David Smiley created SOLR-15260:
-----------------------------------
Summary: Precompute snippet delimiter breaks for the
UnifiedHighlighter
Key: SOLR-15260
URL: https://issues.apache.org/jira/browse/SOLR-15260
Project: Solr
Issue Type: Improvement
Security Level: Public (Default Security Level. Issues are Public)
Components: highlighter
Reporter: David Smiley
The "BreakIterator" implementation inside the UnifiedHighlighter can play a
significant role in the performance of highlighting. The default ones are
based in the JDK and thus we don't have control over them but they may very
well be optimized but have a complicated job to do. I propose that the break
locations be computed at indexing time in a Solr UpdateRequestProcessor and
place them into a pre analyzed common field named maybe
{{\_highlighter_breaks_}} that needs indexed=true plus offsets. In this field,
the term is the actual field name, the position is meaningless, and the offset
pair refers to the span of the break iterator (typically a sentence). This
data can be efficiently stored in Lucene. The UnifiedHighlighter already has a
flexible BreakIterator producer but it's not notified of the current document,
and so changes would be needed there (separate LUCENE issue).
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
