MySQL Query Cache
MySQL has a great feature called “Query Cache” which is quite helpful for MySQL Performance optimization tasks but there are number of things you need to know.
First let me clarify what MySQL Query Cache is - I’ve seen number of people being confused, thinking MySQL Query Cache is the same as Oracle Query Cache - meaning cache where execution plans are cached. MySQL Query Cache is not. It does not cache the plan but full result sets. This means it is much more efficient as query which required processing millions of rows now can be instantly summoned from query cache. It also means query has to be exactly the same and deterministic, so hit rate would generally be less. In any case it is completely different.
Query cache is great for certain applications, typically simple applications deployed on limited scale or applications dealing with small data sets. For example I’m using Query Cache on server which runs this blog. Updates are rather rare so per-table granularity is not the problem, I have only one server and number of queries is small so cache duplication is not the problem. Finally I do not want to hack wordpress to support eaccelerator cache or memcached. Well honestly speaking if performance would be problem I should have started with full page caching rather than MySQL level caching but it is other story.
Lets talk a bit about features and limitations of Query Cache:
Transparent Caching - Caching is fully transparent to the application, and what is very important it does not change semantics of the queries - you always get actual query results. Really there are some chevats - if you’re not using query_cache_wlock_invalidate=ON locking table for write would not invalidate query cache so you can get results even
if table is locked and is being prepared to be updated. So if you’re using query cache in default configuration you can’t assume locking table for write will mean no one will be able to read it - results still can come from query cache unless you enable query_cache_wlock_invalidate=ON.
Caching full queries only - Meaning it does not work for subselects, inline views, parts of the UNION. This is also common missunderstanding.
Works on packet level - This is one of the reason for previous item. Query cache catches network packets as they sent from client to the server, which means it can serve responses very fast doing no extra conversion or processing.
Works before parsing - One more reason for high performance is Query Cache performs query lookup in the cache before query parsing, so if result is served from query cache, query parsing step is saved.
Queries must be absolutely the same As no parsing is done before lookup queries are not normalized (would require parsing) before cache lookup, so they have to match byte by byte for cache hit to happen. This means if you would place dynamic comments in the query, have extra space or use different case - these would be different queries for query cache.
Only SELECT queries are cached SHOW commands or stored procedure calls are not, even if stored procedure would simply preform select to retrieve data from table.
Avoid comment (and space) in the start of the query - Query Cache does simple optimization to check if query can be cached. As I mentioned only SELECT queries are cached - so it looks at first letter of the query and if it is “S” it proceeds with query lookup in cache if not - skips it.
Does not support prepared statements and cursors Query Cache works with query text and want full result set at once. In prepared statements there is query with placeholders and additional parameter values which would need extra care - it is not implemented. Cursors get data in chunks so it is even harder to implement.
Might not work with transactions - Different transactions may see different states of the database, depending on the updates they have performed and even depending on snapshot they are working on. If you’re using statements outside of transaction you have best chance for them to be cached.
Query must be deterministic - Query might provide same result no matter how many times it is run, if data remains the same. So if query works with current data, uses non-deterministic functions such as UUID(), RAND(), CONNECTION_ID() etc it will not be cached.
Table level granularity in invalidation - If table gets modification all queries derived from this table are invalidated at once. Most of them quite likely would not have change their result set but MySQL has no way to identify which one of them would so it gets rid of all of them. This is one of the main features which limits query cache effectiveness - if you have high write application such as forums, query cache efficiency might be pretty low due to this. There is also way to set minimal TTL or anything like it which is allowed by other caching systems. Also note - all queries are removed from cache on table modifications - if there are a lot of queries being cached this might reduce update speed a bit.
Fragmentation over time - Over time Query Cache might get fragmented, which reduces performance. This can be seen as large value of Qcache_free_blocks relatively to Qcache_free_memory. FLUSH QUERY CACHE command can be used for query cache defragmentation but it may block query cache for rather long time for large query caches, which might be unsuitable for online applications.
Limited amount of usable memory - Queries are constantly being invalidated from query cache by table updates, this means number of queries in cache and memory used can’t grow forever even if your have very large amount of different queries being run. Of course in some cases you have tables which are never modified which would flood query cahe but it unusual. So you might want to set query cache to certain value and watch Qcache_free_memory and Qcache_lowmem_prunes - If you’re not getting much of lowmem_prunes and free_memory stays high you can reduce query_cache appropriately. Otherwise you might wish to increase it and see if efficiency increases.
