Ah, the wonderful feeling of being hauled out of the spa/candlelit dinner/Quake game to be told in panicked tones that there is a ‘database error’. You rattle off the usual suspects, determining that their keyboard is actually plugged in, monitor on, monitor’s brightness turned up (yes, it has happened to me), and that they are actually on the right page. Eventually the conclusion is inescapable – there is no other alternative, the database really is behaving strangely.
Some cynics may say it happens more than usual with MySQL, and looking through the list of MySQL bugfixes is enough to strike terror into the hardiest of souls. From the MySQL documentation:
Fixed in 4.0.18
INSERT DELAYED … SELECT … could cause table corruption because tables were not locked properly. This is now fixed by ignoring DELAYED in this context. (Bug #1983)
Fixed in 4.0.16
Fixed bug in overrun check for BLOB values with compressed tables. This was a bug introduced in 4.0.14. It caused MySQL to regard some correct tables containing BLOB values as corrupted. (Bug #770, Bug #1304, and maybe Bug #1295)
Fixed in 4.0.15
Fixed rare bug in MYISAM introduced in 4.0.3 where the index file header was not updated directly after an UPDATE of split dynamic rows. The symptom was that the table had a corrupted delete-link if mysqld was shut down or the table was checked directly after the update.
Fixed in 4.0.14
Comparison/sorting for latin1_de character set was rewritten. The old algorithm could not handle cases like “sä” < "ßa". See section 5.6.1.1 German character set. In rare cases, it resulted in table corruption. But then, has anyone seen a SQL-Server buglist recently? Nevertheless, table corruption should be rare when using MySQL (though an overheating server continually restarting at the most inopportune times has meant I have seen more than my fair share of corruption recently, hence the inspiration for the article). Luckily, MySQL has some easy-to-use tools that can easily repair most cases of table corruption, and this article introduces you to these. You should always look at removing the causes of the corruption of course, but this article only deals with the firefighting aspect - repairing the symptoms. Identifying table corruption Table corruption should be relatively easy to identify. Queries that worked before suddenly stop working, or begin working inconsistently. Your first suspected culprit would be the code of course, but when a query such as UPDATE table_x SET x_key='d' doesn't work for no good reason, it's time to check the tables. If you see any of the following errors, it is also prudent to check the tables for corruption: Record file is crashed
Unexpected end of file
can't find file tablethatshouldbethere.MYI
tablethatwasworking.frm is locked against change
Got error ### from table handler.
The latter error returns an error number, and you can get more details about this error with the perror utility. perror sits in the same directory as all the other MySQL binaries, such as mysqladmin, mysql and those we'll discuss shortly, such as mysqlcheck and myisamchk. Some of the errors, which often indicate table corruption, include: 126: Index file is crashed / Wrong file format
127: Record file is crashed
132: Old database file
134: Record was already deleted (or record file crashed)
135: No more room in record file
136: No more room in index file
141: Duplicate unique key or constraint on write or update
144: Table is crashed and last repair failed
145: Tables was marked as crashed and should be repaired
Checking tables There are three ways to check tables. All of these work with MyISAM tables, the default, non-transactional table type, and one with InnoDB, the most mature of the MySQL transactional table types. Fortunately, MySQL now allows you to check tables while the server is still running, so corruption in a minor table need not affect everything on the server. The CHECK TABLE SQL statement (obviously the server must be running for this)
Running the mysqlcheck command-line utility (the server can be running)
Running the myisamchk command-line utility (the server must be down, or the tables inactive)
Checking tables with CHECK TABLE The first method for checking tables is to run the CHECK TABLE SQL statement while connected to the server. The syntax is: CHECK TABLE tablename[,tablename2...] [option][,option2...], for example: mysql> CHECK TABLE fixtures;
+————————-+——-+———-+———-+
| Table | Op | Msg_type | Msg_text |
+————————-+——-+———-+———-+
| sports_results.fixtures | check | status | OK |
+————————-+——-+———-+———-+
1 row in set (0.01 sec)
There are a number of options to specify as well, which allow you to do a more in-depth, or a more superficial kind of check than normal:
QUICK
The quickest option, and does not scan the rows to check for incorrect links. Often used when you do not suspect an error.
