The Item Store is the brand new index structure available for this release.

The Item Store provides far better performance at indexing time as the index is now fully generational. Previous release limitations for in-memory fields management (any import in the index was requiring a full rebuild of the in-memory index structure) have been tackled, and the Item Store now allows reducing the overall latency of the index import task.

This new store provides a lower memory footprint of indices, due to his columnar-based structure for in-memory fields. Indexing documents mapped to distinct classes does not lead to in-memory sparse matrix anymore, and hence reduces the global memory consumption.

 

• Latency no longer proportionate to the number of documents in the index, due to generational item store

CloudView provides a new kind of field in the data model: Measure. For a field with this data type, 2 new pieces of information are defined:

The unit symbol is the pivot unit in which the value will be stored inside the index.

When a document is sent with this meta value, a Units Of Measurement Normalizer document processor will be applied to it:

When a query is executed on this field, aprefix handler of Units Of Measurement type will handle the query:

 

CloudView is now fully Java 8. Please note that the minimal level of compatibility has been raised to Java 7 for any other Java components (Java clients, Core & Mashup plugins or external jar).
You can also use our latest Eclipse plugin which includes Java 8 packaging (#24812).

This release also contains various enhancements, please read carefully the Other Improvements section below.

• Introspection UI can display a node with Max Depth parameter set to 0. The node will be displayed without its arcs (#24857)

• PushAPITransformationHelpers provides methods to retrieve the type of a document (getType & getTypeInheritanceList) (#25031)

• Default dictionary has a dedicated config and does not depend anymore of the order of the dictionaries declaration (#24935)

• Search performance for queries on a lot of words (thousand words) are up to 10 times faster (#25142).

• If a disjunctive facet has been refined, this state is shown in the facet answer group (#22411).

• Ability to match regex in annotation manager (#23161)

• When using multiple rules in the semantic query analysis configuration (SQA), it is possible to define for each rule the policy for unused words if the default one is not relevant (#24946)

• It is possible to use the dump-attribute-group command for a single attribute and to target all slots (#24192)

Overall enhancement of jrds probes:

ItemStore is activated by default for brand new CloudView installations. However, older installations migrating to R2016x will keep the previous row-oriented structure. If you want to benefit from the new ItemStore, re-indexing is required.

When migrating a CloudView instance, the operation now starts with a full compact of indexes. As this operation may take some time (depending on the size and structure of indexes), the overall migration time will increase accordingly.

CAUTION : When migrating a CloudView cluster containing replicates, the migration will be complete only when the full replication of each replica is achieved (in that case, replica will not be able to serve previous index generations, and thus will not return any hits during the replication operation).