Demand operating mode If you just enable qury cache it will operate in “Cache everything” mode. In certain caches you might want to cache only some of the queries - in this case you can set query_cache_type to “DEMAND” and use only SQL_CACHE hint for queries which you want to have cached - such as SELECT SQL_CACHE col from foo where id=5. If you run in default mode you can also use SQL_NO_CACHE to block caching for certain queries, which you know do not need to be cached.
Counting query cache efficiency There are few ways you can look at query_cache efficiency. First looking at number of your selects - Com_select and see how many of them are cached. Query Cache efficiency would be Qcache_hits/(Com_select+Qcache_hits). As you can see we have to add Qcache_hits to Com_select to get total number of queries as if query cache hit happens Com_select is not incremented. But if you have just 20% Cache hit rate does it mean it is not worth it ? Not really it depends on which queries are cached, as well as overhead query cache provides. One portion of query cache overhead is of course inserts so you can see how much of inserted queries are used: Qcache_hits/Qcache_inserts Other portion of overhead comes from modification statements which you can calculate by (Com_insert+Com_delete+Com_update+Com_replace)/Qcache_hits
. These are some numbers you can play with but it is hard to tell what is good or bad as a lot depends on statement complexity as well as how much work is saved by query cache.
Now lets speak a bit about Query Cache configuration and mainance. MySQL Manual is pretty good on this: Query Cache Query Cache Status Query Cache Configuration
I would just mention couple of points - as protection from one query wiping your all query cache option query_cache_limit was implemented which limits result set which can be stored in query cache. If you need larger queries to be cached you might increase it, if you most important queries are smaller you can decrease it. The other one is Qcache_lowmem_prunes - This one is used to identify if you have enough memory for query cache. Note however due to fragmentation lowmem_prunes can be triggered even if there is some free space, just badly fragmented.
Looking at performance I’ve seen query cache offering about double performance for simple queries with select done by primary key, obviously there is no upper boundary - Very complex queries producing small result set will be offering best speed up.
So when it is good idea to use query cache ?
Third party application - You can’t change how it works with MySQL to add caching but you can enable query cache so it works faster.
Low load applications - If you’re building application which is not designed for extreme load, like many personal application query cache might be all you need. Especially if it is mostly read only scenario.
Why Look for alternatives ?
There are few reasons why Query Cache might be not cache for your application:
It caches queries Application objects might need several queries to compose so it is efficient to cache whole objects rather than individual queries.
No control on invalidation Table granularity is often too bad. With other caches you may implement version based or timeout based invalidation which can offer much better hit ratio for certain application.
It is not that fast Query Cache is fast compared to running the queries but it is still not as fast as specially designed systems such as memcached or local shared memory.
It can’t retrieve multiple objects at the same time You have to query cache object by object which adds latency, there is no way you can request all objects you need to be retrieved at the same time (again memcached has it)
It is not distributed If you have 10 slaves and use query cache on all of them cache content will likely be the same, so you have multiple copies of the same data in cache effectively wasting memory. Distirbuted caching systems can effectively use memory on multiple systems so there is no duplication.
Memcached is probably the most popular distributed caching system and it works great. I should write an article comparing performance of various caching systems some time.
Print This Post
del.icio.us :: 
digg Comment RSS :: Trackback URI
40 Comments
It certainly depends on people’s implementations but I’ve found thhat in the type of apps that I work on the query cache is almost pointless. It just ends up being invalidated all the time so I just set the cache size to zero to avoid even wasting any memory.
I find my real boosts are when I use php with memcached and squid or mod_cache on Apache.
Kevin
Comment :: July 28, 2006 @ 8:30 pm
Thank you Kevin,
Yes sure. In your case you’re getting a lot of updates so queries could be constantly invalidated.
Actually it is good way to estimate if query cache is going to work for you - look at how frequently your tables are invalidated
and how many same queries you’re going to get during this time. In some cases invalidation even once per second is not too bad as there are so many same queries it pays of, in others even once per minute may be too bad.
I’m with you on memcached and server side proxy caching whenever proxy when possible.
Comment :: July 30, 2006 @ 9:46 am
[...] Query Cache is good when there is no other cache If you do not do any other caching for certain object or if you cache on different level (ie single object constructed from multiple query results) MySQL Query Cache may improve performance of your application. [...]
Pingback :: August 9, 2006 @ 8:00 am
is there any tool or application which we can use to bulid Database and use our queries. and then that queries can be Optimiz using that tool or application
Comment :: December 8, 2006 @ 12:04 am
Yogesh,
I have no idea what do you mean by your question.
