Friday Jul 10, 2015

Backing up and restoring tables named with special characters


The names of databases and tables within MySQL are known as identifiers. In the simplest case these identifiers are just strings of certain ASCII characters (the basic Latin letters, the digits 0-9, the dollar sign and the underscore). However, if an identifier is placed in quotes, it can contain any character of the full Unicode Basic Multilingual Plane (except U+0000). We say that a character is a special character if it is permitted in a quoted identifier but not in an unquoted identifier.

MySQL Enterprise Backup (MEB) 3.12.1 introduces support for proper handling of table and database names with special characters. In MEB versions prior to 3.12.1 database and table names were represented as ASCII strings and the same string was used on the command line, internally within MEB and in filenames.  This caused MEB to fail some operations when database/table names contained characters other than those that can appear in unquoted MySQL identifiers.

Starting with version 3.12.1, MEB represents database and table names as strings of Unicode characters (or code points) and three distinct encodings are employed: the command-line character encoding, the internal UTF-8 encoding and the filename encoding used by the MySQL server. Thus, the same string of Unicode characters appears in three different forms depending on where the string is used.

The command-line character encoding is defined by the operating system. On Linux and other Unix-line systems this is done by the command shell and on Windows by the Command Prompt.  In order for the special character support to work, MEB must know what character encoding is used on the command line. The user can tell MEB the character encoding used on the command line by specifying the name of the encoding with the --default-character-set option. If this option is not given, then MEB tries to detect the command-line character encoding by examining the command program from which it was invoked. 

In the following example the database db contains three tables with names that contain special characters: àbc,  à123 and déf. We backup the tables àbc and à123 using the Transportable Tablespaces feature (TTS) of the MySQL Server as follows:

$ mysqlbackup --include-tables='`db`\.`à.*`' --use-tts 
              --backup-dir=/full-backup  backup
MySQL Enterprise Backup version 3.12.1 Linux-3.2.0-80-generic-i686 [2015/06/22]
Copyright (c) 2003, 2015, Oracle and/or its affiliates. All Rights Reserved.

The tables àbc and à123 are included in this partial backup by specifying with the --include-tables option a regular expression that matches the fully-qualified table names `db`.`àbc` and `db`.`à123`.

In the example above, we have specified both the database and table name quoted with backticks. However, it is not necessary to provide these backticks because MEB forms four strings from every fully-qualified table name: DB.TABLE, `DB`.TABLE, DB.`TABLE` and `DB`.`TABLE`, and a regular expression is defined to match a table if it matches any of these four strings generated from the table name.

Special characters can be specified with these four MEB options:

--rename="OLD-NAME to NEW-NAME"

Regular expressions can contain any Unicode character in the Unicode Basic Multilingual Plane and they are matched against strings of Unicode characters.  This means that special characters can be used in regular expressions in exactly the same way as the characters that are not special.

Useful tips

Each command program (or a shell, in the Unix terminology) treats certain characters in a special way. For example, the bash program (which is popular on Linux and other Unix-like systems) uses single and double quotes as quoting characters and assigns a very special meaning to backticks: anything enclosed in backticks is treated as a command line and evaluated. This can be problematic when the user invokes MEB from the bash program. Suppose that the user wishes to restore the table db.à123 from a backup and rename it to db.new_à123. The command-line

$ mysqlbackup ... --include-tables="db\.à123"  
                  --rename="`à123` TO `new_à123`"   copy-back

seems to do just that, but unfortunately when invoked from the bash shell it fails with the following error:

à123: command not found

The problem is that the new table à123 contains a special character and the table name must be therefore quoted. However, the string `à123` is interpreted by the bash shell as a request to run the command à123 which, of course, fails.

There are two ways to remedy this situation. In the first fix, we place the parameter of the --rename option in single quotes:

$ mysqlbackup ... --include-tables="db\.a123" 
                  --rename='`à123` TO `new_à123`'  copy-back

This works because the bash shell treats backticks as normal characters when they appear in a string placed in single quotes.

Another solution is provided by MEB. In addition to the backtick quoting, MEB allows identifiers to be quoted with single quotes and double quotes. Therefore, the following command line (where we replaced backticks with single quotes) works

$ mysqlbackup ... --include-tables="db\.a123"   
                  --rename="'à123' TO 'new_à123'"  copy-back

MEB converts any identifiers on the command line that are quoted with single or double quotes to a standard form where the quoting is done with backticks. This is a very handy feature when MEB is invoked from a command program, such as the bash shell, that assigns special meaning to backticks.

Sometimes the Windows Command Prompt can be a little bit tricky to set up correctly to handle special characters. For example, single quotes do not work with the Windows Command Prompt. Therefore, some users might want to consider using the free Cygwin command program when working with special characters on Windows. Cygwin software provides a command line similar to the one on the Unix-like systems. Using Cygwin should allow easy manipulation of the command-line character encoding.


Sometimes MEB has the wrong idea of the character encoding used on the command line. Here we show two examples of what can happen if the command-line character encoding is different from what MEB thinks it is.

$ mysqlbackup --default-character-set="latin1" --backup-dir=/full-backup 
              --include-tables='db\.a123'  --rename="a123 TO 'new_à123'" copy-back
MySQL Enterprise Backup version 3.12.1 Linux-3.2.0-80-generic-i686 [2015/07/07] 
Copyright (c) 2003, 2015, Oracle and/or its affiliates. All Rights Reserved.
150707 17:01:30 mysqlbackup: INFO: Importing table: db.a123 and 
                                   renaming it to `db`.`new_à123`.
150707 17:01:31 mysqlbackup: INFO: Analyzing table: `db`.`new_à123`.
150707 17:01:31 mysqlbackup: INFO: Copy-back operation completed successfully.
150707 17:01:31 mysqlbackup: INFO: Finished copying backup files to '/sqldata/tts-5.6'
mysqlbackup completed OK!

Everything looks good in the example shown above: it seems that we have successfully restored the table db.a123 and renamed it to db.new_à123, as we intended. However, when we look at the list of tables at the server we see that the new table name is not db.new_à123, but db.new_Ã 123:

$ echo "use db; show tables;" | mysql
new_Ã 123

This problem was caused by specifying an incorrect value for the command-line character encoding with the --default-character-set="latin1" option on the command line invoking the copy-back operation.

In the next example we show another possible outcome when MEB is given a wrong command-line character encoding. We invoke the following MEB restore operation from a command shell that uses UTF-8 character encoding.

$ mysqlbackup --default-character-set="hebrew" --backup-dir=/full-backup 
              --include-tables='db\.a123'  --rename="a123 TO 'new_à123'" 
              --force  copy-back
MySQL Enterprise Backup version 3.12.1 Linux-3.2.0-80-generic-i686 [2015/06/22] 
Copyright (c) 2003, 2015, Oracle and/or its affiliates. All Rights Reserved.
 mysqlbackup: ERROR: Failed to convert "hebrew" characters to "utf8"
 characters in string "a123 TO 'new_à123'". 
 mysqlbackup: ERROR: Unicode character conversion failed.

The conversion from "hebrew" characters to "utf8" characters fails because the character encoding used on the command line is not "hebrew", but "utf8". This conversion fails in the middle of the string because  some of the UTF-8 encoded characters are not valid "hebrew" encoded characters. The error message "Unicode character conversion failed" means always that MEB has the wrong idea of the character encoding used on the command line. To recover from this failure, the user must find out the character encoding used on the command line and specify it to MEB with the --default-character-set option.

