The GitHub Actions job "Build" on 
jackrabbit-oak.git/oak-search-mongot-stored-source has failed.
Run started by GitHub user meticulous-dft (triggered by rishabhdaim).

Head commit for run:
d4d9b18922fce56988d69ded5b62843d03aa40ba / Ming He <[email protected]>
OAK-search-mongot: return stored fields from Mongot instead of loading full 
documents

Every $search hit was followed by an implicit lookup of the full index document
on mongod before the path, type, property, sort and facet stages ran. Those
documents carry the analyzed and full-text fields, including extracted binary
text, so a query with many hits paid for loading all of that only to project
_path and the score.

With storedSource=true on the Oak index definition:

- the Search index stores just the fields the pipeline reads after $search,
  and regular search queries set returnStoredSource so mongod skips the
  lookup. Suggestions and spellchecks project fields that are not stored, so
  they keep the lookup.
- the node-level full-text field is mapped with store:false unless a property
  sets useInExcerpt. Mongot reads a hit's stored fields together, so keeping a
  copy of that field made stored-source queries slower than lookups for large
  documents. Without it Mongot cannot highlight the field, so rep:excerpt(.)
  has no excerpt unless a property sets useInExcerpt, and the query no longer
  requests that highlight.

The flag is read from the stored index definition, so it takes effect only
after a reindex builds a search index generation that stores those fields;
Mongot rejects returnStoredSource against an index built without them. The
include list is emitted in the sorted order Mongot reports back, otherwise
every indexing cycle would resubmit the definition and rebuild the index.

The core and advanced query compatibility tests now also run against a
stored-source index, and a new check asserts an indexing cycle leaves the
search index queryable in both modes.

Report URL: https://github.com/apache/jackrabbit-oak/actions/runs/36199039769

With regards,
GitHub Actions via GitBox

Reply via email to