Solr 6.3 Features


Here’s an overview of some of the new features upcoming in Solr 6.3.
Download Solr 6.3 to try these features out and give us feedback!
You can also check out upcoming features of the next Solr release.

New dvhash faceting method in the JSON Facet API

A new faceting method for the JSON Facet API has been added that utilizes hashing instead of direct indexing by ordinal. Currently, when hashing on a field with docValues enabled, or when specifying method=dv, there are accumulator arrays that are indexed by ordinal. When the cardinality of the domain is low, and the cardinality of the field is high, direct indexing of large arrays both wastes memory and takes longer to scan looking for the highest accumulated entries.

Example:

json.facet={
  top_authors : {
    type : terms,
    field : author,
    limit : 20,
    method : dvhash
  }
}

Since it may be difficult for a client to know when the facet domain will be large vs the field cardinality (esp if the domain is based off of a user query), future work is expected to implement automatically switching between faceting methods.

Optimizing, storing and deploying AI models with Streaming Expressions

The classify streaming expression classifies documents according to a model.

classify(<document_stream>,            // the stream to fetch documents that need to be classified
         <model_stream>,               // the stream to fetch the latest model
         field=<input_text_field>,     // the field on the document to classify
         fieldType=<analysis_fieldType> // (opt) fieldType to use for tokenization of the input field
)

For each input document, the classifier decorates the document with two (currently hardcoded) fields with the result:

  • score_d – the raw score
  • probability_d – the positive probability that the document belongs to the group

The update streaming expression can be used to store classifier models in a SolrCloud collection, and the topic streaming expression can be used with the classify expression to both deploy models as well as stream new data through the models.

The development JIRA SOLR-9258 has examples, and should be consulted until the Solr reference guide is updated.

“executor” Streaming Expression

The executor streaming expression wraps another stream containing streaming expressions. By default the expression will be contained in the expr_s field in each tuple of the wrapped stream. The executor has an internal thread pool so expressions can be executed in parallel on a single worker. This expression can further be wrapped in a parallel streaming expression to enable execution across a cluster of worker nodes.

Example syntax from the JIRA of a work queue… the “topic” expression retrieves expressions to execute (and keeps track of where it left off):

daemon(executor(threads=10, topic(storedExpressions, fl="expr_s", ...)))

“commit” Streaming Expression

The commit streaming expression wraps another stream (normally involving updates) and performs a commit at the end.

Example:

commit(targetCollection, update(targetCollection, search(sourceCollection, q="myquery", ...)))

“fetch” Streaming Expression

The fetch streaming expression wraps another expression and fetches additional fields from documents in batches. It’s very much like an inner join, with one side of the join being very small.

Example that queries books and adds publisher address to each result:

fetch(publisherCollection,
      search(booksCollection, q="*:*", fl="title,author,publisher", sort="publisher asc"),
      fl="publisherAddress",
      on="publisher=publisherId")

Lots More…

See the release notes on the Solr wiki and the CHANGES file for an in-depth list of changes.