_index_entry_set_vals

[Roel Van de Paar]

+bt
#0 0x00007ff3c8961741 in __pthread_kill (threadid=<optimized out>, signo=signo@entry=6) at ../nptl/sysdeps/unix/sysv/linux/pthread_kill.c:61
#1 0x00000000007f9f18 in my_write_core (sig=6) at /git/PS-5.5.52_dbg/mysys/stacktrace.c:433
#2 0x00000000006b9c56 in handle_fatal_signal (sig=6) at /git/PS-5.5.52_dbg/sql/signal_handler.cc:250
#3 <signal handler called>
#4 0x00007ff3c71165f7 in __GI_raise (sig=sig@entry=6) at ../nptl/sysdeps/unix/sysv/linux/raise.c:56
#5 0x00007ff3c7117ce8 in __GI_abort () at abort.c:90
#6 0x0000000000a2d376 in row_ins_index_entry_set_vals (row=0x7ff3b011e1a0, entry=0x0, index=0x7ff3b0026c78) at /git/PS-5.5.52_dbg/storage/innobase/row/row0ins.c:2333
#7 row_ins_index_entry_step (thr=0x7ff3b011e300, node=0x7ff3b011e0e0) at /git/PS-5.5.52_dbg/storage/innobase/row/row0ins.c:2386
#8 row_ins (thr=0x7ff3b011e300, node=0x7ff3b011e0e0) at /git/PS-5.5.52_dbg/storage/innobase/row/row0ins.c:2522
#9 row_ins_step (thr=thr@entry=0x7ff3b011e300) at /git/PS-5.5.52_dbg/storage/innobase/row/row0ins.c:2639
#10 0x000000000089a000 in row_insert_for_mysql (mysql_rec=<optimized out>, prebuilt=0x7ff3b011dc78) at /git/PS-5.5.52_dbg/storage/innobase/row/row0mysql.c:1252
#11 0x0000000000876088 in ha_innobase::write_row (this=0x7ff3b0034010, record=0x7ff3b0044c10 "\375\002") at /git/PS-5.5.52_dbg/storage/innobase/handler/ha_innodb.cc:6200
#12 0x00000000006c2f59 in handler::ha_write_row (this=0x7ff3b0034010, buf=0x7ff3b0044c10 "\375\002") at /git/PS-5.5.52_dbg/sql/handler.cc:5464
#13 0x0000000000580482 in write_record (thd=thd@entry=0x7ff3ba3c1000, table=table@entry=0x7ff3b011ce00, info=info@entry=0x7ff3c8f544a0) at /git/PS-5.5.52_dbg/sql/sql_insert.cc:1797
#14 0x00000000005843bc in mysql_insert (thd=thd@entry=0x7ff3ba3c1000, table_list=0x7ff3b001f0f8, fields=..., values_list=..., update_fields=..., update_values=..., duplic=DUP_ERROR, ignore=false) at /git/PS-5.5.52_dbg/sql/sql_insert.cc:955
#15 0x0000000000598ea1 in mysql_execute_command (thd=thd@entry=0x7ff3ba3c1000) at /git/PS-5.5.52_dbg/sql/sql_parse.cc:3179
#16 0x000000000059df56 in mysql_parse (thd=thd@entry=0x7ff3ba3c1000, rawbuf=<optimized out>, length=24, parser_state=parser_state@entry=0x7ff3c8f55520) at /git/PS-5.5.52_dbg/sql/sql_parse.cc:6062
#17 0x000000000059f954 in dispatch_command (command=command@entry=COM_QUERY, thd=thd@entry=0x7ff3ba3c1000, packet=packet@entry=0x7ff3ba394001 "INSERT INTO t1 VALUES(2)", packet_length=packet_length@entry=24) at /git/PS-5.5.52_dbg/sql/sql_parse.cc:1075
#18 0x00000000005a1825 in do_command (thd=0x7ff3ba3c1000) at /git/PS-5.5.52_dbg/sql/sql_parse.cc:789
#19 0x00000000006500d4 in do_handle_one_connection (thd_arg=thd_arg@entry=0x7ff3ba3c1000) at /git/PS-5.5.52_dbg/sql/sql_connect.cc:1418
#20 0x00000000006501ca in handle_one_connection (arg=0x7ff3ba3c1000) at /git/PS-5.5.52_dbg/sql/sql_connect.cc:1325
#21 0x00007ff3c895cdc5 in start_thread (arg=0x7ff3c8f56700) at pthread_create.c:308
#22 0x00007ff3c71d7ced in clone () at ../sysdeps/unix/sysv/linux/x86_64/clone.S:113