FAST
Only checks tables if they have not been closed properly. Often used when you do not suspect an error, from a cron, or after a power failure that seems to have had no ill-effects.
CHANGED
Same as FAST, but also checks tables that have been changed since the last check.
MEDIUM
The default if no option is supplied. Scans rows to check that deleted links are correct, and verifies a calculated checksum for all keys with a calculated a key checksum for the rows.
EXTENDED
The slowest option, only used if the other checks report no errors but you still suspect corruption. Very slow, as it does a full key lookup for all keys for every row. Increasing the key-buffer-size variable in the MySQL config. file can help this go quicker.
Note that CHECK TABLE only works with MyISAM and InnoDB tables. If CHECK finds corruption, it will mark the table as corrupt, and it will be unusable. See the Repairing tables section below for how to handle this.
Checking tables with mysqlcheck
The second method is to run the mysqlcheck command-line utility. The syntax is:
mysqlcheck [options] dbname tablename [tablename2… ].
The following options pertain to checking (mysqlcheck can also repair, as well as analyze and optimize, which are not covered here).
–auto-repair
Used together with a check option, it will automatically begin repairing if corruption is found.
–check, -c
Checks tables (only needed if using mysqlcheck under another name, such as mysqlrepair. See the manual for more details)
–check-only-changed, -C
Same as the CHECK TABLE … CHANGED option above.
–extended, -e
Same as the CHECK TABLE … EXTENDED option above.
–fast, -F
Same as the CHECK TABLE … FAST option above.
–medium-check, -m
Same as the CHECK TABLE … MEDIUM option above.
–quick, -q
Same as the CHECK TABLE … QUICK option above.
For example:
% mysqlcheck -uuser -ppass sports_results fixtures
sports_results.fixtures OK
Note that you can specify multiple tables, and that mysqlcheck only works with MyISAM tables.
Checking tables with myisamchk
Finally, there is the myisamchk command-line utility. The server must be down, or the tables inactive (which is ensured if the –skip-external-locking option is not in use). The syntax is: myisamchk [options] tablename.MYI, and you must be in, or specify, the path to the relevant .MYI files (each MyISAM database is stored in its own directory). These are the available check options:
–check, -c
The default option
–check-only-changed, -C
Same as the CHECK TABLE … CHANGED option above.
–extend-check, -e
Same as the CHECK TABLE … EXTENDED option above.
–fast, -F
Same as the CHECK TABLE … FAST option above.
–force, -f
Will run the myisamchk repair option if any errors are found
–information, -i
Display statistics about the checked table
–medium-check, -m
Same as the CHECK TABLE … MEDIUM option above.
–read-only, -T
Does not mark the table as checked
–update-state, -U
This option stores when the table was checked, and the time of crash, in .MYI file.
For example:
% myisamchk fixtures.MYI
Checking MyISAM file: fixtures.MYI
Data records: 1415 Deleted blocks: 2
– check file-size
– check key delete-chain
– check record delete-chain
– check index reference
– check data record references index: 1
– check record links
You can also use wildcard to check all the .MYI tables at the same time, for example:
% myisamchk *.MYI
Note that myisamchk only works with MyISAM tables. For those of you still using the old ISAM table types, there is also isamchk, though there is really little reason not to upgrade to MyISAM.
Repairing tables
In most cases, only the index will be corrupted (the index is a separate, smaller, file with records that point to the main data file) – actual data corruption is extremely rare. Fixing most forms of corruption is relatively easy. As with checking, there are three ways to repair tables. These all only work with MyISAM tables – to repair corruption of the other table types, you will need to restore from backup:
The REPAIR TABLE SQL statement (obviously the server must be running for this)
The mysqlcheck command-line utility (the server can be running)
The myisamchk command-line utility (the server must be down, or the tables inactive)
Repairing a table requires twice as much disk space as the original table (a copy of the data is made), so make sure you are not going to run out of disk space before you start.
Repairing a table with REPAIR TABLE
The syntax is, as would be expected, REPAIR TABLE tablename[,tablename1…] [options]. This method only works with MyISAM tables. The following options are available.
