Loadays MySQL

MySQL : maintaining
(too) big tables
Frédéric Descamps

Frédéric Descamps

● lefred on IRC and Twitter
● Senior Linux and Open Source Consultant
@inuits.be
● Certified MySQL DBA since 2007
● First GNU/Linux distro (as far as I remember):
Slackware 3.0 (kernel 1.2.13)
● Blog: http://www.lefred.be

How come we need to
maintain tables ?
● A bit of history first:

once upon a time, there
was a developer who found
a great idea

Then he decided to create a proof of concept
and he coded, coded, coded a lot...

During the period he
unfortunately didn't discuss
with the DBA's and the OPS

But he discussed with
many other people like
marketing manager, sales
manager, regional manager,
product manager and all
were very happy and asked
to code even more !

Then he coded, coded,
coded a lot...

Why ?
● Impossible to stop de database, it's too
important !
● There is too much data, tables are huge !!
• Tables > 400GB with 500M rows
• Primary keys are varchar(32)
● Impossible to optimize, add an index to a table
(>4h needed to add an index)
● How do we alter the schema ?
● How do we migrate to another version of
MySQL ?

How big ?
# ls lh dayview.ibd
rwrw 1 mysql mysql 29G Feb 3 22:20 dayview.ibd

Two very important points to never forget as
developer :
● Don't underestimate the complexity of production
environments

Don't !
● Never say: “but on my
laptop the query is
very fast”

What now ?
● The plan is to create a new server more
manageable with minimal downtime of the
production
• Reduce size if possible
• Make queries faster
● The server will be a replica of the current
production server with some different settings

● Restore last backup

Ze Plan ● Prepare the new
database
● Dump & restore
data using maatkit

databasezilla
databasezilla ● Restart the
restore replication
● Replace the
production db by
the new one (only
downtime here)

Database size
● Deleting historical data sometimes doesn't help
really to save some space... ibdata won't shrink
● Data could be fragmented
• To avoid the use of that huge ibdata we add
this to the configuration of the new server:
innodb_file_per_table

MySQL version
● Maybe it is time to also change the MySQL
version to a newer one and even to change
engine !
● I advice you to try and use XtraDB from
percona. This includes corrections and
interresting patches (like those from Google)
• The list is available here :
http://www.percona.com/docs/wiki/percona-
xtradb:patch:start

Anything else to
change ?
● Maybe other settings aren't optimal for your
production environment. To have an overview
of them, I'd recommand the use of these two
tools :
• MySQLTuner
• tuning-primer

Example 1
-------- General Statistics --------------------------------------------------
[--] Skipped version check for MySQLTuner script
[OK] Currently running supported MySQL version 5.1.37-community-log
[OK] Operating on 64-bit architecture

-------- Storage Engine Statistics -------------------------------------------
[--] Status: -Archive -BDB -Federated +InnoDB -ISAM -NDBCluster
[--] Data in InnoDB tables: 899G (Tables: 149)
[!!] Total fragmented tables: 82

-------- Performance Metrics -------------------------------------------------
[--] Up for: 23h 46m 1s (112M q [1K qps], 70K conn, TX: 44B, RX: 22B)
[--] Reads / Writes: 51% / 49%
[--] Total buffers: 12.4G global + 152.2M per thread (100 max threads)
[!!] Maximum possible memory usage: 27.3G (173% of installed RAM)
[OK] Slow queries: 0% (963/112M)
[OK] Highest usage of available connections: 52% (52/100)
[OK] Key buffer size / total MyISAM indexes: 256.0M/93.0K
[OK] Key buffer hit rate: 100.0% (215K cached / 3 reads)
[OK] Query cache efficiency: 43.3% (23M cached / 54M selects)
[!!] Query cache prunes per day: 3049150
[OK] Sorts requiring temporary tables: 0% (0 temp sorts / 754K sorts)
[!!] Joins performed without indexes: 19547
[OK] Temporary tables created on disk: 0% (1K on disk / 545K total)
[OK] Thread cache hit rate: 99% (53 created / 70K connections)
[OK] Table cache hit rate: 99% (802 open / 809 opened)
[OK] Open file limit used: 1% (59/4K)
[OK] Table locks acquired immediately: 100% (72M immediate / 72M locks)
[!!] InnoDB data size / buffer pool: 899.3G/12.0G