Monday Mar 23, 2015

Renaming tables with MySQL Enterprise Backup 3.12.0


MySQL Enterprise Backup 3.12.0 (MEB) introduces a new feature for restoring an InnoDB table from a backup. Now it is possible to rename the table during restore. This is useful when the user wants to restore a table from a backup without overwriting the existing version of the table in the database.

The following example illustrates how the renaming feature could be used.  Suppose that the DBA has deleted three rows from a table T1 by mistake and he now wishes to get them back from a backup. He wants to leave the database online and to restore the 3 deleted rows from a TTS backup (a backup created with the --use-tts option) that contains the table T1.  He can do this with this feature in three steps:

  1. He restores with MEB the table T1 from a TTS backup renaming it to T2.

  2. He uses MySQL client to issue SQL statements to copy the 3 mistakenly deleted rows from the table T2 to the table T1.

  3. He drops the table T2.

Now the accidentally dropped rows have been restored and the restore took place when the MySQL server was online and the restore did not disturb the normal operation of the server in any way.

User Interface

The command-line interface for restore is extended with the --rename option that specifies a mapping of from the old name to the new name. The --rename option has the following syntax:

--rename="OLD-NAME  to  NEW-NAME

The OLD-NAME and NEW-NAME are either fully-qualified tablenames of the form DB.TABLE, or tablenames without the database part. The OLD-NAME must match the name of the table selected for restore.  

Example 1:

A sample command-line for restoring the table  to  test.abc_new:
$ mysqlbackup --include-tables="test\.abc"
              --rename="abc TO abc_new" 
In this example we assume that the backup contains several tables. Therefore, we have to specify a single table ( with the --include-tables option.

Example 2:

A sample command-line for restoring the table  to

$ mysqlbackup --include-tables="db\.abc"
              --rename="abc to" 

Note that if the database db2 does not exist, the restore will create it.

Example 3: 

Below is an excerpt of the printouts MEB produces when a table is renamed during restore using the command-line from Example 1: 

$ mysqlbackup --backup-dir=/full-backup --include-tables="test\.abc" --rename="abc TO new_abc" copy-back
MySQL Enterprise Backup version 3.12.0 Linux-3.2.0-69-generic-i686 [2015/01/22] 
Copyright (c) 2003, 2014, Oracle and/or its affiliates. All Rights Reserved.
 mysqlbackup: INFO: Starting with following command line ...
        --backup-dir=/full-backup --include-tables=test\.abc 
        --rename=abc TO new_abc copy-back 
 mysqlbackup: INFO: 
IMPORTANT: Please check that mysqlbackup run completes successfully.
           At the end of a successful 'copy-back' run mysqlbackup
           prints "mysqlbackup completed OK!".
150312 11:17:53 mysqlbackup: INFO: MEB logfile created at /full-backup/meta/MEB_2015-03-12.11-17-53_copy_back.log
 mysqlbackup: INFO: MySQL server version is '5.6.11'.
 mysqlbackup: INFO: Got some server configuration information from running server.
 mysqlbackup: INFO: Creating 14 buffers each of size 16777216.
150312 11:17:53 mysqlbackup: INFO: Copy-back operation starts with following threads
		1 read-threads    1 write-threads
150312 11:17:53 mysqlbackup: INFO: Creating table:
150312 11:17:53 mysqlbackup: INFO: Copying /full-backup/datadir/test/abc.ibd.
150312 11:17:53 mysqlbackup: INFO: Completing the copy of all non-innodb files.
150312 11:17:54 mysqlbackup: INFO: Importing table: and renaming it to test.new_abc.
150312 11:17:55 mysqlbackup: INFO: Analyzing table: test.new_abc.
150312 11:17:55 mysqlbackup: INFO: Copy-back operation completed successfully.
150312 11:17:55 mysqlbackup: INFO: Finished copying backup files to '/sqldata/tts-5.6'
mysqlbackup completed OK! 


Renaming works only when restoring a single table from a TTS backup (a backup created with the --use-tts option). If the backup contains multiple tables, then a single table should be specified for restore with the --include-tables and --exclude-tables options. 

Monday Sep 15, 2014

MEB copies binary logs and relay logs to support PITR and cloning of master/slave

With MySQL Enterprise Backup(MEB) 3.9.0 we had introduced full instance backup feature for cloning the MySQL server. Now with MEB 3.11.0 we have enhanced the feature by copying all the master-slave setup files like MySQL server binary logs(will be referred as 'binlogs'), binary log index files, relay logs of slave, relay log index files, master info of slave, slave info files. As part of full instance backup, copying of binlog files is default behavior MEB-3.11.0 onwards. DBA should be aware of the fact that current full instance backup is bigger than the backups with old MEB's.

As every event on MySQL production database goes as a entry to binlog files in particular format, binlog files could be huge. Backing of huge binlog and/or relaylog files should not impact the performance of MySQL server. Hence, all the binlog files, except the current binlog used by server, are copied just like the innodb .ibd files without locking tables. Binlog files currently being used by server and added after backup started, are copied during read the lock which is acquired by MEB for copying meta files and redo logs.

DBA gets the following benefits:


1) Direct cloning of  master and slave possible from backup

Earlier DBA had to copy binlog files manually in order to setup  master/slave. Now, MEB 3.11 by default copies all the files including the global variables needed for setting up master-slave. Hence DBA can clone master or slave with the same state of backed-up server.

Now, DBA need not to use --slave-info option to copy the binlog info for setting up the slave after restore. By copying master and slave info files,  DBA can fetch the information of up to which master binlog position,  slave SQL thread has executed and IO threads has read etc. With this information along with relay logs, binlogs, DBA can easily setup slave from backed-up slave content

2) Backup of binary logs helps in Point In Time Recovery (PITR)

First let us understand what is PITR by above example. Consider DBA has taken full backup on Sunday(assume date as 14-09-2014), and incremental backups on Tuesday(date as 16-09-2014), Thursday(date as 18-09-2014). It means DBA can only restore database up to full backup or incremental backups in other words database can be restored either up to Sunday or up to Tuesday, Thursday,  but not in between let say Monday or Wednesday. Because backup is just a snapshot of data when it was taken. Hence backup taken once can't be restored in between without change log.  That's where binlog helps in restoring to a certain point of time, which is called Point-In-Time-Recovery(PITR). As binlogs captures all the events of a server with timestamps. Therefore to restore in between DBA need to have base data i.e. full backup and incremental binlogs.

Let's look at our example, below are the points to recover server to Wednesday 12 PM(assume date as 17-09-2014)
a) Restore the backup up to latest backup before PITR time(Here, restore Tuesday's incremental)
b) Get the SQL statements using below mysqlbinlog command up to PITR from the immediate next incremental binlogs(Here get SQL statements up to Wednesday from Thursday's incremental binlogs of binlog.000005, binlog.000006, binlog.000007)

mysqlbinlog --start-datetime=<latest backup time before PITR time> \
         --stop-datetime=<PITR point> \
         <incremental binlogs from immediate next backup>  > <SQL file>

For our above example, the command is
mysqlbinlog --start-datetime="2014-09-16 12:00:00" \
         --stop-datetime="2014-09-17 12:00:00" \
         binlog.000005 binlog.000006 binlog.000007  > mysql_PITR_restore.sql

Read Point-in-Time (Incremental) Recovery Using the Binary Log for more details about PITR using Binary logs.

c) Execute the SQL statements obtained on the restored server, server is restored to PITR point

3) Backing up relay-logs from slave server helps avoiding unnecessary pull of logs from master once it is restored

Let us understand this by an example

Slave has 1 relay log with master binlog positions from 1 to 100

SQL thread at slave reads from relaylog and apply events on slave. Now assume SQL thread currently executed statements 1 to 20 and 21 to 100 are yet to be executed.

