mysql优化 How to Optimising MYSQL

(这是从国外一个论坛中看到的一篇帖子,关于mysql优化,思路很不错)

I’d like to share with the interworx-cp community my knowledge about tuning / tweaking mysql.

First of all, sorry if my English is not perfect. Maybe some sentences may be difficult to understand. So do not hesitate to ask me for explanations :-p

Secondly, the default INTEWORX-CP mysql setup works very fine and is setup for all common usage.
DO NOT CHANGE your setting without understanding what you do.
Make a backup of your my.cnf before editing the /etc/my.cnf
DO THESE CHANGES AT YOUR OWN RISKS.

My thread is only to help you to more well understand how tuning mysql.

Optimising mysql is very well commented on the net, and you’ll find huge information on how to do this. There is never “best parameters”, the best parameters is those fits your needs, box hardware, mysql usage…
So, I’ll not give the best parameters but rather how to define these ones. Make some tests, and you’ll quickly find your own parameters.

I’ll give you at the end of this post some web pointers which may help you.

There a lot of available parameters but only few one are very important to tweak your mysql box.

The most important variables are (for me, and it is not exhaustive)

Quote:
- max_connections
- wait_timeout
- thread_cache_size
-
- table_cache
-
- key_buffer_size
- query_cache_size
- tmp_table_size

First of all, how to find your variable, and the mysql usage ?

*VARIABLES

Quote:
from mysql :
show variables;

or from command line :
mysqladmin variables

*PROCESS / STATUS

Quote:
from Mysql :
show status;

or from command line
mysqladmin –i10 processlist extended-status

*SOME USEFUL COMMAND FOR YOU BOX USAGE

Quote:
>Top

>ps –axfu

>vmstat 1

* OPTIMISING MYSQL

To obtain the stat of your mysql server since it has been loaded, run mysqladmin processlist extended-status as mentionned above.

1 – The two most important variables : Table_cache and Key_buffer_size

* If Opened_tables is big, then your table_cache variable is probably
too small.

table_cache 64
Open_tables 64
Opened_tables 544468

This is the first serious problem. “The table_cache is the number of open
tables for all threads. MySQL, being multi-threaded, may be running many
queries on the table at one time, and each of these will open a table.”
Therefore, even though we only have a few tables, we will need many more
open_tables.

The Opened_tables value is high and shows the number of
cache misses. Getting the table_cache size correct is one of the two best
things you can do to improve performance.

* If Key_reads is big, then your key_buffer_size variable is probably
too small. The cache hit rate can be calculated with
Key_reads/Key_read_requests.

key_buffer_size 16M
Key_read_requests 2973620399
Key_reads 8490571
(cache hit rate = 0.0028)

“The key_buffer_size affects the size of the index buffers and the speed
of index handling, particularly reading.” The MySQL manual (and other sources) say that
“Key_reads/Key_read_request ratio should normally be < 0.01.” This is the
other most important thing to get correct. Here the value seems to be correct (< 0.01)

Also check key_write_requests and key_writes.
The key_writes/key_writes_request should normally be < 1 (near 0.5 seems to be fine)

Here is a very interesting web pointer : http://www.databasejournal.com/featu…0897_1402311_3

2 – Others important settings are : Wait_timeout, max_connexion, thread_cache

A little explanation :
Generaly you have a lot of mysql process that are sleeping because wait_timeout are not set low. So I make sure that the wait_timeout is set to a very low value: 15 seconds (for me) . That means MySQL would close any connection that was idle for more than 15 seconds.

The problem is you also have to increment your max_connexion (mine is set to 300) to be sure there is not a lot of idle clients holding connections and blocking out new clients from connecting and getting real work done.
The pbm is that the box has to create new threads (MySQL is a multi-threaded server) at a very high rate. That may sucks up a measurable amount of CPU time.

So the solution is to use the Thread_cache (from mysql doc) :
“How many threads we should keep in a cache for reuse. When a client disconnects, the client’s threads are put in the cache if there aren’t more than thread_cache_size threads from before. All new threads are first taken from the cache, and only when the cache is empty is a new thread created. This variable can be increased to improve performance if you have a lot of new connections. (Normally this doesn’t give a notable performance improvement if you have a good thread implementation.) By examing the difference between the Connections and Threads_created you can see how efficient the current thread cache is for you.”