==================

160929 12:45:46 [Note] /sda/PS280916-percona-server-5.5.52-38.3-linux-x86_64-debug/bin/mysqld: ready for connections.
Version: '5.5.52-38.3-debug' socket: '/sda/PS280916-percona-server-5.5.52-38.3-linux-x86_64-debug/socket.sock' port: 12933 MySQL Community Server (GPL)
InnoDB: DEBUG: update_statistics for data/#sql2ace_1_0.
InnoDB: DEBUG: update_statistics for data/#sql2ace_1_0.
InnoDB: DEBUG: update_statistics for data/#sql2ace_1_1.
InnoDB: DEBUG: update_statistics for data/#sql2ace_1_1.
InnoDB: DEBUG: update_statistics for data/#sql2ace_1_1.
InnoDB: DEBUG: update_statistics for data/#sql2ace_1_1.
InnoDB: DEBUG: update_statistics for data/#sql2ace_1_2.
InnoDB: DEBUG: update_statistics for data/#sql2ace_1_2.
InnoDB: DEBUG: update_statistics for data/#sql2ace_1_2.
InnoDB: DEBUG: update_statistics for data/#sql2ace_1_2.
160929 12:45:51 InnoDB: Assertion failure in thread 140685025306368 in file row0ins.c line 2333
InnoDB: Failing assertion: entry && row

==================

DROP DATABASE test;CREATE DATABASE test;USE test;
SET expand_fast_index_creation=1;
CREATE TEMPORARY TABLE t1(c1 INT)ENGINE=MRG_InnoDB UNION=(t1,t2) INSERT_METHOD=FIRST;
ALTER TABLE t1 ADD INDEX(c1);
INSERT INTO t1 VALUES(3),(4);
ALTER TABLE t1 KEY_BLOCK_SIZE=8;
INSERT INTO t1 VALUES(2);

[Roel Van de Paar (roel11)]

This is a seemingly new, somewhat prominent, bug in percona-server-5.5.52-38.3-linux-x86_64-debug. May be a regression.

[Roel Van de Paar (roel11)]

Not present in upstream.

[Laurynas Biveinis (laurynas-biveinis)]

Reduced testcase

SET expand_fast_index_creation=1;
CREATE TEMPORARY TABLE t1(c1 INT PRIMARY KEY, INDEX c1(c1)) ENGINE=InnoDB;
INSERT INTO t1 VALUES(3);
OPTIMIZE TABLE t1;
INSERT INTO t1 VALUES(2);

[Laurynas Biveinis (laurynas-biveinis)]

By code review this has been here throughout whole PS 5.5 series.

The root cause is row_get_prebuilt_insert_row returning old prebuilt->ins_node, that was built using fewer indexes.

The fix would be along the lines of

commit 241387a2b6b61fb8a4f78dc4ad0aaa289400c694
Author: Marko M<C3><A4>kel<C3><A4> <email address hidden>
Date: Fri Dec 7 11:47:32 2012 +0200

    Bug#15936065 INNODB: FAILING ASSERTION: NODE->ENTRY == __NULL IN ROW0INS.CC

    The prebuilt->ins_node->entry_list should be rebuilt every time when
    indexes are added or dropped. It is not obvious that this was always
    the case.

    dict_table_t: Add def_trx_id, for identifying the transaction that
    added or dropped indexes last. If a table is evicted and again loaded
    from the data dictionary, the def_trx_id would be 0.

    row_merge_drop_indexes(): Add comments and assertions. Update
    table->def_trx_id before freeing any indexes from the data dictionary
    cache.

    row_get_prebuilt_insert_row(): Use table->def_trx_id to determine
    whether the entry_list needs to be rebuilt.

    ha_innobase::prepare_inplace_alter_table(): Do not invoke
    online_retry_drop_indexes() unless we are holding an exclusive
    meta-data lock. This would cause problem with the 'no-op' ALTER in
    the RQG test case.

    rb:1665 approved by Jimmy Yang

[Shahriyar Rzayev (rzayev-sehriyar)]

Percona now uses JIRA for bug reports so this bug report is migrated to: https://jira.percona.com/browse/PS-3570