If DBA takes backup without copying relay log, when he/she restores the backup as slave, it asks master from the binlog position 21. So restored slave need to pull the logs of binlog position 21 to 100 from master. More network I/O needed as usually slave is on different machine.

As MEB takes backup of relay log, slave can avoid pulling the logs for binlog positions 21 to 100. Now restored slave asks master from binlog positions 101 onwards. This way slave don't pull logs from master which are present in slave backup, there by reducing network I/O which is costly than disk I/O.

Unlike binary logs, relaylogs are mostly deleted automatically once applied by SQL thread, as a result few relay logs exist at any point of time. So all the relay logs are copied for all the backup types full, incremental, partial without major impact on backup size and time.

4) Copied binary logs remains consistent with the backup data

Earlier DBA had to copy binlog files manually in order to setup master/slave. Data files are copied by MEB and binlogs are copied by DBA at two different times, so there is a possibility of binlog files not consistent with the backed-up data.

Lets consider following example:
1. MEB takes backup of the server without binlogs at 1 PM
2. DBA has copied binlogs from the server at 1:30 PM
From 1 PM to 1:30, lets say 100 events logged in binlogs

Now to use these binlog files, DBA has to either execute 100 events on server or have to remove 100 events from binlog files.

Consider another example:

1. DBA has copied binlogs from the server at 1:30 PM
2. MEB takes backup of the server without binlogs at 2 PM
From 1:30 PM to 2 PM, lets say 100 events went into backup data

Now DBA has to copy the missing binlog files again from the running server.
With MEB 3.11.0 onwards, binlogs and the data are copied at the same time, so they are consistent with each other.

Options to avoid binlogs/relay logs:
If DBA is not concerned about backing up binlog files then he/she can use --skip-binlog and --skip-relaylog to skip relay log files in backup. It is advisable to use these options if he/she don't plan to clone server or want PITR.

For Master, to skip only binlogs:
./mysqlbackup --skip-binlog --backup-dir=back_dir --socket=server_sock backup

For Slave, to skip relay-logs
./mysqlbackup --skip-relaylog --backup_dir=back_dir --socket=server_sock backup

For Slave which is also a master, to skip both binlogs and relay logs
./mysqlbackup --skip-binlog --skip-relaylog backup_dir=back_dir --socket=server_sock

Options for offline backup:
MEB also supports offline backup. In order to copy binlog and/or relaylog, MEB searches for default values of log-bin-index(default: host_name-bin.index), relay-log-index(default: host_name-relay-bin.index), relaylog-info-file(default:, master-info-file(default: at default location that is in server's 'datadir'. And if MEB finds those files then it successfully backs up those files. In case those files are configured with different values, DBA need to provide --log-bin-index = PATH, --relay-log-index = PATH, --relaylog-info-file = PATH, --master-info-file=PATH options to MEB in order to copy them.

To enrich the full instance backups that MySQL Enterprise Backup has been performing since release 3.9.0, all the replication setup files are included as part of all the backups in 3.11.0. With these files as well as all the global variables, plugin details, MEB now takes the responsibility of giving all the details to DBA for cloning any server. Read MEB 3.11.0 documentation for more details and many other great features.

Thursday Sep 04, 2014

Optimistic Backup


MySQL Enterprise Backup (MEB) is a highly efficient tool for taking backups of your MySQL databases. In the 3.11.0 release we are taking that one step further by introducing a new concept called "optimistic" backup. Optimistic backup leverages the patterns we saw frequently especially as related to very large databases.

For backups the goals are:
1 - Quality and Consistency - the backup and more importantly the restore just "works"
2 - Size, time, and overhead - like in the game of golf - low score wins - for backups and for
3 - Flexibility – It’s not always one size fits all - whether how the backup is run, where it goes,
     how it is recovered.

With optimistic backup - we look at mostly the read aspects of your database to enable us to create a backup that is smaller, faster to backup, faster to restore, and has less overhead when running, but still remains consistent. As we said earlier “optimistic” is not for all databases. Those that benefit the most are databases where modifications(insert/updates/deletes) are focused on a minority of the tables, and store a great deal of data will benefit the most. In our tests (details below) mean time to restore reduced 18%, resulting from a 60% reduction in the size of the redo log that is applied on recovery. What follows is the deals on how this works. We also want to thank the DBAs who provided great feedback and details on their database characteristics during our development of this new feature.

Optimistic Backup

Whole idea is summarized in following two steps.

1.  Divide the InnoDB tables into two groups. In the first group are those tables that are unlikely to be modified during the backup, and in the second group are all the other tables. MEB is optimistic about the first set of tables as they are unlikely to be modified during backup operation hence these tables are termed as optimistic tables and backup operation is named after them as "optimistic backup". Tables in the second group are likely to be modified during backup operation hence are termed as optimistic busy tables.

2.  Split full backup operation into two phases. In the first phase of the backup, tables from the first group are copied to the backup without locking the MySQL server instance. In the second phase tables from second group are copied similar to normal backup i.e.  All InnoDB tables from second group and tables from the first group that have modified since the backup started are copied to backup, then MySQL server instance is locked to copy remaining files. First phase always excludes the undo/redo logs, system table space files and other meta files because these are most likely to be modified. Since optimistic tables are copied in the first phase hence it is called optimistic phase and second phase is called normal phase. 

How to trigger Optimistic backup

Following two new options have been introduced to trigger optimistic backup. Provide either of these two or both options to enable optimistic backup.


This option specifies the cut-off date-time to identify tables to be skipped in first phase. In other words, tables modified before the specified date-time (i.e. optimistic tables), are copied in first phase and tables modified after specified date-time value (i.e. busy tables), are copied in second phase.  Following date-time formats are supported by this option- 

• <Number><unit> - A valid integer number according to its unit. Supported units are years, days, months, hours and minutes. 

For example: ‘--optimistic-time=1years’ means tables modified one year before the backup starts to be copied in first phase. In other words tables modified in last one year from the backup starts to be copied in second phase.

• Date-time - It can be specified in different formats as:  YYMMDD, YYYYMMDD, YYMMDDHHMMSS, YYYYMMDDHHMMSS, YY-MM-DD, YYYY-MM-DD, YY-MM-DD HH.MM.SS,   and YYYYMMDDTHHMMSS where T is a character. 

For example - 16th May 2014 can be specified as 140516 or 20140516. Therefore ‘--optimistic-time=20140516’ means tables modified before this time will be copied in first phase.

• now - This is the default value of this option. If no value is specified against this option then MEB will assign “now” as default value. Internally MEB keeps backup start time as optimistic time and use the same throughout the backup operation. With this option all InnoDB tables are be copied in the 1st step.

