[MDEV-14409] Assertion `page_rec_is_leaf(rec)' failed in lock_rec_validate_page upon SHOW ENGINE INNODB STATUS Created: 2017-11-15  Updated: 2018-12-04  Resolved: 2018-11-26

Status: Closed
Project: MariaDB Server
Component/s: Admin statements, Storage Engine - InnoDB
Affects Version/s: 10.2, 10.3, 10.4
Fix Version/s: 10.4.1, 10.2.20, 10.3.12

Type: Bug Priority: Major
Reporter: Elena Stepanova Assignee: Marko Mäkelä
Resolution: Fixed Votes: 0
Labels: None

Issue Links:
Relates
relates to MDEV-17843 Assertion `page_rec_is_leaf(rec)' fai... Closed
relates to MDEV-14055 Assertion `page_rec_is_leaf(rec)' fai... Closed

 Description   

10.2 3728b11f879

 
mysqld: /data/src/10.2/storage/innobase/lock/lock0lock.cc:5653: ulint lock_rec_validate_page(const buf_block_t*): Assertion `page_rec_is_leaf(rec)' failed.
 
181124 23:15:12 [ERROR] mysqld got signal 6 ;
 
 
 
#7  0x00007f6516333ee2 in __assert_fail () from /lib/x86_64-linux-gnu/libc.so.6
 
#8  0x00005555f9151a5d in lock_rec_validate_page (block=0x7f64f39f2df8) at /data/src/10.2/storage/innobase/lock/lock0lock.cc:5653
 
#9  0x00005555f91520a9 in lock_rec_block_validate (space_id=9, page_no=3) at /data/src/10.2/storage/innobase/lock/lock0lock.cc:5804
 
#10 0x00005555f9152331 in lock_validate () at /data/src/10.2/storage/innobase/lock/lock0lock.cc:5857
 
#11 0x00005555f9150729 in lock_print_info_all_transactions (file=0x5555fb540410) at /data/src/10.2/storage/innobase/lock/lock0lock.cc:5348
 
#12 0x00005555f9277830 in srv_printf_innodb_monitor (file=0x5555fb540410, nowait=0, trx_start_pos=0x7f6500b09500, trx_end=0x7f6500b09508) at /data/src/10.2/storage/innobase/srv/srv0srv.cc:1303
 
#13 0x00005555f90e2f5a in innodb_show_status (hton=0x5555fb4f8730, thd=0x7f64a4000b00, stat_print=0x5555f8dc844a <stat_print(THD*, char const*, size_t, char const*, size_t, char const*, size_t)>) at /data/src/10.2/storage/innobase/handler/ha_innodb.cc:15950
 
#14 0x00005555f90e39a3 in innobase_show_status (hton=0x5555fb4f8730, thd=0x7f64a4000b00, stat_print=0x5555f8dc844a <stat_print(THD*, char const*, size_t, char const*, size_t, char const*, size_t)>, stat_type=HA_ENGINE_STATUS) at /data/src/10.2/storage/innobase/handler/ha_innodb.cc:16364
 
#15 0x00005555f8dc87e4 in ha_show_status (thd=0x7f64a4000b00, db_type=0x5555fb4f8730, stat=HA_ENGINE_STATUS) at /data/src/10.2/sql/handler.cc:5619
 
#16 0x00005555f8b42627 in mysql_execute_command (thd=0x7f64a4000b00) at /data/src/10.2/sql/sql_parse.cc:3725
 
#17 0x00005555f8b4fb5d in mysql_parse (thd=0x7f64a4000b00, rawbuf=0x7f64a40131d8 "SHOW ENGINE INNODB STATUS /* QNO 66 CON_ID 12 */", length=48, parser_state=0x7f6500b0a250, is_com_multi=false, is_next_command=false) at /data/src/10.2/sql/sql_parse.cc:8013
 
#18 0x00005555f8b3d495 in dispatch_command (command=COM_QUERY, thd=0x7f64a4000b00, packet=0x7f64a444b941 "", packet_length=49, is_com_multi=false, is_next_command=false) at /data/src/10.2/sql/sql_parse.cc:1824
 
#19 0x00005555f8b3bdf8 in do_command (thd=0x7f64a4000b00) at /data/src/10.2/sql/sql_parse.cc:1378
 
#20 0x00005555f8c8e537 in do_handle_one_connection (connect=0x5555fb4e4090) at /data/src/10.2/sql/sql_connect.cc:1335
 
#21 0x00005555f8c8e2c4 in handle_one_connection (arg=0x5555fb4e4090) at /data/src/10.2/sql/sql_connect.cc:1241
 
#22 0x00007f6517def494 in start_thread (arg=0x7f6500b0b700) at pthread_create.c:333
 
#23 0x00007f65163f093f in clone () from /lib/x86_64-linux-gnu/libc.so.6

Strangely, MDEV-14055 seems to claim that it was a bogus debug assertion, but it's still there even though MDEV-14055 has been closed.

