Lookups are performed using {@link SegmentHeader}s and {@link SegmentBody}s. Both are immutable and fully serializable.
There are a few ways to declare a SegmentCache implementation in Mondrian. The first one is to set the {@link MondrianProperties#SegmentCache} property.
The second one is to use the Java Services API. This is the preferred mean. You will need to create a jar file, accessible through the same class loader as Mondrian, and add a file called /META-INF/services/mondrian.spi.SegmentCache which contains the name of the segment cache implementation to use.
The third method is to use the {@link SegmentCacheInjector}. This is to be used as a last resort, in environments where the cache implementation is not part of the same class loader as Mondrian. In those cases, Mondrian can't dynamically load the segment cache class. The injector serves as an IoC-like service.
All of the segment caches that Mondrian discovers, throughout all of these means of discovery, will be used simultaneously. It is not possible to register new segment caches for a previously existing instance of a Mondrian server. The caches are scanned and configured when each Mondrian instance gets created.
Implementations are expected to be thread-safe. Mondrian is likely to submit multiple requests at the same time, from different threads. It is the responsibility of the cache implementation to maintain a consistent state.
Implementations must implement a time-out policy, if needed. Mondrian knows that a call to the cache might take a while. (Mondrian uses worker threads to call into the cache for precisely that reason.) Left to its own devices, Mondrian will wait forever for a call to complete. The cache implementation might know that a call to {@link #get} that has taken 100milliseconds already is probably hung, so it should return null or throw an exception. Then Mondrian can get on with its life, and get the segment some other way.
Implementations must provide a default empty constructor. Mondrian creates one segment cache instance per Mondrian server. There could be more than one Mondrian server running in the same JVM. @author LBoudreau
| |
| |
| |