For example:   --optimistic-time=now   OR --optimistic-time


This option specifies the InnoDB tables (i.e. busy tables) to be skipped in the first phase. It requires regular expression as specified in the POSIX 1003.2 standard, similar to – include/exclude-tables=REGEX option. This option expects fully qualified names of database objects, i.e. database_name[.table_name] in regular expression. Tables that match the REGEX criteria will be skipped in the first phase and copied in the second phase. This option can co-exist with 'optimistic-time', 'include-tables' and 'exclude-tables' options. These options will get precedence in ascending order as following 

Include/exclude-tables > optimistic-busy-tables > optimistic-time

Include/exclude-tables options gets higher precedence over optimistic-busy-tables option.

For instance - if same table is specified in the –exclude-tables and –optimistic-busy-tables then that table is not copied at all. 

optimistic-busy-tables gets higher precedence over optimistic-time option. 
For instance - if a table ‘user’ has modified time of ‎ “04/15/‎2014 ‏‎2:19:32”. If --optimistic-busy-tables=^test.user --optimistic-time=04142014 values are specified then file test/user.ibd is not be copied in first step because optimistic-busy-tables value matches the REGEX so test.user.ibd is not evaluated by optimistic-time option.

Using Optimistic backup

Following are samples of command line options need to be specified to perform different types of optimistic backups. 

1. Normal Backup with optimistic backup options

  >mysqlbackup.exe --backup-dir=D:\backup --optimistic-busy-tables="^db2\.t1+" backup

  >mysqlbackup.exe --backup-dir=D:\backup --optimistic-time backup

2. Compress Backup with optimistic backup options

 >mysqlbackup.exe --backup-dir=D:\backup --compress-method=lzma
   --optimistic-busy-tables="^db2\.t1+" backup 

 >mysqlbackup.exe --backup-dir=D:\backup --compress-method=lzma --optimistic-time=now

3. Compress Image Backup with optimistic backup options

 >mysqlbackup.exe --backup-dir=D:\backup --compress-method=lzma
  --backup-image=D:\backup.img  --optimistic-busy-tables="^db2\.t1 +" backup-to-image

 >mysqlbackup.exe --backup-dir=D:\backup --compress-method=lzma
  --backup-image=D:\backup.img  --optimistic-time="2014-08-27 13.25.00" backup-to-image

4. Compress and encrypted Backup with optimistic backup options

 >mysqlbackup.exe --backup-dir=D:\backup --compress-method=lzma
   --backup-image=D:\backup.img --encrypt --key=xxxxxxxxx
   --optimistic-busy-tables="^db2\.t1 +" backup-to-image

 >mysqlbackup.exe    --backup-dir=D:\backup --compress-method=lzma
   --backup-image=D:\backup.img --encrypt --key=xxxxxxxxx
   --optimistic-time=20140827T132500 backup-to-image

5. Partial Backup with optimistic backup options

 >mysqlbackup.exe --backup-dir=D:\backup --optimistic-time=25days
   --include-tables="^db2+"  backup

 >mysqlbackup.exe --backup-dir=D:\backup --optimistic-busy-tables="^db2\.t1+"
   --include-tables="^db2+"  backup 

 >mysqlbackup.exe --backup-dir=D:\backup --optimistic-busy-tables="^db2\.t1+"
   --optimistic-time=10Months --include-tables="^db2+"  backup

Troubleshooting optimistic backup

If optimistic backup is triggered then MEB prints INFO messages in log in following order to know when each phase of backup started and finished. 

First phase of optimistic backup started 



First phase of optimistic backup completed successfully

Second phase of optimistic backup started



Second phase of optimistic backup completed successfully

Turn on --trace option for more diagnostic information in logs. Internally MEB converts optimistic-time value to an epoch time stamp and compares the same with the InnoDB files’ modification time. 

With increased trace levels we see optimistic-time, backup start time and InnoDB table modification time in form of the epoch time stamp. We can easily convert these epoch time stamps in human readable time either using SQL query or through some external tools, for instance epoch converter.

Choose values of Optimistic backup options wisely

It's imperative to choose the suitable values of above options in order to maximize the gain from optimistic backup. Optimistic backup targets large databases in which only a few tables are modified frequently. DBA can find out such tables or time since when most of those tables have modified, either by profiling database or directly from file system. If such tables are identified then those tables can be specified in optimistic-busy-tables option. If duration, since when only small percentage of tables are modified, is identified then that date-time can be specified in optimistic-time option.  

Imagine the value of optimistic-busy table option is not chosen carefully then chances are that some optimistic tables might be copied in second phase which ideally should have been copied in first phase only, as a result redo log size will increase and hence overall backup time.
Similarly if the value of the optimistic-time option is not chosen carefully then chances are that some busy tables might be copied in first phase which ideally should have been copied in second phase only. In second phase those stale tables are copied again, as a result overall backup time will increase. In such cases for image backup, busy tables will be copied in both phases hence disk space and backup time both will increase. 

If only a few tables less in size (e.g. ~2% of the total size of the the database) will be modified during backup time then default value of optimistic-time 'now' should serve the purpose. In this case only those a few tables will be copied twice. 

Performance tests

In our internal tests we created a 1.2 TB of DB which had 25 tables initially. We created 3 tables and updated them continuously until backup operation completed successfully. We observed ~60% reduction in redo log file (ibbackup_logfile) size and ~18%  reduction in overall backup (backup + apply log) time compared to normal backup operation. Of course these numbers highly depends on the hardware, CPU load and types of DDL being executed. Therefore a user might experience better or poor performance than the numbers shared here. Purpose to share the number is to show that if options are specified correctly then optimistic backup performs better compared to normal backup.


The optimistic backup isn't for all cases, but for many it will have significant advantages and benefits over a typical full backup. So we hope you'll give it a try and provide us feedback on how it works for your environment. We shall also follow-up with additional queries and scripts to help you easily determine how values to set for the optimistic backups, although we suspect that many DBAs likely already know their data so well they won't need these. Never forget, good backups, and more importantly successful and timely recovery is the top priority for a DBA. For more details and usage samples, please refer MEB Manual.

Thursday May 22, 2014

MySQL Enterprise Backup Improved Compression Algorithm for 3.10


Prior to version 3.10, MySQL Enterprise Backup (MEB) used zlib compression for in-memory compression of datafiles. The compression worked by splitting the innodb datafiles into fixed size blocks and compressing each block independently.After searching on the web we found there are many compression algorithms available which can be used for compression. This triggered the idea of testing the performance of available compression algorithms. If the benchmark shows improved performance we can make backup and/or restore faster by adding the new compression algorithm to MEB.

Implementation :

The idea to implement the algorithms procceded as follows .

1. Select a "long list" of algorithms based on literature and what Google and other databases are using.
2. Create a prototype of MEB supporting the algorithms in the long list.
3. Run comparison tests of algorithms with the MEB prototype.
4. Select a "short list" of algorithms that will be added to MEB 3.10.

Criteria for Selecting the Algorithm:

The following criteria were used in comparing compression algorithms.

1. Compression speed
2. Decompression speed
3. Compression ratio
4. CPU-usage
5. Licensing model

These criteria have a differing importance. Compression speed, and compression ratio are probably more important to most users than decompression speed.

Performance Test:

We have now completed the performance tests of the new compression algorithms for MEB. See the below table for the list of compression algorithms were evaluated in the test.

Machine and OS Configurations:

OS : Oracle Linux 6 (x86_64)
Memory: 29 GB RAM
Cpu : 8 vCPUs (2 quad-core processors, no HT)
Read speed of the source dir(data directory) : 600 MB/s
Write speed of the destination dir(backup directory) : 300 MB/s

A backup of a 441 GB database was generated using TPC-H datagen tool taken when the mysqld process was not running .

