How-To Knowledge Persistence#
In general, a knowledge base is physically stored as a Jar file (java archive), that contains a collection of (mostly XML) files. Typically, different types of knowledge are represented by different files in the Jar.The knowledge persistence provides three ExtensionPoints:
- KnowledgeReader
- KnowledgeWriter
- FragmentHandler
KnowledgeReader #
The PersistenceManager coordinates the loading of a knowledge base by calling appropriate KnowledgeReader implementations for each file. The extension of the particular KnowledgeReader defines, which file(s) in the Jar can be handled: Here, it is possible to specify a file name (incl. file path) and/or the specification of a regular expression. The collection of matching files (i.e., files that can be handled by the reader) is the determined by evaluating the regular expression on the Jar file structure. If the file path and the regular expression is specified, then the given file as well as the matching files are handled by the FileReader.
<extension plugin-id="KnowledgePersistenceExtensionPoints" point-id="KnowledgeReader" id="BasicPersistenceHandler"> <parameter id="class" value="de.d3web.core.kpers.BasicPersistenceHandler" /> <parameter id="name" value="BasicPersistenceHandler" /> <parameter id="description" value="KnowledgeReader for basic knowledge" /> <parameter id="version" value="1.0" /> <parameter id="priority" value="1" /> <parameter id="filename" value="kb/basic.xml" /> </extension>
The example from above shows the extension for the basic persistence handler, that expects the file name kb/basic.xml for the handling class de.d3web.core.kpers.BasicPersistenceHandler.
In case that two KnowledgeReaders are able to handle the same file, the KnowledgeReader with the higher priority is selected for handling the file (no duplicate file handling!).
The interface KnowledgeReader specifies the read method with the parameters
- InputStream of the file
- the KnowledgeBase instance as input
and writes updates to the KnowledgeBase instance when importing the handled files.
KnowledgeWriter#
The same strategy as for the KnowledgeReader is used when saving a KnowledgeBase instance physically to a file. In contrast to readers, the extension for a writer only takes a file name and no regular expression for saving the knowledge base. A stream and a knowledge base instance is passed to the writer. The writer extracts all handled knowledge slices from the knowledge base instance and the stores them physically.
FragmentHandler#
FragmentHandlers are special in the way, that they do not handle complete (file-based) knowledge collections, but single knowledge objects. This functionality allows for the application of FragmentHandlers as helpers within (multiple) KnowledgeReaders and KnowledgeWriters. Furthermore, FragmentHandlers simplify the integration of new KnowledgeSlices dramatically.
We describe the application of FragmentHandlers by the following example: For saving a rule, we first store their basic properties such as the ID. For the persistence handling of more complex objects of a rule, e.g., the rule condition and the rule action, we call the appropriate FragmentHandlers for conditions and actions. These calls are realized by following procedure: All KnowledgeReaders give permission to the PersistenceManager to select the appropriate handlers for the knowledge objects. With this permission, the PersistenceManager selects the appropriate FragmentHandler according to their priority. For each FragmentHandler (extensions that implement the ExtensionPoint FragmentHandler) it is tested, if the handler can store/load the currently considered knowledge object. The handler with the highest priority is the activated to store/load the considered object. For instance, rule conditions are also used as conditioned findings in XCL models, and thus the same FragmentHandler for conditions can be used by the KnowledgeReaders/KnowledgeWriters of the rule persistence and the XCL persistence.
When extending the domain model of the d3web core, often a new type for a rule condition needs to be defined. For a new rule condition type, we simply need to implement an appropriate FragmentHandler; this FragmentHandler can be used by the rule persistence as well as by the XCL persistence.
Helper Classes#
We defined a collection of useful abstract classes to avoid unnecessary code duplication for the implementation of (new) KnowledgeWriters, KnowledgeReaders and FragementHandlers:
- AbstractRulePersistenceHandler
It implements KnowledgeReader as well as KnowledgeWriter and can be used as super class, when implementing a new RulePersistenceHandler. Then, you need to set the RuleType in the constructor and you will need to implement the method getProblemsolverContent(), which returns a list of problem solvers corresponding to the rules to be handled. - AbstractExpressionHandler
- QuestionComparatorHandler
- QuestionComparatorGroupedHandler