Architecture and Concepts
The query cache plugin is implemented as a PHP extension. It is written in C and operates under the hood of PHP. During the startup of the PHP interpreter, it gets registered as a mysqlnd plugin to replace selected mysqlnd C methods. Hereby, it can change the behaviour of any PHP MySQL extension (mysqli, PDO_MYSQL, mysql) compiled to use the mysqlnd library without changing the extensions API. This makes the plugin compatible with each and every PHP MySQL application. Because existing APIs are not changed, it is almost transparent to use. Please, see the mysqlnd plugin API description for a discussion of the advantages of the plugin architecture and a comparison with proxy based solutions.
Transparent to use
At PHP run time PECL/mysqlnd_qc can proxy queries send from PHP (mysqlnd) to the MySQL server. It then inspects the statement string to find whether it shall cache its results. If so, result set is cached using a storage handler and further executions of the statement are served from the cache for a user-defined period. The Time to Live (TTL) of the cache entry can either be set globally or on a per statement basis.
A statement is either cached if the plugin is instructed to cache all statements globally using a or, if the query string starts with the SQL hint (/*qc=on*/). The plugin is capable of caching any query issued by calling appropriate API calls of any of the existing PHP MySQL extensions.
Flexible storage: various storage handler
Various storage handler are supported to offer different scopes for cache entries. Different scopes allow for different degrees in sharing cache entries among clients.
default (built-in): process memory, scope: process, one or more web requests depending on PHP deployment model used
APC: shared memory, scope: single server, multiple web requests
SQLite: memory or file, scope: single server, multiple web requests
MEMCACHE: main memory, scope: single or multiple server, multiple web requests
user (built-in): user-defined - any, scope: user-defined - any
Support for the APC, SQLite and MEMCACHE storage handler has to be enabled at compile time. The default and user handler are built-in. It is possible to switch between compiled-in storage handlers on a per query basis at run time. However, it is recommended to pick one storage handler and use it for all cache entries.
Built-in slam defense to avoid overloading
To avoid overload situations the cache plugin has a built-in slam defense mechanism. If a popular cache entries expires many clients using the cache entries will try to refresh the cache entry. For the duration of the refresh many clients may access the database server concurrently. In the worst case, the database server becomes overloaded and it takes more and more time to refresh the cache entry, which in turn lets more and more clients try to refresh the cache entry. To prevent this from happening the plugin has a slam defense mechanism. If slam defense is enabled and the plugin detects an expired cache entry it extends the life time of the cache entry before it refreshes the cache entry. This way other concurrent accesses to the expired cache entry are still served from the cache for a certain time. The other concurrent accesses to not trigger a concurrent refresh. Ideally, the cache entry gets refreshed by the client which extended the cache entries lifespan before other clients try to refresh the cache and potentially cause an overload situation.
Unique approach to caching
PECL/mysqlnd_qc has a unique approach to caching result sets that is superior to application based cache solutions. Application based solutions first fetch a result set into PHP variables. Then, the PHP variables are serialized for storage in a persistent cache, and then unserialized when fetching. The mysqlnd query cache stores the raw wire protocol data sent from MySQL to PHP in its cache and replays it, if still valid, on a cache hit. This way, it saves an extra serialization step for a cache put that all application based solutions have to do. It can store the raw wire protocol data in the cache without having to serialize into a PHP variable first and deserializing the PHP variable for storing in the cache again.