Compression Algorithms Time [min] Compr. size [GB] Compr. / Orig. size Avg. CPU usage Avg. CPU Idle Reads [MB/s] Writes [MB/s] Source Disk busy
uncompressed/Normal Backup to Directory 31 N/A 100% 20% 65% 250 250 100%
Zlib (level=1) 34 165 37% 82% 15% 220 90 70%
Zlib (level=9) 720 120 27% - - - - -
LZF 27 222 50% 45% 50% 270 140 100%
LZO 27 224 51% 40% 55% 270 140 100%
Snappy 31 221 50% 55% 40% 260 130 80%
QuickLZ 26 203 46% 35% 55% 280 120 100%
LZ4 26 215 49% 35% 55% 280 130 100%
LZMA (level=1) 90 110 25% 78% 20% 80 22 25%
LZMA (level=9) 360 88 20% - - - - -

Few Important Notes:

•  Some columns are blanks because the test ran for longer duration of time so it was not feasible to collect monitoring stats.

• “Source Disk busy" is the number of I/O per second in percent of what the device can execute. It is not related to the device throughput (MB/s).

• MEB has an internal work queue to process data that is managed by separate read, process, and write threads. Read threads will place data in the process queue where processing threads then process it, and finally after the processing is complete, the data will be placed in the write queue where they will be written out to storage. Due to this design, if writes are slower than reads (which they often are), then the reads will effectively be throttled by the write speeds (write speeds typically being the limiting factor).

Analysis of the Compression Test's:

LZ4 and QuickLZ were the fastest algorithms, while ZLib (level=9) was by far the slowest. For compression ratios, LZMA (level=9) was only able to reach 20%, whereas QuickLZ reached 46%, and LZ4 49%. This illustrates the fact that there is a trade-off between backup speed and the reduction in data size. Nevertheless, we could say that algorithm A is better than algorithm B, if A is faster than B and produces a backup which is not larger than that of B, or if A produces a smaller backup than B and A is not slower than B. Using this criteria we can say that QuickLZ is a better compression algorithm than LZ4, Snappy, LZO, or LZF. Similarly, LZMA (level=1) is superior to Zlib (level=9).                                                                         The summary table shows two limiting factors for the backup speed. The IO speed of the of disk on which the database resides (the source disk) is thelimiting factor for uncompressed backup and compressed backups made with LZF, LZO,QuickLZ and LZ4. For Zlib (level=1), Snappy and LZMA (level=1) the limiting factor is the CPU. After removing the worst performing algorithms, we have four remaining that we can organize into a line where you get higher speeds as you move to the left, and better compression as you move to the right.

BEST SPEED --- QuickLZ --LZ4------ Zlib (lev.=1) ---- LZMA (lev.=1)---- LZMA(lev=9) --- BEST COMPRESSION

Restore Speed:

The restore speed was almost the same for all the algorithms. The restore of uncompressed backup and ZLib compressed backup took 28 minutes, and for all the other algorithms the restore time was 29 minutes.


For licensing reasons QuickLZ cannot be used with MEB. Therefore it was replaced with LZ4. Thus, the new compression algorithms are LZ4 (for fast compression) and LZMA (for high compression ratio).

Thursday Apr 03, 2014

Data Encryption with MySQL Enterprise Backup 3.10


MySQL Enterprise Backup (MEB) 3.10 introduces support for encrypted backups by allowing backup images, or single-file backups, to be encrypted. However, backups stored in multiple files in a backup directory can not be encrypted.

Any MEB command that produces a backup image can be optionally requested to encrypt it. The encrypted backup image can be stored in a file or tape in the same way as an unencrypted backup image. Similarly, any MEB command that reads data from a backup image accepts also an encrypted backup image. This means that encrypted backups can be used in all the same situations as unencrypted backup images.

MEB encrypts data with Advanced Encryption Standard (AES) algorithm in CBC mode with 256-bit keys. AES is a symmetric block cipher which means that the same key is used both for encryption and decryption. The AES cipher has been adopted by the U.S. government and it is now used worldwide.

A new format for the encrypted backup image is introduced. This is a proprietary format developed by Oracle and it allows efficient encryption and decryption in parallel.

Encryption keys

Encryption keys are strings of 256 bits (or 32 bytes) that are represented by strings of 64 hexadecimal digits. The simplest way to create an encryption key for MEB is to type 64 randomly chosen hexadecimal digits and save them in a file. Another method is to use some shell tool to generate a string of random bytes and encode it as hexadecimal digits. For example, one could use the OpenSSL shell command to generate a key as follows:

$ openssl rand 32 -hex

This command uses random data generated on the host for creating the key. Whichever method is used for the creation of the key, the essential point is that the resulting key consists of random bits.

The security of MEB encryption is based on two rules that apply not only to MEB but to all encryption schemes using symmetric block ciphers:

Rule 1: The encryption keys must be random.

Rule 2: The encryption keys must remain secret at all times.

When these rules are followed, it is very difficult for unauthorized persons to get access to the secure data.

Encryption keys can be specified either on the command-line with the

option where KEY is a string of 64 hexadecimal digits, or in a file with the

option where FILENAME is the name of the file that contains a string of 64 hexadecimal digits.

It is important to notice that specifying the key on the command-line with the --key option is generally not secure because the command-line is usually visible to other users on the system and it may even be saved in system log files that may be accessible by unauthorized persons. Therefore, the --key-file option should be preferred over the --key option in all production environments, and the use of the --key option should be limited to testing and software development environments.

Using encryption

Encryption is very simple to use. Any MEB command that produces a backup image can be requested to encrypt it by specifying the --encrypt option with either --key or --key-file option. The following example shows how to make a compressed backup and store it as an encrypted backup image.

$ mysqlbackup --encrypt --key-file=/backups/key --compress --backup-dir=/full-backup  --backup-image=/backups/image.enc  backup-to-image

MySQL Enterprise Backup version 3.10.0 Linux-3.2.0-58-generic-i686 [2014/03/04]

Copyright (c) 2003, 2014, Oracle and/or its affiliates. All Rights Reserved.

 mysqlbackup: INFO: Starting with following command line ...

 /home/pekka/bzr/meb-3.10/src/build/mysqlbackup --encrypt

        --key-file=/backups/key --compress --backup-dir=/full-backup

        --backup-image=/backups/image.enc backup-to-image

 mysqlbackup: INFO:

IMPORTANT: Please check that mysqlbackup run completes successfully.

           At the end of a successful 'backup-to-image' run mysqlbackup

           prints "mysqlbackup completed OK!".

140306 21:40:33 mysqlbackup: INFO: MEB logfile created at /full-backup/meta/MEB_2014-03-06.21-40-33_compress_img_backup.log

 mysqlbackup: WARNING: innodb_checksum_algorithm could not be obtained from config or server variable and so mysqlbackup uses the default checksum algorithm 'innodb'.


                       Server Repository Options:





Backup Image Path = /backups/image.enc

 mysqlbackup: INFO: Unique generated backup id for this is 13941348344547471

 mysqlbackup: INFO: Uses LZ4 r109 for data compression.

 mysqlbackup: INFO: Creating 18 buffers each of size 16794070.

140306 21:40:36 mysqlbackup: INFO: Compress Image Backup operation starts with following threads

        1 read-threads    6 process-threads    1 write-threads

140306 21:40:36 mysqlbackup: INFO: System tablespace file format is Barracuda.