* If Threads_created is big, you may want to increase the
thread_cache_size variable. The cache hit rate can be calculated with
Threads_created/Connections.

thread_cache_size 0
Threads_created 150022
Connections 150023

This is the second problem that should be fixed. A cache size of zero is the default for my-medium.cnf but the recommended size in my-large.cnf is 8.

you may try this formula : table_cache = opened table / max_used_connection

3 – Finally, you may also have a look at : tmp_table_size and Handler_read_rnd / Handler_read_rnd_next

* If Created_tmp_disk_tables is big, you may want to increase the
tmp_table_size variable to get the temporary tables memory-based instead
of disk based.

tmp_table_size 32M
Created_tmp_disk_tables 3227
Created_tmp_tables 159832
Created_tmp_files 4444

Created_tmp_disk_tables are the “number of implicit temporary tables on
disk created while executing statements” and Created_tmp_tables are
memory-based. Obviously it is bad if you have to go to disk instead of
memory. About 2% of temp tables go to disk, which doesn’t seem too bad
but increasing the tmp_table_size probably couldn’t hurt either.

* If Handler_read_rnd is big, then you probably have a lot of queries
that require MySQL to scan whole tables or you have joins that don’t use
keys properly.

Handler_read_rnd 27712353
Handler_read_rnd_next 283536234

These values are high, that we could probably stand to improve
the indexes and queries.

I hope this will help some of you to more understand how it is possible to optimise MYSQL to fit your needs, hardaware box, or mysql current usage.

Maybe there is others tweaks to perform, but I know well only these ones. I did setup using these ones on differents mysql box, and generally it did help us to increase performance without have to change hardware (our boxes have 2GB ram)

Pascal
Used MySQL memory = key_buffer + max_connections * (join_buffer + record_buffer + sort_buffer + thread_stack + tmp_table_size)

Notice the max_connexion and the multiplier.
connexion increase = memory usage increase too.

Notice key_buffer
for a given memory :
more you add mem to key buffer, less connexion is
less is key buffer, more connexion is

If you change one of these settings for a high value, you system may swap.
If you system swap, try lot decrease these values

Also, about table_cache :
Increasing the size of the table cache may really help you.
But you must be careful not to make the value too large. All operating systems have a limit on the number “open file pointer” (sorry in french it is called pointer, maybe descriptors is the good translation) a single process may have.
If MySQL tries to open a lot of files, the OS may refuse it and MySQL will generate error message in the error log.

Pascal

####

Your runtime values don’t point necessarily to a bad mysql server configuration, but instead to poor query design. These values:

Handler_read_rnd_next 769312987
Key_read_requests 94470278
Select_scan 128879
Created_tmp_tables 877

Indicate that you have several queries that either:
- are not using indexes
- are performing full table scans to retrieve values
- are not performing lossless joins

You need to schedule a nice evening with your SQL console, your SQL queries and run EXPLAIN to discover which are creating these temp tables, and where you could use indexes to prevent your mysql server from scanning entire tables to locate single potential keys.

Use the mysql slow query log, and the log queries without indexes directives and then examine your logs to find the culprit.

There’s a saying in our field, tweaking servers can improve your performance by ten, tweaking the queries can improve it by ten thousand.

You might want to add skip-locking to your mysqld section, that’s a huge lock value for what I assume are myisam tables (which perform table locking). If you’re using InnoDB and have set it to a finer grained lock, you can ignore this last statement. If you are not using InnoDB, turn it off.

Disk access also plays a great part, make sure you have hdparm set to have DMA access on drives that can support it. Further, if something is chewing up your filesystem time, ie. scripts that have bad code that loop file locks, this could be consuming your file descriptors. Tweaking your filesystem is an entirely different ballgame – but should be considered. Lots of good tutorials available on the net.

分享到QQ空间

Comments are closed.