Comment :: December 8, 2006 @ 11:04 am
[...] Since WordPress uses MySQL, almost every page load results in various MySQL queries for logins, post content, categories and so on. It is very likely that this data didn’t change since the last request so we begin with enabling the cache within the MySQL server. Basically, all you need to to is to activate it in the my.cnf (or my.ini) MySQL configuration file with setting the query_cache_type variable to 1. The query-cache-size variable is the size of the query cache in MB, for example 20M. To change the MySQL configuration file, you’ll probably need root access which won’t be available on shared hosts. More info about query cache is available on techiecorner.com and the MySQL Performance blog. [...]
Pingback :: February 16, 2007 @ 6:19 pm
sounds like yogesh hasnt heard of the “explain” command in mysql.
Comment :: March 7, 2007 @ 10:32 am
The MyISAM key cache retains index blocks in RAM for fast access, with both default and custom-created caches being available. Its just like the code cache in oracle.
Comment :: March 12, 2007 @ 2:23 am
I am facing a strange problem with Query cache when used with JDBC. I am using MySQL5.0.37. I have Query cache enabled and configured as 64MB.
I perform the following operations
1.SELECT * from tableA
2.UPDATE tableA
3.SELECT * from tableA
When the above mentioned operation is performed in JDBC, step#1 fills the cache and steps#2 fails to invalidate the cache. Hence step#3 gets the old results from the cache.
The same set of operations are working when tried with a query browser.
My application is not a transaction based. Even i tried AUTOCOMMIT=1.
I will appreciate any ideas about this problem.
Comment :: March 26, 2007 @ 3:00 pm
Is there a way to ask mysql the size of the result set prior to actualy querying for it? Some result sets are too large and/or too small for my application and I’d rather not query for them/receive them.
Also, the obvious follow up question is whether or not it will improve overall application performance. =)
Thanks in advance for your responses.
Comment :: May 22, 2007 @ 10:44 pm
Fima, you can of course run count(*) query to see how many results are where but it may not be fast.
you can also add LIMIT 1000 for example to result set and if you got 1000 results you will know you likely got incomplete result in the application.
Comment :: May 23, 2007 @ 12:46 am
if you want to know how many for sure up front, put SQL_CALC_FOUND_ROWS after your select. You can then SELECT FOUND_ROWS() to get the count of all possible items if you would get without a limit. I would do the initial query with LIMIT 0 if you don’t want anything actually returned to you. Realistically, do a limit of 100, and you will also have the max number from the second query, and decide from there how to proceed.
Tony
Comment :: June 20, 2007 @ 10:52 am
[...] Since WordPress uses MySQL, almost every page load results in various MySQL queries for logins, post content, categories and so on. It is very likely that this data didn’t change since the last request so we begin with enabling the cache within the MySQL server. Basically, all you need to to is to activate it in the my.cnf (or my.ini) MySQL configuration file with setting the query-cache-type variable to 1. The query-cache-size variable is the size of the query cache in MB, for example 20M. To change the MySQL configuration file, you’ll probably need root access which won’t be available on shared hosts. More info about query cache is available on techiecorner.com and the MySQL Performance blog. [...]
Pingback :: August 26, 2007 @ 5:00 am
[...] MySQL Query Cache | MySQL Performance Blog MySQL performansi icin optimize ip uclari. [...]
Pingback :: November 29, 2007 @ 4:36 pm
[...] Here you can find the great post about other limitations of the MySQL query cache and MySQL query cache efficiency. [...]
Pingback :: December 9, 2007 @ 5:50 am
[...] Since WordPress uses MySQL, almost every page load results in various MySQL queries for logins, post content, categories and so on. It is very likely that this data didn’t change since the last request so we begin with enabling the cache within the MySQL server. Basically, all you need to to is to activate it in the my.cnf (or my.ini) MySQL configuration file with setting the query-cache-type variable to 1. The query-cache-size variable is the size of the query cache in MB, for example 20M. To change the MySQL configuration file, you’ll probably need root access which won’t be available on shared hosts. More info about query cache is available on techiecorner.com and the MySQL Performance blog. [...]
Pingback :: December 30, 2007 @ 5:00 pm
Different methods of caching
Because we encounter PHP more and more in the enterprise market, the performance demands of websites we build are becoming ever more challenging. The usefulness of caching is often underestimated by developers (myself included). When using a lightweight f
Trackback :: January 30, 2008 @ 6:00 am
Some of the information in this blog entry is out of date and thus incorrect:
ie.