140306 21:40:36 mysqlbackup: INFO: Starting to copy all innodb files...

 mysqlbackup: INFO: Copying meta file /full-backup/backup-my.cnf.

 mysqlbackup: INFO: Copying meta file /full-backup/meta/backup_create.xml.

140306 21:40:36 mysqlbackup: INFO: Copying /sqldata/simple-5.6/ibdata1 (Barracuda file format).

140306 21:40:36 mysqlbackup: INFO: Found checkpoint at lsn 188642964.




140306 21:40:51 mysqlbackup: INFO: Compress Image Backup operation completed successfully.

 mysqlbackup: INFO: Image Path = /backups/image.enc


   Parameters Summary         


   Start LSN                  : 188642816

   End LSN                    : 188642964


mysqlbackup completed OK! with 2 warnings

This resulting encrypted backup image (file "image.enc") can be used with all commands that accept a backup image in the same way as an unencrypted backup image. For example, one could restore the server from the encrypted backup as follows:

$ mysqlbackup --decrypt --key-file=/backups/key --uncompress --backup-image=/backups/image.enc --backup-dir=/full-backup copy-back-and-apply-log

MySQL Enterprise Backup version 3.10.0 Linux-3.2.0-58-generic-i686 [2014/03/04]

Copyright (c) 2003, 2014, Oracle and/or its affiliates. All Rights Reserved.

 mysqlbackup: INFO: Starting with following command line ...

 /home/pekka/bzr/meb-3.10/src/build/mysqlbackup --decrypt

        --key-file=/backups/key --uncompress --backup-image=/backups/image.enc

        --backup-dir=/full-backup copy-back-and-apply-log

 mysqlbackup: INFO:

IMPORTANT: Please check that mysqlbackup run completes successfully.

           At the end of a successful 'copy-back-and-apply-log' run mysqlbackup

           prints "mysqlbackup completed OK!".

 mysqlbackup: INFO: Backup Image MEB version string: 3.10.0 [2014/03/04]

 mysqlbackup: INFO: The input backup image contains compressed backup.

140310 12:51:54 mysqlbackup: INFO: MEB logfile created at /full-backup/meta/MEB_2014-03-10.12-51-54_copy_back_cmprs_img_to_datadir.log



140310 12:52:14 mysqlbackup: INFO: We were able to parse ibbackup_logfile up to

          lsn 188642964.

140310 12:52:14 mysqlbackup: INFO: The first data file is '/home/pekka/sqldata/copyback-simple-5.6/ibdata1'

          and the new created log files are at '/home/pekka/sqldata/copyback-simple-5.6'

140310 12:52:14 mysqlbackup: INFO: Apply-log operation completed successfully.

140310 12:52:14 mysqlbackup: INFO: Full Backup has been restored successfully.

mysqlbackup completed OK!

In these examples we have used the --key-file option for specifying the encryption key because it is more secure than giving the key on the command-line with the --key option.


This section describes two tips that may be useful when working with encrypted backups.

The "Wrong key" error

Encryption and decryption use the same key. If decryption is attempted with a key different from the encryption key, a wrong key error occurs. When this happens, MEB prints an error message like the one shown below.

MySQL Enterprise Backup version 3.10.0 Linux-3.2.0-58-generic-i686 [2014/03/04]

Copyright (c) 2003, 2014, Oracle and/or its affiliates. All Rights Reserved.

 mysqlbackup: INFO: Starting with following command line ...

        mysqlbackup --backup-image=/backups/image.enc --decrypt

        --key-file=/key-file2 list-image

 mysqlbackup: INFO:

IMPORTANT: Please check that mysqlbackup run completes successfully.

           At the end of a successful 'list-image' run mysqlbackup

           prints "mysqlbackup completed OK!".

 mysqlbackup: INFO: Creating 14 buffers each of size 16777216.

 mysqlbackup: ERROR: Failed to decrypt encrypted data in file /backups/image.enc : the file may be corrupted or a wrong encryption key was specified.

For the user, this can be problematic because two possible reasons for the failure are offered in the error message: either the backup is corrupted or a wrong key was supplied. This is not a bug or feature of MySQL Enterprise Backup but, instead, it is a theoretical limitation imposed by the encryption scheme. It is not possible even in theory to distinguish with absolute certainty between these two explanations when decryption fails.

However, these two explanations are not always equally likely. If decryption fails at the very start without decrypting any data, then it is more likely that a wrong key was supplied. On the other hand, if the decryption fails later after some data was successfully decrypted, then it is very likely that the correct key was given but the encrypted backup is broken. Using these two rules it is possible to determine with high probability the cases where decryption fails because of a wrong key.

Recognizing encrypted backups

On Unix-like operating systems "magic numbers" may be used for identifying the type of a file. Magic numbers are patterns in files that allow recognizing the type of a file by examining the first bytes in the file. Both the unencrypted backup images and encrypted backup images have magic numbers that can be used by shell tools to detect the file type. For example, by putting these lines to the /etc/magic file

0   string  MBackuP\n   MySQL Enterprise Backup backup image
0   string  MebEncR\n   MySQL Enterprise Backup encrypted backup

the file command detects the backups images as follows:

$ file /backups/image1 /backups/image2
/backups/image1: MySQL Enterprise Backup backup image
/backups/image2: MySQL Enterprise Backup encrypted backup

Monday Nov 25, 2013

MEB integration with Workbench

This blog talks about MySQL Enterprise Backup integration with Workbench and how the Workbench UI can be used to configure and operate MEB.[Read More]

Sunday Sep 29, 2013

Backing up full server instance using MySQL Enterprise Backup


MySQL Enterprise Backup(MEB) takes fast, consistent backups of MySQL server data, and helps in restoring the server to source server's data at the time of backup. But most of the times it is as much important to have same source server's state(server configuration like server global variables, plugins), as data. As backups become more frequent, server variables modified, plugins added or removed, it is very difficult to keep track of this changing server states for every backup. MEB 3.9.0 helps user in providing a complete backup so that the restored server can run with exactly the same state as that of the source server at the time of the backup.

MEB 3.9.0 performs full server instance backup,  which, on top of the log files and data files, also includes all the global variables and plugins(both internal and external) details. With this feature, backup-content.xml, a meta file under "meta" folder of the backup directory , now additionally contains all the plugins details like name, status, type etc under <plugins> section. In addition, there are two new files created under backup directory

  • server-my.cnf - contains all the global variables with non-default values for that server environment(MySQL server version, Operating System, Hardware Architecture etc).
  • server-all.cnf - contains all the global variables, that includes all the variables with non-default values and the other global variables with default values.

Advantages of Full Server Instance Backup:

  • Create replica - User can clone source server state by using either server-my.cnf or server-all.cnf file in the backup as defaults-file for starting the target server. As most of the global variables default value depend on the server environment, user can create a server with same state even if the target environment is different from source by using server-all.cnf as defaults file. If the target environment is same as source, user can use either of the files.
  • Keep a history of global variables - This new feature reduces the user's task for storing the state of the server, if changed, before every backup or incremental backup, so user need not worry about keeping track of server global variables. With this feature, now we can figure out non-default values of global variables for the running server with server-my.cnf.
  • Full plugins information - With all the plugins information backed up, this information will be used while installing missing plugins on restored server. Plugin details like type, status, library can be used to install missing plugins with the same configuration of the source.

