innochecksum always fail to check table files with KEY_BLOCK_SIZE=4 and ROW_FORMAT=COMPRESSED

[Ilya V. Yesin]

Test case files are attached.
Please, pay attention to following lines:
Converting table in COMPRESSED format... done.
Checking it:.
page 0 invalid (fails new style checksum)
file /var/lib/mysql/test/test/innocheck.ibd = 65536 bytes (4 pages)...
checking pages in range 0 to 3
page 0: log sequence number: first = 0; second = 0
page 0: old style: calculated = 412297167; recorded = 0
page 0: new style: calculated = 2171856543; recorded = 113443075
innochesum return error code 1

[Laurynas Biveinis (laurynas-biveinis)]

We could take https://bazaar.launchpad.net/~mysqlatfacebook/mysqlatfacebook/5.1/revision/3813

[Muhammad Irfan (muhammad-irfan)]

This also affects when COMPRESSED table used with default KEY_BLOCK_SIZE= 8K

mysql [localhost] {msandbox} ((none)) > SHOW GLOBAL VARIABLES LIKE '%version%';
+-------------------------+--------------------------------------------------+
| Variable_name | Value |
+-------------------------+--------------------------------------------------+
| innodb_version | 5.5.40-36.1 |
| protocol_version | 10 |
| slave_type_conversions | |
| version | 5.5.40-36.1 |
| version_comment | Percona Server (GPL), Release 36.1, Revision 707 |
| version_compile_machine | x86_64 |
| version_compile_os | Linux |
+-------------------------+--------------------------------------------------+

mysql [localhost] {msandbox} (test) > SHOW CREATE TABLE t1\G
*************************** 1. row ***************************
Table: t1
Create Table: CREATE TABLE `t1` (
`id` int(11) NOT NULL,
`name` char(20) DEFAULT NULL,
PRIMARY KEY (`id`)
) ENGINE=InnoDB DEFAULT CHARSET=utf8 ROW_FORMAT=COMPRESSED
1 row in set (0.00 sec)

mysql [localhost] {msandbox} (test) > SHOW TABLE STATUS LIKE 't1'\G
*************************** 1. row ***************************
Name: t1
Engine: InnoDB
Version: 10
Row_format: Compressed
Rows: 6
Avg_row_length: 1365
Data_length: 8192
Max_data_length: 0
Index_length: 0
Data_free: 0
Auto_increment: NULL
Create_time: 2015-02-12 02:53:19
Update_time: NULL
Check_time: NULL
Collation: utf8_general_ci
Checksum: NULL
Create_options: row_format=COMPRESSED
Comment:
1 row in set (0.00 sec)

mysql [localhost] {msandbox} (test) > SELECT * FROM t1;
+----+--------+
| id | name |
+----+--------+
| 1 | Aamir |
| 2 | Arif |
| 3 | Essa |
| 4 | Hashir |
| 5 | Umair |
| 6 | Ali |
+----+--------+
6 rows in set (0.00 sec)

$ /home/mirfan/percona5.5.40/bin/innochecksum -vd data/test/t1.ibd
file data/test/t1.ibd = 65536 bytes (4 pages)...
checking pages in range 0 to 3
page 0: log sequence number: first = 1618767; second = 0
page 0 invalid (fails log sequence number check)

===============================
Testing with innochecksum 5.7.5:
===============================
$ /home/mirfan/mysql5.7.5/bin/innochecksum --version
/home/mirfan/mysql5.7.5/bin/innochecksum Ver 5.7.5, for linux-glibc2.5 (x86_64)

$ /home/mirfan/mysql5.7.5/bin/innochecksum --log=/tmp/log.txt --page-type-dump=/tmp/pagetypedump.log data/test/t1.ibd

$ cat /tmp/log.txt
InnoDB File Checksum Utility.
Filename = data/test/t1.ibd
Innochecksum: checking pages in range 0 to 7
page::0; innodb checksum: calculated = 1235628449; recorded = 1235628449
page::0: crc32 checksum: calculated = 4062747429; recorded = 1235628449
page::0: none checksum: calculated = 3735928559; recorded = 1235628449
page::1; innodb checksum: calculated = 1034289170; recorded = 1034289170
page::1: crc32 checksum: calculated = 4179299713; recorded = 1034289170
page::1: none checksum: calculated = 3735928559; recorded = 1034289170
page::2; innodb checksum: calculated = 2700424990; recorded = 2700424990
page::2: crc32 checksum: calculated = 650835105; recorded = 2700424990
page::2: none checksum: calculated = 3735928559; recorded = 2700424990
page::3; innodb checksum: calculated = 3540530163; recorded = 3540530163
page::3: crc32 checksum: calculated = 5797...

[Przemek (pmalkowski)]

Laurynas, instead of patching innochecksum 5.5/5.6, why not just backport the one from 5.7, which works with compressed tables and also provides more features?

[Laurynas Biveinis (laurynas-biveinis)]

Maybe. innochecksum uses InnoDB, so the backport might pull in a bit more than the innochecksum itself, but it's indeed an option to consider.

[Laurynas Biveinis (laurynas-biveinis)]

5.5, 5.6, and 5.7 versions of innochecksum are all different: 5.5 one is a standalone utility, 5.6 links with InnoDB, 5.7 is further rewritten. Thus 5.5 version of the fix is a port from Facebook 5.1 patch, and 5.6 version is a port from Facebook 5.6 patch, which is different from 5.1 version. Both versions backport a small subset of 5.7 innochecksum MTR testcases.

    https://github.com/percona/percona-server/pull/77
    https://github.com/percona/percona-server/pull/78

[Shahriyar Rzayev (rzayev-sehriyar)]

Percona now uses JIRA for bug reports so this bug report is migrated to: https://jira.percona.com/browse/PS-1302