QUICK
The quickest, as the data file is not modified.
EXTENDED
Will attempt to recover every possible data row file, which can result in garbage rows. Use as a last resort.
USE_FRM
To be used if the .MYI file is missing or has a corrupted header. Uses the .frm file definitions to rebuild the indexes.
In most cases, a simple REPAIR without any options should work fine. An unusual case is when the .MYI is missing. Here is what would happen:
mysql> REPAIR TABLE fixtures;
+————————-+——–+———-+———————————————+
| Table | Op | Msg_type | Msg_text |
+————————-+——–+———-+———————————————+
| sports_results.fixtures | repair | error | Can’t find file: ‘fixtures.MYI’ (errno: 2) |
+————————-+——–+———-+———————————————+
The repair has failed because the index file is missing or has a corrupted header. To use the definition file to repair, use the USE_FRM option, as follows:
mysql> REPAIR TABLE fixtures USE_FRM;
+————————-+——–+———-+————————————+
| Table | Op | Msg_type | Msg_text |
+————————-+——–+———-+————————————+
| sports_results.fixtures | repair | warning | Number of rows changed from 0 to 2 |
| sports_results.fixtures | repair | status | OK |
+————————-+——–+———-+————————————+
Everything has gone smoothly this time, as indicated by the OK Msg_text.
Repairing tables with mysqlcheck
The mysqlcheck command-line utility can be used while the server is running, and, like all the methods of repair, only works with MyISAM tables. The syntax is:
%mysqlcheck -r sports_results fixtures -uuser -ppass
sports_results.fixtures OK
You can also repair multiple tables in a database, by listing them after the database name, or all tables in a database by just passing the database name, for example:
%mysqlcheck -r sports_results fixtures events -uuser -ppass
sports_results.fixtures OK
sports_results.events OK
%mysqlcheck -r sports_results -uuser -ppass
sports_results.fixtures OK
sports_results.events OK
…
Repairing tables with myisamchk
The server must be down, or the tables inactive (which is ensured if the –skip-external-locking option is not in use). The syntax is myisamchk [options[ [tablenames]. Remember again that you must be in, or specify, the path to the relevant .MYI files. The following options are available:
–backup, -B
Makes a .BAK backup of the table before repairing it
–correct-checksum
Corrects the checksum
–data-file-length=#, -D #
Specifies the maximum length of the data file, when recreating
–extend-check, -e
Attempts to recover every possible row from the data file. This option should not be used except as a last resort, as it may produce garbage rows.
–force, -f
Overwrites old temporary .TMD files instead of aborting if it encounters a pre-existing one.
keys-used=#, -k #
Can make the process faster by specifying which keys to use. Each binary bit stands for one key starting at 0 for the first key.
–recover, -r
The most commonly used option, which repairs most corruption. If you have enough memory, increase the sort_buffer_size to make the recover go more quickly. Will not recover from the rare form of corruption where a unique key is not unique.
–safe-recover, -o
More thorough, yet slower repair option than -r, usually only used only if -r fails. Reads through all rows and rebuilds the indexes based on the rows. This also uses slightly less disk space than a -r repair since a sort buffer is not created. You should increase the key_buffer_size value to improve repair speed if there is available memory.
–sort-recover, -n
MySQL uses sorting to resolve the indexes, even if the resulting temporary files are very large.
–character-sets-dir=…
The directory containing the character sets
–set-character-set=name
Specifies a new character set for the index
–tmpdir=path, -t
Passes a new path for storing temporary files if you dont want to use the contents of the TMPDIR environment variable
–quick, -q
The fastest repair, since the data file is not modified. A second -q will modify the data file if there are duplicate keys. Also uses much less disk space since the data file is not modified.
–unpack, -u
Unpacks a file that has been packed with the myisampack utility.
An example of its usage:
% myisamchk -r fixtures
– recovering (with keycache) MyISAM-table ‘fixtures.MYI’
Data records: 0
I hope that you will never have to do any repairing, but I am sure at least some of you reading this article (at least those who got this far) are here precisely because you’ve got some corruption. I hope that your repairs will be as painless as all mine have been. Good luck!