Using Full Server Instance Backup:


From MEB 3.9 onwards, this feature is enabled by default for all kinds of backup(i.e normal, incremental, image, compressed etc). That is there is no need to turn on any feature or use any option and all backups are full server instance backups.

Note: Binary logs and *info files used for replication and Innodb buffer pool details, which are also part of server instance, are not included in backups. Server plugins details are copied, but actual plugins binaries are not copied.


After copy-back operation, server-all.cnf, server-my.cnf will be present in restored data directory. If there were any external plugins exist in source server, copy-back operation throws a warning about the missing plugins to install.

Starting Server:

The files server-my.cnf, server-all.cnf, which could be used as defaults-file to start the server on the restored data directory. When source and target environments are same, restoring server using server-my.cnf will be easier than using server-all.cnf as server-my.cnf has fewer global variables to verify or modify.

Note: User has to be careful while starting another server instance on the same host using server-my.cnf, server-all.cnf files without changes. There is a possibility of modifying source server settings or data, as some file paths like innodb_log_group_home_dir, innodb_log_group_home_dir,tmpdir,general-log etc are related to source server.

Incremental Backup:

The files server-my.cnf,server-all.cnf reflect the state of the server at the time of a incremental backup and it is desirable to have the same state after applying the incremental backup. So after apply-incremental-backup operation, full backup's server-my.cnf and server-all.cnf will be overwritten by corresponding incremental backup's server-my.cnf and server-all.cnf files.

Wednesday Jun 26, 2013

MySQL Enterprise Backup 3.8.2 has been released!

MySQL Enterprise Backup v3.8.2, a maintenance release of online MySQL backup tool, is now available for download from My Oracle Support  (MOS) website as our latest GA release.  It will also be available via the Oracle Software Delivery Cloud in approximately 1-2 weeks. A brief summary of the changes in MySQL Enterprise Backup version 3.8.2 is given below.

  A. Functionality Added or Changed: 

  • MySQL Enterprise Backup has a new --on-disk-full command line option. mysqlbackup could hang when the disk became full, rather than detecting the low space condition. mysqlbackup now monitors disk space when running backup commands, and users can now specify the action to take at a disk-full condition with the --on-disk-full option. For more details, refer this page
  • MySQL Enterprise Backup has a new progress report feature, which periodically outputs short progress indicators on its  operations to user-selected destinations (for example, stdout, stderr, a file, or other choices). For more details on progress report options, refer here

  B. Bugs Fixed:

  • When --innodb-file-per-table=ON, if a table was renamed and backup-to-image was in progress, apply-log would fail when being run on the backup. (Bug #16903973) 
  •  MySQL Server failed to start after a backup was restored if  there had been online DDL transactions on partitioned tables during the time of backup. (Bug #16924499) 
  • apply-incremental-backup might fail with an assertion error if  the InnoDB tables being backed up were created in Barracuda format and with their KEY_BLOCK_SIZE  values  different from the innodb_page_size . This fix ensures that different KEY_BLOCK_SIZE  values are handled properly during incremental backup and apply-incremental-backup operations. 
  • If a table was renamed following a full backup, a subsequent incremental backup could copy the .frm file with the new name, but not the associated .ibd file with the new name. After a  restore, the InnoDB data dictionary could be in an  inconsistent state. This issue primarily occurred if the table  was not changed between the full backup and the subsequent  incremental backup. Bug #16262690)
  •  After a full backup, if a table was renamed and modified,  apply-incremental-backup would crash when run on the backup directory. (Bug #16262609)
  • The value of the binary log position in backup_variables.txt  could be different from the output displayed during the   backup-and-apply-log operation. (This issue did not occur if  the backup and apply-log steps were done separately.) (Bug  #16195529)
  • When using the --only-innodb-with-frm option, MySQL Enterprise Backup tried to create temporary files at unintended locations in the file system, which might cause a failure when, for example, the user had no write privilege for those locations.   This fix makes sure the paths for the temporary files are  correct. (Bug #14787324)
  •  A backup process might hang when it ran into an LSN mismatch between a data file  and the redo log. This fix makes sure the process does not hang and it displays an error message showing the  name of the problematic data file (Bug #14791645)

Please post your questions / comments about Backup in forums.


MEB Team

Monday Jun 17, 2013

MySQL Camp at Bangalore - 19 June

As promised we are back with the 2nd edition of Bangalore MySQL User camp this Wednesday 19 June.

As requested we are going into the details of InnoDB and Performance Schema.

See the details here:!topic/bangalore-mysql-user-camp/gnqGPLKDB0s

Venue remains the same as last time : 

Our Kalyani Magnum Office behind Apollo hospital on the Bannerghatta road

Time - 5PM-6PM

Don't miss the chance to meet the MySQL engineering team at Bangalore and learn internals of MySQL.

Hope to see you there.



Friday May 03, 2013

MEB : The journey so far 2010-2013

MySQL Enterprise Backup (MEB) was born 3 years ago as a newly branded avatar of InnoDB Hot backup. Wanted to share what has gone on so far, how we at Oracle think about backup, the milestones that we have achieved and the road ahead. The idea for this blog came to me after looking at Mikael's latest blog. While Mikael talks about MySQL, I want to talk about MEB.

When we started with InnoDB Hot backup the first challenge was to have it adhere to the development, quality and release processes for MySQL. This meant creating a quality plan, getting it into the development trees of MySQL and ensuring that each piece of new code went through architecture and code review. Though the initial implementer and architect of Hot backup continues to work with the MEB team, there were a host of new engineers to be trained. We also needed to ensure that the new (at that time) Barracuda InnoDB file format and incremental backup was supported. MEB 3.5.1 was the release which got these things along with the adherence to the development and quality model of MySQL.

The next challenge we faced was that of ensuring that MEB was on equal footing for both Linux and Windows. InnoDB hot Backup consisted of 2 programs - ibbackup and innobackup; innobackup is a Perl module. The main issue with using the program on Windows was the requirement to install Perl. With multiple Perl implementations and changing Perl versions, we did not want to check MEB compatibility for every implementation and new version of Perl when it was released. Even though the problem is the similar for Linux; the users of Linux are used to hacking around, changing paths and managing multiple versions of software like Perl. Windows users however expect things to just work. So we set about removing the Perl code altogether. This meant that the innobackup functionality had to be re-coded as a C program. Merging these 2 programs meant a major re-think on how the combined command line interface needed to look. The solution we came up with was to let the ibbackup and innobackup command line syntax remain as is, while the combined program had a similar but more logical "mysqlbackup" command line syntax. We were very happy with the new syntax because it freed us from history and MEB syntax became very much in line with the syntax of other MySQL clients. With the release of 3.6 we had a single C program, a more logical syntax , a product which was easier to install and worked exactly the same for all platforms.

We were getting to 2012 and database sizes were commonly approaching the 1 TB size. Such large databases meant the backup should ideally be streamed to tape. Interfacing with tape drives is a complicated and specialized activity. We neither had the bandwidth nor the expertize to handle tapes in MEB. The best solution was to adhere to a good common standard interface that was adopted by software which dealt with tapes. The interface we decided to support was Oracle's System Backup to Tape (SBT) . MEB was modified to be able to to stream the backup output to this interface. A common requirement for these interfaces is that they ideally want to deal with the backup as a single file. A single file can be streamed and restored by any software that speaks SBT. There is a whole ecosystem around SBT because it is the preferred way to backup the Oracle database. Changing MEB to think streaming instead of random access directory output was the challenge we overcame with the release of version 3.7 of MEB. With version 3.7, MEB could interface with Oracle Secure Backup, Symantec Net Backup and Tivoli Storage Manager and any other backup software that understood SBT.