> “Avoid comment (and space) in the start of the query - Query Cache does simple optimization to check if query can be cached. As I mentioned only SELECT queries are cached - so it looks at first letter of the query and if it is “S” it proceeds with query lookup in cache if not - skips it.”
I just wasted a couple of hours on this one only to realise that the statement was incorrect and based on a bug which has long since been fixed.
Other things in this article are wrong too. For instance some prepared statements will in fact work. Check the manual for more accurate info on what will and won’t work.
And the Pro MySqL book that I have always documents issues such as this, but does not mention this problem and the same goes for the MySQL manual. I would have preferred if the author had specified where he was getting his info so I didn’t have to waste so much time looking into this.
This guy tested it:
http://www.petefreitag.com/item/390.cfm
This documents that the bug only exists in older 4.0 version of MySQL:
http://books.google.com/books?id=iaCCQ13_zMIC&pg=PA80&lpg=PA80&dq=white+space+query+cache&source=web&ots=3ED_193Hn1&sig=Ctiy3T5zxWyJHuKnqBtHxuofm3Q
Comment :: February 18, 2008 @ 11:47 pm
Gerry,
At the point this article was written (about 1.5 years ago) the prepared statements did not work with query cache. Even now they are only supported in MySQL 5.1 which is still not production release.
The problem with spaces was indeed partially fixed. Basic whitespace is fixed AFAIR but comments are not.
Comment :: February 19, 2008 @ 11:31 am
“AFAIR but comments are not”
Hmmmm… as far as you remember…. but from where? Where would you have got this information?
The second link I provided says that the comments/space issue was fixed in MySQL 5
Comment :: March 20, 2008 @ 1:27 am
[...] on my previous post on MySQL Query Cache Gerry pokes me as I’m all wrong and both comments and whitespace are fixed in [...]
Pingback :: March 20, 2008 @ 10:56 am
thanks for the informative article.
i would like to know if there is some transparent query cache available for mysql that is not ACID safe. something to work with high update scenario, the cache is NOT dropped with every table change. memcached is one option, but it is not transparent, i.e., i need to modify in application every place where an expensive mysql query is being done. i would rather prefer a transparent way, e.g., something implemented as a wrapper over JDBC classes with local in-memory caching. or a set of APIs that first query the memcached/JCS/EHCache and if nothing is found, query the MySQL database.
thanks
Comment :: May 10, 2008 @ 9:25 am
nawab,
Typically people would use either memcache or query cache. Invalidation is important part of being transparent - if you do not invalidate you get stale data from cache so it is not transparent for application any more - application needs to be aware it is getting the stale data. And what if it needs the most recent one ?
Using memcache or other cache explicitly allows control over how invalidation works.
Comment :: May 10, 2008 @ 11:40 pm
I have created a table which has 1 lakh of records but the problem is ..when m trying to perform a Query on this table … it consists of 3-4 tables joins … so it takes around 7-8 mins to fetch the data … Then i tried to Built Index on few columns…. But when we Explain that Query … thn also it shows ALL instead of the index type of any
Comment :: July 2, 2008 @ 11:31 pm
Please clarify that whether it is possible to cache data of whole table in memory so that queries are fast in MYSQL ?
Comment :: July 16, 2008 @ 4:22 am
Pratap,
This is common misconception. Query Cache caches query Result it does not cache any data from the base tables. Other caches exist for that purpose.
Comment :: July 20, 2008 @ 9:17 am
[...] or if you cache on different level (ie single object constructed from multiple query results) MySQL Query Cache may improve performance of your [...]
Pingback :: August 29, 2008 @ 9:26 am
how to optimize the select query with round() func…
Comment :: September 26, 2008 @ 2:46 pm
as my select query is taking 3 secs time i want to optimize it to 1 secs and
my query is like this “SELECT round(sum(i.withdrawrealwin)) FROM invoice i where i.raceno=’1′”
i want to reduce the execution time
please give me a solution
Comment :: September 26, 2008 @ 2:48 pm
Could have been nice if query cache could have been instructed to avoid certain tables.
In my db a few tables are frequently updated thus not suitable for query cache while others are seldom updated and perfect for it. Current MySQL configuration is really uneasy since I’ll have to specify whether to cache or not per query.
Anyway, using APC (or memcached if on multiple machines) is a much better choice as it’s much faster than MySQL’s query cache.