Example 2
MySQL Version 5.1.37-community-log x86_64

Uptime = 1 days 0 hrs 2 min 31 sec
Avg. qps = 1324
Total Questions = 114605776
Threads Connected = 47

SLOW QUERIES
The slow query log is enabled.
Current long_query_time = 10.000000 sec.
You have 972 out of 114605948 that take longer than 10.000000 sec. to complete
Your long_query_time seems to be fine

BINARY UPDATE LOG
The binary update log is enabled
The expire_logs_days is not set.
The mysqld will retain the entire binary log until RESET MASTER or PURGE MASTER LOGS commands are run
manually
Setting expire_logs_days will allow you to remove old binary logs automatically
See http://dev.mysql.com/doc/refman/5.1/en/purge-master-logs.html
Binlog sync is not enabled, you could loose binlog records during a server crash

WORKER THREADS
Current thread_cache_size = 8
Current threads_cached = 4
Current threads_per_sec = 0
Historic threads_per_sec = 0
Your thread_cache_size is fine

Example 2 (b)
MAX CONNECTIONS
Current max_connections = 100
Current threads_connected = 47
Historic max_used_connections = 52
The number of used connections is 52% of the configured maximum.
Your max_connections variable seems to be fine.

INNODB STATUS
Current InnoDB index space = 0 bytes
Current InnoDB data space = 0 bytes
Current InnoDB buffer pool free = 0 %
Current innodb_buffer_pool_size = 12.00 G
Depending on how much space your innodb indexes take up it may be safe
to increase this value to up to 2 / 3 of total system memory

MEMORY USAGE
Max Memory Ever Allocated : 20.06 G
Configured Max Per-thread Buffers : 14.86 G
Configured Max Global Buffers : 12.33 G
Configured Max Memory Limit : 27.20 G
Physical Memory : 15.67 G

nMax memory limit exceeds 90% of physical memory

Example 2 (c)
KEY BUFFER
Current MyISAM index space = 8 K
Current key_buffer_size = 256 M
Key cache miss rate is 1 : 74223
Key buffer free ratio = 81 %
Your key_buffer_size seems to be too high.
Perhaps you can use these resources elsewhere

QUERY CACHE
Query cache is enabled
Current query_cache_size = 64 M
Current query_cache_used = 32 M
Current query_cache_limit = 2 M
Current Query cache Memory fill ratio = 51.30 %
Current query_cache_min_res_unit = 4 K
MySQL won't cache query results that are larger than query_cache_limit in size

SORT OPERATIONS
Current sort_buffer_size = 64 M
Current read_rnd_buffer_size = 16 M
Sort buffer seems to be fine

JOINS
Current join_buffer_size = 64.00 M
You have had 20057 queries where a join could not use an index properly
join_buffer_size >= 4 M
This is not advised
You should enable "log-queries-not-using-indexes"
Then look for non indexed joins in the slow query log.

Example 2 (d)
OPEN FILES LIMIT
Current open_files_limit = 4096 files
The open_files_limit should typically be set to at least 2x-3x
that of table_cache if you have heavy MyISAM usage.
Your open_files_limit value seems to be fine

TABLE CACHE
Current table_open_cache = 1024 tables
Current table_definition_cache = 256 tables
You have a total of 0 tables
You have 843 open tables.
The table_cache value seems to be fine

TEMP TABLES
Current max_heap_table_size = 64 M
Current tmp_table_size = 256 M
Of 554783 temp tables, 0% were created on disk
Effective in-memory tmp_table_size is limited to max_heap_table_size.
Created disk tmp tables ratio seems fine

TABLE SCANS
Current read_buffer_size = 8 M
Current table scan ratio = 162 : 1
read_buffer_size seems to be fine

TABLE LOCKING
Current Lock Wait ratio = 0 : 114608692
Your table locking seems to be fine

