Mysql Crash Recovery

In the event of database corruption it may be necessary to manually perform database recovery. See Bug 15797 for an example of an issue with mysql that will require database recovery. In that example, a warning message like the following appeared in the mysql error log:

InnoDB: Serious error! InnoDB is trying to free page 716
InnoDB: though it is already marked as free in the tablespace!
InnoDB: The tablespace free space info is corrupt.
InnoDB: You may need to dump your InnoDB tables and recreate the whole
InnoDB: database!

Before beginning a full database recovery, check to see if the corruption may be limited to a single mboxgroup or a single user within an mboxgroup. This type of corruption frequently lets the server run normally for extended periods of time, with crashes occurring only when an affected user attempts to access certain mailbox items. If this is the case, it may be possible to dump, drop and recover only the affected entries without disrupting the database as a whole. Please see the instructions in the Mysql Crash Recovery (alternate method) article.

Overview of Recovery Process

1. Configure mysql to start in recovery mode
2. Generate SQL dumps of all relevant databases
3. Remove all existing (and possibly corrupt) databases
4. Re-create all databases
5. Repopulate the databases with the data from the SQL dumps
6. Test databases and start all ZCS services

Details of Recovery Process

1. Configure mysql to start in recovery mode

1. Edit the file /opt/zimbra/conf/my.cnf and add a line like innodb_force_recovery = 1 under the [mysqld] section (Note that it may be necessary to increase the recovery level depending on the extent of the database corruption, as shown at the end of the database dump step)
2. Save the file and re-start mysqld

mysql.server start

2. Generate SQL dumps of all databases

1. Load some mysql configuration into shell variables (i.e. $mysql_socket and $mysql_root_password; note that you will use these again in step 3)
2. Make a list of the existing databases
3. Create a directory to hold the SQL dumps
4. Generate the SQL dumps from the database list

source ~/bin/zmshutil ; zmsetvars

mysql --batch --skip-column-names -e "show databases" | grep -e mbox -e zimbra > /tmp/mysql.db.list

mkdir /tmp/mysql.sql

for db in `cat /tmp/mysql.db.list`; do
 ~/mysql/bin/mysqldump $db -S $mysql_socket -u root --password=$mysql_root_password > /tmp/mysql.sql/$db.sql
     echo "Dumped $db"
 done

Note: If you encounter any mysql errors while dumping the databases, start over by re-editing /opt/zimbra/conf/my.cnf, incrementing the value for innodb_force_recovery by one, and restarting mysqld. The maximum recovery level is 6. Please see MySQL’s Forcing InnoDB Recovery guide for more information.

Note: An error of “bash: /tmp/mysql.sql/$db.sql: ambiguous redirect” probably indicates your using an apostrophe or single quote ‘ rather than a tick ` — which is one the same key as the tilde ~ .

Note: Do not reboot the machine, as some Operating Systems will remove all contents in the /tmp directory during the reboot sequence, i.e. your /tmp/mysql.sql will be removed.

3. Remove all existing (and possibly corrupt) databases

Note that we drop the zimbra database last because the mboxgroup* databases depend on it

for db in `cat /tmp/mysql.db.list |grep mbox`
do
    mysql -u root --password=$mysql_root_password -e "drop database $db"
    echo -e "Dropped $db"
    sleep 1
done

mysql -u root --password=$mysql_root_password -e "drop database zimbra"

Remove existing InnoDB tablespace and log files

rm -rf /opt/zimbra/db/data/ib*

Note: First, use with caution – this shouldn’t need to be used often. Issue came about because of some rsync issues. Can’t dump db’s because of ‘connection’ issues at this point? One could move the /opt/zimbra/db/data directory – mv /opt/zimbra/db/data /opt/zimbra/db/data-old and then make the db – mkdir /opt/zimbra/db/data w/ ownership of zimbra:zimbra . Remove the innodb_force_recovery line from /opt/zimbra/conf/my.cnf . Then recreate a default mysql db by running /opt/zimbra/libexec/zmmyinit –sql_root_pw $mysql_root_password and then attempt this steps over again to confirm you can drop them. Also note that you may have to reset the zimbra password manually in mysql, then set it again in Zimbra with the instructions from this page: http://wiki.zimbra.com/wiki/Resetting_LDAP_%26_MySQL_Passwords

4. Re-create all databases

1. Run mysql in non-recovery mode
   1. Remove the innodb_force_recovery line from /opt/zimbra/conf/my.cnf
   2. Save the file and restart mysqld
2. Re-create the databases from the database list

mysql.server restart

for db in `cat /tmp/mysql.db.list`
do
    mysql -e "create database $db character set utf8"
    echo "Created $db"
done

5. Repopulate the databases with the data from the SQL dumps

Import the data from the SQL dumps. Note that we import the zimbra database first because the mboxgroup databases depend on it

mysql zimbra < /tmp/mysql.sql/zimbra.sql

for sql in /tmp/mysql.sql/mbox*
do
    mysql `basename $sql .sql` < $sql
    echo -e "Updated `basename $sql .sql` \n"
done

6. Test databases and start all ZCS services

Note that this is an example query. If you know of any particular databases that were corrupt, you may want to construct other queries to verify normal access to the data.

mysql zimbra -e "select * from mailbox order by id desc limit 1"

Once you are satisfied that the databases are restored intact, start the rest of the zimbra services.

zmcontrol start

Check /opt/zimbra/log/mysql_error.log and /opt/zimbra/log/mailbox.log for database errors.