Comment :: October 19, 2008 @ 1:07 am
[...] Since WordPress uses MySQL, almost every page load results in various MySQL queries for logins, post content, categories and so on. It is very likely that this data didn’t change since the last request so we begin with enabling the cache within the MySQL server. Basically, all you need to to is to activate it in the my.cnf (or my.ini) MySQL configuration file with setting the query-cache-type variable to 1. The query-cache-size variable is the size of the query cache in MB, for example 20M. To change the MySQL configuration file, you’ll probably need root access which won’t be available on shared hosts. More info about query cache is available on techiecorner.com and the MySQL Performance blog. [...]
Pingback :: November 22, 2008 @ 6:10 am
[...] http://www.mysqlperformanceblog.com/2006/07/27/mysql-query-cache/ [...]
Pingback :: December 13, 2008 @ 8:02 pm
Thanks for the nice post, Peter.
You mentioned that FLUSH QUERY CACHE needs to be run to defragment the free blocks. The Qcache_free_blocks in my server had some high value (~1500) few days back but when i checked today, it was reduced to 3. No one ran FLUSH QUERY CACHE in the server. I wonder what triggered the defragmentation? Is the query_cache completely reset (so that the fragments are also cleaned up) when it runs out of memory (pruning)? I was planning to have a cron to run FLUSH QUERY CACHE on weekly basis but may be it isn’t required. But i don’t understand how and who takes care of the defragmentation? Thanks for your time.
Comment :: December 23, 2008 @ 1:02 pm
follow up to my previous question..
or the query_cache is completely reset (including the fragmentation) when all the results in the cache are invalidated at once?
Can OPTIMIZE TABLE invalidate the query results of the associated table? If so, it makes sense as all the tables are periodically OPTIMIZEd in my case.
Comment :: December 23, 2008 @ 1:10 pm
Thiru,
The free space is merged so if you invalidate all results from query cache. This is probably the reason in your case.
Comment :: December 23, 2008 @ 2:47 pm
Almost all of our db is written in Stored Procedures and some stored functions too (they are using prepare statement as well). I guess I have been facing the issue of query cache overhead as discussed above so I was thinking to cache only selective queries. Now the problem is when you say that Stored functions still cannot be cached does that include stored procedures too? If yes, then the only solution left is memcached?
THanks for the help
Comment :: March 31, 2009 @ 6:36 am
> Anyway, using APC (or memcached if on multiple machines) is a much
> better choice as it’s much faster than MySQL’s query cache.
What a stoopid comment!
This are two different shoes!
APC is a bytecode-cache and prevent php to parse the whole code on every call
This has nothing to to with query-cache and really NOTHING
If you are smart you use eaccelerator/apc AND query-cache and you get response times with 0.015 seconds for a whole page as we do on our servers
Comment :: April 14, 2009 @ 2:49 am
I am experiencing serious bottlenecks which are crashing Apache on my server and I believe they are related to mass query cache invalidation of frequently modified tables. Also I think this is more sensible as in many cases there would be cache duplication with APC with MySQL invalidating unused cached queries. I have decided to set query_cache_type to DEMAND. Fingers crossed!
Comment :: April 22, 2009 @ 7:22 pm
Reindl Harald,
APC does more then just opcode caching. You can also use it as misc object cache from inside php, somewhat comparable to memcached but without networking and in the same process (eg no context switches) which makes it faster then memcached but less useful when you have more then 1 webserver.
Dieter
Comment :: June 2, 2009 @ 2:10 am
> APC does more then just opcode caching.
> You can also use it as misc object cache from inside php
This are still different shoes
And even if you cache whole pages/parts that way you should combine it with mysql-qery-cache because if your manual cache is outdated you get most time parts of it from query cache without to do anything.
On a good configured server with well designed applications it looks like following
Up for: 2d 16h 57m 3s (7M q [32.419 qps], 224K conn, TX: 8B, RX: 1B)
[OK] Key buffer hit rate: 99.7% (14M cached / 44K reads)
[OK] Query cache efficiency: 87.9% (5M cached / 6M selects)
On this machine are 200 domains with generate times 0.010 seconds for a whole page
Many parts (objects) of the pages are stored in a mysql-caching table because its stoopid to use memcached/shm for caching on a shared-hosting. So you are independent of apc/memcached, have optimal caching, reduced queries and a secure setup with a great performance
Comment :: June 3, 2009 @ 3:57 am