Filesystem
● xfs is much faster to store the binlogs and to
delete them

Start the replication
● Change to the right binlog and the right
position before the backup

mysql> show slave statusG
*************************** 1. row ***************************
               Slave_IO_State: Waiting for master to send event
                  Master_Host: xxxxx
                  Master_User: replica
                  Master_Port: 3306
                Connect_Retry: 10
              Master_Log_File: xxxx.000052
          Read_Master_Log_Pos: 426644104
               Relay_Log_File: mysqld1relaybin.000173
                Relay_Log_Pos: 426644245
        Relay_Master_Log_File: xxxx.000052
mysql> CHANGE MASTER TO MASTER_HOST='xxxxx',
             Slave_IO_Running: Yes MASTER_USER='replica', MASTER_PASSWORD='xxxx',
            Slave_SQL_Running: Yes MASTER_PORT=3306, MASTER_LOG_FILE='xxx.000052',
              Replicate_Do_DB: MASTER_LOG_POS=426644104,
          Replicate_Ignore_DB: MASTER_CONNECT_RETRY=10;
           Replicate_Do_Table:
       Replicate_Ignore_Table:
      Replicate_Wild_Do_Table:
  Replicate_Wild_Ignore_Table:
                   Last_Errno: 0
                   Last_Error:
                 Skip_Counter: 0
          Exec_Master_Log_Pos: 426644104
              Relay_Log_Space: 426644441

Monitor your production
● While you will do any work on the database, I
recommand you to use innotop, you will see
directly if something is going wrong
● Use cacti to see obviously strange behavior

Restore the backup
● I won't explain you how to restore a backup,
but on huge tables usually you don't have a
dump of the data but only the files and the
binlogs
● So restore the last production backup you have
on a new server (most of the time we use flash
copies)

Dump & restore the data
● Use maatkit parallel to dump and restore the
data
● While this operation, be careful to not delete
the binlogs on te master that you will need to
recover !

Remove historical data -
(scale back)
● Now that you have a more performant
database (as the indexes are rewritten) it's
time to archive the old data
● As the tables are still huges I don't recommand
to do a massive delete of many rows matching
a query
● Use (again) maatkit to archive

ptxArchiver
● I wrote a small script to use mk-archiver that
follows a defined strucure to archive all the
records linked together (foreignkeys)
$ ptxArchiver.pl t message w "timestamp <
unix_timestamp(date_sub(now(),interval 17
MONTH))*1000"

<tables>
   <message>
      <message>
        field=id
        p_field=origin
      </message>
   </message>
</tables>

Keep the result of these
commands regularly:

iostat dx 5 5
vmstat 1 10
mpstat 1 10
top b n 1
cat /proc/meminfo

Analyse your MySQL
Check regurarly the output of these statements:
● SHOW ENGINE INNODB STATUS;
• Returns information about semaphores, deadlocks, foreign
keys, transactions, buffer pool, I/O, ...
• This info can be automaticaly stored in the mysql log file every
15secs by creating the following table in mysql schema:

CREATE TABLE innodb_monitor (a INT) ENGINE=INNODB;

● SHOW GLOBAL STATUS;
● SHOW TABLE STATUS;

Analyse your MySQL
● Use mk-query-digest on slow query log, binlogs
or processlist
• mk-query-digest --processlist 127.0.0.1,
● Don't forget innotop
● mysqlresources
● mysqlsla
● mysqlidxchk
● mysqlreport

AnalyZe your tables
● ANALYZE TABLE analyzes and stores the key
distribution for a table.
● For InnoDB the table is locked with a write lock.
● This statement is much faster then OPTIMIZE
TABLE.
● For InnoDB the statistics are updated everytime
that 1/16 of the data has been changed. For
huge tables it is then needed to update those
statistics manually more often.

Partitioning
● MySQL since 5.1 is able to manage horizontal
partitioning (records level). Since 5.5 you can
partition a table using multiple columns in the
range.
● Of course to partition the huge tables a
relevant column should be present in it (which
is not the case when the records are refering to
varchar(32) :'-(

Links
● MySQLTuner - http://blog.mysqltuner.com/
● tuning-primer - http://www.day32.com/MySQL/
● ptxArchiver - http://www.lefred.be/?q=node/105
● maatkit - http://www.maatkit.org
● mysqlreport, mysqlsla, mysqlidxchx - http://hackmysql.com
● mysqlresources - http://datacharmer.org/downloads/mysqlresources.zip
● innotop -
http://www.xaprb.com/blog/2006/07/02/innotop-mysql-innodb-monitor/

Loadays MySQL

Recommended

More Related Content

What's hot (20)

Viewers also liked (6)

Similar to Loadays MySQL (20)

Recently uploaded (20)

Loadays MySQL