White Dragon Search

White Dragon Search

Connect DBCrust to a current White Dragon HTTP server:

dbcrust white-dragon://127.0.0.1:7700

This integration targets White Dragon’s schema API v3 and generic WDSP v6 model. White Dragon has one logical SQL table, documents. Field names, index paths, types, stored values, sort columns, and aggregation columns all come from the mapping embedded in the served splits—there are no built-in Git or patch aliases.

Browse the served schema

\d documents

On connection, DBCrust caches GET /v1/schema. The response contains every distinct complete schema in the current standalone or distributed inventory. DBCrust unions their source fields for browsing and completion while preserving all WDSP types:

• text and keyword as multi-valued TEXT[] and KEYWORD[];
• i64, u64, and f64 without losing width or signedness;
• bool, datetime, and json;
• conflicting definitions across heterogeneous schemas as VARIANT<...>.

\d documents also lists mapping-defined indexes such as body.tokens or patch.secrets, including their index kind, case-folding, tokenizer, transforms, and default-search status. Fields and index paths feed SQL completion; qualifier completion uses only actual indexes plus valid wildcard paths such as body.*.

Qualifier searches

A bare search uses the mapping’s default_search set. Qualifiers name an index path:

dbcrust white-dragon://127.0.0.1:7700 --read-only \
  -c 'headline:storage OR visibility:public'

dbcrust white-dragon://127.0.0.1:7700 --read-only \
  -c 'patch.code:credential AND NOT patch.secrets:false-positive'

dbcrust white-dragon://127.0.0.1:7700 --read-only \
  -c 'body.*:/Immutable\s+splits/ case:sensitive'

A field.* qualifier searches that field’s primary and sub-indexes independently. The only schema-independent qualifier control is case; every searchable name otherwise comes from /v1/schema.

SQL searches and aggregations

White Dragon’s compact SQL front end uses the same mapping-defined field/index paths. DBCrust also supports selecting one or more stored source fields; it expands the hit query for White Dragon and applies the requested projection locally:

SELECT headline, priority FROM documents
WHERE headline MATCH 'Dragon'
LIMIT 20;

SELECT * FROM documents
WHERE body.tokens MATCH 'Immutable'
  AND priority >= 8
  AND visibility IN ('public', 'internal')
LIMIT 20;

SELECT visibility, COUNT(*) FROM documents
WHERE headline MATCH 'Dragon'
GROUP BY visibility
ORDER BY COUNT(*) DESC
LIMIT 10;

SELECT month(published), COUNT(*) FROM documents
GROUP BY month(published)
ORDER BY month(published) DESC
LIMIT 12;

Pattern operators require textual indexes, equality requires an exact-compatible index, and ordered comparisons require a range index. Aggregations require a compatible columnar source field.

Structured request policies

Policies not expressed by hit SQL—such as named-column result ordering—can be sent as the complete /v1/search JSON object:

dbcrust white-dragon://127.0.0.1:7700 --read-only -o json -c '{
  "q": "headline:Dragon",
  "limit": 10,
  "sort": {"named_field": "rating", "direction": "desc"}
}'

Structured aggregations work with qualifier or SQL query strings:

{
  "q": "headline:Dragon",
  "aggregation": {
    "terms": {"field": "visibility", "size": 10, "order": "count_desc"}
  }
}

{
  "q": "headline:Dragon",
  "aggregation": {
    "date_histogram": {
      "field": "published",
      "interval": "month",
      "size": 12,
      "order": "key_desc"
    }
  }
}

Result shape

For SELECT *, DBCrust flattens the hit envelope into split_id, doc_id, id, and score, followed by every stored mapping field, then snippet and the matched index. An explicit select list returns only those stored fields in the requested order. Optional fields missing from a hit are SQL NULL. Text and keyword arrays and JSON values remain JSON text in table cells. Empty result sets retain the appropriate projected or full schema.

If a mapping field collides with an envelope name such as score or index, DBCrust displays it as fields.score or fields.index so both values remain visible.

White Dragon is search-only, so DBCrust’s --read-only guard accepts SQL, qualifier, and structured JSON searches while rejecting an empty request. Servers without schema API v3 are rejected instead of being interpreted with stale aliases or sampled metadata.