After having resolved what we saw as the "basic" requirements for backup , our customers were demanding more performance and usability. We took up the challenge of performance for the 3.8 version of MEB. MEB was a monolithic single threaded program. We decided to internally break up MEB into 3 separate modules. The read phase, the process phase and the write phase. Each of these 3 phases could be multi-threaded. The number of threads dedicated for each phase were also made user configurable. All operations of backup including the "Applylog" and "copyback" were made multi-threaded. Read more details about this design approach and the performance gains in my blog - Truly Parallel backup. Meanwhile the new release of the MySQL Server 5.6 was also out. It was an interesting challenge to ensure that MEB understood the new MySQL 5.6 features and was able to take advantage of them. As on date MEB 3.8.1 is the only online backup solution that is compatible with the new features of MySQL 5.6.

Backup is like buying insurance. When all else fails you need to be sure that there is a working backup that is available to bring back your database. Backup is not something that can fail when it is needed. It is required that we are surefooted when dealing with such a critical activity. We take your trust in our solution very seriously. Thanks for being a part of the MEB journey (and for reading this blog) so far. The MySQL landscape is ever changing and we know that you desire more usability, performance and flexibility from MEB. We will try and ensure that we meet these expectations with the best possible quality. With every new MEB release you will see a more usable, flexible and performant MEB.

Thursday Mar 28, 2013

Report : Bangalore MySQL User Camp (BMUC) 22 Mar

We had a great kickoff to the Bangalore MySQL User camp on 22 March 2013 at the Oracle Kalyani Magnum Office.

There were 17 attendees from the MySQL community besides more than 15 MySQL engineers who attended the meet. Of these 3 community members were from groups inside Oracle. There was a mix of new and veteran MySQL users. The MySQL India teams gave a brief introduction about themselves, followed by a quick overdrew of MySQL and how it is different from other databases. The presentation about 5.6 features was presented last and the one that was most liked. There was a great networking over juice and Donuts after the meeting. There were lots of questions by the community on technical details of 5.6 features. Everyone interacted and there was help offered for ensuring that the next meeting on 19th June 2013 is an even greater success.

Thanks to all the attendees, and organizers. Hope to see you in person again in 3 months – on 19 June.

Meanwhile do keep writing and keep connected via linked-in(MySQL India) , Facebook (MySQL User Camp) and Google group(bangalore-mysql-user-camp).

Tuesday Mar 19, 2013

Bangalore MySQL User Camp - 22 Mar

We are planning the kickoff meeting of the Bangalore MySQL User Camp:
We would like to invite all MySQL enthusiasts at Bangalore to join.
Plan is to have at-least one meeting every quarter and give an opportunity to all to participate, attend and speak.

Agenda for 22 March Meeting:
1700-1715: Introductions
1715-1730 : Meet the MySQL Development team @ Bangalore
1730-1815 : Introduction to MySQL 5.6
1815-1830 : Q&A & Plan for the next meeting
1830-1900 : Tea and informal networking

Conf Rom 0C001,
Ground Floor, B Wing, Oracle,
Kalyani Magnum Software technology Park
1st Main, J P Nagar 7th Phase,
Bangalore 560076

You can also join the google group "bangalore-mysql-user-camp" where we will be announcing further meetings etc.

Friday Feb 08, 2013

Truly Parallel backup (MySQL Enterprise Backup 3.8 and later)

How do you implement a parallel algorithm for a software which needs to be streamed to tapes?
How do you ensure that you have the capability to be able to tune the level of parallelism for varying input and output devices and varying levels of load?
These were some of the questions that we needed to answer when we were trying to implement multi-threading capability for MySQL Enterprise Backup (MEB).
The trivial way of achieving parallelism is by having the multiple threads pick up the different files (in a file per table) scenario. But this did not seem adequate because:
a) The sizes of these files (corresponding to the tables) could be different and then one large file would limit the level of parallelism since it would be processed by a single thread.
b) If you have to stream the backup how do you reconcile these multiple files being streamed by separate threads? Large backups are streamed directly to tape so it is better to have a single file being output and not multiple files.
c) If you buffer each file and wait for a file to be completely processed and then push it to tape then it is not true streaming because you are using intermediate disk space to save the incomplete portions of all the files.
The answer that we found was to implement the parallel algorithm using a horizontal strategy instead of a vertical strategy.

In the vertical strategy, each thread acts on a separate file. This limits streaming since the file sizes can vary.
In the horizontal strategy, each file is broken into a sections (denoted by multiple colors). A separate thread is assigned to operate on a single section.
Parallel operations are then possible for reading , processing and writing of these file subsections because no two threads will be operating on the same section of the file.
This setup is especially useful when using compression since there can be multiple threads performing compression while the read and write continues in parallel.
There may be additional overhead of ensuring that the buffers are in the correct order when they are written out, but since most of the buffers of the same size and having similar operations being performed, the overhead is minimal.
You get truly serialized output that is streamed to tape as it gets processed. If you are streaming to a remote host or to tape, there is almost no additional space required on your main server. We call this new mechanism parallel backup because we are achieving parallelism thereby making the backup faster. Indeed, using parallel backup may see up to 10 times the speed of a normal backup in certain scenarios.
The graph below shows the time it took for backup for MEB 3.7.1 v/s MEB 3.8 using varying number of threads.

Note : This is a 16 GB, 2 x 2000 MHz, 2 RAID DISKS (1027 GB,733.9GB) machine running Oracle Linux.

As you can see above; MEB 3.8 provides options to configure the number of threads you use for reading, writing and processing. Lets denote RT, PT and WT as number of Read, Process and Write threads respectively. Default values for MEB 3.8 is RT=3,PT=3, WT=3 which is changing in MEB 3.8.1 to RT=1, PT=6, WT=1.

This is close to the fastest backup we get in the graph above. The reason for not choosing RT=1, PT=12, WT=1 (which is the fastest) is because the CPU gets very highly utilized in the 1,12,1 configuration.

Remember, the read write throughput depends on your input and output devices. It is possible that multiple threads do not give you a better performance for read or write v/s a single thread.

There are also options available to have a configurable number of buffers used by these threads.

Each buffer is of size 16MB. You should have at-least [RT+PT+WT+ (MAX(RT,PT,WT) ] number of buffers so that you get optimal parallelism.

For Example if RT=1, PT=6, WT=1 then you should configure 1+6+1+6 = 14 buffers (default in MEB 3.8.1)

If for example you configure multiple threads but configure only 1 buffer then your backup is not taking advantage of parallelism at all. The read thread reads into the single buffer, buffer is then processed, written and then freed. The read thread is waiting for a buffer to be free to read into it; so it is like a serial process.

One more thing to note is that the number of buffers is limited by the memory limit configured for backup (default 300MB). Please ensure that you configure enough memory to be able to distribute it to the buffers you have configured. If the memory limit configured is less then what is required for the configured number of buffers; MEB will automatically decrease the number of buffers to fit into the memory limit. Based on the default values, if you are configuring more than 18 buffers you will need to increase the memory limit.

Please look at the previous 3.8 blog for detailed configuration examples :

or into our documentation of this feature at


and remember the wise DBA advise:

If you don't verify your backups periodically it is like not having backups at all


MySQL MEB Team Blog


« November 2015