RQG grammar mdev14409.yy

 
thread2_init:
 
  CREATE TABLE `t1` (
 
    `t1` TEXT,
 
    `v1` VARCHAR(1),
 
    `t2` TEXT,
 
    `v2` VARCHAR(1),
 
    UNIQUE (`v2`),
 
    UNIQUE (`t2`(255))
 
  ) ENGINE=InnoDB ROW_FORMAT=DYNAMIC
 
  ; CREATE TABLE `t2` LIKE `t1`
 
  ; CREATE TABLE `t3` LIKE `t1`
 
  ; CREATE TABLE `t4` LIKE `t1`
 
  ; INSERT IGNORE INTO `t1` VALUES vals,vals,vals,vals,vals,vals,vals,vals,vals,vals 
  ; INSERT IGNORE INTO `t2` VALUES vals,vals,vals,vals,vals,vals,vals,vals,vals,vals 
  ; INSERT IGNORE INTO `t3` VALUES vals,vals,vals,vals,vals,vals,vals,vals,vals,vals 
  ; INSERT IGNORE INTO `t4` VALUES vals,vals,vals,vals,vals,vals,vals,vals,vals,vals 
  ; SELECT * FROM `t1` INTO OUTFILE 'load.data.t1'
 
  ; SELECT * FROM `t2` INTO OUTFILE 'load.data.t2'
 
  ; SELECT * FROM `t3` INTO OUTFILE 'load.data.t3'
 
  ; SELECT * FROM `t4` INTO OUTFILE 'load.data.t4'
 
;
 
 
 
vals:
 
  (_char(1024),_char(1),_char(1024),_char(1));
 
 
 
thread1:
 
  SHOW ENGINE INNODB STATUS;
 
 
 
query:
 
  DELETE FROM my_table LIMIT _digit |
 
  LOAD DATA INFILE { "'load.data.t".$prng->int(1,4)."'" } REPLACE INTO TABLE my_table
 
;
 
 
 
my_table:
 
  t1 | t2 | t3 | t4 ;

Run as

perl ./runall-new.pl  --basedir=/data/bld/10.2 --vardir=/dev/shm/vardir --duration=300 --threads=2 --grammar=mdev14409.yy --skip-gendata

Don't forget to use your basedir and check whether the vardir path is okay for you.

Note: If something does not work with your version of RQG, try https://github.com/MariaDB/randgen branch elenst-mdev14409.

 Comments   
Comment by Marko Mäkelä [ 2018-11-26 ]

The fix for MDEV-14055 was to make lock_validate() ignore record lock bitmaps that are empty. The assertion itself was not removed.

In this case, it crashes on the root page of a table (page number is 3). The root page is no longer a leaf page; there are 2 child pages. There are no bits set in the record lock bitmap:

(gdb) p *lock
 
$1 = {trx = 0x7fe3f9fa2378, trx_locks = {prev = 0x7fe3f9fa2458, 
    next = 0x7fe3f9fa2708}, index = 0x7fe37c031728, hash = 0x0, 
  requested_time = 0, wait_time = 0, un_member = {tab_lock = {
 
      table = 0x300000009, locks = {prev = 0x50, next = 0x0}}, rec_lock = {
 
      space = 9, page_no = 3, n_bits = 80}}, type_mode = 1059}
 
(gdb) p/x *(byte*)&lock[1]@10
 
$2 = {0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0}

The problem is in the code:

	for (i = nth_bit; i < lock_rec_get_n_bits(lock); i++) {
 
 
 
		if (i == 1 || lock_rec_get_nth_bit(lock, i)) {
 
 
 
			rec = page_find_rec_with_heap_no(block->frame, i);
 
			ut_a(rec);
 
			ut_ad(page_rec_is_leaf(rec));
 
…
 
			lock_rec_queue_validate(
 
				TRUE, block, rec, lock->index, offsets);

Here, we have i == 1, that is, the supremum record of the page. That condition already existed in the very first published version of InnoDB.

lock_rec_queue_validate() on the page supremum record would only execute one loop:

	if (!page_rec_is_user_rec(rec)) {
 
 
 
		for (lock = lock_rec_get_first(lock_sys->rec_hash,
 
					       block, heap_no);
 
		     lock != NULL;
 
		     lock = lock_rec_get_next_const(heap_no, lock)) {
 
 
 
			ut_ad(!trx_is_ac_nl_ro(lock->trx));
 
 
 
			if (lock_get_wait(lock)) {
 
				ut_a(lock_rec_has_to_wait_in_queue(lock));
 
			}
 
 
 
			if (index != NULL) {
 
				ut_a(lock->index == index);
 
			}
 
		}
 
 
 
		goto func_exit;
 
	}

It looks like the assertion should be moved below this code, and that the original assertion should be relaxed to the following:

ut_ad(!lock_rec_get_nth_bit(lock, i) || page_rec_is_leaf(rec));

With these changes, the test case does not fail.

Generated at Thu Feb 08 08:13:19 UTC 2024 using Jira 8.20.16#820016-sha1:9d11dbea5f4be3d4cc21f03a88dd11d8c8687422.