Some values in a database are know to change slower than others. To improve performance, many developers like to cache often-used data to avoid making unnecessary trips back to the database. iBATIS provides its own caching system, that you configure through a <cacheModel> element.
The results from a query Mapped Statement can be cached simply by specifying the cacheModel parameter in the statement tag (seen above). A cache model is a configured cache that is defined within your DataMapper configuration file. Cache models are configured using the cacheModel element as follows:
Example 3.51. Configuation a cache using the Cache Model element
<cacheModel id="product-cache" implementation="LRU" readOnly="true" serialize="false">
<flushInterval hours="24"/>
<flushOnExecute statement="insertProduct"/>
<flushOnExecute statement="updateProduct"/>
<flushOnExecute statement="deleteProduct"/>
<property name="CacheSize" value="100"/>
</cacheModel>The cache model above will create an instance of a cache named "product-cache" that uses a Least Recently Used (LRU) implementation. The value of the type attribute is either a fully qualified class name, or an alias for one of the included implementations (see below). Based on the flush elements specified within the cache model, this cache will be flushed every 24 hours. There can be only one flush interval element and it can be set using hours, minutes, seconds or milliseconds. In addition the cache will be flushed whenever the insertProduct, updateProduct, or deleteProduct mapped statements are executed. There can be any number of "flush on execute" elements specified for a cache. Some cache implementations may need additional properties, such as the ‘cache-size' property demonstrated above. In the case of the LRU cache, the size determines the number of entries to store in the cache. Once a cache model is configured, you can specify the cache model to be used by a mapped statement, for example:
Example 3.52. Specifying a Cache Model from a Mapped Statement
<statement id="getProductList" cacheModel="product-cache">
select * from PRODUCT where PRD_CAT_ID = #value#
</statement>The cache model uses a pluggable framework for supporting different types of caches. The choice of cache is specified in the "implementation" attribute of the cacheModel element as discussed above. The class name specified must be an implementation of the ICacheController interface, or one of the three aliases discussed below. Further configuration parameters can be passed to the implementation via the property elements contained within the body of the cacheModel. Currently there are 3 implementations included with the .NET distribution. These are as follows:
The MEMORY cache implementation uses reference types to manage the cache behavior. That is, the garbage collector effectively determines what stays in the cache or otherwise. The MEMORY cache is a good choice for applications that don't have an identifiable pattern of object reuse, or applications where memory is scarce.
The MEMORY implementation is configured as follows:
Example 3.53. Configuring a memory-type cache
<cacheModel id="product-cache" implementation="MEMORY" > <flushInterval hours="24"/> <flushOnExecute statement="insertProduct"/> <flushOnExecute statement="updateProduct"/> <flushOnExecute statement="deleteProduct"/> <property name="Type" value="WEAK"/> </cacheModel>
Only a single property is recognized by the MEMORY cache implementation. This property, named 'reference-type' must be set to a value of STRONG, SOFT, or WEAK.
The following table describes the different reference types that can be used for a MEMORY cache.
Table 3.6. Reference types that can be used for a MEMORY cache
| Type | Description |
|---|---|
| WEAK (default) | This reference type is probably the best choice in most cases and is the default if the reference-type is not specified. It will increase performance for popular results, but it will absolutely release the memory to be used in allocating other objects, assuming that the results are not currently in use. |
| SOFT (currently Java only) | This reference type will reduce the likelihood of running out of memory in case the results are not currently in use and the memory is needed for other objects. However, this is not the most aggressive reference type in that regard and memory still might be allocated and made unavailable for more important objects. |
| STRONG | This reference type will guarantee that the results stay in memory until the cache is explicitly flushed (e.g. by time interval or flush on execute). This is ideal for results that are: 1) very small, 2) absolutely static, and 3) used very often. The advantage is that performance will be very good for this particular query. The disadvantage is that if the memory used by these results is needed, then it will not be released to make room for other objects (possibly more important objects). |
The LRU cache implementation uses an Least Recently Used algorithm to determines how objects are automatically removed from the cache. When the cache becomes over full, the object that was accessed least recently will be removed from the cache. This way, if there is a particular object that is often referred to, it will stay in the cache with the least chance of being removed. The LRU cache makes a good choice for applications that have patterns of usage where certain objects may be popular to one or more users over a longer period of time (e.g. navigating back and forth between paginated lists, popular search keys etc.).
The LRU implementation is configured as follows:
Example 3.54. Configuring a LRU type cache
<cacheModel id="product-cache" implementation="LRU" > <flushInterval hours="24"/> <flushOnExecute statement="insertProduct"/> <flushOnExecute statement="updateProduct"/> <flushOnExecute statement="deleteProduct"/> <property name="CacheSize" value="100"/> </cacheModel>
Only a single property is recognized by the LRU cache implementation. This property, named CacheSize must be set to an integer value representing the maximum number of objects to hold in the cache at once. An important thing to remember here is that an object can be anything from a single String instance to an ArrayList of object. So take care not to store too much in your cache and risk running out of memory!
The FIFO cache implementation uses an First In First Out algorithm to determines how objects are automatically removed from the cache. When the cache becomes over full, the oldest object will be removed from the cache. The FIFO cache is good for usage patterns where a particular query will be referenced a few times in quick succession, but then possibly not for some time later.
The FIFO implementation is configured as follows:
Example 3.55. Configuring a FIFO type cache
<cacheModel id="product-cache" implementation="FIFO" > <flushInterval hours="24"/> <flushOnExecute statement="insertProduct"/> <flushOnExecute statement="updateProduct"/> <flushOnExecute statement="deleteProduct"/> <property name="CacheSize" value="100"/> </cacheModel>
Only a single property is recognized by the FIFO cache implementation. This property, named CacheSize must be set to an integer value representing the maximum number of objects to hold in the cache at once. An important thing to remember here is that an object can be anything from a single String instance to an ArrayList of object. So take care not to store too much in your cache and risk running out of memory